Nuclear Engineering and Design 331 (2018) 222–237

Contents lists available at ScienceDirect

Nuclear Engineering and Design

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

Computational fluid-dynamic modeling of the mono-dispersed T

homogeneous flow regime in bubble columns
⁎
Giorgio Besagni , Gaël R. Guédon, Fabio Inzoli
Politecnico di Milano, Department of Energy, Via Lambruschini 4a, 20156, Italy

A R T I C LE I N FO A B S T R A C T

Keywords: Two-phase bubble columns are equipment used to bring one or several gases into contact with a liquid phase. Despite
Homogeneous flow regime the simple system design, bubble columns are characterized by complex fluid dynamic phenomena at different scales;
Computational fluid dynamics for this reason, their correct design, operation and scale-up rely on the precise estimation of global and local fluid
Validation dynamics properties. In this respect, multi-phase Computational Fluid Dynamics (CFD), in the Eulerian multi-fluid
Interfacial closures
framework, is particularly useful to study the fluid dynamics in multi-phase reactors. Within this approach, the
Euler multi-fluid
accurate prediction of the fluid dynamics depends on the correct modeling of (a) the momentum exchange between
the phases, (b) the effects of the dispersed phase on the turbulence of the continuous phase, and (c) the bubble
coalescence and break-up phenomena. Furthermore, the global and the local fluid dynamic properties are related to
the prevailing flow regime, i.e., the homogeneous flow regime and the heterogeneous flow regime. This paper mainly
focuses on the homogeneous flow regime, which can be classified as “pseudo-homogeneous” or “mono-dispersed
homogeneous”, depending on the prevailing bubble size distribution. The numerical modeling of the “pseudo-homo-
geneous” flow regime has been discussed in our previous papers (i.e., modeling closures and suitable boundary
conditions); conversely, this paper contributes to the existing discussion on the modeling closures by investigating the
“mono-dispersed homogeneous” flow regime in “small-scale” and “large-scale” bubble columns. To this end, two test
cases have been considered: (a) a “small-scale” bubble column (a test case taken from the previous literature); (b) a
large-scale bubble column (a test case experimentally studied within this paper by image analysis, optical probe and
gas holdup techniques). In particular, this paper studies the effects of the interfacial forces and bubble induced
turbulence modeling within the Eulerian two-fluid approach. Three-dimensional transient simulations have been
performed and the numerical results were compared with experimental data (both local and global fluid dynamics
parameters). The results have been critically analyzed and the reasons for the discrepancies between the numerical
results and the experimental data have been identified and may serve as a basis for future studies. Likewise, re-
commendations on suitable closures as well as guidelines for future studies have been provided. In conclusion, this
paper extends the validation of a previously proposed set of closure relations (validated for the “pseudo-homogeneous”
flow regime in a “large-scale” annular gap bubble column) to the “mono-dispersed homogeneous” flow regime in “small-
scale” and large-scale bubble columns.

1. Introduction prevailing flow regime: mainly the homogeneous flow regime and the
heterogeneous flow regime, if “large-diameter” bubble columns are
Bubble columns provide a good experimental setup to study the considered (Besagni et al., 2017b). The homogeneous flow re-
turbulent phenomena in dense bubbly flows and to support the vali- gime—generally associated with small gas superficial velocities, UG—is
dation of numerical approaches. Indeed, the numerical modeling of the referred as the flow regime where only “non-coalescence-induced” bub-
local and the global fluid dynamics in bubble column reactors is a way bles exist (as defined by Besagni and Inzoli (Besagni and Inzoli, 2016)).
of supporting the reactor design and scale-up. In particular, among the The homogeneous flow regime can be classified into the “mono-dis-
different numerical approaches, Computational Fluid Dynamics (CFD) persed homogeneous” flow regime and the “pseudo-homogeneous” flow
is a promising method to study the fluid dynamics in bubble columns regime: the former is characterized by a mono-dispersed bubble size
(See, for example, (Peña-Monferrer et al., 2017)). In this respect, the distribution (BSD), whereas the latter is characterized by a poly-dis-
local and the global fluid dynamic properties are related to the persed BSD. The distinction between mono-dispersed and poly-

⁎
Corresponding author at: Politecnico di Milano, Department of Energy, Via Lambruschini 4a, 21056 Milan, Italy.
E-mail address: [email protected] (G. Besagni).

https://doi.org/10.1016/j.nucengdes.2018.03.003
Received 5 June 2017; Received in revised form 1 March 2018; Accepted 2 March 2018
Available online 20 March 2018
0029-5493/ © 2018 Elsevier B.V. All rights reserved.
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Nomenclature g Acceleration of gravity (m s−2)

Hc Bubble column height (m)
Acronyms HD Height of the liquid free surface after aeration (m)
H0 Height of the liquid free surface before aeration (m)
BSD Bubble Size Distribution UG Gas superficial velocity (m/s)
CFD Computational Fluid Dynamics p Pressure (Pa)
CFL Courant Friedrichs Lewy number u Velocity in governing equations (m/s)
URANS Unsteady Reynolds Averaged Navier-Stokes y Axial distance from the gas sparger (m)
yW Distance to the nearest wall in Eq. (13) (m)
Non-dimensional numbers vG,in Velocity of the gas phase at the gas sparger, Eq. (6) (m/s)
V Volume (m3)
2
g (ρk − ρj ) deq
Eo = σ
Eötvös number
g (ρk − ρj ) μk4
Greek letters
Mo = Morton number
ρk2 σ 3
ρk vb deq α Volume fraction
Re = Reynolds number
μk εG Gas holdup
Symbols εG,Local Local void fraction
μ Dynamic viscosity (kg m−1 s−1)
Ac Cross-sectional area of the column ρ Density (kg m−3)
Ain Gas inlet area σ Surface tension coefficient (N m−1)
CL Lift coefficient in Eq. (8) σTD Schmidt number in Eq. (11)
CTD Turbulent dispersion coefficient in Eq. (11) τ Viscous and Reynolds stresses (kg m s−2)
CVM Virtual mass force coefficient τ Time scale (s−1)
CWL Wall force coefficient in Eq. (12)
Superscripts
CW1 and CW2 Coefficients in Eq. (13)
db Bubble equivalent diameter (mm)
→ Vector quantity
dc Bubble column inner diameter (m)
dlift Bubble equivalent diameter for the change of the sign in
Subscripts
the lift force coefficient (mm)
d⊥ Maximum horizontal dimension of the bubble (mm)
j j-th dispersed phase in governing equations
Eo⊥ Eötvös number considering the maximum horizontal di-
k k-th Continuous phase in governing equations
mension of the bubble d⊥
z Generic phase in governing equations
FD Drag force (kg m−2 s−2)
FL Lift force (kg m−2 s−2)
Turbulence quantities
FTD Turbulent dispersion force (kg m−2 s−2)
FVM Virtual mass force (kg m−2 s−2)
ε Turbulent dissipation rate (m2 s−3)
FWL Wall force (kg m−2 s−2)
ω Specific dissipation rate (s−1)
MI Momentum exchanges (kg m−2 s−2)

dispersed BSDs is based upon the change in the sign of the lift force phases, (b) the effects of the dispersed bubbles on the turbulence of the
coefficient (See Refs. (Besagni and Inzoli, 2016; Lucas et al., 2015)), continuous phase, and (c) the effects of coalescence and break-up phe-
which occurs at, approximately, an equivalent bubble diameter of nomena. The first two aspects are discussed in the following; for the last
db = 5.8 mm (considering air-water systems at ambient conditions). point, the reader may refer to Rzehak et al. (Rzehak et al., 2015).
The “mono-dispersed homogeneous” flow regime is a good test for nu- In the Eulerian multi-fluid modeling approach, correlations for inter-
merical codes, to validate models to be applied for the design of “industrial- facial forces are implemented to model the inter-phase momentum ex-
scale” bubble columns, where the “pseudo-homogeneous” flow regime and changes (i.e., the drag, lift, virtual mass, turbulent dispersion and wall
the heterogeneous flow regime are observed at most. Among the available lubrication forces). The drag force has large effects on the macroscopic
modeling techniques, the Eulerian multi-fluid approach and the Eulerian- flow patterns, i.e., gas the gas holdup, axial velocity profiles and local void
Lagrangian approach can be used. The former is the most common ap- fraction profiles (Laborde-Boutet et al., 2009; Tabib et al., 2008). The lift
proach to simulate bubble columns, as reported in the review of Jakobsen force is responsible for the migration of small bubbles toward the column
et al. (Jakobsen et al., 2005); conversely, the latter is mostly applied to walls in co-current flow, and for the uniform spreading of small bubbles in
simulate “small-scale” reactors with low gas holdup1 (Besbes et al., 2015; counter-current flow or in the batch mode. Conversely, a force that can be
Buwa et al., 2006; Delnoij et al., 1997; Hu and Celik, 2008; Jain et al., 2013; assimilated to the lift force tends to push large and deformed bubbles
Lapin and Lübbert, 1994). The Eulerian multi-fluid approach treats each towards the center of the column (Lucas et al., 2005; Tomiyama et al.,
phase as inter-penetrating continua and relies on an ensemble averaging of 2002). As a result, correlations for the lift force coefficient usually display
the multiphase Navier-Stokes equations; for this reason, this approach needs a change of sign from negative, for smaller diameter bubbles, to positive,
closures for the flow turbulence and the inter-phase exchanges phenomena. for large diameter bubbles (Tomiyama et al., 2002). The bubble dispersion
The latter have to include (a) the exchange of momentum between the due to the liquid turbulent fluctuations is taken into account through the
turbulent dispersion force; this force has an important role on the gas
fraction profiles as it modulates peaks of small bubbles near the walls and
1
The gas holdup (εG) is a dimensionless parameter defined as the volume of the gas spreads out large bubbles from the pipe center (Lucas et al., 2007). Its
phase divided by the total volume of the system. Th gas holdup is a global fluid dynamic magnitude is high near the inlets of the gas sparger (Krepper et al., 2007),
property of fundamental and practical importance. The gas holdup determines the mean
thus supporting the modeling of bubble dispersion near coarse gas spar-
residence time of the dispersed phase and, in combination with the size distribution of the
dispersed phase, the interfacial area for the rate of interfacial heat and mass transfer. gers. The wall lubrication force is intended to model the lift force

