Luo 2011

Chemical Engineering Science 66 (2011) 907–923
Contents lists available at ScienceDirect
Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces
Verification and validation of CFD simulations for local flow dynamics

in a draft tube airlift bioreactor
Hu-Ping Luo n, Muthanna H. Al-Dahhan 1
Chemical Reaction Engineering Laboratory (CREL), Department of Energy, Environmental and Chemical Engineering, Washington University, One Brookings Drive,
Campus Box 1198, St. Louis, MO 63130-4899, USA
a r t i c l e in f o abstract
Article history: Airlift reactors have been recognized as one of the promising photobioreactors for biomass/bio-energy
Received 20 July 2010 production, where mixing has significant impact on the reactor performance. In recent years, using CFD
Received in revised form simulations to track microorganism cells and to generate their trajectories in the reactor for reactor
22 November 2010
performance evaluations becomes more common. However, there is a lack of systematic and rigorous
Accepted 23 November 2010
verifications and validations of the reliability of CFD models in particle tracking against experimental
Available online 1 December 2010
measurements in the open literature, which is vital for the faithful application of CFD in reactor design and
Keywords: scale-ups. In this work, we attempt to evaluate the reliability of using CFD simulations to generate
CFD trajectories of microorganisms in a draft tube column photobioreactor. A computationally promising CFD
Airlift column
simulation model based on CFX5.7 was validated against a benchmark experimental database reported in
Photobioreactor
Luo and Al-Dahhan (2008a, b, 2010). This model was then used to generate typical trajectories of
Multiphase
Flow dynamics microorganisms in the studied airlift column, which was further validated against experimentally
Particle tracking measured tracer trajectories. The results indicated that the CFD model reasonably predicted the
recirculation of the microorganism around the draft tube, however over-estimated the cells’ residence
time in the wall regions. Proper treatment for the wall region such as griding and wall function is needed
to better capture the movement of microorganism cells in such bioreactors.
& 2010 Elsevier Ltd. All rights reserved.
1. Introduction following the Lambert–Beer law, the light-exposure history of a cell

is eventually determined by the movement or the trajectory of this
Airlift column reactors are pneumatically mixed reactors that have cell inside the reactor. Combining the cell’s trajectory with other
been widely used in chemical, petrochemical, and bioprocess indus- physiological information about algae cell’s growth thus can be
tries, such as fermentation (Pollard et al., 1998) and wastewater used to predict or evaluate the reactor performance of a photo-
treatment (Heijnen et al., 1990). Recently, such kind of reactors was bioreactor as many researchers have done (Wu and Merchuk, 2001;
also suggested as a promising photobioreactors for microalgae and Luo and Al-Dahhan, 2004; Pruvost et al., 2008).
cyanobacteria cultures (Merchuk et al., 2000; Garcia Camacho et al., Determining the representative trajectory of an algae cell in the
1999; Luo and Al-Dahhan, 2004). Due to their good mixing intensity reactor, nevertheless, is not trivial. For example, due to the lack of a
that can be achieved, airlift column reactors were recognized of real trajectory, Wu and Merchuk (2001) utilized the multi-circula-
having the potential to significantly enhance the photosynthetic tion model developed by Joshi and Sharma (1979) to calculate a
microorganisms’ abilities in using light energy more efficiently, and conceptual trajectory without knowing its representativeness. On
thus improve the overall performance of the culturing system the other hand, Luo and Al-Dahhan (2004) measured a real
(Sanchez Miron et al., 1999; Janssen, 2002; Luo and Al-Dahhan, 2004). trajectory employing a radioactive particle tracking technology
In a photobioreactor, how light energy is delivered to a micro- (i.e., CARPT or Computer Automated Radioactive Particle Tracking),
algae cell plays a significant role in determining how effectively which however is not widely available. As a result, there is a trend
such energy can be converted into biomass. Since the light intensity in recent years to use CFD (computational fluid dynamics) simula-
always reduces exponentially from the wall to the reactor center tions to generate such trajectories (Perner-Nochta and Posten,
2007; Pruvost et al., 2008). Although such an approach is highly
desirable with low cost and many other benefits, it is also well
n
Corresponding author. Currently at Chevron Corporation, USA. known that CFD simulations need to be validated before applied
Tel.: + 1 510 242 1659; fax: + 1 510 242 2823. faithfully, particularly for multiphase applications where physics is
E-mail addresses: [email protected] (H.-P. Luo), [email protected]
still not fully understood (Jacobsen et al., 1997).
(M.H. Al-Dahhan).
1
Currently at Missouri University of Science and Technology, USA. Unfortunately, although many commercially available CFD
Tel.: +1 573 341 4416. codes, such as Fluent, CFX, Star-CD, etc. have included the particle
0009-2509/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2010.11.038
908 H.-P. Luo, M.H. Al-Dahhan / Chemical Engineering Science 66 (2011) 907–923
tracking capability in their codes in recent years, there is a lack of significantly, the computational cost is also considerably higher,
vigorous verification and validation in the open literature to ensure which has also limited its usage (Chen et al., 2005; Sanyal et al.,
that the numerically obtained particle trajectories are truly repre- 2005). Moreover, it is generally considered that the population
sentative for airlift column applications. Some limited attempts on balance model better suit for the churn-turbulent flow regime,
this front were also relied on indirect approaches. These studies, where gas bubbles interact with each other intimately with sig-
usually based on the Euler–Lagrangian model, compare the bulk nificant bubble break-up and coalesce. For bubbly flow regime as for
gas–liquid flow parameters, such as velocity, gas holdup profiles, most bioprocesses where gas bubbles are usually more uniformly
and mixing time, etc., against experimental data to partially verify distributed, the effects of the population balance model are limited
the models used to track gas bubbles that drove the liquid flows in (Chen et al., 2005; Sanyal et al., 2005).
an airlift column (Anderson and Jackson, 1967; Delnoij et al., 1997a, Therefore, the so-called two-fluid model based on the Eulerian–
b, c, 1999; Lain et al., 1999; Pfleger et al., 1999; Sokolichin et al., Eulerian approach without a population balance model is widely
2004). used in the literature and has been found to be adequate in studying
In this work, we attempt to systematically evaluate the relia- the multiphase flow dynamics in airlift column at bubbly flow
bility of using CFD simulations to generate trajectories of micro- regime. However, although the Eulerian–Eulerian approach is
organisms in a draft tube column photobioreactor. We will start relatively simple, in practice, further simplifying the complete
from forming a multiphase CFD model and selecting its correlations set of the Navier–Stokes equations is still inevitable due to the
and closures by comparing the simulation results against a bench- complexity of the multiphase flow. In these simplifications, the
mark experimental database reported in Luo and Al-Dahhan modeling of the multiphase turbulence and the interfacial momen-
(2008a, b, 2010), where local fluid dynamics of a draft tube column tum transfer are most commonly used. While these simplifications
bioreactor was studied comprehensively using advanced experi- greatly reduced the computational costs, they introduced uncer-
mental techniques: CARPT and CT (Computed Tomography). The tainty and inaccuracy (Sokolichin et al., 2004; Tabib et al., 2008).
validated CFD model will be then used to generate trajectories of For example, the standard k–e turbulence model, the most often
tracers injected in the reactor, which mimics the trajectories of used and extensively studied turbulence model in the literature, is
microorganisms in a bioreactor. Finally, these CFD generated well-known to over-estimates the eddy viscosities that damps out
trajectories will be further verified against the CARPT measured small-scale turbulence (Borchers et al., 1999). Although modifica-
ones by statistic approaches. tions of the k–e model (e.g., the Re-Normalization Group (RNG)
model) and alternative turbulence models (e.g., the Reynolds shear
stress model, large eddy simulations (LES)) have been reported in
2. Computational Fluid Dynamics (CFD) modeling the literature, their applications have been limited in specific cases.
Similarly, most correlations proposed in the literature to model
Computational Fluid Dynamics (CFD) simulations have been the interfacial momentum forces (i.e., drag force, lift force, turbu-
emerged as a widely accepted numerical technique to study local lent dispersion force, added mass force, etc.) can only be applied to
characteristics (e.g., liquid velocity profiles, gas holdup profiles, and certain conditions. For instance, the widely used Schiller–Naumann
shear stress profiles) of the multiphase flow dynamics in airlift correlation (1935) for the drag force between the gas and the liquid
column reactors (Mudde and Van Den Akker, 2001). Comparing to phases was developed from single bubble circumstance. Thus,
traditional experimental techniques, CFD simulations cost much theoretically it cannot be applied to high bubble density flow fields.
less in terms of capital and labor, while are applicable to many On the other hand, model parameters for most correlations vary
different applications if proper closures and reliable correlations case by case. Jakobsen (1993) has pointed out cases with either
are used. positive or negative values of the Saffman lift force coefficient and
Two basic approaches exist in modeling the bubble-driven cases with order of magnitude difference of this coefficient.
flows in airlift bubble columns. Based on the volume-averaged Therefore, systematical verification of these models and closures
Navier–Stokes equations, the Eulerian–Eulerian approach treats against reliable experimental measurements obtained by newly
both the dispersed and the continuous phases as interpenetrating developed non-invasive measurement techniques is vital for the
continuum and describes the motion for each of the two phases in application of CFD in reactor design and scale-ups.
Eulerian frame of references (Pan et al., 1999, 2000; Sokolichin and Using CARPT and CT, Luo and Al-Dahhan (2008a, b, 2010)
Eigenberger, 1994; Sokolichin et al., 2004; Becker et al., 1994). On investigated the flow dynamics in a draft tube airlift column
the other hand, the Eulerian–Lagrangian approach treats these two bioreactor. Both global and local information of flow dynamics,
phases differently. In this approach, the continuous phase is still such as gas holdup distributions, liquid velocity profiles, turbulent
treated in the same way as in the Eulerian–Eulerian approach, kinetic energy field, distributions of shear stresses, etc., have been
while the dispersed phase (i.e., discrete bubbles) is tracked by studied. The 3D flow structures in the whole reactor as well as the
solving the motion equations taking into account a balance of all structure in individual regions, i.e., the top, bottom, the riser, and
relevant forces such as pressure, gravity, drag, lift, and added mass the downcomer, have also been characterized.
(Becker et al., 1994; Delnoij et al., 1997a, b, c, 1999; Lain et al., 1999; In the following sections, the experimental data reported in Luo
Pfleger et al., 1999; Thakre and Joshi, 1999; Jacobsen et al., 1997; and Al-Dahhan (2008a, b, 2010) are used as a benchmark database
Sankaranarayanan et al., 2000). Since a very large number of to identify proper closures and correlations for a CFD model based
discrete bubbles have to be tracked even in the bubbly flow on the two-fluid model. The mathematical equations of the two-
regime, the computational cost is usually very expensive in the fluid Eulerian–Eulerian model are first introduced with emphasis
Eulerian–Lagrangian approach. Moreover, it is not suitable to on discussing the various closures and correlations available in
model churn-turbulent flows as the volume fraction of the literature for the interfacial momentum forces. Then 2D and 3D
dispersed phase is usually very high. simulations are conducted using commercial CFD code CFX5.7
In recent years, population balance model has been also widely (Ansys Inc.) to evaluate the CFD model against the experimental
adopt to better capture the impact of bubble sizes distributions on data obtained from CARPT and CT techniques. Different turbulent
the hydrodynamics in bubble columns, where many different sizes models and correlations for various interfacial momentum forces
of bubbles are encountered (Wiemann and Mewes, 2005; Chen et al., were tested to identify appropriate models and correlations that
2005; Sanyal et al., 2005). Although it has been reported that the can represent the data. The validated CFD model and closures,
population balance model is able to improve the CFD simulations combined with the advanced measurement techniques such as
H.-P. Luo, M.H. Al-Dahhan / Chemical Engineering Science 66 (2011) 907–923 909
CARPT and CT, could form valuable knowledge base for reactor due to turbulence:
design and scale-up for many other applications of airlift column
bioreactors at much lower costs.
ma,eff ¼ ma þ mTa ða ¼ l,gÞ ð5Þ
For turbulent flow, the dynamic viscosity is usually much larger

