You are on page 1of 9

Particuology 19 (2015) 124132

Contents lists available at ScienceDirect

Particuology
journal homepage: www.elsevier.com/locate/partic

Short communication

CFD simulation of internal-loop airlift reactor using EMMS drag model


Tingting Xu a,b , Xuedong Jiang a,c , Ning Yang a, , Jiahua Zhu d
a

State Key Laboratory of Multi-Phase Complex Systems, Institute of Process Engineering, Chinese Academy of Sciences, Beijing 100190, China
University of Chinese Academy of Sciences, Beijing 100049, China
c
Department of Chemical Engineering, Xian Jiaotong University, Xian 710049, China
d
School of Chemical Engineering, Sichuan University, Chengdu 610065, China
b

a r t i c l e

i n f o

Article history:
Received 17 February 2014
Accepted 22 April 2014
Keywords:
Computational uid dynamics
Internal-loop
Airlift
Multi-scale
Multiphase ow
Hydrodynamics

a b s t r a c t
The simulation of internal-loop airlift reactors is challenging because complex meso-scale structures exist
in different sections of the reactor, separated by the draft tube. This paper reports on the computational
uid dynamics (CFD) simulation of internal-loop airlift reactors using a new drag model derived from
the dual-bubble-size (DBS) model, an extended energy-minimization multi-scale (EMMS) approach for
gasliquid ows. Compared with the traditional SchillerNaumann (SN) correlation, the new model
improves the simulation of gas holdup in the riser and downcomer signicantly. In particular, gas holdup
and circulation of two-phase ow can be modeled successfully using the new model, whereas traditional
drag models such as the SN correlation show an absence of gas in the downcomer. The simulation
demonstrates the advantage and potential of this new model for internal-loop airlift reactors.
2014 Published by Elsevier B.V. on behalf of Chinese Society of Particuology and Institute of Process
Engineering, Chinese Academy of Sciences.

Introduction
Internal-loop airlift reactors have been used widely in chemical, biochemical, waste water treatment, and other processes
because of the advantages they offer, such as higher uid circulation, enhanced mass transfer, shorter mixing time, and lower shear
stress and energy consumption. Typical examples include microalgae cultivation for biofuel, heavy or residual oil hydrogenation,
and aerobic fermentation. The reactor contains an inner draft tube,
separating the bubble column into a central riser surrounded by
a downcomer and the top and bottom sections. Particles in these
reactors are either in the form of catalyst or reactant, and a good
knowledge of multiphase ow behavior and phase distribution is
essential for process scale up and optimization.
A number of experimental studies exist on uid dynamics,
mixing, and mass transfer in internal-loop reactors. Lu, Hwang,
and Chang (1995) investigated the liquid mixing of two- and
three-phase ows, and found that the degree of mixing in the
riser was higher than that in the downcomer. By analyzing
the chaotic time series of pressure uctuation signals, Fu, Fan,
and Wu (2007) reported that homogeneous, transitional, and

Corresponding author. Tel.: +86 10 82544913; fax: +86 10 62558065.


E-mail address: nyang@home.ipe.ac.cn (N. Yang).

heterogeneous regimes existed in the riser, whereas there was no


heterogeneous regime in the downcomer. Deng, Wang, Zhang, and
Wang (2010) studied gas holdup, bubble behavior, interfacial area,
and gasliquid mass transfer in an internal-loop airlift reactor with
a non-Newtonian uid. Blazej, Kisa, and Markos (2004) found that
the difference in gas holdup between the riser and downcomer was
important only for a lower supercial input gas velocity, and the
circulation velocity was governed by friction in the reactor vessel
when the input gas ow rate is higher. To avoid unfavorable inuences of the reduction in gas hold-up because of wall effects, larger
reactor volumes are more suited to providing higher circulation
velocities and a better gas phase distribution.
Computational uid dynamics (CFD) has become a routine tool
for multiphase ow simulation. The simulation of bubble column
reactors has been reported to be sensitive to a number of closure
laws such as drag, lift, virtual mass, and turbulence models as well
as numerical algorithms (Monahan, Vitankar, & Fox, 2005). The
simulation of internal-loop airlift reactors is more complex since
the different sections separated by the draft tube exhibit different
hydrodynamic and mixing behavior.
ck, Mota, Ruzicka, Vicente, and Teixeira (2011) reported
Sim
that the sensitivity to drag closure was rather low, but this is
because they evaluated only a limited number of drag formulae.
Agreement between simulation and experiments became deteriorative with an increase in gas holdup in the downcomer. The

