Sei sulla pagina 1di 13

*Corresponding author.

Chemical Engineering Science 54 (1999) 5071}5083


Numerical simulation of gas}liquid dynamics in cylindrical bubble
column reactors
Jayanta Sanyal*, Sergio VaH squez, Shantanu Roy, M. P. Dudukovic
Fluent Inc., Lebanon, NH 03766-1442, USA
Chemical Reaction Engineering Laboratory, Department of Chemical Engineering, Washington University, St. Louis, MO 63130, USA
Abstract
In this paper, we have attempted to validate a transient, two-dimensional axisymmetric simulation of a laboratory-scale cylindrical
bubble column, run under bubbly #ow and churn turbulent conditions. The experimental data was obtained via gamma-radiation
based non-invasive #ow monitoring methods, viz., computer automated radioactive particle tracking (CARPT) provided the data on
liquid velocity and turbulence, and computed tomography (CT) determined the gas holdup pro"les. The numerical simulation was
done using the FLUENT software and compares the results from the algebraic slip mixture model, and the two-#uid Euler}Euler
model. Reasonably, good quantitative agreement was obtained between the experimental data and simulations for the time-averaged
gas holdup and axial liquid velocity pro"les, as well as for the kinetic energy pro"les. The favorable results suggest that the simple
two-dimensional axisymmetric simulation can be used for reasonable engineering calculations of the overall #ow pattern and gas
holdup distributions. 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Bubble columns; Axisymmetric simulation; FLUENT; Euler}Euler model; Algebraic slip mixture model
1. Introduction
Bubble columns are contactors in which a discontinu-
ous gas phase in the form of bubbles moves relative to the
continuous liquid phase. As reactors, they are used in
a variety of chemical processes, such as Fischer}Tropsch
synthesis (Kolbel & Ralek, 1980; Srivastava, Rao,
Cinquegrane & Stiegel, 1990), manufacture of "ne chem-
icals (Smidt, Hafner, Jira, Seiber, Sedimeier & Sabel,
1962), oxidation reactions (Hagberg & Krupa, 1976; Sit-
tig, 1967), alkylation reactions (Gehlawat & Sharma,
1970), e%uent treatment (Takahashi, Miyahara
& Nishizaki, 1979), coal liquefaction (Shah, 1981), fer-
mentation reactions, and more recently, in cell cultures
(Katinger, Scheirer, & Kromer, 1979) and production of
single cell proteins (Rosenzweig & Ushio, 1974). The
principal advantages of bubble columns are the absence
of moving parts, leading to easier maintenance, high
interfacial areas and transport rates between the gas and
liquid phases, good heat transfer characteristics, and
large liquid holdup which is favorable for slow liquid-
phase reactions (Shah, Kelkar, Godbole & Deckwer,
1982). Operation of bubble columns as reactors is a!ec-
ted by global operating parameters like gas-phase super-
"cial velocity (the liquid being processed as a batch in
many commercial applications), operating pressure and
temperature, and the liquid height. The actual variables
that in#uence bubble column performance as reactors
are gas holdup distribution, extent of liquid-phase back-
mixing, gas}liquid interfacial area, gas}liquid mass and
heat transfer coe$cients, bubble-size distributions,
bubble coalescence and redispersion rates, and bubble
rise velocities. In industrial operation, the complex two-
phase #ow and turbulence determines the transient and
time-averaged values of the above variables. The lack of
complete understanding of the #uid dynamics makes it
di$cult to improve the performance of a bubble column
reactor by judicious selection and control of the operat-
ing parameters.
The need to establish a rational basis for the inter-
pretation of the interaction of #uid dynamic variables has
been the primary motivation for active research in the
area of bubble column modeling based on computational
#uid dynamics (CFD) tools in the last decade (Kuipers
& van Swaaij, 1998). Various approaches have been
suggested for solving the same fundamental #ow problem
and modeling may be attempted at various levels
0009-2509/99/$- see front matter 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 9 9 ) 0 0 2 3 5 - 3
of sophistication. One may choose to treat both the
dispersed and continuous phases as interpenetrating
pseudo-continua (viz., the Euler}Euler approach, e.g.
Sokolichin & Eigenberger, 1994; Becker, Sokolichin
& Eigenberger, 1994; Ranade, 1992, 1995, 1997) or the
dispersed phase as discrete entities (viz., the Euler}Lag-
range approach: e.g. Lapin & Lubbert, 1994; De-
vanathan, Dudukovic, Lapin & Lubbert, 1995; Delnoij,
Lammers, Kuipers & van Swaaij, 1997a; Delnoij,
Kuipers & van Swaaij, 1997b,c). The simulation may be
done fully transient (e.g. Becker, Sokolichin & Eigenber-
ger, 1994) or only for the steady-state time-averaged
results (e.g. Torvik & Svendsen, 1990; Svendsen, Jakob-
sen & Torvik, 1992; Jakobsen, Svendsen & Hjarbo, 1993,
Ranade, 1997). An appropriate mesh and a robust nu-
merical solver are crucial for getting accurate solutions
(e.g., Sokolichin, Eigenberger, Lapin & Lubbert, 1997).
Fundamental modeling of the "ne-scale phenomena
needed to predict the #ow pattern at the global scale is
also an issue that confronts the modeler (e.g. see Ranade,
1997; Kumar et al., 1995b).
Finally, it is highly imperative to validate the simula-
tion results against carefully designed non-invasive ex-
periments. Unfortunately, reliable data for the #ow
pattern and its dynamics in three-dimensional laborat-
ory-scale bubble columns is not common in the open
literature, so that most of the validation of the CFD
codes done in the past has only been attempted on simple
two-dimensional systems operating in the bubbly #ow
regime (e.g., Becker et al., 1994; Lin, Reese, Hong & Fan,
1996; Delnoij et al., 1997a}c).
In the present work, the experiments were performed
in a 8 in diameter cylindrical air}water bubble column
using the computer automated radioactive particle track-
ing (CARPT) and the gamma-ray computed tomography
(CT) facilities. Both bubbly and churn-turbulent bubble
column #ow was simulated using FLUENT, and predic-
tions of the two-dimensional axisymmetric approxima-
tion are compared against the experimental results.
2. Experimental
At the Chemical Reaction Engineering Laboratory,
Washington University in St. Louis, USA the unique
CARPT-CT facilities allow non-invasive monitoring of
the #ow of two phases in opaque multiphase reactors on
a single platform (Devanathan, 1991; Yang, Devanathan
& Dudukovic, 1992; Kumar, 1994; Degaleesan, 1997).
Computer Automated Radioactive Particle Tracking