223
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

appearing close to the wall that pushes the bubbles away from it. In this
respect, Rzehak et al. (Rzehak et al., 2012) compared various wall lu-
brication force formulations applied to vertical bubbly flow in a vertical
pipe and concluded that the inclusion of this force into the model is of
fundamental importance. Finally, the virtual mass force arises from the
relative acceleration of an immersed moving object to its surrounding
fluid: as the object accelerates, it must accelerate the adjacent layers of the
surrounding fluid, resulting in an interaction force acting on the object.
Despite its apparent relevance in transient bubbly flows, this force is often
found to be negligible in bubble columns simulations (Deen et al., 2001;
Díaz et al., 2008; Masood and Delgado, 2014; Oey et al., 2003; Tabib et al.,
2008; Zhang et al., 2006) and different studies have neglected it (Chen
et al., 2004; Chen et al., 2005; Cheung et al., 2007a,b; Díaz et al., 2009;
Frank et al., 2008; Krepper et al., 2008; Laborde-Boutet et al., 2009;
Larachi et al., 2006; Liao et al., 2014; Lucas et al., 2007; Masood et al.,
2015; Masood et al., 2014; Pourtousi et al., 2015a; Pourtousi et al.,
2015b); conversely, Ziegenhein et al. (Ziegenhein et al., 2015) demon-
strated that the virtual mass force has an influence on the prediction of the
turbulence intensity at higher flow rates. Concerning the turbulence
modeling, it is worth noting that different frameworks for the description
of turbulence have been compared and a number of possibilities to include
the bubble-induced contribution to the turbulence have been considered in
the previous literature. In particular, the interested reader may refer to the
papers of Rzehak and Krepper (Rzehak and Krepper, 2013a,b) and Parekh
and Rzehak (Parekh and Rzehak, 2017).
In our previous papers, we have proposed and validated Eulerian-
multi phase modeling approaches to simulate the “pseudo-homogeneous”
flow regime in large-scale bubble columns (Besagni and Inzoli, 2016;
Besagni et al., 2017c; Guédon et al., 2017). Thus, this paper contributes Fig. 1. Small-scale bubble column: the experimental setup.
to the existing discussion on the model closures by investigating the
“mono-dispersed homogeneous” flow regime in small- and large-scale
(viz. “fine gas sparger”) was placed in the middle of the base of the bubble
bubble columns. To this end, two test cases have been considered: (a) a
column. The “fine gas sparger” produced a mono-dispersed BSD. The ex-
“small-scale” bubble column (test case taken from the literature, Ref.
perimental data consist of wire-mesh local void fraction measurements
(Krepper et al., 2007)); (b) a “large-scale” bubble column (experimen-
performed at two different axial heights (y = 0.08 m and y = 0.63 m above
tally studied within this paper, by image analysis, optical probe and gas
the gas sparger), and for three different gas superficial velocities in the
holdup techniques). In particular, this paper studies the effects of the
mono-dispersed homogenous flow regime (UG = 0.006, 0.008, 0.010 m/s,
interfacial forces and the bubble induced turbulence modeling within
respectively associated to case A, case B and case C, in the following of this
the Eulerian two-fluid approach. Compared to the previous works (i.e.,
paper). Please note that the data used in the validation procedure are data
refer to the literature survey proposed in (Peña-Monferrer et al., 2017)),
averaged (a) over time and (b) over the depth of the bubble column.
the proposed study suggests a thorough analysis of the various closures
and their relative coupling in both “small-scale” and “large-scale” bubble
columns. To the authors’ best knowledge, this paper presents an ad- 2.2. Large-scale bubble column (novel test case)
vancement compared with the previous published literature in two
main areas: (a) the definition of a baseline approach to simulate bubbly The “large-scale” bubble column benchmark consists of an ad-hoc
flows in “small-scale” and in “large-scale” bubble columns; (b) extend the experimental study, performed by the authors within this paper.
existing experimental dataset to validate numerical models.
This paper is structured as follows. In Section 2 the experimental 2.2.1. The experimental setup
setups and datasets are presented, in Section 3 the numerical approach The experimental facility, displayed in Fig. 2a, is a non-pressurized
is discussed, in Section 4 the numerical predictions are compared with vertical pipe with dc = 0.24 m (bubble column inner diameter) and
the experiments. Finally, conclusions are drawn and future studies are Hc = 5.3 m (bubble column height). A pressure reducer was used to reg-
proposed. ulate the pressure before the two rotameters (namely, [1] and [2] in Fig. 2)
and was set at 3 bars. The rotameters were used to measure the gas flow
2. The experimental datasets rate (accuracy ± 2% f.s.v., E5-2600/h, manufactured by ASA, Italy). The
bubble column was equipped with a needle sparger (Fig. 2b), designed to
In this section, the “small-scale” and “large-scale” test cases are dis- produce small bubbles (owing to the small gas sparger openings and the
cussed. First, the “small-scale” bubble column (viz. the literature test mechanics of the bubble growth and detachment) and to generate stable
case) is presented; second, the “large-scale” bubble column (viz. the new and mono-dispersed bubbly flows (as qualitatively displayed in Fig. 2c). The
experimental dataset) is presented. gas sparger consists of a sparger zone and a plenum. The sparger zone,
displayed in Fig. 3b, consists of plates with 581 holes, which are equipped
2.1. Small-scale bubble column (literature test case) with 0.5 mm inner diameter needles, uniformly distributed across the
column cross-section (the needles were inserted in a perforated plate, and
The “small-scale” bubble column benchmark was taken from the litera- silicone was applied to avoid air leakages). Thus, the gas sparger has an
ture and, in particular, from the study of Krepper et al. (Krepper et al., open area (=sum of areas of all the holes divided by the cross-sectional area
2007). The Krepper et al. (Krepper et al., 2007) experimental set-up consists of the column) equal to 0.26%. The plenum, made of plexiglass, is 0.21 m in
of a rectangular 0.1 × 0.02 m cross section bubble column, 0.82 m in height diameter and 0.21 m in height. In addition, an O-ring was placed between
(Fig. 1). A gas sparger plate composed of 49 holes with diameter of 1 mm the plenum and the perforated plate to avoid air leakages. During the

224
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Fig. 2. Large-scale bubble column: the experimental setup.

experimentation, the gas and the liquid temperatures were examined and openings; H0/dc = 12.5). It is worth noting that the precise definition of
maintained constant at room temperature (295 ± 1 K). Filtered air from H0 is of fundamental importance when studying bubble column fluid
laboratory lines was used as the gaseous phase in all the experiments; the dynamics: in this respect, the influence of the liquid level on the bubble
air-cleaning line consists of filters (mechanical and activated carbon) and column fluid dynamics was recently discussed in Refs. (Besagni et al.,
condensation drying unit in order to clean the gas phase properly and thus 2017a; Sasaki et al., 2016; Sasaki et al., 2017). Uncertainty analysis,
avoid the presence of contaminants in the form of (i) solid particles and (ii) related to the gas holdup measurements, were described in Appendix A
organic substances. Please note that the values of gas density (used to of (Besagni et al., 2017a) and are not repeated here.
compute the superficial gas velocity) are based upon the operating condi-
tions existing at the column mid-point (computed using the ideal gas law). 2. Image analysis. The image analysis was performed by using the image
analysis methods described in our previous studies (see, for example,
2.2.2. The experimental methods (Besagni and Inzoli, 2016)); the image analysis was applied to three gas
The experimental study consisted of global and local fluid dynamic flow rates: (a) UG = 0.0019 m/s, (b) UG = 0.0037 m/s and (c)
measurements: (a) gas holdup measurements, (b) image analysis and (c) UG = 0.0056 m/s. Please note that UG = 0.0037 m/s was used to va-
optical probe measurements. Gas holdup measurements were used to lidate the numerical approach in our previous numerical studies. The
study the global fluid dynamics; digital images were taken to have detailed images were taken by means of a NIKON D5000 camera (general
descriptions regarding BSDs; the local bubble properties were measured by settings as follow: Nikon 10–24 mm lenses, f/3.5; 1/1600s; ISO400;
means of a double fiber optical probe. In the following of this section, 4288 × 2848 pixels, spatial resolution of 11.8 pixel/mm) at 2.4 m from
details concerning the experimental methods are presented: the gas sparger2. A 500 W LED halogen lamp was used as light source.
Visualization sections consist of squared boxes (filled with water)
1. Gas holdup measurement. Measurement of the bed expansion (the around the vertical pipe designed for correcting the distorted image.
height of the liquid free surface when air flows in the column) al- The experimental methods, uncertainties and experimental procedure
lowed the evaluation of the gas holdup, εG, as follows: were described in our previous paper and are not repeated here. To
study the influence of the radial location on BSDs, the camera was
εG definition VG (H −H )
εG = ⟶Constantcross−sectionlaarea D 0
VL + G HD (1)
2
It is worth noting that from preliminary image analysis, it has been observed that
In Eq. (1), HD and H0 are the heights of the free surface after and these BSDs are representative of the major part of the bubble column, owing to the
before aeration, respectively (H0 = 3.0 m above the gas sparger negligible coalescence and break-up phenomena of the present gas sparger.