http://dx.doi.org/10.1016/j.partic.2014.04.016
1674-2001/ 2014 Published by Elsevier B.V. on behalf of Chinese Society of Particuology and Institute of Process Engineering, Chinese Academy of Sciences.

T. Xu et al. / Particuology 19 (2015) 124132

Nomenclature
CD
CD0
db
dS
dL
fS
fL
Ug,S
Ug,L
Nsurf
Nturb
Nbreak
Ug
Ul
u
k
g
G
D
H
P
FD

effective drag coefcient for a bubble around a


swarm
standard drag coefcient for a single bubble in quiescent liquid
bubble diameter, m
bubble diameter of small bubbles, m
bubble diameter of large bubbles, m
volume fraction of small bubbles
volume fraction of large bubbles
supercial gas velocity for small bubbles, m/s
supercial gas velocity for large bubbles, m/s
rate of energy dissipation because of bubble oscillation per unit mass, m2 /s3
rate of energy dissipation in turbulent liquid phase
per unit mass, m2 /s3
rate of energy consumption because of bubble
breakage and coalescence per unit mass, m2 /s3
supercial gas velocity, m/s
supercial liquid velocity, m/s
velocity vector, m/s
turbulent kinetic energy, m2 /s2
gravitational acceleration, m/s2
turbulent kinetic energy production because of
mean velocity gradient, J/(m3 s2 )
distance, diameter, m
distance, diameter, m
pressure, Pa
drag force, N/m3

Greek letters
phase volume fraction, gas holdup

turbulence dissipation rate


molecular dynamic viscosity, Pa s

t
turbulence dynamic viscosity, Pa s
uid density, kg/m3


surface tension, N/s

125

numbers provided better predictions than those using only the


Reynolds number.
Modeling issues for bubble column and internal-loop reactors
are essentially related to the understanding of meso-structures,
such as bubble swarms and non-uniform size populations related
to bubble breakup and coalescence. Meso-scale phenomena may
result physically from the compromise and interaction between
different bubble classes, liquid vortices, wakes, etc. We have
extended the energy-minimization multi-scale (EMMS) approach
from gas-solid uidization to gasliquid bubble column reactors,
and a dual-bubble-size (DBS) model was proposed as the extended
EMMS approach for gasliquid ow (Chen, Yang, Ge, & Li, 2009;
Yang, Chen, Zhao, Ge, & Li, 2007; Yang, Chen, Ge, & Li, 2010). The
model was then integrated into a CFD simulation in a simplied
manner. Compared with traditional drag correlations, the model
is able to predict the overall and local gas holdup, as well as the
normalized axial liquid velocity over a relatively wider range of
supercial gas velocities without tting parameters (Xiao, Yang, &
Li, 2013; Yang, Wu, Chen, Wang, & Li, 2011).
As the original DBS model has so far only been applied to bubble
column reactors, in this paper we aim to provide a simple evaluation of the model for internal-loop airlift reactors. Because the
original model integrated into the CFD simulation is simplied, the
ratio of effective drag coefcient to bubble diameter, Cd /db is only
relevant to the global supercial gas velocity (Ug ). Although it is
reasonable for bubble columns, the adaptability of this simplied
model to internal-loop airlift reactors has not yet been evaluated or
validated. The authors recently extended the model for relevance to
the local hydrodynamic parameters and the work will be published
later. In this paper, we try to use this new model for internal-loop
reactors. We nd that the new model is indeed able to improve the
simulation of gas holdup in the riser and downcomer, and hence is
promising for internal-loop reactors.