(CARPT) is the method employed for measuring the
time-averaged velocity and turbulence parameters of the
liquid phase in the bubble column. In CARPT, one
resorts to tagging the `typical #uid elementa with
a gamma ray source. For instance, in a bubble column
experiment in which the liquid-phase velocity pro"le is of
interest, the liquid phase is tagged with a 2.3 mm dia-
meter, `neutrally-buoyanta, hollow polypropylene
sphere "lled with radioactive Sc-46 (of 250}300 Ci
strength). Subsequently, the motion of this sphere is
monitored using an array of strategically positioned scin-
tillation detectors over a long time span in which the
particle (like any other typical #uid element) visits each
location in the bubble column a large number of times.
Data is acquired over a very long time, typically 18}20 h,
in order to collect su$cient statistics (typically, two mil-
lion or more occurrences in the column) for the tracer
particle to sample instantaneous velocities at all points in
the column. A record of the gamma-ray photon counts at
each detector, and a pre-established calibration between
the detector counts and tracer particle location is used to
reconstruct the precise particle position at each time
instant. Time-di!erencing yields instantaneous velocities,
which when averaged at each spatial location over the
whole time span of the experiment, yield the ensemble-
averaged velocity #ow map. By the ergodic hypothesis,
this is also the time-averaged velocity "eld. The di!erence
between instantaneous and average velocity for each cell
yields the #uctuating component of velocity as a time
series. This time series is then used to construct the
cross-correlation matrix of #uctuating velocity compo-
nents. Trace of this matrix can be interpreted as the total
#uctuating kinetic energy of `turbulencea per unit vol-
ume, while the o!-diagonal components are directly re-
lated to the `turbulent Reynolds' shear stressesa
(Devanathan, 1991; Degaleesan, 1997). Error in recon-
struction of position using the CARPT technique has
been shown to be less than 0.5 cm, and error in spurious
root-mean-square velocities is less than 5 cm s in the
worst case (Degaleesan, 1997).
Computed tomography (CT) can be used to measure
time-averaged phase holdup pro"les in a multiphase re-
actor, and is used here to determine gas holdup pro"les in
the bubble column operated in bubbly and churn}turbu-
lent #ow. By placing a strong gamma-ray source in the
plane of interest and a planar array of scintillation de-
tectors in the same line on the other side of the reactor,
one measures attenuation of the beam of gamma radi-
ation. The attenuation is a function of the line-averaged
holdup distribution along the path of the beam. Many
such `projectionsa (e.g., 4851 for an 8 in. column) are
obtained at di!erent angular orientations around the
reactor. The complete set of projections is then used to
back-calculate the cross-sectional distribution of densit-
ies. Since the density at any point in the cross-section is
a sum of densities of individual phases weighted by their
volume fractions, the cross-sectional density distribution
of any particular phase can be uniquely recovered (if only
two phases are present).
The total scanning time in the CREL scanner is about
2 h, thus the scanned image provides a time-averaged
cross-sectional distribution of mixture density. The CT
5072 J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083
setup at CREL provides a spatial resolution of better
than 5 mm, and a density resolution of about
0.04 g cm` (Kumar, Moslemian & Dudukovic, 1995).
The results of the two-dimensional holdup distribution
can be subsequently averaged azimuthally for direct
comparison with the results of a suitable axisymmetric
simulation. For cylindrical bubble columns, the axisym-
metric pro"le of the gas holdup measured via CT is
typically parabolic, with the highest gas holdup in the
center of the column. At lower super"cial gas velocity, the
pro"le is #atter across the cross-section and becomes
signi"cantly steeper with the increase in gas velocity.
This fact is used to discriminate between bubbly #ow
regime and churn-turbulent regime of operation of
a bubble column. The liquid circulation in a bubble
column is driven by buoyancy of the gas, thus the circula-
tion velocities are signi"cantly higher (with steeper mean
axial liquid velocity pro"les) in the churn}turbulent #ow
regime as compared to the bubbly #ow regime.
Details of the experimental setup and procedure
for the particle tracking and the tomographic techniques
may be found elsewhere (Devanathan, Moslemian
& Dudukovic, 1990; Devanathan, 1991; Moslemian,
Devanathan & Dudukovic, 1992; Yang et al.,
1992; Kumar, 1994; Kumar et al., 1995; Kumar, Moslem-
ian & Dudukovic, 1997; Degaleesan, 1997; Roy,
Chen, Degaleesan, Gupta, Al-Dahhan & Dudukovic,
1998).
3. Numerical Simulation
In the present work, the #ow in the bubble column
reactor was modeled using two di!erent approaches
incorporated in the FLUENT software * the Eulerian
multiphase model and the algebraic slip mixture
model (ASMM). Although both models are used
to predict multiphase #ows, there are fundamental
di!erences in their respective approaches which are
outlined here.
3.1. The Eulerian multiphase model
In the Eulerian two-#uid approach, the di!erent
phases are treated mathematically as interpenetrating
continua. The derivation of the conservation equations
for mass, momentum and energy for each of the
individual phases is done by ensemble averaging
the local instantaneous balances for each of the phases
(Anderson & Jackson, 1967). The basic assumptions of
this formulation used in the present computations are as
follows:
E All phases are treated as interpenetrating continua and
the probability of occurrence of any one phase in
multiple realizations of the #ow is given by the instan-
taneous volume fraction of that phase at that point.
Sum total of all volume fractions at a point is identi-
cally unity.
E Both #uids are treated as incompressible, and a single
pressure "eld is shared by all phases.
E Continuity and momentum equations are solved for
each phase.
E Momentum transfer between the phases is modeled
through a drag term, which is a function of the local
slip velocity between the phases. A characteristic dia-
meter is assigned to the dispersed phase gas bubbles,
and a drag formulation based on a single sphere sett-
ling in an in"nite medium is used.
E Turbulence in either phase is modeled separately.
The conservation equations can be written as follows:
Continuity (kth phase):
c
ct
(:
I
j
I
)#V) (:
I
j
I
u
I
)"
L

