Академический Документы
Профессиональный Документы
Культура Документы
157 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
8.1. Introduction
The current design and scale-up procedures for multiphase systems (flow systems, reactors,
separation equipment, etc.) need long commercialization time for generating confidence through the
extensive data generation in the laboratory, pilot and demonstration scales. It also results into high
capital as well as operating expenditure, long start-up and shut down times, etc. The genesis of these
drawbacks lie in the empiricism which originates from the lack of knowledge of fluid mechanics and
its relationship with the design objectives. In order to improve the lack of understanding, over the
past 50 years, continuous efforts are being made using Experimental Fluid Dynamics (EFD) and
Computational Fluid Dynamics (CFD). The progress in CFD has been remarkable considering the
complexity of the three dimensional turbulent multiphase flows. This has been made possible by the
additional work is needed in understanding the physics of turbulence (of multiphase systems) so that
equations of conservation of mass, momentum, energy and scalar (enthalpy, tracer, reactant etc.) can
be solved with the first principles. Therefore, with the current status of knowledge, a good number
of assumptions are needed in the low order models such as k-and RSM. The standard k- model
has performed satisfactory in many flows, but the applicability of this model is limited due to the
production and dissipation, turbulent convective transport, etc. (discussed later). It is known that the
number of assumptions decreases as the order of turbulence models becomes higher in the sequence
of Reynolds Stress Model (RSM), Large Eddy Simulation (LES) and Direct Numerical Simulation
(DNS). Though the accuracy of predictions increases from k- to DNS, the demand on computational
time increases by orders of magnitude. In fact, the DNS simulation of even moderate size multiphase
equipment is possible only at places of super computational facilities. Therefore, it was thought
desirable to compare understand the k- RSM and LES approaches for gas-liquid bubble columns in
158 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
At this stage, it may be pointed out that k- and RSM approaches are based on RANS and LES is
another completely different approach. RANS is based on Reynolds-average that transforms a chaotic
field (that is obtained by solving the governing equations with DNS) into a non-chaotic field. LES
instead, by filtering the equations, takes a chaotic field and transforms it into another chaotic field,
The conservation equations for k, and Reynolds stress consist of convective transport, diffusive
transport, turbulent transport, production and dissipation. These terms are derived from the equations
of continuity and motion (through Reynolds averaging procedure) and are in the form of double and
triple correlations of fluctuating velocities (u′i ), fluctuating phase fractions (ϵ′G , ϵ′L) and their
gradients. These correlations were estimated by using instantaneous fluctuating values u′i , ϵ′G , ϵ′L
obtained from large eddy simulations (LES). In this paper, each of these correlations has been
quantified and an attempt has been made to give a clearer insight into each modeling assumption.
It is known that the LES is also simplified version as compared to DNS because the LES also needs
the modeling of subgrid scale turbulence. Therefore, though it is desirable that the comparison of low
order models (k- RSM and LES) be made with DNS, we still need to wait till the available
computational speed permits the DNS of moderate size bubble columns. Until then, we thought
desirable to use LES for understanding the gravity of assumptions made in k- and RSM. In order to
give some credibility for such comparison, we have implemented some care in LES simulations
which essentially is in the form of energy balance. The care has been taken at two levels: (1) while
implementing LES, it was ensured that, by proper selection of the grid size, more than 95% of the
turbulent kinetic energy is accounted in large scale fluctuations and (2) the LES simulation gives the
values of at all the grid points. The volume integral of this field gives the total energy dissipation
rate (ED) in the column. We have ensured that ED is within 100±5% of the energy input rate given
by:
π 2
Ei = D VG HD (ρL − ρG )g (8.1)
4
159 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Thus the process of filtering loses the accuracy of less than 5% of the input energy. Even then, we
realize that the use of DNS would have given accurate comparisons. However, with the current status
of computational speed and the care taken for LES simulations, we are expected to get some insight
(into the gravity of assumptions made in the k- and RSM) and is the subject of the present paper.
Though the activity of CFD simulations began in late 70s, it got firm ground in mid 90s. Joshi (2001),
Sokolochin et al. (2004) and Rafique et al. (2004) have critically analyzed the published literature
then and have made coherent presentation of the status. Since 2001, substantial development has
occurred in terms of the usage of turbulence models (k-, RSM, LES) and Table 8.1 shows the utility
pattern of these models. Dhotre et al. (2013) have presented a comprehensive review of LES (both
E-E and E-L) of dispersed bubbly flows and the same subject will not be repeated here. Since the
subject of this paper is comparison of three turbulence models, only those papers will be reviewed
Deen et al. (2001) have carried out pioneering work for the comparison of k- and LES models for
gas-liquid flow in a square cross-sectional bubble column. They found that the k- model
overestimates the turbulent viscosity and could only predict low frequency unsteady flow. However,
LES has been shown to reproduce high frequency experimental data. Further, LES could predict the
Milelli (2002) simulated a free bubble plume which was experimentally investigated by Anagbo and
Brimacombe (1990). The simulations have revealed that the mean quantities as well as the bubble
induced turbulence affected by the different SGS models. Further, Milelli (2002) has observed that
the lift coefficient value plays a major role in capturing the plume spreading. He further observed
that the values of lift coefficient may differ for LES as compared to the one which is justified in a
160 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Table 8.1. Literature review for the turbulence models used in CFD simulations
Turbulence
No References
Model
1 k- Delnoij et al. (1999), Glover et al. (2000), Krishna and van Batan (2001),
van Batan and Krishna (2002), Buwa and Ranade (2002), Bertola et al.
(2002), Rampure et al. (2003), Ekambara and Joshi (2003a, 2003b),
Rigopoulos and Jones (2003), van Batan and Krishna (2003), Dhotre et al.
(2004), van Batan and Krishna (2004a, 2004b), Dhotre and Joshi (2004),
Chen et al. (2004), Ekambara and Joshi (2005), Dhotre et al. (2005),
Monaham et al. (2005), Ekambara et al. (2005), Rahimi et al. (2006),
Kulkarni and Joshi (2006), dos Santos et al. (2007), Dhotre and Joshi
(2007), Kulkarni et al. (2007), Wang and Wang (2007), Rampure et al.
(2007), Roy and Joshi (2008), Bhole et al. (2008), Tabib et al. (2008),
Simonnet et al. (2008), Ekambara et al. (2008), Rampure et al. (2009),
Chen et al. (2009), Cachaza et al. (2009), Laborde-Boutet et al. (2009),
Kulkarni et al. (2009), Li et al. (2009), McGuffie et al. (2010), Ekambara
and Dhotre (2010), Selma et al. (2010), Wang (2011), Yang et al. (2011),
Ribeiro et al. (2012), Studley and Battaglia (2011), Qi et al. (2013), Xing
et al. (2013), Xiao et al. (2013), Mehrnia et al. (2013), Silva et al. (2014),
McClure et al. (2014), Masoosd et al. (2014), Liu and Hinrichsen (2014),
Pourtousi et al. (2015a, 2015b), McClure et al. (2015a, 2015b, 2015c),
Zeigenhein et al. (2015), Yancheshme et al. (2016)
2 RSM Ekambara et al. (2008), Ekambara and Dhotre (2010), Silva et al. (2014),
Masood et al. (2014), Liu and Hinrichsen (2014)
3 LES Deen et al. (2001), Milelli et al. (2001), Milelli PhD Thesis (2002), Niceno
et al. (2009), Tabib et al. (2008),Dhotre et al. (2009), Dhotre et al. (2008),
Zhang et al. (2006a, 2006b), Lain (2009), Bove et al. (2004), Tabib and
Schwarz (2011), van den Hengel et al. (2005), Hu and Celik (2008),
Darmana et al. (2009), Sungkorn et al. (2011).