than the molecular viscosity. It is a flow property and is usually
3. Mathematical equations
modeled by various turbulent models as described below.
!
The fourth term on the right-hand side of Eq. (4), M a , takes into
In this work, a two-fluid model (Drew, 1983; Mudde and Van
account of the inter-phase momentum transfer between different
Den Akker, 2001) was used to study the multiphase flow dynamics
phases due to interfacial forces:
at bubbly flow regime in a draft tube airlift column reactor
! X!
investigated by Luo and Al-Dahhan (2008a, b, 2010). A commer- Ma ¼ M l:g
cially available CFD software package CFX5.7 (Ansys Inc.) was used 0 D 1
! !L !VM
to solve the obtained governing equations. X B M l:g ðDrag forceÞ þ M l:g ðLift forceÞ þ M l:g ðVirtual mass forceÞ C
¼ @ !TD A
baa þ M l:g ðTurbulence dispersion forceÞ þ
3.1. Two-fluid model assumptions and conservation equations
ð6Þ
Basic assumptions of the two-fluid model are:
!D !L !VM !TD
where M l:g , M l:g , M l:g , and M l:g , stands for the momentum transfer
All phases are treated as interpenetrating continua. The prob- between the liquid phase and the gas phase due to, respectively,
ability of any one phase occurs in the multiphase flow field is drag force, lift force, virtual mass force, and turbulent dispersion
given by the instantaneous volume fraction of that phase at that force. Correlations needed to model these interfacial forces will be
point. The sum total of all volume fractions at a point thus discussed in the following sections.
should be unity.
Both fluids are treated as incompressible with uniform pressure
field: 3.1.3. Multiphase turbulent equations
The dynamic viscosities shown in Eq. (5) for different phases are
p ¼ pl ¼ pg ð1Þ
flow properties and need to be modeled by turbulent models. In
where p is the pressure, subscripts l and g stands for the liquid gas–liquid flows, since the density for the dispersed gas phase are
phase and the gas phase, respectively. much smaller than those of the continuous liquid phase, it is well-
The gas phase is assumed to be spherical bubbles of same size. known that turbulence inside of bubbles has limited effect on the
Turbulence in the dispersed gas phase is insignificant thus it is liquid flow (Drew, 1983), especially for bubbly flow regime where
correlated to the turbulence in the continuous liquid phase, the gas phase’s characteristic length (i.e., bubble diameter) is
which is modeled via turbulent models. usually much smaller than that of the liquid phase (i.e., reactor
Mass exchange between phases is neglected. diameter). Therefore, a simple algebraic model is adopted in this
The whole fluid domain is assumed to be under isothermal work to compute the dynamic viscosity of the dispersed phase,
condition, thus heat exchange is not considered. which relates its dynamic viscosity to the continuous phase’s
viscosity as (Markatos, 1986; CFX5.7 manual, 2004):
Based on these assumptions, the conservation equations can be
written as follows (CFX5.7 Manual). A more comprehensive rg mTl
mTg ¼ ð7Þ
derivation of these equations can be found elsewhere (Jakobsen, rl s
1993; Joshi, 2001). where s is the interfacial tension between the gas and liquid
phases.
3.1.1. Continuity equations The situation is much more complex in computing the con-
tinuous liquid phase’s dynamic viscosity. In gas–liquid flow,
@ ! turbulence could be stemmed from either the shear stresses within
ðra ra Þ þ rUðra ra U a Þ ¼ 0 ða ¼ l,gÞ ð2Þ
@t the continuous phase or the rising bubbles. The dynamic viscosity
! thus contains two components: the shear-induced turbulence, ms,T
where t is the time, ra, U a and ra are the density, the velocity a ,
vector, and the volume fraction of phase a, respectively. The and the bubble induced turbulence, mb,T
a :
constraint for volume fractions of both phases has to be imposed: mTa ¼ ms,T b,T
a þ ma ð8Þ
rl þ rg ¼ 1 ð3Þ
Many approaches have been proposed to compute the shear-
induced dynamic viscosity (Markatos, 1986), such as the standard
3.1.2. Momentum
equations

@ ! ! ! ! k–e model (after modified for multiphase flow) and the Reynolds
ra ra U a þ rU ra ra U a U a ¼ ra rpa þ ra ra g
@t |fflfflfflfflffl{zfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} Pressure |fflfflfflffl{zfflfflfflffl} shear stress model. For example, the standard k–e turbulent model,
gradient term Body force term
Time variance term Convection term proposed by Launder and Spalding (1974), calculates the shear-
! ! ! induced dynamic viscosity by
þ rUðra ma,eff ðr U a þ ðr U a ÞT ÞÞ þMa ða ¼ l,gÞ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflffl{zfflffl} 2
Interfacial momentum transfer term ka
Reynolds stress force term
ms,T
a ¼ C m r a ð9Þ
ð4Þ
ea
where Cm is a constant, ka is the turbulent kinetic energy and ea is
! the turbulent dissipation rate for phase a. The conservation
where g is the gravity vector, ma,eff is the effective viscosity of
! equations used to solve ka and ea are:
phase a, the superscript T indicates the turbulent term, and M a
stands for the interfacial momentum transfer term. @ ! ms,T
ðra ra ka Þ þ rUðra ra U a ka Þ ¼ r U ra ma þ a Urka ra ðPa ra ea Þ
In this work, the effective viscosity includes contributions from @t sk
both the molecular viscosity (ma) and the dynamic viscosity (mTa ) ð10Þ
and where the terminal velocity for the bubble, Ut, is given by
mg
@ ! ms,T Ut ¼ M 0:149 ðJ0:857Þ
ðra ra ea Þ þ rUðra ra U a ea Þ ¼ rU ra ma þ a Urea r db
@t se ( a
ea 0:94H0:757 , 2 o H o 59:3
þra ðCe1 Pa Ce2 ra ea Þ ð11Þ J¼
ka 3:42H0:441 , H 459:3
4 m 0:14
l
H ¼ Eo M 0:149
where Ce1, Ce2, sk and se are constants; and the production term, Pa, 3 0:0009
is defined as g m4l ðrl rg Þ
M¼ 2 3
ð20Þ
! ! !T ms,T ! rl sl
Pa ¼ ma r U Uðr U þ r U Þ a g Urr
s,T
ð12Þ
rPrt
where Prt is the turbulent Prantdl constant which is usually 3.2.2. Lift force
approximate to 0.85 for gas–liquid system. Lift force is a term referring to interfacial forces acting on the
As for the bubble induced dynamic viscosity, the model lateral direction, the direction perpendicular to the major flow
proposed by Sato and Sekoguchi (1975) was used direction. Lateral forces with different mechanisms have been
! ! proposed, such as the Magnus force due to bubbles’ rotations, and
ml b,T ¼ 0:6rl rg db 9 U g U l 9 ð13Þ the Saffman force due to the shear flow or the velocity gradient
around the bubbles (Jacobsen et al., 1997). In bubbly flow regime,
where db is the bubble diameter.
Saffman (1965, 1968) found that the Saffman force is usually
an order of magnitude larger than the Magnus force. Therefore,
3.2. Interfacial momentum forces only the Saffman force was tested in this work (Jacobsen et al.,
1997)
Interfacial momentum forces in the governing momentum ! ! !
Eq. (4) account for the interactions between the continuous phase Fl ¼ rg rl CL ð U g U l Þ ðr U l Þ ð21Þ
and the dispersed phase. For gas–liquid flows, important interfacial where CL is non-dimensional lift force coefficient, l stands for the
forces include drag, lift, turbulent dispersion, and added mass or liquid phase, and g stands for the gas phase.
virtual mass forces. Closures are required to model these forces,
which have been extensively discussed in the literature (Jacobsen
3.2.3. Turbulent dispersion force
et al., 1997; Joshi, 2001), and are briefly introduced below.
Turbulent dispersion force takes into account of the turbulent
diffusion of the dispersed phase in the continuous phase. It is
3.2.1. Drag force derived by Favre-averaging the multiphase Navier–Stokes
Drag force arises when bubbles move at different velocities to equation (Lopez de Bertodano, 1998):
the surrounded liquid phase. It is the most important interfacial !TD !TD
force, and is a function of the local slip velocity between the M l ¼ M g ¼ CTD rl kl rrl ð22Þ
continuous and the dispersed phases
where CTD is a constant, usually in the range 0.1–0.5. kl and rl are,
!D ! 3 CD ! ! ! ! respectively, the turbulent kinetic energy and the holdup of the
M l:g ðDrag forceÞ ¼ nb D b ¼ rg rl 9 U g U l 9ð U g U l Þ ð14Þ
4 db liquid phase.
where nb is the number of bubbles, CD is a drag coefficient required