N
m
NI
. (1)
Momentum (kth phase):
c
ct
(:
I
j
I
u
I
)#V) (:
I
j
I
u
I
u
I
)"!:
I
Vp#V) t

I
#F
I
#
L

N
(K
NI
(u
N
!u
I
)#m
NI
u
NI
), (2)
where t

is the kth phase stress}strain tensor, whose


components are given by
t
IGH
":
I
j
I

cu
IH
cx
H
#
cu
IH
cx
G

!2
3
:
I
j
I
o
GH
cu
IG
cx
G
. (3)
The fourth term on the right-hand side of Eq. (2) repres-
ents the interphase drag term, with K
NI
being the mo-
mentum exchange coe$cient between the pth and the kth
phases. The evaluation of the needed drag coe$cient
requires the bubble Reynolds number which is based on
the local slip velocity for a single sphere of constant
diameter sedimenting in stagnant #uid. In the present
computations, the drag coe$cient, K
NO
is based on the
generalized correlations (Morsi & Alexander, 1972).
The turbulence in the continuous phase is modeled
through a set of modi"ed k}c equations with extra terms
that include interphase turbulent momentum transfer
(Launder & Spalding, 1974; Elghobashi & Abou-Arab,
1983), supplemented with extra terms that include the
interphase turbulent momentum transfer. This term can
be derived exactly from the instantaneous equation of the
continuous phase and involves the continuous-dispersed
velocity covariance. The turbulence quantities for the
dispersed phase in FLUENT are based on characteristic
particle relaxation time and Lagrangian time scales (Sim-
onin & Viollet, 1990). For the dispersed gas phase, the
turbulence closure is e!ected through correlations from
J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083 5073
the theory of dispersion of discrete particles by homo-
geneous turbulence (Tchen, 1947).
The equations discussed above are solved using an
extension of the SIMPLE algorithm (Patankar, 1980).
The momentum equations are decoupled using the full
elimination algorithm (FEA). Using SIMPLE-FEA
(Spalding, 1980), the variables for each phase are elimi-
nated from the momentumequations for all other phases.
The pressure correction equation is obtained by sum-
ming the continuity equations for each of the phases. The
equations are then solved in a segregated, iterative
fashion and are advanced in time. At each time step, with
an initial guess for the pressure "eld, the primary- and
secondary-phase velocities are computed. These are used
in the pressure correction equation (continuity), and
based on the discrepancy between the guessed pressure
"eld and the computed "eld, the velocities, holdups and
#uxes are suitably modi"ed to obtain convergence in an
iterative manner.
3.2. The Algebraic Slip Mixture Model
The Algebraic Slip Mixture Model (ASMM) has an
underlying philosophy that is quite distinct from the
Euler}Euler two-yuid model (Manninen, Taivassalo
& Kallio, 1996). The principal assumptions in this formu-
lation are as follows:
E It models the phases as two interpenetrating continua,
with the probability of existence of each phase at
a point in the computational domain given by its
respective volume fraction (holdup). In general, the
two phases move at di!erent velocities.
E A single equation is solved for continuity of the mix-
ture and a single equation is solved for the momentum
of the mixture.
E Motion of each phase relative to the center of mass of
the mixture in any control volume is viewed as a di!u-
sion of that phase; this introduces the concept of a dif-
fusion velocity of each phase (which is analogous and
directly related to the slip velocity and the drift velo-
city, as referred to in the classical drift #ux model for
a mixture, Wallis, 1969).
E The Reynolds'-averaged mixture momentum equation
has a term, called the diwusion stress, which originates
because of the relative slip between the two phases.
This requires closure in terms of the di!usion velocity
of each phase (or, equivalently, the drift or the slip
velocity between the phases). In the ASMM, this is
supplied by assuming that the phases are in local
equilibrium over short spatial length scales. This means
that the dispersed phase entity (bubble, particle) al-
ways slips with respect to the continuous phase at its
terminal Stokes' velocity in the local accelaration "eld.
E The diwusion stress term is also the only term in which
the phase volume fractions appear explicitly. In order
to back out the individual phase velocities and volume
fraction at the end of the computation at each time
step, it is necessary to solve a di!erential equation for
the volume fraction of the dispersed phase coupled
with the solution of the mixture equations. This equa-
tion is obtained from the equation of continuity for the
dispersed phase.
E Finally, the turbulent stress term in the mixture equa-
tion is closed by solving a k}c model for the mixture
phase.
Based on these assumptions, the "nal equations of the
ASMM model are formulated as follows.
Equation of continuity for the mixture:
c
ct
(j
K
)#
c
cx
G
(j
K
u
KG
)"0. (4)
Equation for mixture momentum ( jth component):
c
ct
(j
K
u
KH
)#
c
cx
G
(j
K
u
KG
u
KH
)"!
cp
cx
H
#
c
cx
G
;j
K