4 Bubble van Batan and Krishna (2004)6,7, Li et al. (2009)5,6, McClure et al. (2014)1,
induced Masoosd et al. (2014)2, Liu and Hinrichsen (2014)4, Zeigenhein et al.
turbulence (2015), Pourtousi et al. (2015a2, 2015b3), McClure et al. (2015a1, 2015b1)
Models for bubble induced turbulence:
1
Pfleger and Becker (2001), 2Sato and Sekoguchi (1975), 3Sato, Sadatomi and Sekoguchi (1981),
4
Arnold, Drew and Lahey (1989), 5BelF’dhilla and Simonin (1992), 6Elgobashi and Abou-arab
(1983), 7Launder and Spalding (1974)
Tabib et al. (2008) have performed 3D transient CFD simulations of a demonstration size bubble
column over a wide range of superficial gas velocities. In addition, the authors have simulated a pilot
scale bubble column with three different designs of spargers (single hole, perforated plate and
sintered plate) using three different turbulence closures (k-, RSM, LES) and the predictions have
been compared with the experimental data of Bhole et al. (2006). Near the sparger, none of the
161 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
turbulence model has been able to predict the axial velocity profiles and the gas holdup profiles as
reported experimentally. The predictive capability improves at higher axial locations, where the flow
gets developed. Contrary to the expectations, the RSM has not been able to show better predictive
performance than k–in predicting the average axial velocity profiles. The profiles of Reynolds stress
and kinetic energy for all the three turbulence models are in partial agreement with some deviations.
In predicting turbulent kinetic energy, RSM does a better job than k–, which could be due to its
intrinsic ability in capturing the anisotropic energy transfer mechanism. LES has been successful in
capturing averaged behavior of the flow. Though, at some locations, it slightly over predicts the
kinetic energy and stress profiles, which could be due to the consideration of a single Smagorinsky
LES has been able to capture the instantaneous phenomena quite well. Further, the LES has been
able to identify the flow structures and also the flow regions of the vertical-spiral flow regime in the
sieve plate column. Further, LES has been shown to capture the bubble plume dynamics along with
vertical structures for the single hole sparger. Perhaps, the incorporation of bubble coalescence and
break up model in LES could help to distinguish the two regions (fast bubble region and central
bubble plume region) in the vertical-spiral regime, and thus help in better prediction of the flow fields
in these regions. For a column provided with a sintered plate distributor, LES gives nearly
homogeneous flow conditions with an absence of dynamic eddies. It has been shown by the authors
that LES can be effectively used for the study of coherent structures and instantaneous flow profiles.
Ekambara and Dhotre (2010) have analyzed the performance of different turbulence models
namely, k–ɛ, k–ω, RNG k–ɛ, Reynolds stress model (RSM) and large eddy simulation (LES) and the
predictions have been assessed against the experimental data of Bhole et al. (2006). All the non-drag
forces [turbulent dispersion (only RANS), virtual mass, lift and wall lubricant force] and drag force
were incorporated in the model. The comparison of prediction of axial liquid velocity, fractional gas
hold-up, turbulent kinetic energy and turbulent eddy dissipation rate, with the experimental data is
162 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
presented.
Bhole et al. (2006) have given pertinent details of the experimental set-up and the measurement
techniques. The same has been briefly described below: An acrylic column of 150 mm diameter and
1 m height was employed as a bubble column. A schematic diagram for the experimental setup is
Figure 8.1. Schematic of experimental setup: (1) 5W Ar-ion laser, (2) bubble column, (3)
photomultiplier tubes, (4) burst spectrum analyzer, (5) personal computer, (6) rotameter, and
(7) compressor.
To avoid the laser beam distortion, the cylindrical column was enclosed by a square column, and the
gap between the two was filled with water. An oil-free diaphragm compressor was used to sparge the
air through the gas sparger. Air flow rate was monitored using a rotameter, which was precalibrated
using a soap-film meter. Ordinary tap water was used as the liquid phase, and oil-free compressed
air was used as the gas phase. The height of gas-liquid dispersion was kept about 900 mm. The bubble
column was mounted on a traverse, which allowed the accurate vertical movement so that the
163 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
measurements at various axial locations in the column are possible. At a fixed axial location, the
radial variation of the measurement volume was obtained by accurate movement of the laser-focusing
front lens along a guided horizontal platform. The measurements were made from the center of the
column up to the wall. Axisymmetry was ensured by circular symmetric sparger plate design and the
perfect vertical orientation of the column. A slight departure from the vertical orientation leads to a
Thus, the issue of the column orientation is not trivial. In fact, in our measurements, we ensured
axisymmetry by comparing the mean axial velocity at two points equidistant from the center (r=0).
The average gas holdup in the column was measured by noting the height of liquid with and without
gas dispersion. All the Laser-Doppler Anemometer (LDA) measurements were carried out at
superficial gas velocity of 20 mm/s. However, the average gas holdup was measured at various
superficial gas velocities. A multipoint perforated plate with 2 mm 25 holes was used as the sparger.
To maximize the data rate, the photomultiplier tube is placed almost collinear with the front lens and
measurement volume. This is because the intensity of scattered laser light (in the forward scatter
mode) is highest at the receiving angle of 0°. Data acquisition in bubbly flows is relatively difficult
compared to the single phase flows. Laser beams are blocked by bubbles rising through their path;
hence, the data rate gets reduced substantially. Various modes of bubble-beam interactions are
elucidated by Kulkarni et al. (2001a, 2001b, 2001c). In the center of the column, the lowest data rate
is observed. This is mainly due to the fact that the laser beams have to travel a larger distance through
the bubbly fluid and the frequency of beam interruption by the bubbles increases. Furthermore, the
gas holdup is also higher in the center of a typical bubble column. Bubbles also pose an important
difficulty due to multiple reflections of the laser light from their surfaces. In our work, we have
obtained the data rate as high as 1000 Hz near the wall and about 100 Hz at the center. To study the
turbulence quantities such as power spectra and Reynolds shear stresses, a higher data rate is
preferable. However, it must be noted that the data validity must also be high, and the photomultiplier
164 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
anode current must not exceed a certain value in order to keep the noise level in the acquired data to
a minimum level. Although, the mean velocities may not get affected, the RMS velocities increase
considerably if the data validation is poor and/or photomultiplier anode current is high. Thus, a
judicious optimization between the data rate, data validity and photomultiplier anode current must
be carried out to obtain a good signal from the LDA. In this work, we have maintained the data
validity of more than 90%. Typically, the data acquisition at a particular location is carried out for a
period of 300 s to ensure that the mean velocity is not affected by the total measurement time. LDA
measurements were carried out at four axial locations in the column (viz., z/D = 1, 2, 3, and 4). At
each axial location, the measurements were made from the center up to the wall with 5 mm gap
between the successive points. Due to extreme curvature effects, the LDA measurements very close
to the wall are not possible. In our column of 150 mm diameter, the measurement point closest to the
wall is about 3-4 mm away from the wall. This corresponds to the dimensionless radial distance (r/R)
of about 0.95.
The momentum conservation equation for the liquid phase, under steady state conditions, for
∂ ∈L ∂P ∂ ∂ui
(∈L ui uj ) = − ( )+ (ν ∈L ( )) + FI (8.2)
∂xj ρL ∂xi ∂xj ∂xj
The conservation equation for k is obtained from Eq. (8.2) by the following stepwise procedure
[Celik (1999)]: (a) convert Eq. (8.2) into two conservation equations in terms of mean variables
(u̅𝑖 , u̅𝑗 , p̅, ϵ̅𝐿 ) and fluctuating variables (u′i , u′j , p′ , ϵ′L , etc. ).