to compute the drag force, Db. Many different correlations can be 3.3. Particle (cell) tracking
found in the literature to compute the drag coefficient, CD. In this
In a photobioreactor, the algal cells’ movement determines the
work, the following drag correlations were tested:
pattern of how they are exposed to the light energy, and thereby
Constant drag coefficient (Jacobsen et al., 1997; Joshi, 2001):
their growth rate. Hence, it is very important to track the cells’
CD ¼ 0:44 ð15Þ trajectories for photobioreactor analysis, design, and process inten-
sification. These cells’ trajectories should be able to statistically
Ishii–Zuber correlation (1979):
represent the behavior of all cells in the reactor for a meaningful
gðrl rg Þd2b analysis. Thus, the trajectories should cover everywhere in the whole
CD ðellipseÞ ¼ 2=3 Eo1=2 , Eo ¼ ð16Þ
s flow domain with statistically plenty of times. This requires to trace
either a single algal cell for a very long time (e.g., 24 h as in CARPT
where Eo is the Eotvos number considering the bubble shape (i.e.,
experiments discussed in Luo and Al-Dahhan (2008a)), or to trace a
the ratio between the gravitational and the surface tension forces)
large number of cells for a relatively short time. In CFD simulation,
and s is the interfacial tension between the gas and liquid phases.
the first approach is usually very time-consuming. Therefore, the
Schiller–Naumann correlation (1935):
second approach was used in this work.
24 The movement of algal cells in the reactor was simulated by
CD ¼ ð1 þ0:15 Re0:678
b Þ ð17Þ
Reb introducing particles with same characteristics (i.e., density and
size) as algal cells into the fluid domain. In a Lagrangian reference,
where the bubble Reynolds number, Reb, is defined as
the motion of the particles was computed by solving the famous
! ! Basset–Boussinesq–Oseen motion equation which takes into
rl 9 U g U l 9db
Reb ¼ ð18Þ account of the forces acting on the particle (CFX5.7 manual,
ml
2004; Delnoij et al., 1997a)
Grace correlation (Grace et al., 1976): !
dvp 1 ! ! ! ! pd3 rf d! vf
4 gdb rr mp ¼ prd2 CD 9 v f v p 9ð v f v p Þ þ
CD ¼ ð19Þ dt 8 6 dt
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
3 UT2 r Term I Term II
pd3 rf d! ! ! manual, 2004):

vf dvp pd3 !
þ þ ðrp rf Þ g ð23Þ
12 dt dt |fflfflfflfflfflfflfflfflfflfflfflfflfflffl
6 ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl} qffiffiffiffiffiffiffiffiffiffiffi 3=4
Cm k3=2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} ! le
Term IV v f u ¼ G 2k=3, 5te ¼ pffiffiffiffiffiffiffiffiffiffiffi , and le ¼ ð24Þ
Term III 2k=3 e
where k and e are, respectively, the continuous phase’s local
where mp is the particle mass, d is its diameter, r is its density, and turbulent kinetic energy and the dissipation rate obtained from
!
v is the velocity vector, with the subscripts p stands for particle the turbulence model, Cm is the turbulent constant in Eq. (9), and
and f for the fluid (the continuous phase). The terms on the right- variable G is a randomly chosen number. This random number has
hand side of the equation is explained as follows. a normal distribution and takes into account of the non-determi-
nistic nature of the turbulent flow.
Term I is the drag force due to the slip velocity between the
continuous phase and the particle. The drag coefficient CD is 3.4. Mathematical solution
calculated by Schiller–Naumann’s correlation as incorporated in
the CFX5.7 code. A commercially available CFD code, i.e., CFX of version 5.7, was
Term II considers the force applied on the particle due to the used in this work to solve the governing equations of the two-fluid
pressure gradient surrounding the particle caused by the fluid model described above. CFX5.7 employs a finite control volume
acceleration. approach to discretize the governing Navier–Stokes equations into
Term III is the virtual mass force due to the acceleration of the a set of linear equations. A high resolution scheme was used in this
particle. Since the density difference between the particle and work to compute the advection terms. This scheme, combining the
the continuous phase is very small, there is no much difference first-order upwind difference scheme with a numerical advection
between their velocity accelerations. Therefore, this term is not correction term, is believed to be both accurate (close to second-
important and neglected in this study. order scheme) and bounded (CFX5.7 manual, 2004).
Term IV is the body force due to buoyancy.
Other forces, such as the Basset force and Saffman force, are not
important with relatively smaller magnitude, and thus 4. CFD simulations
neglected in this work.
Numerical Computational Fluid Dynamics (CFD) simulations
were carried out to study the flow dynamics in the draft tube airlift
Due to the random nature of turbulent flow, the motion column and to evaluate various closures for CFD simulations using
equation, i.e., Eq. (23), needs to be solved in a Eulerian–Lagrangian commercial code CFX5.7 (Ansys Inc.) based on the two-fluid
reference in transient simulations. However, tracking a large amount mathematical model discussed above. The following sections first
of particles (e.g., more than 4000) in a 3D domain could be introduce the configuration and operating conditions of the draft
computationally extremely expensive. Therefore, this work uses tube column that will be modeled in this work. Luo and Al-Dahhan
a pseudo Eulerian–Lagrangian approach (Rammohan, 2002), (2008a, b, 2010) have extensively studied this airlift column using
solving the motion equation (23) in a steady state simulation but CARPT (Computer Automated Radioactive Particle Tracking) and CT
considering the effects of turbulence on the particle motion. This (Computed Tomograph) techniques, providing a benchmark data-
approach first uses an Euler/Euler approach to generate the liquid base for evaluation and validation of the selected CFD closures.
flow field by steady state simulations, and then introduces particles How this airlift column will be modeled in 2D and 3D simulations
into the simulated flow field, by which the particles’ motions will then be discussed, followed by simulation results and
(i.e., trajectories) within the flow domain are numerically computed discussions.
by solving Eq. (23).
Since a steady state simulation gives a time independent flow
4.1. Reactor configuration and operating conditions
field (e.g., dvf/dt¼0), the particle velocities, vp, computed by
Eq. (23) are actually a time-averaged quantity. Under such conditions,
Fig. 1 illustrates a Plexiglas draft tube airlift column studied in
the computed particle trajectories are deterministic. In other words,
Luo and Al-Dahhan (2008a, b, 2010) where rich gas–liquid flow
the trajectory of a particle injected at a given location is unique,
dynamic information was obtained using CARPT and CT techni-
which is not true due to the chaotic nature of turbulence. Therefore,
ques. The diameter and height of this airlift column are, respec-
a fluctuating component has to be added to the mean particle velocity
tively, 0.13 m and 1.5 m. Inside the column, an inner cylinder with a
to consider the turbulent dispersion of the tracer particles.
diameter of 0.09 m and a height of 1.05 m is co-axially mounted.
To consider the turbulent dispersion of the tracer particles,
In all experiments, tap water was used as the liquid phase and
following assumptions are made in CFX5.7 code:
was operated in a batch mode, while cylinder compressed air was
used as the gas phase and was operated in a once-through mode.
A particle is always within a single turbulent eddy with Since the transition from bubbly flow to the turbulent flow regime
characteristic fluctuating velocity (vf0 ), lifetime (te), and length for this studied airlift column occurs at 2 cm/s superficial gas
scale (le). velocity as reported in Luo and Al-Dahhan (2010), five different
A particle immediately acquires the eddy’s fluctuating velocity superficial gas velocities (i.e., Ug of 0.076, 0.29, 0.82, 1.0, 5.0 cm/s),
when the particle enters the turbulent eddy. calculated based on the whole column cross-section, were
Only when the particle–eddy interaction time exceeds the chosen to cover the bubbly to turbulent flow regimes. These
eddy’s lifetime, te, or the displacement of the particle relative superficial gas velocities represent the typical operating range
to the eddy is larger than the eddy’s length, le, the particle is for bioprocesses.
assumed to be entering a new turbulent eddy and acquires a
new fluctuating velocity.
4.1.1. 2D simulations
Due to its simplicity and low computational costs, 2D sym-
Based on the mean local turbulence properties, the fluctuating metric steady state simulations have been widely used in the
velocity, the eddy lifetime and length were calculated as (CFX5.7 literature. In this work, such simulations were first conducted to
Draft Tube
Support
Top Clearance
(From the static liquid level to
the top edge of the draft tube)
0.09 m
0.13 m 15 holes (φ1mm) equidistantly