225
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

optoelectronic module that emits the laser to the probe tip and
converts the reflected optical signal into a digital signal. From the
digital signal, the bubble frequency f (bubble number per unit time)
and void fraction εG,Local (assuming it equals the proportion of time
when the tip is surrounded by gas) can be obtained. In this study, all
the measurements have been obtained using a sampling period
equal to Δtsampling = 1000 s, which is large enough to produce reli-
able time-averaged values and is far above the typical values of
1–5 min for similar optical probes (Chang et al., 2003; Chaumat
et al., 2005; Lima Neto et al., 2008; Zhang and Zhu, 2013).

2.2.3. The experimental dataset

In this section, the main outcomes of the experimental study are
presented and discussed. Please note that the aim of this section is to
describe and define a test case for the numerical simulations rather than
providing a comprehensive description of the bubble column fluid dy-
namics. To this end, Fig. 3 summarizes the experimental results, further
described in the following of this section: the gas holdup curve (Fig. 3a),
the BSDs (Fig. 3b and c) and the local void fraction profiles (Fig. 3d).
Fig. 3a displays the gas holdup curve. It is observed a peak in the gas
holdup curve leading to the characteristic S-shaped curve: as UG increases,
the gas holdup curve increases and attains the maximum, decreases with
further increase in UG until it attains the minimum, and then, increases
again with UG. The S-shaped gas holdup curve is the macroscopic outcome
of the mono-dispersed BSDs (displayed in Fig. 3b and c) as discussed in our
recent paper (Besagni et al., 2018) as well as in the previous literature
(Ruzicka et al., 2001; Sharaf et al., 2016). The S-shaped curve can be in-
terpreted by the Ledinegg instability (Ledinegg, 1938), which is re-
presentative of the pressure-drop–flow-rate (e.g., εG and UG) instabilities
(see (Boure et al., 1973)). The reader may refer to our recent paper for a
precise discussion on this concept (Besagni et al., 2018).
The “mono-dispersed homogeneous” flow regime is quantitatively
described by the image analysis and, in particular by the BSDs sampled
in the wall region (Fig. 3b) and in the center cross-section (Fig. 3c) of
the bubble column. It is observed that most of the bubbles have an
equivalent diameter below 5.81 mm, which is the value for the change
of the sign of the lift force (Tomiyama et al., 2002); hence, the present
homogeneous flow regime is classified as “mono-dispersed-homogeneous”
flow regime. Fig. 4 presents some flow visualizations, to provide in-
sights into this flow regime. The BSDs near the wall are unimodal
(Fig. 3b) and the first peak of frequency appears in the range of
deq = 2.5 mm and 3.5 mm (for the three gas flow rates). Similarly, the
BSDs at the center of the column are unimodal with a peak of frequency
the range of deq = 2.5 mm and 3.5 mm (UG = 0.0019 m/s) and the
Fig. 3. Large-scale bubble column: the experimental results. range of deq = 3.5 mm and 4.5 mm (UG = 0.0037 and 0.0056 m/s).
Increasing UG, BSD slightly shifts towards higher equivalent diameters,
focused on (a) a ruler along the external wall and (b) a ruler inside the as expected and observed in our previous studies. It is observed that
column (see (Besagni and Inzoli, 2016) for further details on this BSDs near the wall are, on average, slightly smaller than the ones in the
method). central section, possibly because of the effect of the lift force (please
3. Optical probe measurements. A double fiber optical probe system refer to the discussion proposed in Ref. (Besagni and Inzoli, 2016)).
(manufactured by RBI, France) was used to measure the local fluid Indeed, the larger bubbles (even if limited in number), having a nega-
dynamic properties at UG = 0.0037 m/s. Please note that tive lift coefficient, tend to migrate from the low liquid velocity region
UG = 0.0037 m/s was also to validate the numerical approach in our towards the high liquid velocity regions: in the batch mode, the larger
previous numerical studies. The optical probe was inserted, via an bubbles tend to migrate toward the center of the column.
access port, into the flow at y = 1.9 m from the sparger. Each probe Fig. 3d displays the local void fraction profiles (UG = 0.0037 m/s):
is made of two 40-μm glass fibers whose tips are re-enforced by two as expected the local void fraction profile is flat and seems slightly wall
sharp sapphire pins: further information concerning this equipment peaked. This experimental observation is the consequence of the “mono-
can be found in the paper of Boes and Hager (Boes and Hager, dispersed homogeneous” flow regime: the small bubbles, having a posi-
1998). Similar optical probe systems were used by different authors tive lift coefficient spread in the cross-section of the bubble column and,
(Chaumat et al., 2005; Kiambi et al., 2003; Lima Neto et al., 2008; thus, stabilize the “bubble bed”. This conclusion was also observed by
Simonnet et al., 2007; Zhang and Zhu, 2013). Optical probes dis- Lucas et al. (Lucas et al., 2005) and is a further proof of the definition of
tinguish the gas and liquid phases by measuring the intensity of a the “mono-dispersed homogeneous” flow regime.
laser light that is reflected and/or refracted at the probe tip on the In conclusion, the proposed experimental dataset represents the
basis of the refractive indexes of the probe tip, gas and liquid phases “mono-dispersed homogeneous” flow regime in a “large-diameter” and “large-
(Barrau et al., 1999). The probe signal is measured via an scale” bubble column, as summarized in the following. The gas sparger
produces a uniform bubble bed having a small equivalent diameter, thus

226
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

equations for each phase are ensemble-averaged, and the effects of

turbulence and interphase phenomena have to be taken into account
using closure models (Ishii, 1975). For an isothermal flow without mass
transfer, the URANS governing equations for the generic k-th phase are:
∂
(αk ρk ) + ∇ ·(αk ρk →
uk ) = 0
∂t (2)

∂ ⎯→
⎯
(αk ρk →
uk ) + ∇ ·(αk ρk →
uk →
uk ) = −αk ∇p + ∇ ·(αk τk ) + αk ρk→
g + MI ,k
∂t
(3)
In the following, the dispersed phase is listed as j and the continuous
phase as k. The terms on the right-hand side of Eq. (3) represent the
pressure gradient, the stresses (viscous and Reynolds), the body forces
and the interfacial momentum exchanges between the phases including
the closure models, respectively.
The closure models for the momentum exchange are formulated as
forces acting on the liquid and dispersed phases. Development and
validation of such models is an active field of research; in the present
paper, a model set that has been applied recently with promising suc-
cess (Besagni and Inzoli, 2016; Guédon et al., 2017) is used, which is
from now on named as baseline model. Subsequently, the influence of
different models has been investigated by varying the different forces
one by one, keeping constant the others. The main idea is to identify all
the relevant phenomena including a complete set of interfacial force
and bubble-induced turbulence models. In this respect, this paper aims
to further develop our previously validated numerical approach (see
Refs. (Besagni and Inzoli, 2016; Guédon et al., 2017)). It is worth noting
that modeling coalescence and break-up has not considered in this
study, in agreements with the nature of the “mono-dispersed homo-
geneous” flow regime. Further details on modeling coalescence and
break-up were discussed by Besagni et al. (Besagni et al., 2017c).
A transient formulation has been applied; this is taken into account
by using the unsteady Reynolds Averaged Navier-Stokes equations with
the SST k-ω two equation model.

3.2. Interfacial momentum exchanges

The interactions between the continuous and the dispersed phases

are taken into account by means of source terms in the momentum
equation. In particular, considering the continuous phase k, the last
term in Eq. (2) comprises several independent physical mechanisms:
drag, lift, virtual mass, turbulent dispersion and wall lubrication forces:
⎯→
⎯ ⎯→
⎯ ⎯→
⎯ ⎯→
⎯ ⎯→
⎯ ⎯→
⎯
MI ,k = FD,k + FL,k + FVM ,k + FTD,k + FWL,k (4)

The source term for the continuous phase is related to the sum of the
Fig. 4. Large-scale bubble column: flow visualizations (UG = 0.0037 m/s). dispersed phase source terms as follows:
2
⎯→
⎯ ⎯→
⎯
reducing bubble coalescence and generating the typical S-shaped gas MI ,k = − ∑ MI ,j
holdup curve. The reduced bubble coalescence results in BSDs character- j=1 (5)
ized by bubbles having an equivalent diameter below 5.81 mm (viz. the
Modeling and validation of forces acting on a bubble have been
value for the change of the sign of the lit force, (Tomiyama et al., 2002)):
intensively studied over the last decade. All forces act together to
thus, they have a positive lift coefficient. Bubbles having a positive lift
produce observable phenomena, such as, for example, the distribution
coefficient, in the batch mode, spread in the cross section of the bubble
of the void fraction. Therefore, a complete set of interfacial forces
column, thus resulting in a flat void fraction profile.
should be used, as reviewed in the introduction and described in the
paper of Rzehak and Krepper (Rzehak and Krepper, 2013b). A summary
3. The numerical model of the closure models analyzed in proposed in Table 1. For the math-
ematical description of the closures the reader should refer to the
The numerical model, detailed in the following, was implemented in above-mentioned references, for their implementation the readers
the commercial software ANSYS Fluent Release 15.0.7. should refer to Refs. (Fluent, 2013; Frank et al., 2004).