(b) taking dot product of (u′𝑖 ) with each term in the conservation equation of fluctuating variables
165 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
∂ ′
∂ ′ ′
∂ ′ ′ ′
∂ ′ ′ ′ ′
∂2 u̅i
̅ ̅
(∈L u̅i k) + ∈L u̅i 〈ui (u )〉 + u̅j 〈ui (∈ u )〉 + u̅i 〈ui (∈ u )〉 + 〈ui ∈L 〉ν 2
∂x
⏟j ⏟ ∂xj i ⏟ ∂xj L i ⏟ ∂xj L j ⏟ ∂x𝑗
1.1 1.2 1.3 1.4 2.1
∂ ∂k ∂〈u′j 𝑘〉 ′ ′
∂ ′
∂ ∈′L ̅L
∂∈
+ ̅ ̅
(∈L ν ) + ∈L − 〈ui ∈L 〉 (u̅i u̅j ) − u̅i u̅j 〈ui 〉 − 〈u′i u′i u′j 〉
∂xj
⏟ ∂xj ⏟ ∂xj ⏟ ∂xj ⏟ ∂xj ⏟ ∂xj
2.2 3.1 3.2 3.3 3.4
+ S⏟k (8.3)
6.1
The above equation can be written in the following form every term is rate of energy (W/m3):
As per Eq. (8.4), the terms in Eq. (8.3) such as 1.1, 1.2, 1.3 and 1.4 correspond to convective transport,
terms 2.1, 2.2 correspond to diffusive transport, terms 3.1, 3.2, 3.3, 3.4 correspond, to the turbulent
transport, terms 4.1, 4.2, 4.3, 4.4 correspond to the production and term 5.1 corresponds to dissipation
of turbulent kinetic energy. Term 6.1 is the source term arising out of bubble induced turbulence
(BIT). This above mentioned classification of terms in k equation is shown in Table 8.2. The last
column shows the corresponding modelled terms in a standard k- ε model under steady state
∂ ∂ ∂k ∂ νt ∂k Gk,L
(ϵ̅L kui ) = [ϵ̅L ν ] + [ϵ̅L ] + ϵ̅L + Sk − ϵ̅L ε (8.5)
∂xi ∂xj ∂xj ∂xj σk ∂xj ρL
The turbulent viscosity νt is given by Sato and Sekoguchi model (1975) as follows:
k2
νt = Cμ ϵG B |v̅𝑖 − u̅i |
+ Cμ,G ̅̅̅d (8.6)
ε
The production of turbulent kinetic energy from mean flow (Gk,L) is computed as:
∂u̅i
̅̅̅̅̅
Gk,L = −ρL u ′ u′ (
i j ) (8.7)
∂xj
166 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Table 8.2. Budget for the turbulent kinetic energy (k) in k-model
Terms arising out of derivation of k equation [Eq. (8.3)] from the equation of motion and continuity for two- Modelled Terms in
phase flow standard k- model
Rate of ∂ ∂〈u′i u′j 〉 ∂ ∂ ∂
̅ u̅ k)
(∈ ̅L u̅i
+∈ +u̅j 〈u′i (∈′ u′ )〉 +u̅i 〈u′i (∈′ u′ )〉 ̅ ku̅ )
(∈
Convective ∂xj L j ∂xj ∂xj L i ∂xj L j ∂xj L j
Transport
Term 1.1 Term 1.2 Term 1.3 Term 1.4
167 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
The term Sk is the source term that is included to model the turbulent interaction between gas and
liquid phases. For instance, Troshko and Hassan (2001) have modelled this term as follows:
Where, Cke (=0.75) is the constant for source term arising from bubble induced turbulence. K GL is the
∂ ∂ ∂k ∂ νt ∂k ∂ϵ̅L u̅i
(ϵ̅L kui ) = [ϵ̅L ν ] + [ϵ̅L ] − τij ( ) − ϵ̅L ε + Sk (8.9)
∂xi ∂xj ∂xj ∂xj σk ∂xj ∂xj
At this stage, we thought desirable to provide brief explanation on bubble generated turbulence [the
details are given in Joshi et al. (2016)]. The discussion is based on the energy balance approach
[Freedman and Davidson (1968), Joshi and Sharma (1976), Joshi (2001)]. For bubble column, the
energy supply is due to the introduction of gas and the rate is given by:
π 2
EiG = D VG (ϵ̅L ρL + ̅̅̅ρ
ϵG G )HD g (8.10)
4
The gas and liquid leaving the column have the following rates, respectively:
π 2
EOG = D VG ρG HD g (8.12)
4
π 2
EOL = D VL ρL HD g (8.13)
4
From Eqs. (8.10) to (8.13), the net energy input rate is given by:
π 2 ϵG L
̅̅̅V
EN = D HD g ϵ̅L (ρL − ρG ) (VG − ) (8.14)
4 ϵ̅L
From the above net energy input, some energy is used for overcoming drag at the gas-liquid interface
168 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
E = EN − ED
π 2 ϵG L
̅̅̅V
E= D HD g ϵ̅L (ρL − ρG ) (VG − − ̅̅̅
ϵG VS ) (8.16)
4 ϵ̅L
The energy given by Eq. (8.16) is used for generating convective circulatory motion in the column
which is eventually converted into turbulent kinetic energy. The estimation of turbulent kinetic
energy production [given by Eq. (8.16)] is represented by Eq. (8.7). The energy spent for bubble
generated turbulence is given by Eq. (8.15) and is represented by Eq. (8.8) in the revised manuscript.
It may be pointed out that, there are many other alternative ways to represent Eq. (8.15) and Eq. (8.8)
The bracketed term in Eq. (8.16) gives us the relative estimate of energy available for bubble
is available for generating convective flows and the gas –liquid dispersion becomes homogeneous.
Under these conditions, the energy spent for bubble generated turbulence is maximum. On contrast,
the energy spent for liquid circulation increases when the value of bracketed term increases or with
a decrease in ̅̅̅
ϵG and the sip velocity (VS ).
8.4.2. Conservation equation for turbulent energy dissipation rate [Celik (1999)]
Similar to the conservation equation for turbulent kinetic energy (k), by taking derivative of the
conservation equation for fluctuating variables with respect to 𝜕𝑥𝑘 and then multiplying it by
(2𝜈(∂u′i ⁄∂xk ))we get the conservation equation for the turbulent energy dissipation rate (ɛ) as
follows:
169 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
2
2
∂2 u′i ∂〈u′j ε〉 ∂u′i ′ ∂2 (u̅i ϵ̅L ) ∂u′i ′ ∂3 u′i
̅ L 〈(
+ 2ν ∈ ̅L
) 〉+∈ − 2ν 〈 ui 〉 −ν〈 ϵL 〉 2
⏟ ∂xk ∂xj ⏟ ∂xj ⏟ ∂xk ∂xj ∂xk ⏟ ∂xk ∂xj ∂xk
2.2 3.1 3.2 3.3
∂2 (u̅i 𝜖L′ ) ∂u′i ∂u̅i ∂u′i ∂(u̅j 𝜖L′ ) ∂u′j ∂u′i ∂(u̅j 𝜖L′ )
− 2ν 〈u′j 〉 − 2ν 〈 〉 − 2ν 〈 〉
⏟ ∂xk ∂xk ∂xk ⏟ ∂xk ∂xk ∂xj ⏟ ∂xk ∂xk ∂xj
3.4 3.5 3.6
∂u′i ′ ∂2 (u̅i u̅j ) ∂u′i ∂u′j ∂(u̅i ϵ̅L ) ∂u′i ′ ∂2 (u̅i ϵ̅L )
= − 2ν 〈 ϵL 〉 − 2ν 〈 〉 − 2ν 〈 uj 〉
⏟ ∂xk ∂xj ∂xk ⏟ ∂xk ∂xk ∂xk ⏟ ∂xk ∂xj ∂xk
4.1 4.2 4.3
∂u′i ∂u′i u′j ∂ϵ̅L ∂u′i ∂u′i u′j ∂2 ϵ̅L ∂(u̅i ϵ̅L ) ∂u′i ∂u′i ∂u′j ∂2 u′i
− 2ν 〈 〉 + 2ν 〈 〉 − 〈 〉 − 2ν 〈 (u̅i 𝜖L′ )〉
⏟ ∂x k ∂x k ∂x k ∂x k ∂x k ∂x k ∂x k ∂x
⏟ k ∂x k ∂x j ⏟ ∂x k ∂x k ∂x j
4.4 5.1 5.2
+ S⏟ε (8.17)
6.1
As per Eq. (8.4), the terms in Eq. (8.17) such as 1.1, 1.2 and 1.3 correspond to convective transport