opened upward on the ring tube
120°
1.05 m
φ5cm
1.5 m
10 cm
Ring tube
Riser
(φ5mm)
Straight tube
(φ5mm)
Air Ring Sparger
Draft Tube
Support
3cm
Bottom Clearance
(From lower edge of the
draft tube to column base)
Ring Sparger
Fig. 1. Configuration of the studied airlift column reactor.
test their capability on capturing the mean flow within the bubbly was implemented by adding an extra sink term to both phases on the
flow regime (i.e., at superficial gas velocity of 1 cm/s) and to top domain surface, so that the liquid elements touched this surface
compare with the 3D simulation results. Fig. 2 shows the geometry bounce back while the gas phase can be eliminated in a way as if it
and the structural hexahedra griding used in the computations. The goes out of the domain. Such degassing boundary condition can
computational domain is consists of a 301 wedge cut from the column considerably reduce the computational cost because only the part of
with the draft tube and the sparger regions being further cut out. This the column below the gas–liquid interface needs to be modeled. It
computational domain has a height of 1.16 m, same to the dynamic also has an advantage of enhance solution convergence as observed
liquid level observed in experiments at superficial gas velocity of in the numerical experiments when a pseudo-steady state can be
1 cm/s, and a radius of 0.065 m, same to the column dimension. An assumed. All gas bubbles were also assumed to be the same size with
equally spaced mesh in both the radial and the axial directions were a diameter of 3 mm, which is based on experimental observations
used, with the mesh size of 2.5 mm in a fine griding case and 5 mm in that most of bubbles are about 3 mm. This assumption was also used
a coarse griding case (to evaluate the effects of mesh size on the for all other simulations conducted in this study.
converged solution). On the tangential direction, only one mesh cell In summary, the boundary conditions for this 2D steady state
was considered in the 2D simulations as shown in Fig. 2. simulation are:
Neumann boundary conditions with zero gradients were applied
!
to the center axis and the front and back planes as shown in Fig. 2, Top surface of the sparger (inlet): Ug ¼(0, 0, 0.338) m/s and
where symmetric has to be imposed due to the nature of 2D rg ¼0.5 (for superficial gas velocity of 1 cm/s);
simulations. On the upper surface of the sparger, a Dirichlet boundary Top surface of the column (outlet): Degassing boundary
condition was applied with a gas upward velocity of 0.38 m/s and a condition;
volume fraction (rg) of 0.5 to give a superficial gas velocity of 1 cm/s Center axis (r¼0, symmetric):
(for other superficial gas velocities, these two values need to be ! !
specified). On the top surface of the whole column, a degassing @ Ug @Ul @rg @r
¼ ¼ ¼ l ¼ 0; ð25Þ
boundary condition was applied. Such degassing boundary condition @r @r @r @r
Fig. 2. Geometry and griding for the 2D steady state simulations.
Front and back planes (symmetric): Fig. 3c shows the griding on a cross-sectional plane, where finer
! ! griding at the wall regions was used. This is also true for the griding
@ Ug @Ul @rg @r on the axial direction as shown in Fig. 3d. Cases with two different
¼ ¼ ¼ l ¼ 0; ð26Þ
@y @y @y @y mesh sizes were tested for the 3D simulations to assess the effects
!
where U is the velocity vector; rg and rl are the volume fraction of mesh size on the results. The cases with coarse griding have
of gas and liquid phases; and subscript g and l stands for, totally 48,132 elements with a mesh size of around 5mm, while
respectively, gas and liquid phase, r is the radius. cases with fine griding have totally 271,360 elements with a mesh
The initial conditions for the 2D simulations were simply set to size of around 2.5 mm.
! ! Similar to the 2D simulations, a Dirichlet boundary condition was
rg ¼ 0:01; Ug ¼ Ul ¼ ð0,0,0:01Þ m=s ð27Þ applied to the upper face of the sparger, i.e., the gas inlet velocity and
its volume fraction are specified. The product of these two variables
4.1.2. 3D simulations gives the desired superficial gas velocity of the simulations. On the
Both 3D transient and steady state simulations were carried top domain surface, a degassing boundary condition was applied for
out in this work to capture the flow dynamics. However, due to the 3D steady state simulations, while a constant pressure boundary
high computational cost of the transient flow simulations, 3D condition (i.e., P¼1 atm) was applied on the top domain surface for
steady state simulations were mainly used to test the turbulent 3D transient simulations, in which a pseudo-steady state does not
models and the closures. This is justified by the low superficial gas need to be assumed. In summary, the boundary conditions are:
velocity used in this work, thus a pseudo-steady state could be
!
reached. Upper surface of the sparger (inlet): Ug ¼(0, 0, 0.38) m/s and
Fig. 3 shows the geometry and the griding used in the 3D rg ¼0.5 (for superficial gas velocity of 1 cm/s);
simulations. In 3D transient simulations, the whole column was Upper surface of the column (outlet):
considered (i.e., H¼1.5 m), while in 3D steady state simulations, J Steady state simulation: Degassing boundary condition;
only the part of the column below the gas–liquid interface was J Transient simulation:
modeled (i.e., H¼1.16 m). In all 3D simulations, the draft tube and
p ¼ 1 atm ð28Þ
the sparger were cut out from the computational domain. More-
over, the real circular ring sparger as shown in Fig. 1 was modeled
as a rectangular ring (shown in Fig. 3b). Such modification of the As for the initial conditions, conditions close to a pseudo-steady
sparger in the simulation is justified by the high aspect ratio (i.e., state result were assigned to facilitate the convergence of the
the height to diameter ratio) of the studied column, which is 13. As solution in 3D steady state simulations; while in the 3D transient
discussed in Chisti (1998), sparger configuration has limited effects simulations, the initial conditions were assigned to be the same as
on the global flow dynamics in the airlift column reactor for column the real conditions before experiments start:
with an aspect ratio greater than 8. This modification also has an
advantage of applying structured hexahedra griding, which is Steady state simulation:
important in reducing the computational error and facilitating
! !
the solution convergence. rg ¼ 0:01; Ug ¼ Ul ¼ ð0,0,0:01Þm=s ð29Þ
Degassing boundary conditions are applied on

the top surface for steady state simulations
Draft Tube (having been cut out from the

computational domain)
Rectangular Sparger
(4cm×4cm, Inlet boundary
conditions is applied on
the upper surface)
Fig. 3. Configurations of the computational domain in 3D simulations: (a) overview of the computational geometry for a 3D steady state simulation with height of 1.16 m (the
whole column with height of 1.5 m is modeled in a 3D transient simulation); (b) sparger configuration; (c) mesh on the cross-sectional plane for the fine griding; and (d) mesh
in the axial direction for the fine griding.
Transient simulation: It is clear from these figures that the 2D simulation results
noticeably deviate from the experimental data, especially for the
1 H 4 1:16m ! !
rgas ¼ ; Ug ¼ Ul ¼ ð0,0,0Þm=s ð30Þ gas holdup profiles, although results with finer griding are slightly
0 H o ¼ 1:16 m
better as shown in Fig. 4b. This is consistent with the literature that
2D symmetric steady state simulations might hard to capture the
5. Simulation results and discussion mean multiphase flows (Bertola et al., 2003). On the other hand, the
3D steady state simulation results well match the axial liquid
In the following sections, the CFD simulation results for operat- velocity, reasonably agree with the local gas holdup profiles, but
ing conditions of 1 cm/s superficial gas velocity, 3 cm top clearance, considerably under-estimate the experimentally measured turbu-
and 5 cm bottom clearance are discussed. Under the default options lent kinetic energies.
provided in CFX5.7 code (i.e., standard k–e turbulence model, Grace Considering the experimental error, such an agreement between
drag law, Lopez de Bertodano’s turbulent dispersion force with the experimental and the simulated results are encouraging, although
coefficient of 0.3), different types of simulations (i.e., 2D steady certainly there is room for improvement. It is not surprising that the
state, 3D steady state, and 3D transient simulations) with different calculated turbulent kinetic energies are much lower than the ones
griding sizes were first conducted to identify a robust simulation measured by CARPT experiments. As mentioned earlier, k–e turbulent
with low computational cost but good accuracy. model is well-known to damp turbulent intensity resulting in
Such type of simulation then was relied on to test the turbulent considerable under-estimation. On the other hand, CARPT experi-
models and various closures for the interfacial momentum forces (i.e., ments could also over-estimate the turbulent kinetic energies. As
the drag force, the lift force, the turbulent dispersion force). To assess discussed in Luo and Al-Dahhan (2008b), in CARPT experiments,
models and closures, the simulation results were compared against white noises and the fluctuation nature of g-ray radioactivity produce
the benchmark database obtained by CARPT and CT experiments in nontrivial errors in reconstructing the tracer’s location, and thus in
Luo and Al-Dahhan (2008a, b, 2010), specifically against the experi- computing the fluctuating velocities. Since the errors are squared and
mental data on axial liquid velocity, turbulent kinetic energy, and summed in the calculation of the turbulent kinetic energy, CARPT
local gas holdup profiles. These parameters are the most sensitive usually over-estimates the turbulent kinetic energy.
ones to the simulation. The simulation accuracy of other flow dynamic In addition, since uniform bubble size is used in the CFD simulations,
parameters, such as the shear stresses, is usually within the range of it is hard for the simulation to capture the gas holdups in the down-
the simulation accuracy of the above mentioned three parameters. comer regions where many tiny bubbles were observed in the experi-
The identified models and closures were further utilized to simulate ments since only small bubbles can be carried down to this region. In
the flow dynamics at different superficial gas velocities. that sense, a different bubble size should be used for the downcomer
region or using population balance models to better account for that
5.1. Effects of types of simulations and griding size phenomena, which will be investigated in future studies.
Fig. 4 also indicates that the 3D steady state simulation results
Three types of simulations (i.e., 2D steady state, 3D steady state, based on the coarse griding are not very different from the results
and 3D transient simulations) with fine and coarse grindings were based on the fine griding. This is due to the k–e turbulent model
tested to find the balance between the computational cost (in terms used in the simulations to compute the dynamic viscosity in the
of the CPU time or the net time a computer CPU spends on the liquid phase and indirectly in the gas phase via Eq. (7). The k–e
simulation job) and the simulation accuracy. Fig. 4a–c shows the turbulent model is well known to over-estimate the eddy viscosity
comparison of these simulations with the experimental data and damp out the small-scale turbulences. Therefore, decreasing
obtained by CARPT (for axial liquid velocity and turbulent kinetic the griding size after certain value does not affect the simulation
energy profiles) and CT (for local gas holdup profile at a level of results, but considerably increases the computational cost. For
67.7 cm from the bottom). example, the convergence CPU time associated with the fine
CARPT Results 2D Coarse Griding CT Results 2D_Coarse Griding