3.1. Governing equations 3.3. Turbulence modeling

An Eulerian two-fluid approach has been adopted in the present Due to the small density and the small scales of the dispersed gas, in
numerical simulations. Within such a framework, the Navier-Stokes the homogeneous flow regime, it suffices to consider turbulence in the

227
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

continuous phase (Rzehak and Krepper, 2013b). In this study, we used an the present cases, a CFL < 0.3 is considered: the resulting time step
eddy diffusivity approach for the turbulence modeling. In particular, the size is Δt = 0.001 s. The iterations within each time step are stopped
SST k-ω model is used to include the effects of turbulence. In the literature, when the residuals fall below 10−5.
the SST k-ω model has proven to be slightly superior to the k-ε model to The simulation procedure is similar to the typical one employed for
simulate upward bubbly flow in the studies of Cheung et al. (Cheung et al., studying transient bubble column flows (Masood and Delgado, 2014;
2007a,b) and has also been used successfully in recent works to simulate Masood et al., 2015; Ziegenhein et al., 2013; Ziegenhein et al., 2015): the
two-phase flow in vertical pipes and bubble columns (Duan et al., 2011; sequence followed includes an initial run to reach a statistical steady
Frank et al., 2008; Liao et al., 2014; Liao et al., 2015; Lucas et al., 2015; temporal convergence of the solution. For the “small-scale” bubble column
Rzehak and Krepper, 2013b; Rzehak et al., 2015; Rzehak and Kriebitzsch, the first run has a duration of 10 s in physical time, and a second run of
2015). The constants of the model follow their single-phase values. In the 140 s is performed with data sampling to collect temporal averages. For
present paper, bubble induced turbulence was considered with the Sato the “large-scale” bubble column, the first run has a duration of 20 s in
et al. (Sato et al., 1981) and the Simonin et al. (Simonin and Viollet, 1990) physical time, and a second run of 250 s has been performed with data
approaches. The former does not add an explicit source terms to the tur- sampling to collect temporal averages. The choice of the duration of the
bulence equations and relies on the assumption that the turbulent viscosity first run is dictated by the temporal evolution of the total gas holdup, i.e.,
may be modeled as the sum of the single phase shear viscosity and particle within the entire computational domain. When this quantity stabilizes, it
induced turbulent viscosity (this assumption is valid at only low secondary means that the flow is developed and that data sampling operations can be
phase volume fractions, which well applies in the present cases); con- performed, as discussed by Besagni et al. (Besagni et al., 2014).
versely, the latter includes source terms in the momentum equations. The The different numerical schemes have been chosen to reduce the
interested reader may refer to the papers of Rzehak and co-workers discretization error as much as possible. A second-order Euler implicit
(Parekh and Rzehak, 2017; Rzehak and Krepper, 2013a; Rzehak and temporal discretization scheme is adopted. Gradients are estimated
Krepper, 2013b) for a detailed analysis concerning the role of bubble in- using a least squares cell-based method. The QUICK scheme is used to
duced turbulence modeling. discretize the convection term of each scalar solved. A Phase Coupled
SIMPLE algorithm guarantees the coupling between pressure, velocity
and volume fraction.
3.4. Numerical settings and simulation procedure
3.5. Boundary conditions and working fluids
Three-dimensional and transient simulations were carried out.
Indeed, three-dimensional simulations are required for accurate pre- First, the boundary conditions for the dispersed phase have been
dictions of the gas-liquid flows in bubble columns (Ekambara et al., implemented. The BSDs in both the experimental setups (viz., (Krepper
2005; Mudde and Simonin, 1999; Pfleger et al., 1999; Sokolichin and et al., 2007) and Fig. 3) are “mono-dispersed” and below the critical
Eigenberger, 1999). The simulations were performed using a geometric equivalent diameter value for the change of the sign of the lift force
representation of the real column and, in both the numerical setups, the coefficient in air–water systems (db,lift = 5.8 mm, (Tomiyama et al.,
gas sparger was modeled as a uniform surface on the bottom. For 2002)). For this reason, the fixed mono-dispersed approach (as defined
transient simulations, it is known that a sufficiently fine mesh is re- in (Besagni et al., 2017c)), is used. Within this approach, the dispersed
quired to capture the main transient phenomena of two-phase pro- phase is modeled by using a single gas phase (considering only “small
blems. If the mesh resolution is too low, the accuracy of gradients es- bubbles”): (a) in the “small-scale” experimental setup, accordingly with
timates required in the computation of bubble forces, for instance, is the experimental data, the dispersed phase is represented by a bubble
compromised and flow instabilities are not resolved, as observed by equivalent diameter equal to db = 3 mm; (b) in the “large-scale” ex-
Guédon et al. (Guédon et al., 2017). Therefore, it must be guaranteed perimental setup a bubble equivalent diameter equal to db = 4.2 mm
that the solution is independent of the time step and of the mesh size; has been applied, accordingly with the experimental dataset.
thus, a grid resolution study was conducted to ensure that convergence Despite the air phase having a slightly varying density from the
with respect to the spatial resolution was achieved. In particular, the bottom to the top of the column, both fluids are considered in-
same procedure as described by Guédon et al. (Guédon et al., 2017) has compressible. At the inlet, the gas velocity is computed by using Eq. (6):
been applied in this paper: (a) in the “small-scale” bubble column the
vG,in = UG Ac / αG Ain (6)
numerical simulations were performed on a mesh with uniform cubic
computational cells (15 × 15 × 45 – please note that the same space where UG is the gas superficial velocity, Ac is the cross-sectional area of
discretization was also applied in (Krepper et al., 2007) and the same the column, Ain is the gas inlet area, and αG is the gas void fraction at
discussion as in Section 4.2 of (Krepper et al., 2007) apply here); in the the inlet. At the outlet, a degassing condition is used. Along the walls,
“large-scale” bubble column, the numerical simulations were performed wall functions are applied. At the walls, a no-slip boundary condition is
on a mesh composed by 205,131 hexahedral cells (the mesh in- applied for the continuous phase and a free slip condition for the dis-
dependency study of this case is presented in Appendix A). The time persed phase. It is worth noting that the boundary conditions for the
discretization is characterized by the CFL number and, in particular, in turbulence quantities have been implemented according to the method
illustrated in Guédon et al. (Guédon et al., 2017).
Table 1
Summary of the closure models analyzed. 3.6. Method
Interfacial force References
A baseline closure of interfacial forces is considered and the influence of
Drag force Grace et al. (1976) different models is investigated by varying forces one by one, keeping
Tomiyama et al. (1998) constant the others. A summary of closures implemented is given in Table 2
Lift force Tomiyama et al. (2002)
Wall lubrication force Antal et al. (1991))
(“small-scale” bubble column) and in Table 3 (“large-scale” bubble column).
Tomiyama et al. (1995) Further details on the closures implemented in the baseline approaches can
Hosokawa et al. (2002) be found in the previous papers as well as in Appendix B. For the literature
Frank et al. (2008)) benchmark, every closure (Table 2) is tested at the three gas velocities.
Turbulent dispersion force Lopez de Bertodano (1991)
Conversely, for the “large-scale” bubble column, the closures listed in Table 3
Burns et al. (2004))
Virtual mass force Implemented with CVM = 0.5. were applied. It is worth noting that, owing to the preliminary numerical
results obtained for the “large-scale” bubble column, the inclusion of bubble

228
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Table 2
Summary of the closure models implemented (small-scale bubble column).

Force Model Baseline Influence of the wall lubrication force Influence of the turbulent Influence of the Influence of the BIT
approach dispersion force virtual mass force

Frank Hosokawa Tomiyama No wall Lopez No turbulent CVM = 0.5 Simonin No BIT
force dispersion force

Drag Grace X X X X X X X X X X
Lift Tomiyama X X X X X X X X X X
Wall lubrication Antal X X X X X X
Turbulent Burns X X X X X X X X
dispersion
Virtual Mass No model X X X X X X X X X
BIT Sato X X X X X X X X

Table 3
Summary of the closure models implemented (large-scale bubble column).