of ε. The terms 2.1 and 2.2 correspond to diffusive transport of ε. Terms 3.1, 3.2, 3.3, 3.4, 3.5, 3.6
correspond to the turbulent transport of ε. The terms 4.1, 4.2, 4.3, 4.4 correspond to production of ε
and terms 5.1, 5.2, 5.3, 5.4 correspond to dissipation of ε, Sε is the source term. The above mentioned
classification of different terms in equation is shown in Table 8.3. The last column shows the
∂ ∂ ∂ε ∂ νt ∂ε ε ̅L u̅i
∂∈ ε2
(ϵ̅L εui ) = [ϵ̅ ν ] + [ϵ̅ ] + ϵ̅L Cε1 τij − ϵ̅L Cε2 + Sε (8.18)
∂xi ∂xj L ∂xj ∂xj L σε ∂xj 𝑘 ∂xj 𝑘
Ctd
Sε = S (8.19)
τp k
170 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Table 8.3. Budget for the kinetic energy dissipation rate () in k-model
Terms arising out of derivation of equation [Eq. (8.17)] from the equation of motion and continuity for two-phase flow Modelled terms in
standard k- model
Rate of ∂ ∂u′i ∂ϵ′L ∂ ∂u′i ∂2 ϵ′L ∂
̅ u̅ ε)
(∈ −2ν 〈 〉 (u̅ u̅ ) −2ν(u̅i u̅j ) 〈 〉 (ϵ̅ εu )
Convective ∂xj L i ∂xk ∂xk ∂xk i j ∂xk ∂xj ∂xk ∂xi L i
Transport Term 1.2
Term 1.1 Term 1.3
2
Rate of ∂ ∂ε ∂2 u′i ∂ ∂ε
Viscous + ̅L ν )
(∈ +2ν2 ∈
̅L 〈( ) 〉 (ϵ̅L ν )
∂xj ∂xj ∂xk ∂xj ∂xj ∂xj
Transport
Term 2.1 Term 2.2
Rate of ∂〈u′j ε〉 ∂u′i ′ ∂2 (u̅i ϵ̅L ) ∂u′i ′ ∂3 u′i ∂2 (u̅i 𝜖L′ ) ∂u′i ∂ νt ∂ε
Turbulent ̅L
∈ −2ν 〈 u〉 −ν 〈 ϵ 〉 −2ν 〈u′j 〉 (ϵ̅L )
∂xj ∂xk i ∂xj ∂xk ∂xk L ∂xj2 ∂xk ∂xk ∂xk ∂xk ∂xj σε ∂xj
Transport
Term 3.1 Term 3.2 Term 3.3 Term 3.4
∂u̅i ∂u′i ∂(u̅j 𝜖L′ ) ∂u′j ∂u′i ∂(u̅j 𝜖L′ )
2ν 〈 〉 2ν 〈 〉
∂xk ∂xk ∂xj ∂xk ∂xk ∂xj
Term 3.5 Term 3.6
Rate of ∂u′i 2
∂ (u̅i u̅j ) ∂u′i ∂u′j ∂(u̅i ϵ̅L ) ∂u′i ′ ∂2 (u̅i ϵ̅L ) ∂u′i ∂u′i u′j ∂ϵ̅L ε ̅ L u̅i
∂∈
Production −2ν 〈 ϵ′L 〉 −2ν 〈 〉 −2ν 〈 u〉 −2ν 〈 〉 ϵ̅L C1ε τij
∂xk ∂xj ∂xk ∂xk ∂xk ∂xk ∂xk j ∂xj ∂xk ∂xk ∂xk ∂xk 𝑘 ∂xj
Term 4.1 Term 4.2 Term 4.3 ∂u′i ∂u′i u′j ∂2 ϵ̅L
− 2ν 〈 〉
∂xk ∂xk ∂xk ∂xk
Term 4.4
Rate of ∂(u̅i ϵ̅L ) ∂u′i ∂u′i ∂u′j ∂2 u′i ∂u′i ∂u′i ∂ ∂2 (u̅j 𝜖L′ ) ∂u′i ε2
Dissipation − 〈 〉 −2ν 〈 ′
(u̅i 𝜖L )〉 −2ν 〈 (u̅j 𝜖L′ )〉 −2ν 〈u′j 〉 −ϵ̅L C2ε
∂xk ∂xk ∂xj ∂xk ∂xk ∂xj ∂xk ∂xj ∂xk ∂xk ∂xk ∂xk 𝑘
Term 5.1 Term 5.2 Term 5.3 Term 5.4
Source Sε Term 6.1 Sε
Term
The overall modelled equation for turbulent kinetic energy dissipation rate is given by:
∂ ∂ ∂ε ∂ νt ∂ε ε ̅L u̅i
∂∈ ε2
(ϵ̅L εui ) = [ϵ̅L ν ] + [ϵ̅L ] + ϵ̅L Cε1 τij − ϵ̅L Cε2 + Sε
∂xi ∂xj ∂xj ∂xj σε ∂xj 𝑘 ∂xj 𝑘
171 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
where, Ctd (=0.45) is the constant for source term in Eq. (8.19) arising from bubble induced
turbulence. τp is the characteristic time of the induced turbulence and is given by:
2CVM dB
τp = (8.20)
3CD |v̅i − u̅i |
Where, CVM and CD are virtual mass coefficient and drag coefficient, respectively.
Table 8.3 represents the governing equation for the turbulent energy dissipation rate. The production
and dissipation terms for are proportional to the production and dissipation terms of k. The higher
order fluctuating terms arising due to the gas hold up are neglected in the RANS approach. The
transport of turbulent kinetic energy and the dissipation rate of turbulent kinetic energy is primarily
modeled as a diffusive process that is a spatial derivative of order two. The analytic solution of this
transport process is Gaussian in nature, having the effect of smoothing fluctuations and sharp spikes.
Correctly handling the transport term so as to prevent this smoothing effect can be attributed to the
inclusion of the higher order fluctuating terms, which increases the computational requirement.
The transport equation for Reynolds stresses is derived by multiplying u′j with the momentum
equation for ith velocity fluctuating component (u′i ) and multiplying u′i with the momentum equation
for jth velocity fluctuating component (u′j ). Addition of these equations will lead to the conservation
172 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
̅̅̅̅̅
∂u ′ u′
i j ∂ϵ̅L ∂ϵ̅L ∂ϵ′L u′i ∂ ∂u′i ′ ∂ ∂u′i
′ ′
2(ϵ̅L u̅k ) − u̅i τij − u̅j τik −〈 u̅k ui 〉 + (ϵ̅ ν 〈 u 〉) + 〈uj (ϵ̅L ν )〉
⏟ ∂xk ⏟ ∂xk ⏟ ∂xk ⏟∂xk ∂xk L ∂xk j
⏟ ⏟ ∂xk ∂xk
Term 1.1 Term 1.2 Term 1.3 Term 1.4 Term 2.1 Term 2.2
′
∂ ∂u̅j ′
∂ ∂u′j ∂τik ∂τjk
〈u
− i ′
(ϵL ν 〈u
)〉 + i ′
(ϵL ν )〉 − ν 〈u′j ϵ̅L 〉 − ν 〈u′i ϵ̅L 〉
⏟ ∂xk ∂xk ⏟ ∂xk ∂xk ⏟ ∂xk ⏟ ∂xk
Term 2.3 Term 2.4 Term 3.1 Term 3.2
P ′ ∂u′i ∂u′j 2
− ϵ′L (
〈 + 〉) − ϵ̅L εδij (8.21)
⏟ ρ ∂xj ∂xi ⏟
3
Term 5.2 Term 6.1
In Eq. (8.21), terms 1.1, 1.2, 1.3 and 1.4 correspond to rate of convective transport. Terms 2.1, 2.2,
2.3 and 2.4 correspond to the rate of diffusive transport. Terms 3.1, 3.2, 3.3 correspond to turbulent
transport of Reynolds stresses. Terms 4.1, 4.2 and 4.3 correspond to the rate of production of
Reynolds stresses. Terms 5.1 and 5.2 represent pressure strain terms. Term 6.1 represents the rate of
dissipation of Reynolds stresses. The above mentioned classification of different terms in equation
is shown in Table 8.4. The last column shows the corresponding modelled terms in a stress equation
̅̅̅̅̅
∂u ′ u′
i j ∂ ϵ̅L k ′ ′ ∂u ̅̅̅̅̅
′ u′
i j ∂(ϵ̅L u̅j ) ∂(ϵ̅L u̅i ) ϵ̅L ε ′ ′
(ϵ̅L u̅i ) = CS ( ̅̅̅̅̅̅̅
uk um ) − ̅̅̅̅̅̅̅
u′i u′m − ̅̅̅̅̅̅̅
u′j u′m CR1 ̅̅̅̅̅
u u
∂xi ∂xj ε ∂xk ∂xm ∂xm k i j
The pressure strain term in Eq. (8.22) is classified as: (i) interactions of fluctuating components
[terms 3 and 4 on RHS of Eq. (8.22)], and (ii) interaction of mean strain and fluctuating quantity
173 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Table 8.4. Budget for the turbulent kinetic energy (k) in Reynolds Stress Model
Terms arising out of derivation of k equation [Eq. (8.21)] from the equation of motion and continuity Modelled Terms in RSM
for two-phase flow