Abbreviations
DBS
dual-bubble-size
EMMS energy-minimization multi-scale
Subscripts
bubble
b
g
gas
l
liquid

CFD simulation they conducted predicted existence of only limited


amounts of gas in the downcomer. Oey, Mudde, and Van den
Akker (2003) used the drag coefcient of the SchillerNaumann
(SN) correlation (Schiller & Nauman, 1935) and compared the predicted period of oscillation in the downcomer with experimental
results. Simulated results were not compared with experimental
data in terms of local or global gas fraction or liquid velocity. Simulations by Van Baten, Ellenberger, and Krishna (2003) showed
a signicant reduction in gas holdup, which was attributed to
a signicantly higher liquid recirculation. Mohajerani, Mehrvar,
and Ein-Mozaffari (2012) tested seven kinds of drag coefcients
and found that the drag force model proposed by Sokolichin,
Eigenberger, and Lapin (2004) did not t well with experimental
data, and predicted a lower liquid velocity in the riser and downcomer. Drag correlations as a function of the Eotvos and Reynolds

ck et al., 2011).
Fig. 1. Airlift geometry (Sim

126

T. Xu et al. / Particuology 19 (2015) 124132

Table 1
Drag model formulae.
Drag models
Schiller and Naumann (SN) (1935):
p
CD = C
D0 (1 g ) ,
24
(1 + 0.15Re0.687 ) Re 1000
.
CD0 =
Re
Re > 1000
0.44
In this paper, p = 1.
(DBS-local):
Dual-bubble-size-local

560.25574 + 1108.91813Ug + 2812.57093Ul + 3681.46603U12 + 316.77038U13

CD

db

1 + 39.93939Ug + 201.51498Ug2 + 246.35488Ug3 + 5.7683Ul + 14.51696U12


10,
500,
Ug = ug,axial g , Ul = ul,axial (1 g )

Model description
The DBS model is the extended EMMS approach for gas-liquid
ows (Yang et al., 2007, 2010). The gasliquid system is resolved
into a liquid phase and two bubble classes, i.e., into small and large
bubbles, by using a number of structure parameters, namely, bubble diameters (dS , dL ), volume fractions (fS , fL ), and supercial gas
velocities (UgS , UgL ). The mass and force balance equations are formulated in terms of the small and large bubbles. In addition to
the structure resolution and mass and force balance equations, the

(1)
(2)

Ug 0&Ul 0.5 m/s


Ug 0&Ul 0.5 m/s
Ug < 0

(3)

energy fed into the system is partitioned into micro- (Nsurf , Nturb )
and meso- (Nbreak ) scale parts. The equations are closed by a stability condition that the so-called microscale energy dissipation tends
to its minimum. By solving the non-linear optimization problems,
the six structural parameters can be obtained at a given supercial gas velocity (Ug ). Here we give only a shorter description of
the DBS model, and focus on the model framework as well as the
integration of the new drag model into a CFD simulation. For the
details and a thorough understanding of the DBS model, interested
readers are referred to our previous publications (Chen et al., 2009;

Fig. 2. Case A: simulated vs. experimental results: (a) riser liquid interstitial velocity, (b) downcomer liquid interstitial velocity, (c) riser gas holdup, and (d) downcomer gas
holdup.

T. Xu et al. / Particuology 19 (2015) 124132

127

Fig. 3. Gas holdup distribution at different riser supercial gas velocities (Case A,
SN drag model).

Fig. 5. Liquid velocity vector plot (Case A, SN drag model, Ug,riser = 0.075 m/s).