Force Model Baseline approach Influence of the drag force Influence of the wall lubrication force Influence of the virtual mass force

Grace Frank Hosokawa Tomiyama CVM = 0.5

Drag Tomiyama X X X X X
Lift Tomiyama X X X X X X
Wall lubrication Antal X X X
Turbulent dispersion Burns X X X X X X
Virtual Mass No model X X X X X
BIT Sato

induced turbulence leads to instabilities in the solution and are conse- respect to the baseline model have been investigated: no wall
quently neglected. This issue is a matter of future studies. In addition, a lubrication force (Fig. 5a and b), Hosokawa wall lubrication force model
preliminary influence of the drag force model has been included. (Fig. 5c and d), Frank wall lubrication force model (Fig. 6e and f), and
Tomiyama wall lubrication force model (Figs. 6g and 5h). First of all, it is
4. Results important to notice that without a wall lubrication force model the void
fraction profile is flat far from the gas sparger: this means that bubbles are
In the following, the experimental data for both the test cases are equally distributed also near to the wall, not respecting the real behavior
compared with the numerical results. It is worth noting that the com-
parison concerns both the global (i.e., plume frequency and gas holdup)
and the local (i.e., void fraction data, averaged over time and over the
depth of the bubble column) fluid dynamic quantities. In particular, it is
worth noting that a validation of a numerical approach against local
void fraction profiles is particularly useful in density/gravity driven
flows as it summarizes the local- and global-scale fluid dynamics.

4.1. Small-scale bubble column

4.1.1. Local void fraction profiles

In the following, the local time averaged void fraction profiles mea-
sured at two different heights from the sparger, i.e., y = 0.63 m and
y = 0.08 m, are compared with experimental data. The data have been
obtained from the simulations using the baseline model and a variation
from the baseline model in terms of wall lubrication force, turbulent dis-
persion force, virtual mass force and bubble induced turbulence models.

4.1.1.1. Baseline model. The local void fraction profiles obtained using
the baseline model have been compared with the experimental data in
Fig. 5. Far from the gas sparger (y = 0.63 m), the void fraction profile
is, in average, close to the experimental one. Conversely, near the gas
sparger (y = 0.08 m), the void fraction profile is very similar for the
cases A and B, while for the highest gas superficial velocity, case C, the
central peak in the profile is underestimated. Overall, the baseline
model is performing quite well and there is some room for improvement
for case C near the gas sparger. Analyzing the different closure models
one-by-one will therefore allow determining if the singular
modification may improve or worsen the predictions.
Fig. 5. Comparison between experimental and numerical results: Baseline model (small-
scale bubble column).
4.1.1.2. Influence of the wall lubrication force. Four variations with

229
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Fig. 6. Comparison between experimental and numerical results: Influence of the wall lubrication force model (small-scale bubble column).

shown by the experimental observations. Instead, the wall lubrication force results in slight variations of the void fraction profiles, as
force pushes the gas phase away from the wall, resulting in a decreasing displayed in Fig. 7. The most noticeable changes are close to the
profile near to it. The baseline closure with the Antal wall lubrication force walls at the sparger level (y = 0.08 m), where void fractions are over-
model seems to give the best prediction. Indeed, the Frank wall lubrication predicted.
force and Tomiyama wall lubrication force models provide a too strong
drop of the profile close to the wall, while the Hosokawa wall lubrication 4.1.1.4. Influence of the virtual mass force. The inclusion of the virtual
force model over-predicts void fractions near the lateral walls at the mass force leads to the most significant changes in the void fraction
sparger level (y = 0.08 m) for the cases A and B. profiles among the model variations investigated. The results are
illustrated in Fig. 8 and higher void fractions are noticed in averaged.
4.1.1.3. Influence of the turbulent dispersion force. Two variations of the In particular, for case C, the void fraction profile is peaked at the center
turbulent dispersion force have been investigated: no turbulent of the bubble column and indicates a prediction closer to the
dispersion force and Lopez turbulent dispersion force model. No experiments. This is probably due to the fact that near to the gas
differences are found using the Lopez model with respect to the sparger the effect of relative acceleration is acting strongly, and this is
baseline model using the Burns turbulent dispersion force model, so noticed more with the increasing gas velocity. In this respect, the
results are not reported. Conversely, omitting the turbulent dispersion inclusion of the virtual mass force may improve the predictions for the

230
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Fig. 7. Comparison between experimental and numerical results: influence of the tur-
bulent dispersion force (small-scale bubble column).

Fig. 9. Comparison between experimental and numerical results: influence of bubble

induced turbulence, BIT (small-scale bubble column).

(BIT) model. In general, the inclusion or exclusion of BIT model has a

Fig. 8. Comparison between experimental and numerical results: influence of the virtual low impact on the results. The largest differences are observed near the
mass force (small-scale bubble column). gas sparger (y = 0.08 m) and the best performances are given by the
baseline model, implementing the Sato bubble induced turbulence
case C. However, for the cases A and B the virtual mass force is slightly model, for case A and B while slightly better results are obtained
deteriorating the predictions. with the Simonin bubble induced turbulence model for case C.

4.1.1.5. Influence of the bubble induced turbulence. Fig. 9 illustrates the 4.1.2. Plume frequency
void fraction profiles obtained varying the bubble induced turbulence Experimental observation of the column shows that oscillating plumes

231
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Table 4
Plume frequencies measured in Hz for the various cases investigated (small-scale bubble column).

Gas superficial velocity Baseline Influence of the wall lubrication force Influence of the turbulent Influence of the virtual Influence of the
approach dispersion force mass force BIT

Frank Hosokawa Tomiyama No wall No turbulent dispersion force CVM = 0.5 No BIT
force

UG = 0.006 m/s 0.285 0.291 0.247 0.284 0.235 0.300 0.194 0.284
UG = 0.008 m/s 0.272 0.314 0.234 0.301 0.227 0.308 0.190 0.311
UG = 0.010 m/s 0.220 0.343 0.218 0.316 0.218 0.299 0.000 0.192

Experimental data (Figure 3d) Baseline (Tomiyama drag force) Grace drag force
Table 5
0.025 Experimental and numerical gas holdup for the various cases investigated (large-scale
bubble column).
0.020
Case Gas holdup [−] Relative errore
ɸG,Local [-]

0.015 Baseline 0.015115 1.93%

Grace drag force 0.014847 3.67%
0.010 Hosokawa wall force 0.015192 1.43%
Tomiyama wall force 0.014749 4.30%
Virtual mass force 0.016234 5.33%
0.005

Experimental gas holdup 0.0154.

0.000
0 0.2 0.4 0.6 0.8 1
Dimensionaless axial posiƟon [-] possible to calculate the plume oscillation frequency. As a result, in Table 4,
the plume frequency, calculated in Hz, is reported for the different cases: not
Fig. 10. Comparison between experimental and numerical results: influence of drag force
model (large-scale bubble column).
a univocal trend is found. In fact sometimes the frequency increases with the
increasing flow rate, sometimes it decreases: unfortunately experimental
data are not available for a comparison, so it is not possible to draw certain
Experimental data (Figure 3d) Baseline (Antal wall force)
Hosokawa wall force Tomiyama wall force
conclusions. However, it is interesting to notice that when virtual mass is
0.025 included in the model the oscillation frequency greatly decreases and even
no oscillation is found at UG = 0.01 m/s. This happens because the virtual
0.020 mass has a damping effect when the relative acceleration between gas and
liquid becomes higher.
ɸG,Local [-]

0.015

4.2. Large-scale bubble column

0.010

4.2.1. Local void fraction profiles

0.005
In the following, the local time averaged void fraction profiles
measured at y = 1.9 m are compared with experimental data. The data
0.000
0 0.2 0.4 0.6 0.8 1 have been obtained from the simulations using the baseline model and
Dimensionaless axial posiƟon [-] a variation from the baseline model in terms of drag force, wall lu-
Fig. 11. Comparison between experimental and numerical results: influence of wall lu- brication force and virtual mass force.
brication force model (large-scale bubble column).
4.2.1.1. Baseline model and influence of the drag force. The local void
Experimental data (Figure 3d) Baseline (no virtual mass force) Virtual mass force fraction profile obtained using the baseline model have been compared
0.025 with experimental data in Fig. 10. In particular, the Grace and the
Tomiyama drag force models have been compared. For both the drag
0.020 models, the numerical void fraction profile is close to the experimental
one. In particular, the Tomiyama drag force model provides numerical
ɸG,Local [-]

0.015 results in a better agreement compared with the Grace drag force
model. For this reason, the Tomiyama drag force model has been used
0.010 in the following of this analysis. This results is somehow in agreement
with the screening of drag models performed by the authors for an
0.005 annular gas bubble column, see Ref. (Besagni et al., 2014). In the
following, analyzing different closure models one-by-one will therefore
0.000 allow determining if the singular modification may improve or worsen
0 0.2 0.4 0.6 0.8 1
Dimensionaless axial posiƟon [-] the predictions.

