Chemical Engineering Science 56 (2001) 673}681

Simulation of a slurry airlift using a two-#uid model

R. S. Oey*, R. F. Mudde, L. M. Portela, H. E. A. van den Akker
Kramers Laboratorium voor Fysische Technologie, Delft University of Technology, Prins Bernhardlaan 6, 2628 BW Delft, Netherlands

Abstract

A two-#uid approach is used to simulate the #ow of a three-phase mixture in an internal-loop airlift reactor. The loop consists of
a riser, in which gas is injected and two downcomers on both sides. The airlift loop is at "rst treated as a two-phase system (gas}liquid).
A full two-#uid formulation is used to describe the dynamics of the two-phase #ow. Subsequently, the distribution of the third, solid
phase is considered by solving a mass balance in which the solids velocity is the superposition of the liquid velocity, a gravitational
settling and turbulent dispersion. An extended version of Tchen's theory is applied for the dispersed phases; turbulence in the
carrier-phase is modelled by a modi"ed k}e equation. The model is implemented in the in-house "nite-volume code DISSIM
(Lathouwers, D. (1999). Modelling and simulation of turbulent bubbly #ow. Ph.D. Thesis), a 2-D pressure-based code. Variation of the
gas #ow rate revealed the existence of di!erent #ow regime with respect to the gas fraction in the downcomer, in agreement with
literature. In case the gas separation at the top is complete, the circulation velocity simulated agreed well with a simple mechanical
energy balance. The solid phase is found to accumulate in the dead corners if the turbulent dispersion is ignored. The settling process is
very slow and part of the solids are trapped in the circulation of the liquid. When the turbulent dispersion is taken into account
a smooth solids distribution is found with a higher volume fraction of solids in the lower part of the downcomers, where the liquid
velocity is minimal.  2001 Elsevier Science Ltd. All rights reserved.

Keywords: Internal-loop airlift reactor; CFD; Two-#uid model; Turbulence; k}e model; Three-phase #ow

1. Introduction methods for CFD are the Eulerian}Eulerian (Sokolichin

& Eigenberger, 1999) and the Eulerian}Lagrangian
Internal-loop airlift reactors are widely used in bio- approach (e.g. Lain et al., 1999) both with their advant-
technology and chemical industry. They are suitable for ages and disadvantages and their range of applicability.
processes where good mixing and close contact between In the Eulerian method, both the continuous and disper-
the phases are required at large volumes and/or low sed phase(s) are mathematically modelled as interpenet-
shear rates. A good example is the nitri"cation/denitri"- rating #uids, represented by sets of mass, momentum and
cation of waste water (see e.g. Van Benthum et al., 1999; energy balances. In the Lagrangian approach, a large
Van Benthum, 1998; Chisti, 1989). Inside the reactor, number of particles is tracked individually, while the
which in many cases consists of two concentric cylinders, liquid phase is treated as a continuum. The interaction
a recirculation #ow is generated when gas is injected from between the particles and the liquid shows up as a source
a sparger below the draft tube. In three-phase applica- term in the momentum equations. The advantage of the
tions small solid particles (e.g. as catalysts) are used to Eulerian method above the Lagrangian way is seen when
improve the process. The generated recirculation deter- the void fraction of the dispersed phase is high. While the
mines, e.g. the residence time of the gas phase. Further- computational time in the latter approach depends high-
more, it controls the spatial distribution of the catalyst ly on the number of particle trajectories to be calculated,
particles that is of importance for the performance of the in the Eulerian method the number of equations to be
reactor (Lain, Broder & Sommerfeld, 1999 and Cockx, solved remains the same. Interaction terms describing
Roustan, Do-Quang & Lazarova, 1997). There are drag, virtual mass and e!ects of lift forces appear in
various approaches to the mathematical and physical momentum balances of both phases (see e.g. Lahey
modelling of multiphase #ows. The most widely used & Drew, 1992). Turbulence modelling in two-phase #ow
is done using the same approach as in single-phase #ow.
Similar to single-phase #ow one can derive Reynolds
* Corresponding author. stress models for multiphase #ows, which may be

0009-2509/01/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 0 ) 0 0 2 7 5 - X
674 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681

simpli"ed to the k}e model. Because of its computa- phases with separate mass and momentum balances and
tional expenses, turbulence modelling in multiphase #ow it circumvents the modelling of even more complicated
computations, is restricted to the carrier phase. Turbu- closure laws.
lence of the dispersed phase is mostly computed from the
continuous turbulence "eld by using algebraic relations.
3. Model description