cu
KG
cx
H
#
cu
KH
cx
G

#j
K
g
H
#F
H
#
c
cx
G
L

I
:
I
j
I
u
"IG
u
")H
. (5)
Volume fraction equation for the secondary phase:
c
ct
(:
Q
j
Q
)#
c
cx
G
(:
Q
j
Q
u
KG
)"!
c
cx
G
(:
Q
j
Q
u
"QG
). (6)
The above equations are formulated in terms of the
mixture density, j
K
, mixture viscosity, j
K
, and the mass-
averaged mixture velocity, u
K
, which are de"ned as fol-
lows:
j
K
"
L

I
:
I
j
I
, j
K
"
L

I
:
I
j
I
, u
K
"

L
I
:
I
j
I
u
I
j
K
. (7)
u
"I
is the drift velocity of the kth phase with respect to
the mixture center of mass, and is related to the slip
velocity with respect to the continuous phase in the
following manner:
u
"I
"u
I
!u
K
"v
IA
!
1
j
K
L

G
:
G
j
G
v
GA
, (8)
where v
IA
is the slip velocity of the kth phase with respect
to the continuous phase. In the ASMM, the slip velocity
is calculated based on the assumption of local equilib-
rium between the phases over short spatial scales. The
5074 J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083
Table 1
Parameters used in the numerical simulation
Grid 300 (axial);19 (radial)
Cell size 0.66 cm (axial);0.5 cm (radial) (uniform grid)
Time step 0.01 s (Euler}Euler); 0.005 s (ASMM)
Iterations per time step for convergence around 25
Under-relaxation parameters Euler}Euler: 0.6 (pressure), 0.4 (velocities); ASMM: 0.3 (pressure), 0.7 (velocities)
Bubble size 0.5 cm
Drag formulation Single sphere drag correlation (Morsi & Alexander, 1972)
Turbulence model Standard k} model (Elghobashi & Abou-Arab, 1983)
convection terms of the dispersed phase are assumed to
be of similar magnitude as the convection terms of the
mixture (Manninen et al., 1996). Thus:
v
IA
"
(j
K
!j
I
)d`
I
18j
A
f
g!
Du
K
Dt
, (9)
where
f"