Xiao et al., 2013; Yang et al., 2007, 2010, 2011). The ratio of effective drag coefcient to bubble diameter CD /db is then derived as a
function of the supercial gas velocity (Yang et al., 2011).
The DBS model can be integrated into a CFD simulation in different ways. Here we apply the DBS-local model, which relates
the CD /db of each computational cell to the local hydrodynamic
parameters. To compare the drag model, we also use the S-N drag
Table 2
Two-uid model formulae and turbulence model.
Model formula
Conservation equations:
(k k )
(4)
+ (k k uk ) = k , (k = liquid or gas)
t
(k k uk )
T
D
+

(

u
u
)
=

P
+


u
+
(

u
)
]
+


g
+
F
.
(5)
k
k
k
k
k
k,eff
k
k
k
k
k
t
k
Drag model:


FgD = FlD =

C
3
D
4 g db

l ug ul  (ug ul ).

(6)

 t,m


(m km ) + (m um km ) =
km + Gk,m m m ,
k
t

 m 
t,m

Turbulence model:
Fig. 4. Gas velocity vector plot (Case A, SN drag model, Ug,riser = 0.075 m/s).

(m m ) + (m um m ) =

k

m +

km

C1 Gk,m C2 m m .

(7)
(8)

128

T. Xu et al. / Particuology 19 (2015) 124132

Fig. 6. Gas holdup at different riser supercial gas velocities (Case A, DBS-local drag
model).

Fig. 8. Liquid velocity vector plot (Case A, DBS-local drag model, Ug,riser = 0.075 m/s).

correlation. All models are listed in Table 1. Table 2 shows the conservative equations of the two-uid and turbulence models in the
CFD simulation.
ck et al. (2011) is simThe internal-loop airlift reactor of Sim
ulated in this paper, as illustrated in Fig. 1. The geometrical
parameters are listed in Table 3. The only difference in the two cases
is the inner and outer diameter of the draft tube. Initially the liquid
level is at 1.7 m (Case A). The gas dispersed by a porous sparger
in the experiment is simulated as a circular plane with diameter
equal to the inner draft tube diameter, and allows the gas to enter
over the entire riser cross-section. A pressure boundary condition is
imposed at the top. Air and water are used as the working uid. The
computational grid contains 100,704 (Case A) and 61,422 (Case B)
Table 3
Airlift dimensions for Cases A and B.

Fig. 7. Gas velocity vector plot (Case A, DBS-local drag model, Ug,riser = 0.075 m/s).

Case A (mm)
Case B (mm)

D1

D2

D3

D4

D5

H1

H2

62
87

62
87

70
92

142

420

23

1200 200

H3

H4

H5

170

240

T. Xu et al. / Particuology 19 (2015) 124132

129

Fig. 9. Case B: simulated vs. experimental results: (a) riser liquid interstitial velocity, (b) downcomer liquid interstitial velocity, (c) riser gas holdup, and (d) downcomer gas
holdup.

mesh cells. The time step in all simulations is 0.005 s. Quasi-steady


state is achieved after 55 s of simulation. A computation time of
55 s is used to obtain the time-averaged quantities.

deecting and turns back to the downcomer. The gas therefore


bypasses the downcomer in the simulation. Jet ow of gas is often
observed in many CFD predictions of pilot and industrial-scale
internal-loop airlift reactors using commercial CFD software. These

Results and discussion


Fig. 2 depicts the gas holdup and liquid velocity in the riser or
downcomer as a function of riser supercial gas velocity (Ugr ). As
shown in Fig. 2(a) and (b), the SN model performs well in terms
of liquid axial velocity prediction in the riser or downcomer sections, but underpredicts gas holdup (shown in Fig. 2(c) and (d)).
In contrast, the DBS-local model is better than the SN model in
predicting gas holdup, although it underpredicts the liquid axial
velocity slightly. In particular, the gas holdup in the downcomer
section is almost zero when using the SN model. This result is
ck et al. (2011). This demonconsistent with the simulation of Sim
strates the failure of the SN model for predicting behavior in
the downcomer. The DBS-local model improves predictions signicantly.
The reason why it is not possible to entrain gas in the downcomer in the simulation using the SN drag model is further
illustrated by the contour plot of gas holdup (Fig. 3) and the velocity vector (Figs. 4 and 5). These three gures indicate that the gas
phase motion in the riser is more like a jet ow in the SN drag
model simulation; because the gas phase erupts out of the liquid
layer of the riser with less resistance, it has little possibility of

Fig. 10. Gas holdup distribution at different riser supercial gas velocities (Case B,
SN drag model).