2. Setup and CFD-Approach The equations used can be divided into three parts.
The "rst part, the two-#uid formulation, models the
In this paper, an Eulerian}Eulerian approach is pre- hydrodynamics of the (pseudo-) two-phase system. The
sented to simulate and study the #ow in a three-phase turbulence modelling forms the second part of the model
internal-loop airlift column. The column is shown sche- equations used and can be subdivided in continuous- and
matically in Fig. 1. The airlift consists of a rectangular dispersed-phase turbulence. The last part describes the
vessel (H;=;D"200;32;12 cm) "lled with water. distribution of the third phase in the airlift loop, i.e. the
Two vertical plates (height 124 cm, horizontal separation solid particles. All three parts will be discussed brie#y in
12 cm) create a riser and two downcomers. Gas is injec- the following and are listed in Table 1.
ted via the bottom over an area equal to the cross-section
of the riser. The riser starts 16 cm above the bottom of 3.1. Two-yuid formulation
the reactor. The total level of the bubbly mixture is kept
at 160 cm above the bottom. At "rst, the system is treated The equations of the two-#uid formulation are derived
as a pseudo-two-phase: the carrier-phase contains liquid by ensemble averaging the local instantaneous equations
with small solid particles, which is modelled as a pseudo- of single-phase #ow (Delhaye, 1974; Drew, 1983). Two
single-phase. The gas acts as the dispersed phase. It sets of balance equations for mass, momentum and en-
enters directly into the riser (see Fig. 1). A standard ergy are obtained. When assuming the #ow to be isother-
single-phase k}e model, including source terms due to mal and ignoring interfacial mass transfer, the energy
coupling of the phases, is used for the turbulence and equations can be skipped and no source or di!usion
solved simultaneously with the mass and momentum terms appear in the mass balances. The balance equa-
balances. The turbulent kinetic energy and dissipation tions for mass and momentum are shown in Table 1, with
rate of the dispersed phase are obtained by algebraic D /Dt representing the total derivative for phase k, a the
I I
relation and are functions of turbulent kinetic energy and volume fraction, o the density and v the velocity of
I I
dissipation rate of the continuous phase. In the present phase k. The mean pressure of the continuous phase is
simulations, interest is merely in steady-state solutions of denoted by p, g is the gravitational acceleration, v the
I
the two-#uid formulations. The obtained #ow "eld of the
continuous phase is then used to compute the solids
Table 1
distribution inside the mixture, by solving a transport Model equations
equation for the solid volume fraction. This pseudo-two-
phase approach is from a computational prospective 2-Fluid formulation
much cheaper than treating the gas, liquid and solid
Mass bal. Da o
I I I "0
Dt
Momentum bal. D a o v
I I I I"
Dt
!a
p#a o g
I I I
!
) (a 1ovv2 )#I ,  I "0
I G H I I I I
Turbulence model (cont. phase)

 
Kin. energy D a o k kR
    "
a  #k
k
Dt  p 
I
# a o (P  !e )#% 
  I  I

 
Dissipation D a o e kR
    "
a  #k

Dt  p C
C
e
# a o  (C  P !C  e )#% 
 k C I C  C

Solid transport
Mass bal. *a
 #
(a (v #v #v ))"0
*t      
Fig. 1. Airlift loop: dimensions in cm.
 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681 675

#uctuating part of the local instantaneous velocity and The second term on the r.h.s. of Eq. (1) represents the
I the interfacial momentum transfer. The two last terms virtual mass, whereas the e!ects of turbulent pressure is
I
arise in the momentum equation as a result of averaging given by the last term, accounting for correlation be-
the local instant formulation and are a priori unknown. tween the instantaneous distribution of the particles and
Both terms need to be modelled in order to obtain undisturbed #uid pressure #uctuations.
closure of the set of equations. The contribution to the
interfacial momentum transfer can be subdivided in drag, 3.1.1. Continuous-phase turbulence
lift and virtual mass and turbulent pressure. In the pres- For the continuous-phase turbulence, a standard k}e
ent simulations the lift force is ignored. The drag force model with additional source terms (see Table 1) is solved
depends not only on the mean velocity di!erence of the simultaneously with the other equations. The additional
two phases. Due to averaging it also depends on a drift terms %  and %  account for the e!ects the interfaces
velocity v , representing dispersion of the particles/ I C
B have on the continuous phase turbulence. The "rst is
bubbles due to #uid turbulence transport. The drift modelled according the work of Bel F'Dhila and Simonin
represents the correlation between particle distribution (1992)
and turbulent #uid motion of the continuous phase (see
Bel F'Dhila & Simonin, 1992; Viollet & Simonin, 1994). o
% "  a o F [1v ) v 2
Summarising, the interfacial momentum transfer can be I o #o C   "   
  