1#0.05Re""`` Re(1000,
0.018Re Re*1000.
(10)
The complete set of the above equations are discretized
in an unstructured domain, and solved in a segregated
fashion using an implicit time stepping algorithm. The
pressure correction is e!ected through the SIMPLEC
algorithm (van Doormal & Raithby, 1984). At each time
step and in each computational cell, the volume fraction
of the dispersed phase is used to back-calculate from the
mixture velocity the velocities of the individual phases.
4. Results and Discussion
The simulations were performed for a 8 in. (19.0 cm
i.d.) air}water bubble column. Experimental data using
the CARPT technique (Degaleesan, 1997) and the CT
scanner (Kumar, 1994) were available at the conditions of
interest. Gas super"cial velocities simulated were 2.0 and
12.0 cm s, characteristic of bubbly #ow and churn}tur-
bulent #ow regime, respectively. The column contained
a batch liquid with unexpanded height of 104.5 and
95.0 cm, respectively. The distributor used in the experi-
ments was a perforated plate, with 0.33 mm diameter
holes in a square pitch, with an open area of 0.1%.
Distributor e!ects have been shown to a!ect (Kumar,
1994; Degaleesan, 1997) the overall #ow pattern in the
bubble column. However, in the present simulations,
distributor e!ects have been ignored and the perforated
plate has been modeled as a uniform surface source of the
gas phase.
The numerical simulation was performed on a 300
(axial);19 (radial) rectangular grid. Eqs. (1), (2), (4)}(6)
were solved in a transient fashion with a time step of
0.01 s for the Euler}Euler simulations and 0.005 s for the
ASMM. A number of sub-iterations were performed
within each time step to ensure continuity. It took ap-
proximately 25 iterations per time time step to converge
the residuals by three orders of magnitude or more for
both the Euler}Euler and the ASMM. The under-relax-
ation parameters were set to 0.6 for pressure and 0.4 for
velocities for the Euler}Euler case and 0.3 for pressure
and 0.7 for velocities for the ASMM. The power-law
discretization scheme was used for the Euler}Euler
model and a "rst-order upwinding scheme was used for
the ASMM. Inlet boundary conditions are assigned at
the distributor, and outlet conditions at the free surface.
No-slip conditions were applied at the wall, and sym-
metry conditions at the central axis of the column.
A bubble size of 5 mm, typical of air}water bubble col-
umns under atmospheric pressure operating at these
super"cial velocities with a perforated plate sparger, was
used in the simulations. Parameters used in the simula-
tion are summarized in Table 1.
In Fig. 1, a set of vector plots of the time-averaged
liquid velocity pro"le obtained via the CARPT technique
is shown (Degaleesan, 1997), at a gas super"cial velocity
of 12 cm s. The results are presented in four vertical
planes at four di!erent angular orientations. Fig. 2 shows
vector plots for the same experiment, at di!erent cross-
sectional planes. One notes that irrespective of angular
orientation (Fig. 1), in a time-averaged sense, the liquid
goes up at the center of the column and descends at the
walls in a single circulation loop. The circulation shows
a symmetric r}z dependence alone, with the instan-
taneous azimuthal component of velocity not being sig-
ni"cant in determining the time-averaged #ow pro"le
(Degaleesan, 1997). Further, azimuthal}radial compo-
nents of time-averaged liquid velocity (Fig. 2) are signi"-
cantly smaller than the axial components (Fig. 1). Slight
asymmetries seen just above the distributor and below
the free surface are ascribed to asymmetric distributor
and gas disengagement e!ects, respectively. Data col-
lected using the CARPT technique under a variety of
operating conditions, distributors and column sizes, and
across various #ow regimes con"rms this picture (De-
galeesan, 1997); Figs. 1 and 2 being representative cases.
The message is that while the #ow in bubble columns is
highly turbulent and chaotic, and the transient #ow is
J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083 5075
Fig. 1. Velocity vector plots (longitudinal views) for column diameter of 19 cm: ;
E
"2.0 cm s (Degaleesan, 1997).
asymmetric about the central axis, the time-averaged
#ow is not. This motivates the use of a two-dimensional
axisymmetric coordinates (with the governing equations
averaged in the azimuthal direction) as a simpli"ed
means of simulating the three-dimensional #ow.
Fig. 3 presents a vector plot of the simulated #ow
pro"le, using the Euler}Euler two-#uid model at the gas
super"cial velocities of the 2.0 and 12.0 cm s (the vec-
tor plots are `mirroreda about the axis of symmetry for
direct qualitative comparison with Fig. 1). Similar quali-
tative results are seen with the ASMM simulations as
well. The time-averaged simulation pro"les (axisymmet-
ric) are qualitatively similar to the experimental #ow
pro"les observed from the CARPT experiments (Fig. 1).
One observes a single circulation loop with the liquid
ascending at the center of the column and descending at
the walls. Naturally, no asymmetries are observed at the
distributor because the gas introduction in the simula-
tion is through an uniform plate, and a real sparger has
not been modeled.
5076 J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083
Fig. 2. Velocity vector plots (cross-sectional views) for column diameter of 19 cm: ;
E
"2.0 cm s (Degaleesan, 1997).
It is worth noting here that imposition of the symmetry
boundary condition at r"0 causes the liquid #ow to
develop very quickly (around "rst 5 s of real time) and
reach its long-time pattern. In contrast, the real experi-
ment is characterized by a highly dynamic #ow, with
three-dimensional vortical bubble swarms which cause
smaller bubbles and liquid to be trailed in their wakes.
Instantaneous #ow is never truly axisymmetric in reality,
while the simulation is tailored to be so. Thus, one is
unable to capture realizations of the true instantaneous
velocities (which can have signi"cant azimuthal compo-
nents), and hence all the true time-scales in the problem.
Consequently, the true swirling bubble swarms motions,
as well as time scales of such phenomena, will be missed