Rate of ∂u ̅̅̅̅̅
′ u′
i j ∂ϵ ̅L ∂ϵ̅L ∂ϵ′L u′i ̅̅̅̅̅
∂u ′ u′
i j
Convective 2(ϵ̅L u̅k ) −u̅ τ
i ij −u ̅ τ
j ik − 〈 u̅k u′i 〉 (ϵ̅L u̅k )
∂xk ∂x k ∂x k ∂xk ∂xk
Transport Term 1.1 Term 1.2 Term 1.3 Term 1.4
Rate of ∂ ∂u′i ′ ∂ ∂u′i ∂ ∂u̅j ∂ ∂u′j
Viscous (ϵ̅L ν 〈 uj 〉) 〈u′j (ϵ̅L ν )〉 − 〈u′i (ϵ′L ν )〉 〈u′i (ϵ′L ν )〉
∂xk ∂xk ∂xk ∂xk ∂xk ∂xk ∂xk ∂xk
Transport Term 2.1 Term 2.2 Term 2.3 Term 2.4
∂τ ∂τ ′ ′ ′ ̅̅̅̅̅
′ u′
Rate of ik jk ∂ϵ̅L ui uj uk ∂ ϵ̅L k ′ ′ ∂u i j
Turbulent −𝜈 〈u′j ϵ̅L 〉 −𝜈 〈u′i ϵ̅L 〉 −2𝜈 〈 〉 CS ( ̅̅̅̅̅̅
uk u1 )
∂xk ∂xk ∂xk ∂xj ε ∂xk
Transport Term 3.1 Term 3.2 Term 3.3 [CS = 0.20-0.25]
Rate of ∂(ϵ̅L u̅k ) ∂(u̅i ) ∂(u̅j ) ∂(ϵ̅L u̅k )
Production −2u̅̅̅̅̅
′ u′
i j −ϵ̅L̅̅̅̅̅̅
u′k u′j −ϵ ̅
L
̅̅̅̅̅̅
u ′ ′
ui
̅̅̅̅̅
−u ′ u′
i j
∂xj ∂xk k
∂xk ∂xj
Term 4.1 Term 4.2 Term 4.3
′ ′ ′ ′ ′ ′
Pressure P ∂ui ∂u j P ∂ui ∂u j ∂(ϵ̅L u̅i ) ϵ̅L ε ′ ′
−ϵ̅L ( 〈 + 〉) −𝜖L′ ( 〈 + 〉) ̅̅̅̅̅̅
−u ′ u′
j 1 CR1 ̅̅̅̅̅
uu
Strain Term ρ ∂xj ∂xi ρ ∂xj ∂xi ∂x1 k i j
Term 5.1 Term 5.2 2 ′ ′ ∂(ϵ̅L u̅i ) ϵ̅L ε
+ ̅̅̅̅̅̅
uj u1 CR1 ϵ̅ δ
3 ∂x1 k L ij
2
− CR1 ϵ̅L (Pij − Pδij )
3
Rate of 2 2
− ϵ̅L εδij (Term 6.1) − ϵ̅L εδij
Dissipation 3 3
The overall modelled equation for Reynolds stresses is given by:
̅̅̅̅̅
∂u ′ u′
i j ∂ ϵ̅L k ′ ′ ∂u̅̅̅̅̅
′ u′
i j ∂(ϵ̅L u̅j ) ∂(ϵ̅L u̅i ) ϵ̅L ε ′ ′ 2 ′ ′ ∂(ϵ̅L u̅i ) ϵ̅L ε 2
(ϵ̅L u̅k ) = CS ( ̅̅̅̅̅̅
uk u1 ) − ̅̅̅̅̅̅
u′i u1′ − ̅̅̅̅̅̅
u′j u1′ CR1 ui uj + ̅̅̅̅̅̅
̅̅̅̅̅ uj u1 CR1 ϵ̅L δij − CR1 ϵ̅L (Pij − Pδij )
∂xk ∂xj ε ∂xk ∂x1 ∂x1 k 3 ∂x1 k 3
2
− ϵ̅L εδij
3
174 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Along with Eq. (8.22), the conservation equation for is also solved, which is given by Eq. (8.18).
Table 8.3 represents the true terms arising out of derivation of equation (Section 8.4.2) from the
equation of motion and continuity for two-phase flow and the corresponding modelled terms for
RSM.
For the LES simulation, Eq. (8.2) was solved along with Smagorinsky model for
Where, μ is the molecular viscosity, CS is the Smagorinsky constant, Sij is the characteristic strain
tensor of filtered velocity and ∆ is the filter width (usually taken as the cubic root of the cell volume).
Further details pertaining to the LES model have been given by Tabib et al. (2008) and Dhotre et al.
(2013). The terms for equation for LES are given in Table 8.5.
Bubble column of 150 mm i.d. was simulated at three different superficial gas velocities (20, 40 and
100 mm/s). The perforated plate sparger was replicated by injection points at the base of the column.
An Eulerian-Eulerian approach was used. Water was taken as the continuous phase and air bubbles
of 5mm diameter was the dispersed phase. In the present work the drag model of Ishii and Zuber
(1979) has been employed. The choice of dB to be 5 mm is based on the slip velocity which was
found to be 330 mm/s by iterations in CFD simulations. A no-slip boundary condition was considered
for all the walls and the degassing boundary condition was used for the outlet that allowed selectively
only the gas phase to escape while the liquid recirculated back into the column. The standard k-
model [given by Eqs. (8.9) and (8.18)] and RSM model [given by Eq. (8.22) and (8.18)] were used
to simulate the bubble column depicting the RANS approach. For the LES simulation Eq. (8.2) was
solved along with Smagorinsky model [Eq. (8.22)]. The flow was developed on a coarse grid of 0.6
175 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Table 8.5. Budget for the kinetic energy dissipation rate () in LES model
Rate of ∂u′i ′ ∂2 (u̅i u̅j ) ∂u′i ∂u′j ∂(u̅i ϵ̅L ) 𝜕𝑢𝑖′ ′ 𝜕 2 (𝑢̅𝑖 𝜖̅𝐿 ) ∂u′i ∂u′i u′j ∂ϵ̅L
Production −2ν 〈 ϵ 〉 −2ν 〈 〉 −2𝜈 〈 𝑢〉 −2ν 〈 〉
∂xk L ∂xj ∂xk ∂xk ∂xk ∂xk 𝜕𝑥𝑘 𝑗 𝜕𝑥𝑗 𝜕𝑥𝑘 ∂xk ∂xk ∂xk
Term 4.1 Term 4.2 Term 4.3 Term 4.4
∂2 u′i ∂(u̅i ϵ̅L ) ∂2 u′i u′j ∂(u̅i ϵ′L ) ∂u′i ∂u′i u′j ∂2 ϵ̅L
−2ν 〈u′i u′j 〉 −2ν 〈u′i 〉 −2ν 〈 〉
∂xk ∂xj ∂xk ∂xk ∂xj ∂xk ∂xk ∂xk ∂xk ∂xk
Term 4.5 Term 4.6 Term 4.7
Rate of ∂(u̅i ϵ̅L ) ∂u′i ∂u′i ∂u′j ∂2 u′i ∂u′i ∂u′i ∂ ∂2 (u̅j 𝜖L′ ) ∂u′i
− 〈 〉 −2ν 〈 (u̅ 𝜖 ′ )〉 −2ν 〈 (u̅ 𝜖 ′ )〉 −2ν 〈u′j 〉
Dissipation ∂xk ∂xk ∂xj ∂xk ∂xk ∂xj i L ∂xk ∂xj ∂xk j L ∂xk ∂xk ∂xk
Term 5.1 Term 5.2 Term 5.3 Term 5.4
176 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
million and the results were interpolated on a much finer mesh with 3.1 million grid points with a
As per the recommendations of Tomiyama et al. (2002), the lift coefficient works out to be 0.025.
However, it may be noted that these authors used a model flow pattern for the recommendation of
lift coefficient. Tabib et al. (2008) have comprehensively analyzed the parametric sensitivity of lift
coefficient and have recommended a value of -0.05 for lift coefficient for the same geometry and VG
of the present work. The same value has been used in the present work. The Smagorinsky model
constant, (CS) was set at 0.5. Lopez de Bertodano et al. (1991) has suggested a value of 0.3 for the
simulation of the turbulent dispersion force and was selected for the present simulations. The Sato
Enhanced Eddy Viscosity model was used for the modeling the interaction of the bubble with its
surrounding liquid. It may be pointed out that the Sato and Sekoguchi (1975) model was used (given
by Eq. (8.6)) for the estimation of eddy viscosity which determines the mean flow pattern. It is known
that the turbulence is also generated because of the motion of bubbles and is called Bubble Induced
Turbulence (BIT). This subject has been comprehensively researched by several investigators during
the past 40 years. A critical review of this published work has been recently presented by Joshi et al.