130

T. Xu et al. / Particuology 19 (2015) 124132

Fig. 11. Gas velocity vector plot (Case B, SN drag model, Ug,riser = 0.075 m/s).

Fig. 12. Liquid velocity vector plot (Case B, SN drag model, Ug,riser = 0.075 m/s).

predictions are less representative of practical situations, largely


because of the incorrect calculation of forces between gas and liquids in different sections. The SN model only relates the drag
coefcient with local slip velocity and gas volume fraction, and does
not consider the more complicated mesostructures that may need
to be reected in the models.
The current DBS model can capture the return of gas from the
riser to the downcomer quite realistically. Fig. 6 shows the gas contour plot, and Figs. 7 and 8 depict the gas and liquid phase ow
elds. There is indeed a signicant amount of gas returning into
the downcomer other than that released into the freeboard from
the top. To further improve predictions, one needs to consider the
turbulence interaction of the two phases. In this work, we use the
standard k mixture turbulence model to eliminate the turbulence
interaction between phases.
Fig. 9 compares the simulation of SN and DBS drag models
for the airlift reactor of Case B. The riser diameter of Case B is
larger and the downcomer region is narrower than that of Case

A. For the lower riser supercial gas velocity, the prediction of liquid velocity using the DBS-local model is in reasonable agreement
with experimental results, but the model underpredicts the liquid velocity at relatively higher riser supercial gas velocity. This
may be relevant to the simplied gas inlet condition for the real
porous sparger. The gas is modeled to enter into the riser via the
entire riser cross-section and therefore the inlet gas velocity may
be lower than the real case. It may also be related to the neglected
turbulence interaction between phases since the standard k mixture model is used. The gas holdup of the DBS-local model is in good
agreement with experiments both in the riser and downcomer. In
contrast, the SN model underpredicts gas holdup in the riser, and
the simulated gas holdup in the downcomer is almost zero. This
large improvement in prediction of gas holdup further demonstrates the advantage of the DBS-local model. Fig. 10 also shows
an absence of gas in the downcomers when using the S-N drag
model, and Figs. 11 and 12 illustrate the corresponding gas or liquid
velocity.

T. Xu et al. / Particuology 19 (2015) 124132

131

Fig. 13. Gas holdup distribution at different riser supercial gas velocities (Case B,
DBS-local drag model).

Fig. 15. Liquid velocity


Ug,riser = 0.075 m/s).

Fig. 14. Gas velocity vector plot (Case B, DBS-local drag model, Ug,riser = 0.075 m/s).

vector

plot

(Case

B,

DBS-local

drag

model,

Fig. 13 demonstrates that by using the DBS-local drag model,


the gas exists in both the riser and downcomer when the riser
supercial gas velocity increases, and Figs. 14 and 15 show the corresponding gas or liquid velocity vector plots for Case B. It can be
seen that the gas and liquid motion is downwards and a large circulation of two-phase ow is established between the riser and
downcomer of the loop reactor. This is different from Case A where
the liquid circulates but gas in the downcomer is not fully circulated (Figs. 7 and 8). This difference actually corresponds to various
regimes in the downcomer. Bubbles in the downcomer may be
absent, remain stationary or be circulated into the riser, depending
on the reactor geometry, bubble size, and operating conditions, as
reported by Fu et al. (2007) using the chaotic analysis of pressure
uctuation signals. Further investigation of the correlation of the
regime map and drag models as well as turbulence interaction is
required.