by axisymmetric models.
In Fig. 4 the time-averaged liquid axial velocity pro-
"les are compared against the time-averaged velocity
pro"les obtained by CARPT. Fig. 5 shows the compari-
son for gas holdup pro"les obtained from the simulations
and from computed tomography (CT). Results presented
are at a height of 53 cm above the distributor, typical of
the fully developed region of the #ow. All the simulations
J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083 5077
Fig. 3. Velocity vector plots obtained from simulations: (a) ;
E
"2.0 cm s (two #uid) (b) ;
E
"12.0 cm s (two #uid). (The r}z plane has been
`mirroreda along the axis of symmetry to show the vector "eld across the diameter.)
showed that there is no signi"cant axial variation in any
of the variables, in the zone of developed #ow. The
centerline axial velocity is overpredicted by both the
Euler}Euler two-#uid model, as well as the ASMM.
However, the general shape of the pro"le is well captured
and the discrepancy in the model predicted and
CARPT-measured time-averaged axial liquid velocity di-
minishes as one moves radially outwards in the column.
5078 J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083
Fig. 4. Comparison of axial liquid velocity pro"les.
One is also able to predict the cross-over point (i.e., the
radial location where the axial velocity component be-
comes zero) reasonably well.
In Fig. 5, one observes that in the bubbly #ow regime
(;
E
"2 cm s), both the ASMM and the two-#uid
models agree in their prediction of the gas holdup pro"le
and seem to predict the mean holdup pro"le reasonably
well. (Unfortunately, the experimental data for gas hold-
up at this condition were not of the highest accuracy.) In
the churn}turbulent regime, however, there is still some
discrepancy between the predictions of the two models
and the data seems to be bracketed by the predictions.
The roughly parabolic pro"le observed in the experi-
mental data is seen in the simulation results as well. The
higher volume fraction of gas at the center of the column
drives the liquid #ow upwards at a high velocity, due to
gas buoyancy, and the liquid, being in batch mode, re-
turns downwards at the periphery.
Comparison of kinetic energy pro"les (obtained by
solution of the k}c model in the simulations, and those
measured via CARPT, also shows good agreement in the
shape of the curves (Fig. 6). Kinetic energy pro"les typi-
cally exhibit a maximum around the cross-over point, due
to large gradients and large #uctuations in the liquid
velocity. In the bubbly #ow regime this e!ect is not very
signi"cant (because of suppressed turbulence) and the
turbulent kinetic energy is practically #at as a function of
radius. These e!ects are clearly captured in the simula-
tion results presented in Fig. 6. It may be noted that for
the ASMM, the kinetic energy plotted in Fig. 6 is the
mixture kinetic energy in contrast to the liquid phase
turbulent kinetic energy. However, it can be readily
shown that due to the signi"cantly lower gas-phase den-
sity, liquid-phase inertia predominates and is the domi-
nant contributor to the mixture kinetic energy.
Quantitative agreement between the two-#uid model and
data is particularly good at churn}turbulent conditions.
Based on what we see in Fig. 6 one may argue that in
any bubble column simulation in which the mean velo-
city pro"le and the gas holdup pro"le is reasonably
predicted, one should also expect to observe a reasonable
comparison of the overall kinetic energy pro"les. The
total energy input to the systemis through the gas in#ow,
and a simple overall energy balance reveals that this total
input energy is dissipated as work done against the
distributor, walls, and the #uid-phase viscosity and tur-
bulence. If the "rst two e!ects are small (distributors may
be properly modeled in a more sophisticated simulation),
then a reasonable prediction of the mean velocity and
holdup pro"les necessarily implies that the remaining
kinetic energy (both in the simulation as well as in the
real experiment) must be accounted for as the `turbulenta
kinetic energy. If the pertinent physics is properly
modeled, then this should result in good prediction of the
kinetic energy proxles as well.
Turbulent kinetic energy referred to in models like the
two-phase k}c formulation arises from the turbulence
microscale, while that obtained from experiments like
J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083 5079
Fig. 5. Comparison of gas holdup pro"les.
CARPT (Fig. 6) lumps energies from all the #uctuations
about the time-averaged (over 18}20 h) mean pro"le,
sampling more e$ciently the larger scales. (For example,
it is estimated that the `CARPT tracer particlea cannot
respond to the turbulence #uctuations above 20}25 Hz in
frequency.) In the present type of axisymmetric simula-
tions (in which we attempt to predict the 18}20 h time-
averaged pro"les), the `turbulenta kinetic energy (i.e., all
the #ow energy that is not due to the mean #ow) is forced
to represent the large-scale turbulent kinetic energy (as
measured by CARPT). In other words, the 2D axisym-
metric simulation imposes all the turbulence time scales
smaller than the total averaging time to contribute to the
turbulent kinetic energy at the `microscalea (hence, cap-
tured by the k}c model). In general, however (e.g. from
other more sophisticated simulations), one would need to
compare the right scales between the simulation and the
experiment for a meaningful comparison of kinetic ener-
gies. Good comparison in Fig. 6 is not merely fortuitous,
but is actually very consistent with our understanding,
expectation and development of the the two-dimensional
axisymmetric model, as well as the CARPT technique.
Similar assertions would hold for steady-state model
simulations (e.g., Ranade, 1997) as well.
It however, does not necessarily follow that the entire
turbulence "eld is well captured, even if the mean
pro"les are. This is because turbulence is a multi-scale
phenomenon, with extremely complex energy cascading
in multiphase #ows. Models like the k}c formulation
phenomenologically model turbulent kinetic energy pro-
duction at large scales and its dissipation as small scales