Fig. 12. Comparison between experimental and numerical results: influence of virtual
mass force (large-scale bubble column).
4.2.1.2. Influence of the wall lubrication force. Two variations with
respect to the baseline model have been investigated (Fig. 11): (a) the
Hosokawa wall lubrication force model and (b) Tomiyama wall
are seen in a small region right above the gas injection point (Krepper et al., lubrication force model. Conversely, the Frank wall lubrication force
2007). This is caused by the liquid recirculation that has a strong effect in a model, unfortunately, lead to convergence issues and it was not
column of such small dimensions (as the one in Fig. 1). This peculiarity is considered here. Indeed, as previously observed, a wall lubrication
shown by numerical simulations as well: from a transient analysis it is force model is essential to be able to predict the wall-peaked behavior

232
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

near the wall. As previously observed, Tomiyama wall lubrication force force modelling approaches have been tested. The relative influences of
model provides a too strong drop of the profile close to the wall; drag and non-drag and non-lift forces are investigated by changing the
conversely, the Hosokawa wall lubrication force model slightly under- interfacial closure models one-by-one starting from a baseline set-up.
predicts the void fractions near the lateral walls. Consequently, the Overall, the baseline model set-up is found to perform quite well for all the
Antal wall lubrication force model is preferred. gas superficial velocities investigated and the two test cases.
The main results are as follows:
4.2.1.3. Influence of the virtual mass force. As observed in the “small-
scale” bubble column, the inclusion of the virtual mass force leads to the • The influence of the drag force has been tested comparing the Grace
most significant changes in the void fraction profiles among the model and the Tomiyama drag force models. No significant differences
variations investigated (Fig. 12). Indeed, the local void fraction profiles have been observed. We suggest using the Tomiyama drag force
are over-predicted in the whole cross section of the bubble columns. model implemented in the baseline set-up, owing to the better
This is probably due to the strong effects arising from the bubble plum prediction of the gas holdup in the “large-scale” bubble column.
dynamics and lateral accelerations. • The influence of the wall lubrication force has been tested in both
the “small-scale” and “large-scale” bubble columns. The inclusion of
4.2.2. Gas holdup the wall lubrication force has been found to improve the predictions
Table 5 proposes a comparison between the numerical and the ex- and no large differences were found between the four models tested.
perimental global fluid dynamic parameters. In particular, the experi- We suggest using the Antal wall lubrication force model im-
mental and the numerical gas holdups are compared. The gas holdup is a plemented in the baseline set-up since it demonstrated slightly
global fluid dynamic property of fundamental and practical importance as better performances.
it determines the mean residence time of the dispersed phase and, in • The inclusion of the turbulence dispersion force has been tested in
combination with the size distribution of the dispersed phase, the inter- the “small-scale” bubble column and has been found to be relevant as
facial area for the rate of interfacial heat and mass transfer. Therefore, a well, since the predictions were much better near the gas sparger.
precise prediction of the gas holdup allows a correct estimation of reactor No significant differences have been observed among the available
design and reduces operation costs. In this respect, Table 5 provides a turbulence dispersion closure model and either the Lopez or Burns
synthetic comparison of the modeling approaches. The baseline approach model can be implemented.
with the Tomiyama drag model allows a correct estimation of the gas • The inclusion of the virtual mass force has been tested in both the
holdup (relative error equal to 1.93%). Conversely, the Grace drag model “small-scale” and “large-scale” bubble columns. It has been found to
over predicts the gas holdup and the relative error increases to 3.69%. The improve the prediction of the void fraction profile near the sparger
application of the Hosowaka wall lubrication force improves the predic- for the highest gas superficial velocity investigated while the pre-
tion of the gas holdup, as it under-predicts the wall peaked behavior (thus, dictions have been slightly worst for the lower gas superficial ve-
it decreases the performance of the Baseline approach). Conversely, the locities analyzed.
Tomiyama wall lubrication further over-predicts the gas holdup (as also • In the “small-scale” bubble column, the inclusion of bubble induced
observed in Fig. 11). Similarly, as also observed in Fig. 12, the application turbulence has been found to slightly improve the results and the
of the virtual mass force further over-predicts the local void fraction model of Sato demonstrated slightly better predictions with respect
profiles and, consequently, the gas holdup. to the Simonin model; conversely, in the “large-scale” bubble
column, the inclusion of a bubble induced model lead to con-
5. Conclusions vergence issues, which is a matter of future studies.
• In the “small-scale” bubble column, the prediction of the global fluid
This paper deals with interfacial force modeling in “small-scale” and dynamics has been compared in terms of the plume frequency. Large
“large-scale” bubble column simulation. The gas–liquid “mono-dispersed differences have been observed with respect to the bubble plume
homogeneous” flow regime in a “small-scale” rectangular cross-sectioned frequency in the column, suggesting that a more thorough in-
bubble column and in a “large-scale” circular bubble column has been si- vestigation is required.
mulated. The former is a well-known test case taken from the literature; • In the “large-scale” bubble column, the prediction of the global fluid
conversely, the latter is a new benchmark, which has been proposed dynamics has been compared in terms of the gas holdup, thus pro-
within this paper. The Eulerian-Eulerian approach and different interfacial viding a ranking of the modeling approach.

Appendix A: Mesh sensitivity study

It is known that a sufficiently fine mesh is required to capture the main transient phenomena of two-phase problems. If the mesh resolution is too
low, the accuracy of gradients estimates required in the computation of bubble forces, for instance, is compromised and flow instabilities are not
resolved. To obtain a mesh independent solution a mesh study has been performed and an extract of this study for the case with a superficial velocity
of 0.0037 m/s is presented. In particular, the large scale-bubble column case is considered Three meshes have been tested and compared, namely (i)
coarse (48132 cells), (ii) medium (205131 cells), and (iii) fine (368016 cells). The characteristics of these meshes are summarized in Table 6 and
displayed in Fig. 13. Owing to the gas sparger design (Fig. 2), the gas sparger has been modeled as a uniform surface on the bottom. Comparing the
obtained values for the local gas volume fraction (Fig. 14), the medium grid gives similar results as the finest mesh. Conversely, the coarse mesh
differs from the medium and fine mesh, especially for the local values near the wall of the column, suggesting that the dynamics predicted on the

Table 6
Characteristics of the meshes implemented (large-scale bubble column).

Mesh Transversal mesh size Vertical mesh size Number of cells

[m] [m] [−]

Coarse 0.010 0.016 48,132

Medium 0.010 0.010 205,131
Fine 0.009 0.008 368,016

233
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Fig. 13. Cross-sectional views of the three meshes implemented in the simulations (large-scale bubble column).

coarse mesh are different from the ones obtained with the medium and fine meshes. Otherwise, the numerical gas holdup (Table 7) shows low
sensitivity to the mesh element size as the results display relative errors below 5% for the three meshes investigated. Summarizing, the solution has a
low mesh dependency, however the results given by the medium and fine meshes are able to reproduce the slightly wall peaked profile of void
fraction. Consequently, the medium mesh has been employed in the simulations as it provides the best performance-accuracy trade-off. Please note
that these results agrees with the ones observed in the comprehensive analysis performed by Guédon et al. (Guédon et al., 2017).

234
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

Experimental data (Figure 3d) Baseline - Medium Mesh

Baseline - Coarse Mesh Baseline - Fine Mesh
0.025

0.020

ɸG,Local [-]
0.015

0.010

0.005

0.000
0 0.2 0.4 0.6 0.8 1
Dimensionaless axial posiƟon [-]

Fig. 14. Mesh element size sensitivity study for the Baseline model (large-scale bubble column) – (UG = 0.0037 m/s).

Table 7
Experimental and numerical gas holdup for the various meshes investigated (large-scale bubble column).

Case Gas holdup [−] Relative errore

Baseline – Fine Mesh 0.015574 1.05%

Baseline – Medium Mesh 0.015115 1.93%
Baseline – Coarse Mesh 0.014969 2.88%

Experimental gas holdup 0.0154.

Appendix B: Non-drag forces in the baseline approach

Herein, a brief description of the non-drag interfacial momentum exchange forces implemented in the baseline model is provided.

Lift force

The lift force is a lateral force originating in a shear flow. It is implemented as follows:
⎯→
⎯
FL,j = −CL αj ρk (→
uj−→
uk ) × (∇ × →
uk ) (7)
The lift coefficient CL depends mainly on the shape of the bubble. For small bubbles, it is positive; for deformed bubbles, it changes its sign. To
account for this dependency, the model of Tomiyama (Tomiyama et al., 2002) is used to compute the lift coefficient together with the use of two
bubble classes as introduced previously. One of the bubble classes represents small and nearly spherical bubbles with a positive CL, while the other
one models large and deformed bubbles with a negative CL. For the air-water system at ambient conditions, the bubble diameter at which the change
in sign occurs is 5.8 mm. The lift coefficient according to Tomiyama et al. is given as:

⎧ min[0.288tanh(0.121Reb),f (Eo⊥)] Eo⊥ ⩽ 4

CL = f (Eo⊥) 4 < Eo⊥ ⩽ 10
⎨
⎩ − 0.27 10 < Eo⊥ (8)
with
f (Eo⊥) = 0.00105Eo⊥3−0.0159Eo⊥2−0.0204Eo⊥ + 0.474 (9)
Eo⊥ is the Eötvös number considering the maximum horizontal dimension of the bubble d⊥, given by the empirical correlation for the aspect ratio by
Wellek et al. (Wellek et al., 1966):
d⊥ = db (1 + 0.163Eo0.757)1/3 (10)

Turbulent dispersion force

The diffusion effect of the turbulent fluctuations of the liquid phase on the bubbles is modeled through the turbulent dispersion force. It is derived
by Favre averaging the inter-phase drag term. The model of Burns et al. (Burns et al., 2004) is implemented and reads as:

⎯→
⎯ 3 C μ turb ∇αj ∇αk ⎞
FTD,j = − CTD αj (1−αj ) D |→
uj−→
uk | k ⎜⎛ − ⎟
4 dj σTD ⎝ αj αk ⎠ (11)
where CTD = 1 and σjk = 0.9 .