2D Fine Griding 3D_Coarse Griding 2D_Fine Griding 3D_Coarse Griding
3D_Fine Griding 3D_Transient 3D_Fine Griding 3D_Transient
Axial Liquid Velocity, m/s 0.5 0.06
0.3
Gas Holdup
0.04
0.1
0.02
-0.1 0 0.2 0.4 0.6 0.8 1

r/R
0
-0.3 0 0.5 1
r/R
CARPT Results 2D_Coarse Griding

2D_Fine Griding 3D_Coarse Griding
3D_Fine Griding 3D_Transient
Turbulent Kinetic Energy, m2/s2
0.04
0.02
0
0 0.5 1
r/R
Fig. 4. Effects of griding on the CFD simulations: (a) effects on the axial liquid velocity profile; (b) effects on the gas holdup profile; and (c) effects on the turbulent kinetic
energy profile.
griding in 3D steady state simulations is four times more than the experimental data obtained by CARPT and CT. Although the turbulent
time associated with the coarse griding. kinetic energies predicted by the Reynolds shear stress model are
Similar results were also obtained from the 3D transient closer to CARPT data comparing to the results of the k–e model, its
simulation using the coarse griding. These results were time- predictions on the mean multiphase flow, i.e., the axial liquid and
averaged for quantitative comparison. However, such 3D transient particularly the gas holdup profiles, are much worse. Moreover, due to
simulation usually takes more than one week to get any quanti- the complexity of this model, the convergence is not as good as the k–e
tative meaningful results. Therefore, it is not practical to use 3D model (the momentum residuals are almost one order of magnitude
transient simulations for closure tests. Instead, 3D steady state larger). Considering these, the standard k–e model will be used to
simulations with a coarse griding were used in the following further test the closures for interfacial momentum transfer.
sections to test the turbulent models and closures.
5.3. Coefficient for shear-induced dynamic viscosity, Cm

5.2. Turbulent models
A simple and widely used method to compensate the standard
Since the standard k–e model as outlined above considerably k–e model’s limitations on the turbulent intensity prediction
under-estimates the turbulent intensity in the studied airlift for two-phase flow is to adjust the coefficient in calculating the
reactor, a Reynolds shear stress model incorporated in CFX5.7 shear-induced dynamic viscosity, Cm in Eq. (8) (Sokolichin and
was also tested in this work to compute the dynamic viscosity in Eigenberger, 1994). When decreases this coefficient, the dynamic
the liquid phase and also in the gas phase indirectly via Eq. (7). This viscosity calculated by Eq. (8) decreases, resulting in a higher
Reynolds shear stress model expresses each component of the turbulent intensity. This simple method is attempted in this work to
Reynolds shear stresses with separate transportation equations minimize the discrepancy between the experimentally measured
derived from the Navier–Stokes equations. It is believed that such and the CFD predicted turbulent kinetic energy as mentioned above.
treatments can better resolve the small-scale turbulence. However, Fig. 6 shows the effects of reducing the turbulent dispersion
due to the introduction of second and third order correlations, coefficient in 3D steady state simulations. When Cm decreases for 10
further modeling of these high order correlations are inevitable, times, almost one order of magnitude larger turbulent kinetic
which impose considerable challenges to apply such turbulent energies were obtained at r/R of 0.3–0.4. The obtained kinetic
model in multiphase flow simulations. energies are closer to the CARPT measurement particularly in the
Fig. 5 shows the simulation results of the Reynolds shear stress riser section comparing to the simulations using the standard Cm
model together with the simulation results of the k–e model and the value of 0.09.
a b
CARPT Results 0.12 CT Results
0.6 Standard k-ε Standard k-ε
Reynolds Shear Stress 0.1 Reynolds Shear Stress
Axial Liquid Velocity, m/s
0.4
0.08
Gas Holdup
0.2 0.06
0.04
0
0 0.2 0.4 0.6 0.8 1
0.02
-0.2 r/R
0
-0.4 0 0.2 0.4 0.6 0.8 1
r/R
c
0.05 CARPT Results
Standard k-ε
Reynolds Shear Stress

0.04
0.03
0.02
0.01
0
0 0.5 1
r/R
Fig. 5. CFD simulation results by different turbulent models: (a) results of the axial liquid velocity profile; (b) results of the gas holdup profile; and (c) results of the turbulent
kinetic energy profile.
However, the effects of reducing Cm are twofold: as the turbulent quite different. Simulations based on a constant drag coefficient and
intensity increases, the turbulent diffusion of the dispersed phase is Schiller–Nauman’s correlation considerably under-estimate the gas
also enhanced, resulting in a less steep gas holdup profile as shown holdups in both the riser and the downcomer region. On the other
in Fig. 6b. Due to the enhanced interfacial momentum transfer, the hand, simulations based on both Grace’s and Ishii–Zuber’s correla-
slip velocity between the gas and the liquid phases decreases, tions properly predicted the CARPT measured axial liquid velocities,
which in turn results in a steeper axial liquid velocity profile as and reasonably agree with the CT measured gas holdups. Among
shown in Fig. 6a. Therefore, although reducing the turbulent these simulations, clearly the simulation based on Ishii–Zuber’s
dispersion coefficient results in a larger turbulent intensity pre- correlation agrees with the experimental data best. Hence, the
dicted by the k–e model, such improvement can be greatly offset by Ishii–Zuber correlation is chosen for further closure verification.
the loss of mean flow prediction capability. Indeed, when using a
low Cm in a 3D transient simulation, a highly dynamic multiphase
flow was observed in one attempt, which however could be very 5.5. Effects of turbulent dispersion force
misleading. Accordingly, the standard Cm value of 0.09 will be used
throughout the rest of numerical experiments. Turbulent dispersion force takes into account of the turbulent
diffusion of the dispersed phase in the continuous phase. It is
therefore can move the dispersed phase on the lateral directions. As
5.4. Correlations for drag force shown in Fig. 8, the dispersed phase spreads more obviously in the
radial directions as the coefficient increases due to the higher
Due to the significance of the drag force on multiphase flow simu- turbulent diffusion. With a coefficient (CT) of 0.3, the predicted gas
lations, the following drag coefficient correlations were tested: con- holdup profile agrees best with the experimental data. How-
stant coefficient (CD¼0.44), Schiller–Naumann’s (1935), Ishii–Zuber’s ever, although the gas holdup profiles change considerably as
(1979), and Grace et al.’s (1976) correlations. The standard k–e model the coefficient increases, the axial liquid velocity and the turbulent
and a Cm of 0.09 were used to test these correlations, which were kinetic energy profiles change only slightly. Moreover, considering
developed for bubbly flow regime. the turbulent dispersion force does not computationally introduce
Fig. 7 shows the simulation results together with the experi- convergence problem, these features make the turbulent disper-
mental data. It is clear from Fig. 7 that, the predicted axial liquid sion force great in fine tuning the CFD simulations. The turbulent
velocity and turbulent kinetic energy profiles are quite close to each dispersion force with a standard coefficient (0.3) thus will be
other for all drag laws, while the obtained gas holdup profiles are considered for further closure verification in this work.
0.8 CT = 0.3 exp CT = 0.3 exp

CL = 0.0 Cμ = 0.09 CL = 0.0 Cμ = 0.09
Cμ = 0.009 0.06 Cμ = 0.009
0.6
Gas Holdup
0.4
0.04
0.2
0.02
0
0 0.2 0.4 0.6 0.8 1
-0.2 r/R
0
-0.4 0 0.2 0.4 0.6 0.8 1
r/R
0.05
CT = 0.3 exp
Turbulent Kinetic Energy,m2/s2
CL = 0.0 Cμ = 0.09
0.04 Cμ = 0.009
0.03
0.02
0.01
0
0 0.5 1
r/R
Fig. 6. CFD simulation results using different magnitude of turbulent dispersion coefficient, Cm: (a) results of the axial liquid velocity profile; (b) results of the gas holdup
profile; and (c) results of the turbulent kinetic energy profile.
5.6. Effects of Saffman force flow dynamics in the draft tube airlift column reactor studied in
this work:
Saffman force has been extensively discussed in the literature for
the lateral movements of the dispersed bubbles in the liquid flows
(Jakobsen, 1993; Jacobsen et al., 1997; Sokolichin et al., 2004). 3D steady state simulation using a coarse griding;
However, its physical existence and the mathematical representation the k–e model with Cm of 0.09;
are still under debate (Sokolichin et al., 2004), especially considering Ishii–Zuber correlation (1979) for the drag force;
the fact that bubbles deform in the flow domain (Jacobsen et al., Lopez de Bertodano correlation (1998) with coefficient, CT, of 0.3
1997). The direction of the force acted on a deformable bubble has for the turbulent dispersion force);
been observed to be opposite to the direction on a spherical bubble. neglect Saffman lift force.
Such arguments imply that the coefficient of the Saffman lift force
could be a flow property, thus it is hard to predict. 6. Simulations for different superficial gas velocities
Different coefficients with opposite signs were tried in 3D
steady state simulations. As shown in Fig. 9, a positive Saffman Based on the selected turbulent models and the correlations, 3D
force coefficient tends to move the bubbles to the walls in both the steady state CFD simulations were performed to simulate the flow
riser and the downcomer flow regions. This results in flatter profiles dynamics at superficial gas velocities of 0.3 cm/s (i.e., bubbly flow
for all three profiles shown. In contrast, a negative coefficient has regime) and 5 cm/s (i.e., churn-turbulent flow regime) to further
exactly the opposite effects. As for the turbulent kinetic energy, test its prediction capability in different flow regimes. The simula-
moving the bubbles to the wall region obviously reduces the tion results are shown in Fig. 10. The results for superficial gas
turbulent due to the wall limitation, while moving the bubbles to velocity of 0.3 cm/s, still within the bubbly flow regime, agree well
the reactor center promote the turbulence. However, it is obvious with the experimental data in terms of the axial liquid velocity and
that the simulation without considering the Saffman lift force best the gas holdup profiles, although the turbulent kinetic energies are
matches the experimentally measured mean flow in the currently still under-estimated. However, the simulation results for super-
studied case, when the turbulent dispersion force was considered. ficial gas velocity of 5 cm/s, within the churn-turbulent flow
Moreover, it was found that the convergence is much harder to regime, fail to capture even the mean flows in the studied airlift
reach when the Saffman force is considered. Therefore, the Saffman column.
lift force was not recommended for the applications in this study. Such results suggest that the correlations and closures identi-
Based on the above evaluation against CARPT and CT data, the fied above cannot be applied to the churn-turbulent flow regime
following closures were identified to investigate the multiphase but only to the bubbly flow regime. In fact, it is well-known that the
exp: 0.08 CT = 0.3 exp:

Grace Cμ = 0.09 Grace
0.6 Ishii-Zuber Ishii-Zuber
Axial Liquid Velocity, m/s Schiller Naumann Schiller Naumann
Constant: 0.44 0.06 Constant: 0.44
0.4
Gas Holdup
CT = 0.3
0.2 0.04
Cμ = 0.09
0
0 0.2 0.4 0.6 0.8 1 0.02
r/R
-0.2
0
-0.4 0 0.2 0.4 0.6 0.8 1
r/R
exp:
0.05 CT = 0.3
Grace
Cμ = 0.09
Ishii-Zuber
0.04 Schiller Naumann
Constant: 0.44
0.03
0.02
0.01
0
0 0.5 1
r/R
Fig. 7. Effects of drag correlations on the CFD simulations: (a) effects on the axial liquid velocity profile; (b) effects on the gas holdup profile; and (c) effects on the turbulent
break-up and coalescence of bubbles are very prominent in the introduced from the bottom of the airlift column and traced for
churn-turbulent flow regime, resulting in a wide range of bubble up to 40 s for each particle in a 3D steady state simulation. By this
size distribution. Population balance model will be needed in this approach, the simulation results provided particle trajectories with
case to better account for the bubble break-up and coalescing total effective tracing time of around 11 h, and total effective
phenomena. The drag force correlations used in this work, which particle positions of more than 2.5 million in the studied airlift
were developed from homogeneous spherical bubble distribution, column. With such a large amount of particle trajectories, it was
thus cannot describe the complex bubble dynamics under such found that further increasing the tracer particle numbers does not
churn-turbulent condition. On the other hand, due to the highly affect the statistical results of the particle trajectories.
dynamic feature at the churn-turbulent flow regime, it is obvious Due to the non-deterministic nature of the particle trajectories
that a transient simulation is required for CFD simulations under in a turbulent multiphase flow, it is impossible to directly compare
such conditions. Indeed, the convergence of the simulation at the CFD simulated particle trajectories against the CARPT measured
superficial gas velocity of 5 cm/s was also much worse than the ones. Therefore, two statistically based methods are used in this
simulations for lower superficial gas velocities. study to evaluate the CFD generated particle trajectories.
The first method is based on a single trajectory concept
introduced in Luo et al. (2003), derived from the fact that liquids
7. Particle (cell) trajectory evaluation circulate around the riser and downcomer continuously in an airlift
column. A single trajectory is defined as the trajectory of a particle
Since the information of the microorganism cells’ movement in that performed one such circulation. The concept is illustrated in
a photobioreactor is important for the bioreactor performance Fig. 11 where a few typical single trajectories obtained from CARPT
evaluation as mentioned earlier, the two-fluid model with the measurements are shown (Luo and Al-Dahhan, 2008a). Usually, the
above identified closures and correlations was further adopted to duration time of each single trajectory is defined as the circulation
simulate the cells’ movements in the studied draft tube airlift time, and the total distance the particle traveled is defined as the
column. This is justified by the fact that photobioreactors are trajectory length. The probability density functions (PDF) of the
usually operated under bubbly flow regime. circulation time and the trajectory length for these identified single
To mimic the movement of microorganism cells, 4000 small trajectories provide information for a genuine residence time
(i.e., with sizes evenly distributed between 5 and 10 mm) and distribution analysis which reveals the macro-mixing in the airlift
neutrally buoyant particles (i.e., density of 1000 g/cm3) were column reactor.
0.6 exp: 0.1 CL = 0.0 exp:

CT = 0.0 CT = 0.0
0.5 CT = 0.1 Cμ = 0.09
CT = 0.1

0.4 CT = 0.3 0.08 CT = 0.3
0.3 CL = 0.0
Gas Holdup
0.06
0.2 Cμ = 0.09
0.1
0.04
0
0 0.2 0.4 0.6 0.8 1
-0.1 0.02
r/R
-0.2
-0.3 0
-0.4 0 0.2 0.4 0.6 0.8 1
r/R
0.05 exp:
CL = 0.0
CT = 0.0
Cμ = 0.09 CT = 0.1
0.04 CT = 0.3
0.03
0.02
0.01
0
0 0.5 1
r/R
Fig. 8. Effects of the turbulent dispersion force on the CFD simulations: (a) effects on the axial liquid velocity profile; (b) effects on the gas holdup profile; and (c) effects on the
turbulent kinetic energy profile.
From the CFD generated trajectories, thousands of single Fig. 13 shows the radial profiles of the occurrence numbers of a
trajectories were identified. Fig. 12 compare the probability density microorganism cell inside the reactor for both the CFD generated and
functions (PDF) of circulation time and trajectory length for both the CARPT measured trajectories under same operating conditions.
the CFD simulated single trajectories and the CARPT measured Please note the occurrence numbers have been normalized by total
trajectories at superficial gas velocity of 0.82 cm/s. As it can be seen, number of occurrences and the volume of each region, given a unit of
the PDFs of the CFD simulated trajectories are similar to but slightly ml 1, for better comparison. The rather flat profile of the CARPT
narrower than the PDFs of the CARPT measured data, especially for measured trajectories indicates that a microorganism cell will travel
the trajectory length distribution. The means are also slightly everywhere in the whole reactor for almost same amount of time.
different: 11.9% for mean trajectory length and 2% for the mean However, based on the CFD simulated trajectories, the microorgan-
circulation time. This is due to the lower turbulent intensity ism cell will stay considerably longer time in the wall regions.
predicted in the CFD simulations. With less turbulence, the This quite substantial discrepancy is likely due to the treatment
particles are more likely to follow the mean liquid flow which of wall regions in CFD simulations, such as the griding and the wall
orderly circulates around the draft tube in the column, resulting in functions. In reality, the turbulence boundary layer is usually very
a flow closer to plug flows. thin, i.e., in microns, and therefore its impacts on the cells’
The second method to validate the CFD generated trajectory is to movements are relatively small. However, in CFD simulations,
evaluate its capability in capturing the relative residence time in since it is not practical to use a griding so thin due to computational
each region of the reactor, particularly in the wall region. This is cost issue, the wall region becomes much larger with mesh size
important for applications about photobioreactor since light usually in millimeters. Although a wall function is usually used to
intensity reduces exponentially as light penetrates from the reactor relate the flow dynamics within the wall region to better reflect the
wall into the center. The residence time of a microorganism cell in real world, the impacts of wall region on the cells’ movement are
the wall region therefore directly relates to the energy it is exposed usually still much more prominent than in reality. Moreover, since
to and thus its growth rate. The relative residence time distribution the k–e turbulent model tends to damp out turbulence, the
in different regions can be extracted from the cell’s trajectory by fluctuating velocities of particles in the wall regions are signifi-
counting the amount of time it travels through each region of the cantly depressed. This makes it hard for a tracer to escape the wall
whole reactor, or simply by counting how many times the cell region once it enters in a CFD simulation. Therefore, to better
occurs (i.e., occurrence number) in different regions as the time capture the cell’s trajectory in a draft column reactor for photo-
between any consecutive points of the trajectories are equal. bioreactor applications using CFD simulation, better treatment in
0.7 CT = 0.3 exp: CT = 0.3 exp:
CL = -0.1 Cμ = 0.09 CL = -0.1