(Tennekes & Lumley, 1972). They do not "x the scale
information, neither spatial correlations between velocity
#uctuations between two directions, nor autocorrela-
tions (in time).
Thus, good prediction of kinetic energy pro"les does
not necessarily imply good predictions of Reynolds' stres-
ses (i.e., quantities dependent on correlations between
various components of velocity #uctuations), for in-
stance, or of turbulent eddy di!usivities (i.e., quantities
dependent on autocorrelations in time). For either of
these quantities, our present models showed reasonable
order-of-magnitude comparisons and the radial trends
were also captured, but exact values were not. More
5080 J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083
Fig. 6. Comparison of turbulent kinetic energy pro"les.
sophisticated models with transport equations for the
entire second or higher-order turbulence velocity correla-
tions may be required for this purpose.
5. Conclusions
In summary, the usefulness of the two-dimensional
axisymmetric models has to be acknowledged. They pro-
vide good engineering descriptions, and can be used
reliably for approximately predicting the time-averaged
#ow and holdup patterns in bubble columns. We have
validated the models in di!erent #ow regimes, as well as
with di!erent fundamental modeling approaches to de-
scribe the #ow (i.e., ASMM versus Euler}Euler two #uid
models). We have also shown that a reasonable choice of
turbulence description is able to predict the kinetic en-
ergy pro"les, and must do so in a self-consistent model.
Further improvements in the turbulence model, or in
description of two-point correlations, is likely to improve
the description of the other turbulence parameters such
as shear stresses and turbulent di!usivities.
A fully transient three-dimensional model is necessary
to capture the transient #ow structures in the bubble
column, which are in general, not axisymmetric and have
a signi"cant azimuthal component. It would be interest-
ing to examine the relative importance of these phe-
nomena in determining the overall #ow pattern, and
more importantly, the overall reactor performance.
A satisfactory answer to these issues can only be deter-
mined through a detailed comparison of #ow, turbulence
and reactor performance variables between experiment,
two-dimensional codes and three-dimensional models.
Presently, work is in progress in this area and we hope to
present some of our results in this "eld in a subsequent
communication.
Notation
a local acceleration, cm s`
C
"
drag coe$cient, dimensionless
d particle (bubble) diameter, cm
f friction factor, dimensionless
J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083 5081
F external body force per unit volume, dyne cm`
g acceleration due to gravity ("981 cm s`)
K momentum exchange coe$cient, g cm` s
m mass transfer rate per unit volume between
phases, g cm` s
p pressure, dyne cm`
Re Reynolds number, dimensionless
u velocity, cm s
v slip velocity (mixture model), cm s
t time, s
x spatial coordinate, cm
Greek letters
: volume fraction (holdup) of phase
o Kronecker's delta
j density of phase, g cm`
j e!ective viscosity, g cm s
t stress tensor, dyne cm`
Superscripts and subscripts
c continuous phase (mixture model)
D di!usion variable (mixture model)
i coordinate index
j coordinate index
k phase index
m mixture variable
n number of phases
p phase index (distinct from k)
s secondary phase (mixture model)
6. For Further Reading
The following reference is also of interest to the reader:
Lin et al., 1996.
Acknowledgements
The authors would like to thank Dr. Sailesh B. Kumar
and Dr. Sujatha Degaleesan for sharing their experi-
mental data, and Fluent, Inc. for their code. We express
our sincere appreciation to Dr. S. Subbiah of Fluent, Inc.
for o!ering constructive suggestions to improve the paper.
References
Anderson, T. B., & Jackson, R. (1967). A #uid dynamical description of
#uidized beds. Industrial Engineering and Chemistry Fundamentals, 6,
527}534.
Becker, S., Sokolichin, A., & Eigenberger, G. (1994). Gas}liquid #ow in
bubble columns and loop reactors: Part II: Comparison of detailed
experiments and #ow simulation. Chemical Engineering Science, 49,
5747.
Delnoij, E., Lammers, F. A., Kuipers, J. A. M., & van Swaaij, W. P. M.
(1997a). Dynamic simulation of dispersed gas}liquid two-phase #ow
using a discrete bubble model. Chemical Engineering Science, 52(9),
1429}1458.
Delnoij, E., Kuipers, J. A. M., & van Swaaij, W. P. M. (1997b). Dynamic
simulation of gas}liquid two-phase #ow: e!ect of column aspect
ratio on the #ow structure. Chemical Engineering Science, 52(21/22),
3759}3772.
Delnoij, E., Kuipers, J. A. M., & van Swaaij, W. P. M. (1997c). Com-
putational #uid dynamics applied to gas}liquid contactors. Chem-
ical Engineering Science, 52(21/22), 3623}3638.
Degaleesan, S. (1997). Fluid dynamic measurements and modeling of
liquid mixing in bubble columns. D.Sc. thesis, St. Louis, Missouri,
USA: Washington University.
Devanathan, N., Moslemian, D., & Dudukovic, M. P. (1990). Flow
mapping in bubble columns using CARPT. Chemical Engineering
Science, 45, 2285}2291.
Devanathan, N. (1991). Investigation of liquid hydrodynamics in
bubble columns via computer automated radioactive particle track-
ing (CARPT). D. Sc. thesis, St. Louis, Missouri, USA: Washington
University.
Devanathan, N., Dudukovic, M. P., Lapin, A., & Lubbert, A. (1995).
Chaotic #ow in bubble column reactors. Chemical Engineering
Science, 50, 2661.
Elghobashi, S. E., & Abou-Arab, T. W. (1983). A two-equation
turbulence model for two-phase #ows. Physics of Fluids, 26(4),
931}938.
Lin, T.-J., Reese, J., Hong, T., & Fan, L.-S. (1996a). Quantitative
analysis and computation of two-dimensional bubble columns. The
American Institute of Chemical Engineers Journal, 42(2), 301}318.
Gehlawat, J. K., & Sharma, M. M. (1970). Alkylation of phenols with
isobutylene. Journal of Applied Chemistry, 20, 93.
Hagberg, C. G., & Krupa, F. X. (1976). A mathematical model for
bubble column reactors and applications of it to improve a cumene
oxidation process. Proceedings of Fourth International Symposium on
Chemical Reactor Engineering, Heidelberg, Germany.