(2016). Thus, it is clear that the BIT needs to appear in the conservation equation of turbulent kinetic
energy and is given by Eq. (8.8). CFX Expression Language (CEL) step functions were embedded
to define the initial conditions for the volume fraction, liquid height and bubble diameter. The initial
volume fraction was set within a range of 0 to 0.15 and the initial liquid height was set at 0.9m. For
convergence, a high resolution advection scheme and the second order backward Euler transient
scheme were employed. Convergence was achieved with a residual target set at 1x10 -6. A control
was set on the fluid timescale, where the solution for convergence was looped in a range of 2 to 10.
The solver relaxation factor was set at 0.7. ANSYS CFX 13 was used to simulate the flow dynamics
in the bubble column. Simulations were run on the SGI Altix XE 371 which comprised of 16 compute
nodes. Each node has eight cores; two Intel Xeon Quad core linked via L2 cache (8 MB per node)
177 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
supported by 8 GB of RAM. Data were extracted in the CFD general notation system format (.CGNS)
and stored in a binary format. MATLAB 8.0 (2012b) was used to read in the data and all the
calculations were carried out on a quad core Intel i7, 3.13GHz with a minimum requirement of 12GB.
The API incorrectly estimates outliers and NaNs in the process of conversion and transference. Hence
pretreatment of data is an important step for the removal of outliers and singleton values. An
interpolation scheme was devised based on a predictive intelligent learning algorithm that locates the
The instantaneous velocities, pressure and volume fraction were ensemble averaged on a time basis
(spatially statistically invariant) on all the node points. Table 8.2 to 8.5 lists second order, third order
correlation terms and these were also ensemble averaged temporally. The data are arranged three
dimensionally in a four dimensional array; the fourth dimension representing the physical quantity
and the first three representing the coordinate locations with respect to their positions in the column.
A gradient operator method was written to calculate the first order derivatives with a fourth order
accuracy of the given physical quantity in all its respective dimensions and a fourth order accurate,
The predictive nature of the turbulence models has been compared for a bubble column at superficial
gas velocity of 20 mm/s. The radial profiles of mean axial velocity, fractional gas hold-up (ϵ̅G ) and
turbulent kinetic energy (k) have been compared with the experimental results of Bhole et al. (2006)
and Kulkarni et al. (2007). Figure 8.2 shows the comparison of axial velocity for the sieve plate
sparger. The cross over point for the flow was observed to be at 0.7. At this stage, a material balance
was established between the upflow and downflow. In principle the two flow rates need to be
178 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
identical and the difference was found to be within 2.5 per cent. Figure 8.3 shows the comparison of
experimental values of gas hold-up and the CFD predictions. It can be seen that all the three models
favorably predict radial profiles of axial velocity and fractional gas hold-up. However, the predictive
capability of turbulence models becomes demanding for turbulence parameters. For instance, Figure
0.3
0.25 3
0.2
Axial Liquid Velocity, u (m/s)
2
0.15
0.1
1
0.05
0
0 0.2 0.4 0.6 0.8 1
-0.05
-0.1
-0.15
-0.2
Dimensionless Radial Distance, r/R
Figure 8.2. Comparison between the simulated and experimental profiles of axial liquid
velocity at axial position HD/D=4 with sieve plate sparger at VG = 20 mm/s; (Δ) Experimental
data [Bhole et al. (2006)]; (1) LES model; (2) RSM and (3) k–ε model.
The fluctuating components of three dimensional velocities, gas hold-up and pressure were obtained
from LES. The time series data has been ensemble averaged and the ensemble velocities, turbulence
179 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Figure 8.3. Comparison between the simulated and experimental profiles of gas holdup at
axial position HD/D=4 with sieve plate sparger at VG = 20 mm/s; (Δ) Experimental data [Bhole
et al. (2006)]; (1) LES model; (2) RSM and (3) k–ε model.
intensities and turbulent stresses have been calculated. Also double and triple correlations of
velocity, volume fraction and stresses have been calculated. Further, as per Table 8.2, convection
term, viscous transport term, turbulent transport term, turbulent energy dissipation term and the
production of turbulent kinetic energy term have been estimated (as per correlations given in Table
8.2) using the extracted velocities from each model. The volumetric conservation of the energy across
each of the cells has been calculated. The results are shown in Figures 8.5 to 8.7. Also shown is the
residual (subplot F in all the Figures) of the conservation of energy, [Eq. (8.3)] for all the models.
The LES model satisfies the energy balance. For the LES model the residual can be seen to be zero,
180 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Figure 8.4. Comparison between the simulated and experimental profiles of turbulent kinetic
energy at axial position HD/D=4 with sieve plate sparger at VG = 20 mm/s; (Δ) Experimental
data [Bhole et al. (2006)]; (1) LES model; (2) RSM and (3) k–ε model.
Figures 8.5A, 8.6A and 8.7A represent the rate of convective transport of k at 20, 40 and 100 mm/s,
respectively. From Figure 8.5A it was observed that the rate of convective transport predicted by
using standard k- model (line 3) showed deviation (with the SD of 141%) from the LES predictions
(line 1). In case of RSM (line 2), though the rate of convective transport predicted is better than that
of the standard k- model, it still shows deviation (with the SD of 102%) from LES results. Terms
1.1 to 1.4 in Table 8.2, on addition, contributes to the net rate of convective transport of k. the
resulting sum is named as convection by true k-equation [conservation equation for k (Eq. (8.3))
without any modelling assumptions). In Figure 8.5A, this term is shown by line 4. When plotted
against the convection of k by using LES (line 1), it was found to be in good agreement with LES.
From Table 8.2 it can be seen that the convective transport term in standard k- model is same as that
181 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
of term 1.1. In the standard k- model, terms 1.2, 1.3 and 1.4 are neglected. But the approach for true
k- model shows that these terms (terms 1.2, 1.3 and 1.4) also effectively contribute to the net rate of
convective transport in order to achieve the accuracy of the order of LES. Thus, it is clear that the
deviation in line 3 is mainly because of modelling assumptions, which can be excluded by using true
k- model. Similarly, in case of RSM terms 1.1 to 1.4 in Table 8.4 were added to get the net rate of
convective transport of k by RSM. The resulting term is called as convection by true-RSM [Eq.
(8.21)], which is represented by line 5 in Figure 8.5A. It also shows good agreement with that of the
LES data. Similar procedure was followed for the calculation of the rate of convective transport by
using true k-equation and true RSM at the superficial gas velocities of 40 mm/s and 100 mm/s and
results were plotted in Figure 8.6A and 8.7A, respectively. In all the cases, the rate of convection was
found to be maximum in the region between flow reversal point and the wall, where mean and the
fluctuating velocity gradients are dominating. At this stage, we emphasize one important point. The
lines 4 (true k- model) and 5 (true RSM) in all the plots show an excellent agreement (within 3
percent) with line 1 for LES. This observation validates the procedure for (a) constructing Tables 8.2
The rate of viscous transport is shown in Figures 8.5B, 8.6B and 8.7B at the superficial gas velocities
of 20, 40 and 100 mm/s, respectively. Line 1 to 3 shows the results for diffusive transport of k by
using LES model, RSM and standard k- model, respectively. Lines 2 and 3 show deviation (SD of
72.5% and 132% for standard k- model and RSM, respectively) from LES. The rate of diffusion of
k using true k- model was obtained by adding terms 2.1 and 2.2 in Table 8.2. The resulting term was
plotted in Figure 8.5B as line 4. Similar to line 4, line 5 was also plotted as rate of diffusive transport
arising out of true RSM. True RSM diffusive term is obtained by adding terms 2.1, 2.2, 2.3 and 2.4
in Table 8.4. Lines 4 and 5 show good agreement with the LES results (line 1). In this case also
(similar to convective transport) the contribution of term 2.1 in Table 8.2 to get net rate of diffusive
transport is very important. Similar procedure was followed for the calculation of the rate of diffusion
182 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
0.2 0.01
0.005 2
Rate of convective transport of k
-0.02 2
0.16
Rate of production of k
C 5
0.14
-0.04
0.12
-0.06 5 0.1 1
4 0.08 4
-0.08 1
0.06
0.04
-0.1 3 0.02
2
0
-0.12
0 0.2 0.4 0.6 0.8 1
0 6.0E-06
0 0.2 0.4 0.6 0.8 1
-0.01
4.0E-06 F
E 2.0E-06
Rate of dissipation of k
-0.02
0.0E+00
-0.03
Residual
2 -2.0E-06
4 5 1
-0.04 -4.0E-06 1
3 2
-6.0E-06
-0.05
-8.0E-06
-0.06 -1.0E-05
3
-0.07 -1.2E-05
0 0.2 0.4 0.6 0.8 1
Dimensionless Radial Distance, r/R
Figure 8.5. Turbulent kinetic energy (k) budget in a bubble column at axial position HD/D=4
with sieve plate sparger at VG = 20 mm/s (1) LES model (2) RSM (3) k–ε model (4) True k–ε
model and (5) True RSM; (A) Rate of convective transport (B) Rate of viscous transport (C)
Rate of turbulent transport (D) Rate of production (E) Rate of dissipation and (F) Residual.