0.6 Cμ = 0.09 0.08
CL = 0.0
Axial Liquid Velocity, m/s 0.5 CL = 0.0
CL = 0.1 CL = 0.1
0.4 0.06
Gas Holdup
0.3
0.2
0.04
0.1
0
0.02
-0.1 0 0.2 0.4 0.6 0.8 1
-0.2 r/R
-0.3 0
-0.4 0 0.2 0.4 0.6 0.8 1
r/R
0.05
CT = 0.3 exp:
Cμ = 0.09 CL = -0.1
0.04 CL = 0.0
CL = 0.1
0.03
0.02
0.01
0
0 0.5 1
r/R
Fig. 9. Effects of the Saffman lift force on the CFD simulations: (a) effects on the axial liquid velocity profile; (b) effects on the gas holdup profile; and (c) effects on the turbulent
the wall region in terms of griding, wall functions and even the measured ones by statistic methods. Following conclusions can be
turbulent model is needed. Experimental data with more precise reached:
measurements in the wall region might be needed to develop such
treatment, which is out of scope of this work.
A computationally promising CFD simulation model was iden-
tified to study the multiphase flow dynamics in a draft tube
airlift column reactor under the bubbly flow regime in air–water
system. This model uses 3D steady state simulations, the
8. Remarks
standard k–e turbulent model with Cm of 0.09, and closures
such as Ishii–Zuber’s drag force correlation and Lopez de
CFD simulations based on a two-fluid Eulerian–Eulerian model
Bertodano’s turbulent dispersion force with coefficient of 0.3,
have been attempted in this work to verify and validate the
while neglect the Saffman lift force.
reliability of CFD simulations in particle tracking in airlift bioreac-
The identified CFD model and closures properly captured the
tors under bubbly flow regime, particularly in photobioreactors for
mean multiphase flow field, but considerably under-estimated
algae/cyanobacterial cultures. Using commercially available CFD
the turbulent kinetic energy in the studied airlift column.
code CFX5.7 (Ansys Inc.), different types of simulations, turbulent
The identified CFD model and closures predicted reasonably the
models, and correlations for various interfacial momentum forces
microorganism trajectories in a draft tube column reactor
were tested and evaluated against the experimental data obtained
recirculation in terms of the recirculation behavior, however
from CARPT and CT techniques by Luo and Al-Dahhan (2008a, b,
over-estimated the cells’ residence time in the wall region.
2010) in an air–water system. The bases of the evaluations include
Proper treatment for the wall region such as griding and wall
the mean multiphase flow fields such as the time-averaged axial
function is needed to better capture the movement of micro-
liquid velocities, the local gas holdups, and the turbulent kinetic
organism cells in such bioreactors.
energy. After identifying appropriate closures and correlations for
the two-fluid CFD model, it was further utilized to generate
Lagrangian trajectories of microorganism cells in a photobioreactor. It may also worthy to point out that the closures mentioned
The CFD generated trajectories were also evaluated against CARPT above were selected for air–water system in bubbly flow regime.
1 Ug = 1cm/s 0.35 Ug = 5cm/s

Ug = 5cm/s
Ug = 1cm/s
0.8

Ug= 0.29cm/s 0.3 Ug = 0.29cm/s
0.6 0.25
Gas Holdup
0.4 0.2
0.2 0.15
0
0.1
0 0.2 0.4 0.6 0.8 1
-0.2
r/R 0.05
-0.4
0
-0.6 0 0.2 0.4 0.6 0.8 1
r/R
0.12 Ug = 5cm/s
Ug = 1cm/s
0.1 Ug = 0.29cm/s
0.08
0.06
0.04
0.02
0
0 0.5 1
r/R
Fig. 10. CFD simulations for different superficial gas velocities: (a) results of the axial liquid velocity profile; (b) results of the gas holdup profile; and (c) results of the turbulent
kinetic energy profile. Note: solid symbols stand for experimental results while the empty symbols stand for the corresponding CFD simulation results.
systems are non-Newtonian and are likely change with time in

batch operations. In such cases, further examination of correlations
and closures against high quality experimental data are needed for
proper selection of closures for CFD simulations. This is also true for
industrial processes, e.g., organic systems, where the rheology is
significantly different from an air–water system.
Nomenclature
!
g gravity vector, m/s2
!
Ua velocity vector of phase a, m/s
!
v fu fluctuating velocity of tracer particle in Eq. (24), m/s
!
v velocity vector in Eq. (23), m/s
!
Ma interfacial momentum transfer acted on phase a in Eq. (4)
!
M l:g interfacial momentum transfer between the liquid and
gas phases
!D
M l:g interfacial momentum transfer due to drag force between
the liquid and gas phases
!L
M l:g interfacial momentum transfer due to lift force between
the liquid and gas phases
!TD
M l:g interfacial momentum transfer due to turbulence disper-
sion force between the liquid and gas phases
!VM
M l:g interfacial momentum transfer due to virtual mass force
Fig. 11. Typical single trajectories obtained from CARPT measurements (a) 3D between the liquid and gas phases
visualization; (b) projected on the radius–height plane. Different colors are used to Ce1, Ce2 constants in Eq. (10) for standard k–e model,
present each circulation of the particle around the draft tube (Luo and Al-Dahhan, dimensionless
2008a) (for interpretation of the references to color in this figure legend, the reader
CD drag coefficient, dimensionless
is referred to the web version of this article).
CL Saffman lift force coefficient, dimensionless
CTD coefficient of the turbulent dispersion force,
Although most bioprocess systems are also operated in bubbly flow dimensionless
regime, the rheology of these bioprocess systems might be very Cm coefficient for dynamic viscosity in Eq. (9), dimensionless
different from an air–water system. For example, many bioprocess d tracer particle diameter in Eq (23), m
H axial position in the column, m

0.2 ka turbulence kinetic energy of phase a, m2/s2
Mean = 225.3 cm le length scale of the eddy size in Eq. (24), m
Probability
0.15
STD = 8.2 cm mp mass of the tracer particle, kg
0.1 nb number of bubbles in Eq. (14) for drag force calculation
M = 1.073
0.05 p pressure, Pa
0 Prt turbulent Prandtl constant, approximate 0.85
0 100 200 300 400 500 600 700 800 Pa production of turbulence kinetic energy for phase a
Trajectory Length, cm defined in Eq. (12)
r radial position in the column, m
0.08 R radius of the column
Mean = 9.8 s
0.06 Reb bubble Reynolds number, dimensionless
STD = 2.1 s
0.04 Pe = 42.3 ra volume fraction of phase a, dimensionless
t time, s
0.02 Ut terminal velocity of bubble, m/s
0
0 5 10 15 20 25 30 35 40 Greek letters
Circulation Time, s
ma molecular viscosity of phase a, kg/m s
ma,eff effective dynamic viscosity of phase a, kg/m s
0.2 mTa turbulent dynamic viscosity of phase a, kg/m s
Mean = 255.7 cm
Probability
0.15 mb,T
a dynamic viscosity of phase a due to bubble induced
STD = 31.0 cm turbulence, kg/m s
0.1 M = 1.2
ms,T
a dynamic viscosity of phase a due to shear-induced
0.05 turbulence, kg/m s
0 y tangential direction for 2D simulation
0 100 200 300 400 500 600 700 800 ea turbulent dissipation rate of phase a, m2/s3
Trajectory Length, cm ra density of phase a, kg/m3
s surface tension of the gas phase, N/m
0.08 sk, se constants in the standard k–e model in Eq. (10), N/m
Mean = 10.0 s G a random number for fluctuating velocity of tracer
Probability
0.06 STD = 1.3 s particles in Eq. (24)