Jakobsen, H. A., Svendsen, H. F., & Hjarbo, K. W. (1993). On the
prediction of local #ow structures in internal loop and bubble
column reactors using a two-#uid model. Computer Chemical Engin-
eering (Supplement to European Symposium on Computer Aided Pro-
cess Engineering) 17, S531}S536.
Katinger, H. W. D., Scheirer, W., & Kromer, E. (1979). Bubble column
reactor for mass propagation of animal cells in suspension culture.
German Chemical Engineering, 2, 31.
Kolbel, H., & Ralek, M. (1980). The Fischer}Tropsch synthesis in the
liquid phase. Catalysis in Review-Science Engineering, 27(2), 225.
Kuipers, J. A. M., & van Swaaij (1998). Computational #uid dynamics
applied to chemical reaction engineering. Advances in Chemical
Engineering, 24, 227}328.
Kumar, S. B. (1994). Computed tomographic measurements of void frac-
tion and modeling of the yow in bubble columns. Ph.D. thesis, Florida
Atlantic University: Boca Raton, FL, USA.
Kumar, S. B., Moslemian, D., & Dudukovic, M. P. (1995). A -ray
tomographic scanner for imaging voidage distribution in two-phase
systems. Flow Measurement Instrumentation, 6(1), 61}73.
Kumar, S. B., Moslemian, D., & Dudukovic, M. P. (1997). Gas-holdup
measurements in bubble columns using computed tomography.
The American Institute of Chemical Engineers Journal, 43(6),
1414}1425.
Kumar, S., VanderHeyden, W. B., Devanathan, N., Padial, N. T.,
Dudukovic, M. P., & Kashiwa, B. A. (1995b). Numerical simulation
and experimental veri"cation of the gas}liquid #ow in bubble col-
umns. The American Institute of Chemical Engineers Symposium
Series, 42(9), 11.
Launder, B. E., & Spalding, D. B. (1974). The numerical computation of
turbulent #ows. Computer Methods in Applied Mechanical Engineer-
ing, 3, 269}289.
5082 J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083
Lapin, A., & Lubbert, A. (1994). Numerical simulations of the dynamics
of two-phase gas}liquid #ows in bubble columns. Chemical Engin-
eering Science, 49, 3661.
Lin, T.-J., Reese, J., Hong, T., & Fan, L.-S. (1996). Quantitative analysis
and computation of two-dimensional bubble columns. The Ameri-
can Institute of Chemical Engineers Journal, 42(2), 301}318.
Morsi, S. A., & Alexander, A. J. (1972). An investigation of particle
trajectories in two-phase #ow systems. Journal of Fluid Mechanics,
55(2), 193}208.
Moslemian, D., Devanathan, N., & Dudukovic, M. P. (1992). Radioac-
tive particle tracking technique for investigation of phase recircula-
tion and turbulence in multiphase systems. Review Science Instru-
mentation, 63(10), 4361}4372.
Manninen, M., Taivassalo, V., & Kallio, S. (1996). On the mixture model
for multiphase yow. Technical Research Center of Finland: VIT
Publications.
Patankar, S.V. (1980). Numerical heat transfer and two-phase yow. Wash-
ington DC: Hemisphere.
Ranade, V. V. (1992). Numerical simulation of dispersed gas}liquid
#ows. Sadhana, 17, 237}273.
Ranade, V. V. (1995). Computational #uid dynamics for reactor engin-
eering. Reviews in Chemical Engineering, 11, 229}289.
Ranade, V. V. (1997). Modeling of turbulent #ow in a bubble column
reactor. Transactions of the Institution of Chemical Engineers, 75A,
14}23.
Rosenzweig, M., & Ushio, S. (1974). Protein from methanol Chemical
Engineering, 62.
Roy, S., Chen, J., Degaleesan, S., Gupta, P., Al-Dahhan, M. H.,
& Dudukovic, M. P. (1998). Non-invasive #ow monitoring in
opaque multiphase systems with CARPT and CAT. In Proceedings
of the FEDSM'98 * ASME yuids engineering division summer meet-
ing. Washington DC: USA.
Sokolichin, A., & Eigenberger, G. (1994). Gas}liquid #ow in bubble
columns and loop reactors: Part I: Detailed modeling and numer-
ical simulation. Chemical Engineering Science, 52, 5735.
Simonin, C., & Viollet, P. L. (1990). Predictions of an oxygen droplet
pulverizaton in a compressible subsonic co#owing hydrogen #ow.
Numerical Methods for Multiphase Flows, FED-91, 65.
Sokolichin, A., Eigenberger, G., Lapin, A., & Lubbert, A. (1997). Dy-
namic simulation of gas}liquid two-phase #ows * Euler/Euler
versus Euler/Lagrange. Chemical Engineering Science, 52, 611.
Spalding, D. B. (1980). Numerical computation of multi-phase
#uid #ow and heat transfer. In C. Taylor, & K. Margar, Recent
Advances in Numerical Methods in Fluids (pp. 139}167). UK:
Pineridge Press.
Svendsen, H. F., Jakobsen, H. A., & Torvik, R. (1992). Local #ow
structure in internal loop and bubble column reactors. Chemical
Engineering Science, 47(13/14), 3297}3304.
Shah, Y. T. (1981). Reaction engineering in direct coal liquefaction. USA:
Addison-Wesley.
Shah, Y. T., Kelkar, B. G., Godbole, S. P., & Deckwer, W.-D.
(1982). Design parameter estimations for bubble column
reactors. The American Institute of Chemical Engineers Journal, 28(3),
353}379.
Sittig, M. (1967). Organic chemical process encyclopaedia. USA: Noyes
Dev. Corp.
Smidt, J., Hafner, W., Jira, R., Seiber, R., Sedimeier, J., & Sabel, A.
(1962). Olefunoxydation mit Palladiumchlorid-Katalysatoren.
Angew. Chemie, 74, 93.
Srivastava, R. D., Rao, V. U. S., Cinquegrane, G., & Stiegel, G. J. (1990).
Catalysts for Fischer}Tropsch. Hydrocarbon Proceedings, February
(pp. 59}68).
Takahashi, T., Miyahara, T., & Nishizaki, Y. (1979). Separation of oily
water by bubble columns. Journal of Chemical Engineering Japan,
12, 394.
Tchen, C. M. (1947). Mean value and correlation problems connected with
the motion of small particles suspended in a turbulent yuid. Ph.D.
thesis, TU Delft, Netherlands.
Tennekes, H., & Lumley, J.L. (1972). A ,rst course in turbulence.
Cambridge, MA, USA: MIT Press.
Torvik, R., & Svendsen, H. F. (1990). Modeling of slurry reactors:
A fundamental approach. Chemical Engineering Science, 45,
2325.
Van Doormal, J. P., & Raithby, G. D. (1984). Enhancements of the
SIMPLE method for predicting incompressible #uid #ows. Numer-
ical Heat Transfer, 7, 147}163.
Wallis, G. (1969). One-dimensional two-phase yow. New York, USA:
McGraw-Hill.
Yang, Y. B., Devanathan, N., & Dudukovic, M. P. (1992). Liquid
backmixing in bubble columns via computer automated radioactive
particle tracking (CARPT). Chemical Engineering Science, 47,
2859}2864.
J. Sanyal et al. / Chemical Engineering Science 54 (1999) 5071}5083 5083

Potrebbero piacerti anche