183 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
0.35 0.02
2
Rate of convective transport of k
0.3 3 B
A
1 0.35 D 2
0.05 Rate of production of k
0.3 5
0
-0.05 0 0.2 0.4 0.6 0.8 4 1 0.25
0.2 1
-0.1
-0.15 4
0.15
-0.2 3 0.1
-0.25 C
2 0.05
-0.3
-0.35 0
0 0.2 0.4 0.6 0.8 1
0 4.E-06
0 0.2 0.4 0.6 0.8 1 F
-0.02 3.E-06
Rate of dissipation of k
-0.04 E 2.E-06
1 2
-0.06 1 3
1.E-06
Residual
-0.08 4 5 0.E+00
-0.1 -1.E-06
-0.12 -2.E-06
-0.14
-3.E-06
-0.16 3
2 -4.E-06
-0.18 0 0.2 0.4 0.6 0.8 1
Dimensionless Radial Distance, r/R
Figure 8.6. Turbulent kinetic energy (k) budget in a bubble column at axial position HD/D=4
with sieve plate sparger at VG = 40 mm/s (1) LES model (2) RSM (3) k–ε model (4) True k–ε
model and (5) True RSM; (A) Rate of convective transport (B) Rate of viscous transport (C)
Rate of turbulent transport (D) Rate of production (E) Rate of dissipation and (F) Residual.
184 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
1 0.05
1
Rate of convective transport of k
0.8 A
0.4 1
2 3
5 0.9
Rate of turbulent transport of k
0.2 D 1
0.8
Rate of production of k
0 0.7
5
0 0.2 0.4 0.6 0.8 1 0.6
-0.2 1
0.5
-0.4 C 0.4
4 4
0.3
-0.6 0.2
3
-0.8 0.1
2
0
-1 0 0.2 0.4 0.6 0.8 1
0 5.E-06
0 0.2 0.4 0.6 0.8 1
-0.05
Rate of dissipation of k
-0.1 E 0.E+00
-0.15 1
Residual
5 -5.E-06
-0.2 4 2
-0.25
-1.E-05
3
-0.3
-0.35 2 F 1
-2.E-05
-0.4 3
-0.45 -2.E-05
0 0.2 0.4 0.6 0.8 1
Dimensionless Radial Distance, r/R
Figure 8.7. Turbulent kinetic energy (k) budget in a bubble column at axial position HD/D=4
with sieve plate sparger at VG = 100 mm/s (1) LES model (2) RSM (3) k–ε model (4) True k–ε
model and (5) True RSM; (A) Rate of convective transport (B) Rate of viscous transport (C)
Rate of turbulent transport (D) Rate of production (E) Rate of dissipation and (F) Residual.
185 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
by using true k-equation and true RSM at the superficial gas velocities of 40 mm/s and 100 mm/s
Figures 8.5C, 8.6C and 8.7C represent the rate of turbulent transport of k at the superficial gas
velocities of 20, 40 and 100 mm/s, respectively. In Figure 8.5C, lines 1, 2 and 3 represent the rate of
turbulent transport calculated using LES model, RSM and standard k- model. Both RSM and
standard k- model show deviation (with SD of 2% and 4%, respectively) from LES results. On the
other hand, true k- model (line 4 obtained by adding terms 3.1, 3.2, 3.3 and 3.4 in Table 8.2) and
true RSM (line 5 obtained by adding terms 4.1, 4.2 and 4.3) show good agreement with the LES
results. The turbulent transport (line 3) is calculated by using term given in the last column of Table
8.2, which neglects the terms 3.2, 3.3 and 3.4 and considers the modelling of term 3.1, according to
∂〈u′j k〉 ∂ ν ∂k
̅L
∈ = ̅L t
(∈ ) (8.24)
∂xj ∂xj σk ∂xj
This results into an inaccurate prediction of turbulent transport (line 3 in Figure 8.5C) as compared
with LES predictions (line 1). Again we point out that the accurate predictions need the inclusion of
neglected terms, which is possible by using true k- model (line 4) and true RSM (line 5). Thus the
assumption of Boussinesq’s approximation can be excluded in order to get the accuracy of LES
The rate of production of k at the superficial gas velocity of 20 mm/s are shown in Figures 8.5D.
Lines 1, 2 and 3 in Figure 8.5D represent production by LES model, RSM and standard k- model,
respectively. Deviation is observed between RSM and LES as well as standard k- model and LES.
This is mainly because of the modelling assumptions. While modelling rate of production of k, the
terms corresponding to pressure strain (terms 4.1 and 4.4) as well as 4.3 are neglected in the standard
186 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
∂u̅i ̅L u̅i
∂∈
̅L 〈u′i u′j 〉
∈ = τij (8.25)
∂xj ∂xj
The consequence of all the assumptions is the deviation (SD as mentioned in Table 8.6) in predictions
of standard k- model (line 3 in Figure 8.5D) and LES (line 1). Above mentioned assumptions were
made in deriving RSM also and hence the deviation (SD as mentioned in Table 8.6) is observed in
RSM and LES model. All the above assumptions were eliminated in true k- model (line 4) and true
RSM (line 5) and were plotted in Figure 8.5D. Both the models show very good agreement with the
LES results. Similar procedure was followed for the calculation of the rate of production by using
true k-equation and true RSM at the superficial gas velocities of 40 mm/s and 100 mm/s and results
Figure 8.5E represents the rate of dissipation of k at the superficial gas velocity of 20 mm/s. It is
observed that standard k- model (line 3) shows much more deviation (SD of 31% as mentioned in
Table 8.6) from LES results (line 1). As compared to standard k- model, RSM (line 2) shows better
agreement with the LES with the SD of 16%, as shown in Table 8.6. The discrepancy arises in the
result of standard k- model in spite of the fact that only one term exists under dissipation. However,
no assumptions are made in this case and term 5.1 is exactly the same as that in the last column. This
discrepancy, however, is just apparent because the k equation in k- model also needs to be conserved
as per equation (4). True k- model (line 4) and true RSM (line 5) also shows very good agreement
with the LES. Thus, from the above discussion it is clear that we can get the accuracy of LES model
by properly deriving the k- model [Eqs. (8.9) and (8.18)] and RSM [Eq. (8.22)] in the form of true
k- model and true RSM. It is possible to get the desired accuracy with small computational demands
by using the new approach of writing true equations. In all the cases, the residuals were found to be
of the order of 10-6 as shown in Figures 8.5F, 8.6F and 8.7F. The residuals for LES model (line 1),
RSM (line 2) and standard k- model (line 3) almost of the same order. This is because all the
187 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Figures 8.8, 8.9 and 8.10 represents the budget for the energy dissipation rate at the superficial gas
velocities of 20, 40 and 100 mm/s, respectively. Figure 8.8A shows the rate of convective transport
of at 20 mm/s. The predictions of standard k- model (line 3) and RSM (line 2) show deviations
(SD of 112% and 72%, respectively) from the LES predictions (line 1) in the central region of the
column. Near the wall, the predictions for both the models are comparatively in good agreement with
the LES model. True k- model and true RSM convective transport term can be obtained by adding
the terms 1.1, 1.2 and 1.3 in Table 8.3. Both the models, true k- model (line 4) and true RSM (line
5), show good agreement with LES predictions. Similar results are plotted in Figures 8.9A and 8.10A
at the superficial gas velocities of 40 and 100 mm/s, respectively. Figures 8.8B, 8.9B and 8.10B show
the comparison of the rate of diffusive transport of at the superficial gas velocities of 20, 40 and
∂〈u′j ε〉
̅L
∈ (8.26)
∂xj
νt ∂ε
〈u′j ε〉 = − (8.27)
σε ∂xj
∂〈u′j ε〉 ∂ ν ∂ε
̅L
∈ =− ̅L t
(∈ ) (8.28)
∂xj ∂xj σε ∂xj
188 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Which is modelled turbulent transport term in the last column. Thus, the contribution of terms 3.2 to
3.6 is neglected in standard k- model and hence results into difference (SD as shown in Table 8.6)
in the predictions of LES (line 1) and standard k- (line 3) in Figure 8.8C. Similar trend is found
between the predictions of RSM (line 2) and LES model. When all the above modeling assumptions
are excluded, which is indication of the true k- model (line 4) and true RSM (line 5), the models
show very good agreement with the LES predictions. The results are plotted for the superficial gas
In case of the rate of production of the term 4.2 in production is simplified as:
u′2 ̅ i ϵ̅L )
i ∂(u
4ν 2
𝑙 ∂xk
u′3 ′
̅ i ϵ̅L )
i 1 ui ∂(u
= 4ν
𝑙 u′2
i 𝑙 ∂xk
ε ∂(u̅i ϵ̅L )
= Cε1 ν (8.31)
k ∂xk
189 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
Eq. (8.31) is modelled production term in the last column. The terms 4.1 and 4.2 are thus neglected.