0.04 Pe = 128 te time scale of the eddy’s lifetime in Eq. (24)
0.02
0 Superscripts and subscripts
0 5 10 15 20 25 30 35
Circulation Time, s b bubble
f fluid (the continuous phase) in Eq. (23)
Fig. 12. PDFs of the circulation time distributions based on single trajectory g gas phase
analysis: (a) CFD simulation results and (b) CARPT results. Operating conditions: l liquid phase
Ug of 0.82 cm/s, top clearance of 3 cm, and bottom clearance of 5 cm.
p tracer particle
T turbulence
a phase index, stands for the liquid or the gas phase
Normalized Occurence Probability
0.18
CFD Simulated Results
0.16 CARPT measured Results
0.14
0.12 Acknowledgements
0.1
The authors greatly appreciate Ansys Inc. for providing the CFX
0.08 license which made this study possible. Part of this work is also
0.06 supported by sponsors of Chemical Reaction Engineering Labora-
0.04 tory (CREL).
0.02
References
0
0 0.2 0.4 0.6 0.8 1
r/R Anderson, T.B., Jackson, R., 1967. A fluid dynamical description of fluidized beds.
Industrial Engineering and Chemistry Fundamentals 6, 527–534.
Becker, S., Sokolichin, A., Eigenberger, G., 1994. Gas–liquid flow in bubble columns
Fig. 13. Radial profile of the normalized occurrence probability for the tracer
and loop reactors: Part II. Comparison of detailed experiments and flow
particle in the airlift column reactor (normalized by the total number of occur-
simulations. Chemical Engineering Science 49, 5747–5762.
rences) at superficial gas velocity of 0.82 cm/s. Bertola, F., Vanni, M., Baldi, G., 2003. Application of computational fluid dynamics to
multiphase flow in bubble columns. International Journal of Chemical Reactor
Engineering v1 (Article A3).
db bubble diameter, cm Borchers, O., Busch, C., Sokolichin, A., Eigenberger, G., 1999. Applicability of the
Db drag force between gas bubbles and the liquid phase standard k–e turbulence model to the dynamic simulation of bubble columns.
Part II: Comparison of detailed experiments and flow simulations. Chemical
Eo Eotvos number, dimensionless Engineering Science 54, 5927–5935.
Fa Saffman lift force for phase a, N CFX5.7 manual, 2004. Solver Theory. Ansys Inc..
Chen, P., Sanyal, J., Dudukovic, D.P., 2005. Numerical simulation of bubble columns Luo, H.-P., Al-Dahhan, M.H., 2010. Local gas holdup in a draft tube airlift bioreactor.
flows: effect of different breakup and coalescence closures. Chemical Engineer- Chemical Engineering Science 65 (15), 4503–4510.
ing Science 60 (4), 1085–1101. Markatos, N.C., 1986. The mathematical modeling of turbulent flows. Applied
Chisti, Y., 1998. Pneumatically agitated bioreactors in industrial and environmental Mathematical Modeling 10, 190–220.
bioprocessing: hydrodynamics, hydraulics, and transport phenomena. Applied Merchuk, J.C., Gluz, M., Mukmenev, I., 2000. Comparison of photobioreactors for
Mechanics Reviews 51 (1), 33–112. cultivation of the microalga Porphyridium sp. Journal of Chemical Technology
Delnoij, E., Kuipers, J.A.M., Van Swaaij, W.P.M., 1999. A three-dimensional CFD and Biotechnology 75 (12), 1119–1126.
model for gas–liquid bubble columns. Chemical Engineering Science 54 (13–14), Mudde, R.F., Van Den Akker, H.E.A., 2001. 2D and 3D simulations of an internal airlift
2217–2226. loop reactor on the basis of a two-fluid model. Chemical Engineering Science 56
Delnoij, E., Kuipers, J.A.M., Van Swaaij, W.P.M., 1997a. Dynamic simulation of (21–22), 6351–6358.
dispersed gas–liquid two-phase flow: effect of column aspect ratio on the flow Pan, Y., Dudukovic, M.P., Chang, M., 1999. Dynamic simulation of bubbly flow in
structure. Chemical Engineering Science 52 (21/22), 3759–3772. bubble columns. Chemical Engineering Science 54 (13–14), 2481–2489.
Delnoij, E., Lammers, F.A., Kuipers, J.A.M., Van Swaaij, W.P.M., 1997b. Dynamic Pan, Y., Dudukovic, M.P., Cheng, M., 2000. Numerical investigation of gas-driven flow
simulation of dispersed gas–liquid two-phase flow using a discrete bubble in 2-D bubble columns. AIChE Journal 46, 434–449.
model. Chemical Engineering Science 52 (9), 1429–1458. Perner-Nochta, I., Posten, C., 2007. Simulations of light intensity variation in
Delnoij, E.A., Kuipers, J.A.M., Van Swaaij, W.P.M., 1997c. Computational fluid photobioreactors. Journal of Biotechnology 131 (3), 276–285.
dynamics applied to gas–liquid contactors. Chemical Engineering Science 52 Pfleger, D., Gomes, S., Gilbert, N., Wagner, H.–G., 1999. Hydrodynamic simulations of
(21/22), 3623–3638. laboratory scale bubble columns fundamental studies of the Eulerian–Eulerian
Drew, D.A., 1983. Mathematical modeling of two-phase flow. Annual Review of Fluid modeling approach. Chemical Engineering Science 54, 5091–5099.
Mechanics 15, 261–291. Pollard, D.J., Ison, A.P., Shamlou, P.A., Saez, A.E., 1998. Reactor heterogeneity with
Garcia Camacho, F., Gomez, Contreras, Fernandez, Acien, Sevilla, Fernandez, Grima, Saccharopolyspora erythraea airlift fermentations. Biotechnology & Bioengineer-
Molina, 1999. Use of concentric-tube airlift photobioreactors for microalgal ing 58, 453–463.
outdoor mass cultures. Enzyme and Microbial Technology 24, 164–172. Pruvost, J., Cornet, J.-F., Legrand, J., 2008. Hydrodynamics influence on light
Grace, J.R., Wairegi, T., Nguyen, T.H., 1976. Shapes and velocities of single drops and conversion in photobioreactors: an energetically consistent analysis. Chemical
bubbles moving freely through immiscible liquids. Transactions of the Institu- Engineering Science 63 (14), 3679–3694.
tion of Chemical Engineers 54, 67. Rammohan, A.R., 2002. Characterization of single and multiphase flows in stirred
Heijnen, J.J., Mulder, A., Weltevrode, R., Hols, J., van Leeuwen, H.L.J.M., 1990. Large tank reactors. D.Sc. Thesis, Washington University, St. Louis, Missouri.
scale anaerobic/aerobic treatment of complex industrial wastewater using Saffman, P.G., 1965. The lift on a small sphere in a slow shear flow. Journal of Fluid
immobilized biomass in fluidized bed gaslift suspension bioreactors. Chemical Mechanics 22, 385.
Engineering & Technology 13, 145–220. Saffman, P.G., 1968. Corrigendum to ‘The lift on a small sphere in a slow shear flow’.
Ishii, M., Zuber, N., 1979. Drag coefficient and relative velocity in bubbly, droplet or Journal of Fluid Mechanics 31, 624.
particulate flows. AIChE Journal 25, 843–855. Sanchez Miron, A., Contreras Gomez, A., Garcia Camacho, F., Molina Grima, E., Chisti,
Jakobsen, H.A., 1993. On the modeling and simulation of bubble column reactors Y., 1999. Comparative evaluation of compact photobioreactors for large-scale
using two-fluid model. Dr. Ing. Thesis 97, NTH, Trondheim. monoculture of microalgae. Journal of Biotechnology 70, 249–270.
Jacobsen, H.A., Sannæs, B.H., Grevskott, S., Svendsen, H.F., 1997. Modeling of vertical Sankaranarayanan, K., Shan, X., Kevrekidis, I.G., Sundaresan, S., 2000. Analysis of
bubble-driven flows. Industrial and Engineering Chemistry Research 36, drag and virtual mass forces in bubbly suspensions using an implicit formula-
4052–4074. tion of the Lattice Boltzmann method, July 20, 2000. In: Proceedings of the
Janssen, M., 2002. Cultivation of microalgae: effect of light/dark cycles on biomass Engineering Foundation Conference on CFD, Quebec.
yield. D.Sc Thesis, Wageningen University, Wageningen, The Netherlands. Sanyal, J., Marchisio, D.L., Fox, R.O., Dhanasekharan, K., 2005. On the comparison
Joshi, J.B., Sharma, M.M., 1979. A circulation cell model for bubble columns. between population balance models for CFD simulation of bubble columns.
Transactions of the Institution of Chemical Engineers 57, 244–251. Chemical Engineering Science 44 (14), p5063–5072.
Joshi, J.B., 2001. Computational flow modeling and design of bubble column Sato, Y., Sekoguchi, K., 1975. Liquid velocity distribution in two-phase bubble flow.
reactors. Chemical Engineering Science 56, 5893. International Journal of Multiphase Flow 2, 79.
Lain, S., Broder, D., Sommerfeld, M., 1999. Experimental and numerical studies of the Schiller, L., Naumann, A., 1935. A drag coefficient correlation. Zeitschrift des Vereins
hydrodynamics in a bubble column. Chemical Engineering Science 54, Deutscher Ingenieure 77, 318.
4913–4920. Sokolichin, A., Eigenberger, G., 1994. Gas–liquid flow in bubble columns and loop
Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent flows. reactors: Part I: Detailed modeling and numerical simulation. Chemical
Computer Methods in Applied Mechanics and Engineering 3, 269–289. Engineering Science 52, 5735.
Lopez de Bertodano, M., 1998. Two fluid model for two-phase turbulent jet. Nuclear Sokolichin, A., Eigenberger, G., Lapin, A., 2004. Simulation of buoyancy driven bubbly
Engineering and Design 179, 65–74. flow: established simplifications and open questions. AIChE Journal 50 (1),
Luo, H.-P., Kemoun, A., Al-Dahhan, M.H., Fernández Sevilla, J.M., Garcı́a Sánchez, J.L., 24–45.
Garcı́a Camacho, F., Molina Grima, E., 2003. Analysis of photobioreactors for Tabib, M.V., Roy, S.A., Joshi, J.B., 2008. CFD simulation of bubble column—an analysis
culturing high value microalgae and cyanobacteria via an advanced diagnostic of interphase forces and turbulence models. Chemical Engineering Journal 139
technique: CARPT. Chemical Engineering Science 58 (12), 2519–2527. (3), 589–614.
Luo, H.-P., Al-Dahhan, M.H., 2004. Analyzing and modeling of photobioreactors by Thakre, S.S., Joshi, J.B., 1999. CFD simulation of bubble column reactors: importance
combining first principles of physiology and hydrodynamics. Biotechnology & of drag force formulation. Chemical Engineering Science 54, 5055–5060.
Bioengineering 85 (4), 382–393. Wiemann, D., Mewes, D., 2005. Calculation of flow fields in two and three-phase
Luo, H.-P., Al-Dahhan, M.H., 2008a. Macro-mixing in a draft tube airlift column via bubble columns considering mass transfer. Chemical Engineering Science 60,
the CARPT technique. Chemical Engineering Science 63 (6), 1572–1585. 6085–6093.
Luo, H.-P., Al-Dahhan, M.H., 2008b. Local characteristics of hydrodynamics in draft Wu, X., Merchuk, J.C., 2001. A model integrating fluid dynamics in photosynthesis
tube airlift bioreactor. Chemical Engineering Science 63 (11), 3057–3068. and photoinhibition processes. Chemical Engineering Science 56, 3527–3538.

Luo 2011

Uploaded by

Copyright:

Available Formats

Luo 2011

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Luo 2011

Uploaded by

Copyright:

Available Formats

Chemical Engineering Science 66 (2011) 907–923

Contents lists available at ScienceDirect

Chemical Engineering Science

Veriﬁcation and validation of CFD simulations for local ﬂow dynamics

1. Introduction following the Lambert–Beer law, the light-exposure history of a cell

For turbulent ﬂow, the dynamic viscosity is usually much larger

where nb is the number of bubbles, CD is a drag coefﬁcient required

pd3 rf d! ! ! manual, 2004):

0.13 m 15 holes (φ1mm) equidistantly

Air Ring Sparger

Fig. 1. Conﬁguration of the studied airlift column reactor.

Fig. 2. Geometry and griding for the 2D steady state simulations.

Degassing boundary conditions are applied on

Draft Tube (having been cut out from the

CARPT Results 2D Coarse Griding CT Results 2D_Coarse Griding

-0.1 0 0.2 0.4 0.6 0.8 1

CARPT Results 2D_Coarse Griding

5.3. Coefﬁcient for shear-induced dynamic viscosity, Cm

Reynolds Shear Stress

0.8 CT = 0.3 exp CT = 0.3 exp

exp: 0.08 CT = 0.3 exp:

0.6 exp: 0.1 CL = 0.0 exp:

Axial Liquid Velocity, m/s

0.7 CT = 0.3 exp: CT = 0.3 exp:

CL = -0.1 Cμ = 0.09 CL = -0.1

1 Ug = 1cm/s 0.35 Ug = 5cm/s

Axial Liquid Velocity, m/s

systems are non-Newtonian and are likely change with time in

H axial position in the column, m

0.06 STD = 1.3 s particles in Eq. (24)

You might also like