Wall lubrication force

A bubble moving near a wall is subject to a lift force that pushes it away from the wall. This force is often mentioned as the wall lubrication force
and is implemented as:

235
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

⎯→
⎯
FWL,j = −CWL ρk αj |(→
uk−→
uj )|||2 →
nw (12)
→ → →
where ( uk− uj )|| is the relative velocity component parallel to the wall and n w is the unit normal to the wall pointing toward the fluid. CWL is the wall
lubrication coefficient, which depends mainly on the distance to the wall and is given here by the Antal et al. (Antal et al., 1991) model

C C
CWL = max ⎛⎜0, W 1 + W 2 ⎞⎟
⎝ d b yW ⎠ (13)
where CW1 = −0.01 and CW2 = 0.05 are dimensionless constants and yW is the distance to the nearest wall.

References Guédon, G.R., Besagni, G., Inzoli, F., 2017. Prediction of gas-liquid flow in an annular gap
bubble column using a bi-dispersed eulerian model. Chem. Eng. Sci. 161, 138–150.
Grace, J.R., Wairegi, T., Nguyen, T.H., 1976. Shapes and velocities of single drops and
Antal, S.P., Lahey Jr, R.T., Flaherty, J.E., 1991. Analysis of phase distribution in fully bubbles moving freely through immiscible liquids. Trans. Inst. Chem. Eng. 54,
developed laminar bubbly two-phase flow. Int. J. Multiph. Flow 17, 635–652. 167–173.
Barrau, E., Rivière, N., Poupot, C., Cartellier, A., 1999. Single and double optical probes in Hu, G., Celik, I., 2008. Eulerian-Lagrangian based large-eddy simulation of a partially
air-water two-phase flows: real time signal processing and sensor performance. Int. J. aerated flat bubble column. Chem. Eng. Sci. 63, 253–271.
Multiph. Flow 25, 229–256. Hosokawa, S., Tomiyama, A., Misaki, S., Hamada, T., 2002. Lateral migration of single
Besagni, G., Di Pasquali, A., Gallazzini, L., Gottardi, E., Colombo, L.P.M., Inzoli, F., 2017a. bubbles due to the presence of wall. In: Proc. ASME Joint U.S.–European Fluids
The effect of aspect ratio in counter-current gas-liquid bubble columns: experimental Engineering Division Conference, FEDSM2002, Montreal, Canada, p. 855.
results and gas holdup correlations. Int. J. Multiph. Flow 94, 53–78. Ishii, M., 1975. Thermo-fluid dynamic theory of two-phase flow. NASA STI/Recon
Besagni, G., Gallazzini, L., Inzoli, F., 2018. Effect of gas sparger design on bubble column Technical Report A. 29657.
hydrodynamics using pure and binary liquid phases. Chem. Eng. Sci. 176, 116–126. Jain, D., Lau, Y.M., Kuipers, J.A.M., Deen, N.G., 2013. Discrete bubble modeling for a
Besagni, G., Guédon, G., Inzoli, F., 2014. Experimental and Numerical Study of Counter- micro-structured bubble column. Chem. Eng. Sci. 100, 496–505.
Current Flow in a Vertical Pipe. V002T011A009. Jakobsen, H.A., Lindborg, H., Dorao, C.A., 2005. Modeling of bubble column reactors:
Besagni, G., Inzoli, F., 2016. Comprehensive experimental investigation of counter-cur- progress and limitations. Ind. Eng. Chem. Res. 44, 5107–5151.
rent bubble column hydrodynamics: holdup, flow regime transition, bubble size Kiambi, S.L., Duquenne, A.-M., Dupont, J.B., Colin, C., Risso, F., Delmas, H., 2003.
distributions and local flow properties. Chem. Eng. Sci. 146, 259–290. Measurements of bubble characteristics: comparison between double optical probe
Besagni, G., Inzoli, F., De Guido, G., Pellegrini, L.A., 2017b. The dual effect of viscosity on and imaging. Canad. J. Chem. Eng. 81, 764–770.
bubble column hydrodynamics. Chem. Eng. Sci. 158, 509–538. Krepper, E., Lucas, D., Frank, T., Prasser, H.-M., Zwart, P.J., 2008. The inhomogeneous
Besagni, G., Inzoli, F., Ziegenhein, T., Lucas, D., 2017c. Computational Fluid-Dynamic MUSIG model for the simulation of polydispersed flows. Nucl. Eng. Des. 238,
modeling of the pseudo-homogeneous flow regime in large-scale bubble columns. 1690–1702.
Chem. Eng. Sci. 160, 144–160. Krepper, E., Reddy Vanga, B.N., Zaruba, A., Prasser, H.-M., Lopez de Bertodano, M.A.,
Besbes, S., El Hajem, M., Ben Aissia, H., Champagne, J.Y., Jay, J., 2015. PIV measure- 2007. Experimental and numerical studies of void fraction distribution in rectangular
ments and Eulerian-Lagrangian simulations of the unsteady gas–liquid flow in a bubble columns. Nucl. Eng. Des. 237, 399–408.
needle sparger rectangular bubble column. Chem. Eng. Sci. 126, 560–572. Laborde-Boutet, C., Larachi, F., Dromard, N., Delsart, O., Schweich, D., 2009. CFD si-
Boes, R.M., Hager, W.H., 1998. Fiber-optical experimentation in two-phase cascade flow. mulation of bubble column flows: investigations on turbulence models in RANS ap-
Int. RCC Dams Seminar, Denver, USA. proach. Chem. Eng. Sci. 64, 4399–4413.
Boure, J.A., Bergles, A.E., Tong, L.S., 1973. Review of two-phase flow instability. Nucl. Lapin, A., Lübbert, A., 1994. Numerical simulation of the dynamics of two-phase ga-
Eng. Des. 25, 165–192. s—liquid flows in bubble columns. Chem. Eng. Sci. 49, 3661–3674.
Burns, A.D., Frank, T., Hamill, I., Shi, J.-M., 2004. The Favre averaged drag model for Larachi, F.ç., Desvigne, D., Donnat, L., Schweich, D., 2006. Simulating the effects of liquid
turbulent dispersion in Eulerian multi-phase flows. 5th international conference on circulation in bubble columns with internals. Chem. Eng. Sci. 61, 4195–4206.
multiphase flow, ICMF. Ledinegg, M., 1938. Instability of flow during natural and forced circulation. Die Warme
Buwa, V.V., Deo, D.S., Ranade, V.V., 2006. Eulerian-Lagrangian simulations of unsteady 48, 891–898.
gas–liquid flows in bubble columns. Int. J. Multiph. Flow 32, 864–885. Liao, Y., Lucas, D., Krepper, E., 2014. Application of new closure models for bubble
Chang, K.-A., Lim, H.-J., Su, C.B., 2003. Fiber optic reflectometer for velocity and fraction coalescence and breakup to steam–water vertical pipe flow. Nucl. Eng. Des. 279,
ratio measurements in multiphase flows. Rev. Sci. Instrum. 74, 3559–3565. 126–136.
Chaumat, H., Billet-Duquenne, A.M., Augier, F., Mathieu, C., Delmas, H., 2005. Liao, Y., Rzehak, R., Lucas, D., Krepper, E., 2015. Baseline closure model for dispersed
Application of the double optic probe technique to distorted tumbling bubbles in bubbly flow: bubble coalescence and breakup. Chem. Eng. Sci. 122, 336–349.
aqueous or organic liquid. Chem. Eng. Sci. 60, 6134–6145. Lima Neto, I., Zhu, D., Rajaratnam, N., 2008. Air injection in water with different nozzles.
Chen, P., Sanyal, J., Dudukovic, M.P., 2004. CFD modeling of bubble columns flows: J. Environ. Eng. 134, 283–294.
implementation of population balance. Chem. Eng. Sci. 59, 5201–5207. Lopez de Bertodano, M., 1991. Turbulent Bubbly Flow in a Triangular Duct (Ph.D. Thesis)
Chen, P., Sanyal, J., Duduković, M.P., 2005. Numerical simulation of bubble columns In: Rensselaer Polytechnic Institute, Troy, New York.
flows: effect of different breakup and coalescence closures. Chem. Eng. Sci. 60, Lucas, D., Krepper, E., Prasser, H.M., 2007. Use of models for lift, wall and turbulent
1085–1101. dispersion forces acting on bubbles for poly-disperse flows. Chem. Eng. Sci. 62,
Cheung, S.C.P., Yeoh, G.H., Tu, J.Y., 2007a. On the modelling of population balance in 4146–4157.
isothermal vertical bubbly flows—Average bubble number density approach. Chem. Lucas, D., Prasser, H.M., Manera, A., 2005. Influence of the lift force on the stability of a
Eng. Process. Process Intensif. 46, 742–756. bubble column. Chem. Eng. Sci. 60, 3609–3619.
Cheung, S.C.P., Yeoh, G.H., Tu, J.Y., 2007b. On the numerical study of isothermal vertical Lucas, D., Rzehak, R., Krepper, E., Ziegenhein, T., Liao, Y., Kriebitzsch, S., Apanasevich,
bubbly flow using two population balance approaches. Chem. Eng. Sci. 62, P., 2015. A strategy for the qualification of multi-fluid approacheds for nuclear re-
4659–4674. actor safety. Nucl. Eng. Des. 299, 2–11.
Deen, N.G., Solberg, T., Hjertager, B.H., 2001. Large eddy simulation of the Gas-Liquid Masood, R.M.A., Delgado, A., 2014. Numerical investigation of the interphase forces and
flow in a square cross-sectioned bubble column. Chem. Eng. Sci. 56, 6341–6349. turbulence closure in 3D square bubble columns. Chem. Eng. Sci. 108, 154–168.
Delnoij, E., Lammers, F.A., Kuipers, J.A.M., van Swaaij, W.P.M., 1997. Dynamic simu- Masood, R.M.A., Khalid, Y., Delgado, A., 2015. Scale adaptive simulation of bubble
lation of dispersed gas-liquid two-phase flow using a discrete bubble model. Chem. column flows. Chem. Eng. J. 262, 1126–1136.
Eng. Sci. 52, 1429–1458. Masood, R.M.A., Rauh, C., Delgado, A., 2014. CFD simulation of bubble column flows: an
Díaz, E.M., Montes, F.J., Galán, M.A., 2009. Influence of the lift force closures on the explicit algebraic Reynolds stress model approach. Int. J. Multiph. Flow 66, 11–25.
numerical simulation of bubble plumes in a rectangular bubble column. Chem. Eng. Mudde, R.F., Simonin, O., 1999. Two- and three-dimensional simulations of a bubble
Sci. 64, 930–944. plume using a two-fluid model. Chem. Eng. Sci. 54, 5061–5069.
Díaz, M.E., Iranzo, A., Cuadra, D., Barbero, R., Montes, F.J., Galán, M.A., 2008. Numerical Oey, R.S., Mudde, R.F., van den Akker, H.E.A., 2003. Sensitivity study on interfacial
simulation of the gas–liquid flow in a laboratory scale bubble column: influence of closure laws in two-fluid bubbly flow simulations. AIChE J. 49, 1621–1636.
bubble size distribution and non-drag forces. Chem. Eng. J. 139, 363–379. Parekh, J., Rzehak, R., 2017. Euler-Euler multiphase CFD-simulation with full Reynolds
Duan, X.Y., Cheung, S.C.P., Yeoh, G.H., Tu, J.Y., Krepper, E., Lucas, D., 2011. Gas–liquid stress model and anisotropic bubble-induced turbulence. Int. J. Multiph. Flow.
flows in medium and large vertical pipes. Chem. Eng. Sci. 66, 872–883. Peña-Monferrer, C., Monrós-Andreu, G., Chiva, S., Martínez-Cuenca, R., Muñoz-Cobo,
Ekambara, K., Dhotre, M.T., Joshi, J.B., 2005. CFD simulations of bubble column reactors: J.L., 2017. A CFD-DEM solver to model bubbly flow. Part II: critical validation in
1D, 2D and 3D approach. Chem. Eng. Sci. 60, 6733–6746. upward vertical pipes including axial evolution. Chem. Eng. Sci.
Fluent, A., 2013. Ansys Fluent 14.5.7 Users Guide. Fluent Inc., Lebanon, NH. Pfleger, D., Gomes, S., Gilbert, N., Wagner, H.G., 1999. Hydrodynamic simulations of
Frank, T., Shi, J.M., Burns, A.D., 2004. Validation of Eulerian Multiphase Flow Models for laboratory scale bubble columns fundamental studies of the Eulerian-Eulerian mod-
Nuclear Safety Applications. Third International Symposium on Two-Phase Flow elling approach. Chem. Eng. Sci. 54, 5091–5099.
Modeling and Experimentation. pp. Sept. 22–24. Pourtousi, M., Sahu, J.N., Ganesan, P., Shamshirband, S., Redzwan, G., 2015a. A com-
Frank, T., Zwart, P.J., Krepper, E., Prasser, H.M., Lucas, D., 2008. Validation of CFD bination of computational fluid dynamics (CFD) and adaptive neuro-fuzzy system
models for mono- and polydisperse air–water two-phase flows in pipes. Nucl. Eng. (ANFIS) for prediction of the bubble column hydrodynamics. Powder Technol. 274,
Des. 238, 647–659.