132

T. Xu et al. / Particuology 19 (2015) 124132

Conclusions
CFD simulation results from two cases of an internal-loop airlift reactor demonstrate that the SN drag correlation usually used
in commercial CFD packages is not suitable for internal-loop airlift
reactors. The SN model predicts that no gas exists in the downcomer. The new drag model developed using the EMMS approach
can improve the prediction of gas holdup in the downcomer and
riser signicantly, and hence predict gas and liquid circulation in
the loop reactor. Further investigation of turbulence interaction
may need to be incorporated.
Acknowledgements
The authors wish to thank the National Natural Science Foundation of China (21222603, U1162107), Ministry of Science and
Technology of China (2013BAC12B01), and the Strategic Priority
Research Program of Chinese Academy of Sciences (XDA07080301)
for their long-term support.
References
Blazej, M., Kisa, M., & Markos, J. (2004). Scale inuence on the hydrodynamics of an
internal loop airlift reactor. Chemical Engineering and Processing, 43, 15191527.
Chen, J., Yang, N., Ge, W., & Li, J. (2009). Modeling of regime transition in bubble
columns with stability condition. Industrial & Engineering Chemistry Research,
48, 290301.
Deng, Z., Wang, T., Zhang, N., & Wang, Z. (2010). Gas holdup, bubble behavior and
mass transfer in a 5 m high internal-loop airlift reactor with non-Newtonian
uid. Chemical Engineering Journal, 160, 729737.

Fu, C. C., Fan, L. S., & Wu, W. T. (2007). Flow regime transitions in an internal-loop
airlift reactor. Chemical Engineering & Technology, 30, 10771082.
Lu, W. J., Hwang, S. J., & Chang, C. M. (1995). Liquid velocity and gas holdup in
three-phase internal loop airlift reactors with low-density particles. Chemical
Engineering Science, 50, 13011310.
Mohajerani, M., Mehrvar, M., & Ein-Mozaffari, F. (2012). CFD analysis of two-phase
turbulent ow in internal airlift reactors. The Canadian Journal of Chemical Engineering, 90, 16121631.
Monahan, S. M., Vitankar, V. S., & Fox, R. O. (2005). CFD predictions for ow-regime
transitions in bubble columns. AIChE Journal, 51, 18971923.
Oey, R. S., Mudde, R. F., & Van den Akker, H. E. A. (2003). Numerical simulations
of an oscillating internal-loop airlift reactor. The Canadian Journal of Chemical
Engineering, 81, 684691.
Schiller, L., & Nauman, A. (1935). A drag coefcient correlation. VDI Zeitung, 77,
318320.
ck, M., Mota, A., Ruzicka, M. C., Vicente, A., & Teixeira, J. (2011). CFD simulation
Sim
and experimental measurement of gas holdup and liquid interstitial velocity in
internal loop airlift reactor. Chemical Engineering Science, 66, 32683279.
Sokolichin, A., Eigenberger, G., & Lapin, A. (2004). Simulation of buoyancy driven
bubbly ow: Established simplications and open questions. AIChE Journal, 50,
2445.
Van Baten, J. M., Ellenberger, J., & Krishna, R. (2003). Hydrodynamics of internal
air-lift reactors: Experiments versus CFD simulations. Chemical Engineering and
Processing, 42, 733742.
Xiao, Q., Yang, N., & Li, J. (2013). Stability-constrained multi-uid CFD models for gasliquid ow in bubble columns. Chemical Engineering Science, 100,
279292.
Yang, N., Chen, J., Ge, W., & Li, J. (2010). A conceptual model for analyzing the stability condition and regime transition in bubble columns. Chemical Engineering
Science, 65, 517526.
Yang, N., Chen, J., Zhao, H., Ge, W., & Li, J. (2007). Explorations on the multi-scale
ow structure and stability condition in bubble columns. Chemical Engineering
Science, 62, 69786991.
Yang, N., Wu, Z., Chen, J., Wang, Y., & Li, J. (2011). Multi-scale analysis of gasliquid
interaction and CFD simulation of gasliquid ow in bubble columns. Chemical
Engineering Science, 66, 32123222.

You might also like