In the above simplification, one Boussinesq’s approximation has been employed as follows:
∂u̅i
u′i = 𝑙 (8.32)
∂xj
Figure 8.8D shows that all the four terms have significant values (true k- model). These four lines
add up to contribute to the net rate of production of (lines 4 and 5) which excellently agrees with
LES simulations as represented by line 1. However, the k- predictions (line 3) deviates (SD given
in Table 8.6) from LES simulations because of several simplifications explained above. The rate of
production of at superficial gas velocities of 40 and 100 mm/s is shown in Figures 8.9D and 8.10D,
respectively.
The rate of dissipation of using standard k- model (line 3) and RSM (line 2) was found to deviate
from LES prediction (line 1) at lower superficial gas velocity (VG = 20 mm/s) as shown in Figure
8.8E. At higher superficial gas velocities (VG = 40 and 100 mm/s), the deviation was found to be
reduced and all of them show the predictions in the range of 3% as shown in Figures 8.9E and 8.10E.
As expected, true k- model (line 4) and true RSM (line 5) show very good agreement with the LES
Another important simplification in the k- model is the assumption of isotropy which has already
been used while getting Eqs. (8.29), (8.30) and (8.31). The isotropy is also assumed while deriving
k2
νt = Cμ (8.33)
ε
Another set of great simplifications is the methodology of the estimation of standard turbulence
parameters Cε1 , Cε2 and Cμ . While assigning the respective values of 1.44, 1.92 and 0.009, we
consider simple model flows, the details of which are known to CFD community.
190 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
0.004 0.00016
3 3
Rate of convective transport of ε
0.002 2 0.00014
C 0.03 3
0.015 D
Rate of production of ε
0.025
0.01 5
4 3 0.02
2
0.005
0.015 4
0 0.01 1
0 0.2 0.4 0.6 0.8 1
-0.005 0.005
5 1
-0.01 0
0 0.2 0.4 0.6 0.8 1
0 3.E-06
-0.001 0 0.2 0.4 0.6 0.8 1
3.E-06 F
-0.002 3 1
Rate of dissipation of ε
2.E-06 3
-0.003
Residual
-0.004 2.E-06
2
-0.005 1.E-06
-0.006 5
E 5.E-07
-0.007
1 0.E+00 2
-0.008 4
-0.009 -5.E-07
0 0.2 0.4 0.6 0.8 1
Dimensionless Radial Distance, r/R
Figure 8.8. Turbulent energy dissipation rate (ε) budget in a bubble column at axial
position HD/D=4 with sieve plate sparger at VG = 20 mm/s (1) LES model (2) RSM (3) k–ε
model (4) True k–ε model and (5) True RSM; (A) Rate of convective transport (B) Rate of
viscous transport (C) Rate of turbulent transport (D) Rate of production (E) Rate of
dissipation and (F) Residual.
191 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
0.25 0.0015
3
0.2 2
Rate of convective transport of ε
A B
0.5 0.6
3 3
0.4
Rate of turbulent transport of ε
-0.2 4.E-05
Residual
-0.3 2.E-05
2
-0.4 4 0.E+00
2
-0.5 E 5 -2.E-05
1
-0.6 -4.E-05
1 3
-0.7 -6.E-05
0 0.2 0.4 0.6 0.8 1
Dimensionless Radial Distance, r/R
Figure 8.9. Turbulent energy dissipation rate (ε) budget in a bubble column at axial
position HD/D=4 with sieve plate sparger at VG = 40 mm/s (1) LES model (2) RSM (3) k–ε
model (4) True k–ε model and (5) True RSM; (A) Rate of convective transport (B) Rate of
viscous transport (C) Rate of turbulent transport (D) Rate of production (E) Rate of
dissipation and (F) Residual.
192 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
0.2 0.025
3
Rate of convective transport of ε
0 B
2.5 2.5
3
Rate of turbulent transport of ε
C 3 D
2 2
2 Rate of production of ε
2
1.5 1
1.5
1 4
5 1 4
0.5 5
0.5 1
0
0 0.2 0.4 0.6 0.8 1
-0.5 0
0 0.2 0.4 0.6 0.8 1
0 2.0E-04
0 0.2 0.4 0.6 0.8 1 1.5E-04
-0.5
-1 1.0E-04 2
Rate of dissipation of ε
-1.5 5.0E-05
4 5
Residual
0.0E+00
-2 3
1 -5.0E-05
-2.5 2
-1.0E-04 3
-3 1 F
-1.5E-04
-3.5 E
-2.0E-04
-4 -2.5E-04
-4.5 -3.0E-04
0 0.2 0.4 0.6 0.8 1
Dimensionless Radial Distance, r/R
Figure 8.10. Turbulent energy dissipation rate (ε) budget in a bubble column at axial
position HD/D=4 with sieve plate sparger at VG = 100 mm/s (1) LES model (2) RSM (3) k–ε
model (4) True k–ε model and (5) True RSM; (A) Rate of convective transport (B) Rate of
viscous transport (C) Rate of turbulent transport (D) Rate of production (E) Rate of
dissipation and (F) Residual.
193 | P a g e
Chapter 8: Development of new turbulence modeling approach for bubble columns
8.8. Conclusion
1. Using Reynolds averaging procedure, the k and equations have been derived from the
equations of continuity and motion. The resulting equations have been termed as true k and
true equations. The resulting equations involve double and triple correlations of fluctuating
u′i , ϵ′G , ϵ′L , P ′ and their gradients. In Tables 8.2 and 8.3, these are classified into rates of
convective transport, diffusive transport and turbulent transport, rate of production and rate
of dissipation [as per Eq. (8.4)]. Similar exercise was performed for RSM and the results are
2. Procedure has been developed for the estimation of all the terms in Tables 8.2 to 8.4.
3. One important conclusion of this paper is the excellent agreement between the LES
predictions and those of true k- model and true RSM. This observation validates the
procedure for constructing Tables 8.2 to 8.4 (in fact true k, and RSM equations) and
4. The predictions of standard k- model and RSM have been compared with those of LES.
Substantial deviations occur in all these cases. An attempt has been made to explain these
5. The present work is concerned with understanding the relationship between the flow patterns
and mixing/stratification. For better understanding of transport phenomena, the future work
should include the identification and dynamics of flow structures [Thorat et al. (1998),
Kulkarni et al. (2001c), Bhole et al. (2008), Joshi et al. (2009), Mathpati et al. (2009), Ranade
et al. (1989), Ranade and Joshi (1990), Nere et al. (2003) and Kumaresan and Joshi (2006)].
194 | P a g e