236
G. Besagni et al. Nuclear Engineering and Design 331 (2018) 222–237

466–481. Flows, FED91, pp. 65–82.

Pourtousi, M., Zeinali, M., Ganesan, P., Sahu, J.N., 2015b. Prediction of multiphase flow Simonnet, M., Gentric, C., Olmos, E., Midoux, N., 2007. Experimental determination of
pattern inside a 3D bubble column reactor using a combination of CFD and ANFIS. the drag coefficient in a swarm of bubbles. Chem. Eng. Sci. 62, 858–866.
RSC Adv. 5, 85652–85672. Sokolichin, A., Eigenberger, G., 1999. Applicability of the standard k–ε turbulence model
Ruzicka, M.C., Zahradni’k, J., Drahoš, J., Thomas, N.H., 2001. to the dynamic simulation of bubble columns: Part I. Detailed numerical simulations.
Homogeneous–heterogeneous regime transition in bubble columns. Chem. Eng. Sci. Chem. Eng. Sci. 54, 2273–2284.
56, 4609–4626. Tabib, M.V., Roy, S.A., Joshi, J.B., 2008. CFD simulation of bubble column—An analysis
Rzehak, R., Krepper, E., 2013a. CFD modeling of bubble-induced turbulence. Int. J. of interphase forces and turbulence models. Chem. Eng. J. 139, 589–614.
Multiph. Flow 55, 138–155. Tomiyama, A., Kataoka, I., Zun, I., Sakaguchi, T., 1998. Drag coefficients of single bubbles
Rzehak, R., Krepper, E., 2013b. Closure models for turbulent bubbly flows: A CFD study. under normal and micro gravity conditions. JSME Int. J. Ser. B Fluids Therm. Eng. 41,
Nucl. Eng. Des. 265, 701–711. 472–479.
Rzehak, R., Krepper, E., Liao, Y., Ziegenhein, T., Kriebitzsch, S., Lucas, D., 2015. Baseline Tomiyama, A., Sou, A., Zun, I., Kanami, N., Sakaguchi, T., 1995. Effects of Eötvös number
model for the simulation of bubbly flows. Chem. Eng. Technol. 38, 1972–1978. and dimensionless liquid volumetric flux on lateral motion of a bubble in a laminar
Rzehak, R., Krepper, E., Lifante, C., 2012. Comparative study of wall-force models for the duct flow. In: Proc. 2nd Int. Conf. on Multiphase Flow, Kyoto, Japan, p. 3.
simulation of bubbly flows. Nucl. Eng. Des. 253, 41–49. Tomiyama, A., Tamai, H., Zun, I., Hosokawa, S., 2002. Transverse migration of single
Rzehak, R., Kriebitzsch, S., 2015. Multiphase CFD-simulation of bubbly pipe flow: a code bubbles in simple shear flows. Chem. Eng. Sci. 57, 1849–1858.
comparison. Int. J. Multiph. Flow 68, 135–152. Wellek, R.M., Agrawal, A.K., Skelland, A.H.P., 1966. Shape of liquid drops moving in
Sasaki, S., Hayashi, K., Tomiyama, A., 2016. Effects of liquid height on gas holdup in liquid media. AIChE J. 12, 854–862.
air–water bubble column. Exp. Therm. Fluid Sci. 72, 67–74. Zhang, D., Deen, N.G., Kuipers, J.A.M., 2006. Numerical simulation of the dynamic flow
Sasaki, S., Uchida, K., Hayashi, K., Tomiyama, A., 2017. Effects of column diameter and behavior in a bubble column: a study of closures for turbulence and interface forces.
liquid height on gas holdup in air-water bubble columns. Exp. Therm. Fluid. Sci. 82, Chem. Eng. Sci. 61, 7593–7608.
359–366. Zhang, W., Zhu, D.Z., 2013. Bubble characteristics of air–water bubbly jets in crossflow.
Sato, Y., Sadatomi, M., Sekoguchi, K., 1981. Momentum and heat transfer in two-phase Int. J. Multiph. Flow 55, 156–171.
bubble flow—I Theory. Int. J. Multiphase Flow 7, 167–177. Ziegenhein, T., Lucas, D., Rzehak, R., Krepper, E., 2013. Closure relations for CFD si-
Sharaf, S., Zednikova, M., Ruzicka, M.C., Azzopardi, B.J., 2016. Global and local hy- mulation of bubble columns. 8th International Conference on Multiphase Flow. ICMF
drodynamics of bubble columns – effect of gas distributor. Chem. Eng. J. 288, 2013, Jeju (Korea).
489–504. Ziegenhein, T., Rzehak, R., Lucas, D., 2015. Transient simulation for large scale flow in
Simonin, C., Viollet, P., 1990. Predictions of an oxygen droplet pulverization in a com- bubble columns. Chem. Eng. Sci. 122, 1–13.
pressible subsonic coflowing hydrogen flow. In: Numerical Methods for Multiphase

237