written as
!2k #v ) v]. (6)
 B P

 
*v
I "!I "a o F v #a o C P #v
v The latter term is modelled the same way as done by
    " P    *t  P
Elgobashi and Abou-Arab (1983)
#
) a o C 1v v2 !o 1v v 2
a , (1)
    P       e
%  "C   %  , (7)
where the drag is given by the "rst term on the right-hand C C k I
side and F can be represented by 
" where here C  was set to 1.20. For a more complete
3C C
F " " 1"v "2, 1"v "2"(v ) v #1v ) v 2 , (2) discussion see Thai Van, Minier, Simonin, Freydier
" 4 d P P P P P P  and Olive (1994), Bel F'Dhila and Simonin (1992) or
where d represents the bubble diameter and is in the Lathouwers (1999).
order of 1 cm. The drag coe$cient C for Re(1000 is
" 3.1.2. Dispersed-phase turbulence
modelled as follows
Assuming steady and homogeneous #uid turbulence,
24 o "v "d Tchen's theory (Tchen, 1947 see also Simonin, 1990;
C " (1#0.15Re ) a\ , Re"  P . (3)
" Re  k Viollet & Simonin, 1994) is extended to predict turbu-

lence in the dispersed phase. Summarised, there are three
Due to the correlation between the instantaneous distri-
time scales characterising the interaction between the
bution of the particles and the turbulent #uid motion at
#uctuating motions. First, the characteristic time of the
large scales, the average value of the local relative velo-
energetic turbulent eddies, qR ; second, the characteristic
city between each particle and the surrounding #uid, v , 
P time of bubble entrainment by the #uid motion, or
can be written as the sum of the relative mean velocity
bubble relaxation time q$ . The last characteristic time is
v !v and a drift velocity v 
  B given by qR , the eddy}bubble interaction time. The time


 

a
a scales are given by
v "[v !v ]!v , v "!DR !  . (4)
P   B B  a a


  3 k 1 o
qR " C  , q$ "  #C ,
The drift velocity v is a statistical quantity due to condi-  2 Ie  F o 
B  " 
tional averaging and may not be negligible. The drifting
velocity accounts for the dispersion e!ects due to particle qR
qR "  [1#C f]\. (8)
transport by turbulent #uid motion and may be modelled  p @ P
I
when both phases are assumed incompressible as follows
The parameter C relates to the Lagrangian over Euler-
(Bel F'Dhila & Simonin, 1992) The #uid}particle disper- @
ian characteristic length-scale ratio and comparison with
sion tensor is represented by DR which is written in
 experimental results leads to a practical value (Wells
terms of the #uid}particle velocity correlation tensor
& Stock, 1983) of C "0.45 in the direction parallel to
1v v 2 and the #uid}particle turbulent interaction @
   the mean relative velocity and C "1.80 in the ortho-
time-scale qR . For practical computation the dispersion @
 gonal directions. The ratio between the characteristic
tensor is restricted to its diagonal form
time of the energetic turbulent eddies and the character-
DR "qR 1v v 2 . (5) istic bubble entrainment by the carrier #uid motion is
     
676 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681

given by f a high concentration of particles near the wall. In the

P
simple slip-velocity model this is just neglected.
"v "
f" P . (9)
P ((2/3)k 3.2. Numerical method

The turbulent kinetic energy and #uid-bubble velocity
The code DISSIM, originally developed in-house by
covariance of the dispersed phase is then obtained as
Lathouwers (1996a, b; 1999), is a "nite-volume 2D #ow-
a function of the carrier-phase kinetic energy
slover, able to simulate single- and two-phase #ow fea-
b#g b#g tures from the two-#uid equations. In combination of
k " P k , 1v ) v 2" P 2k ,
 1#g    1#g  a structured mesh, the storage of the #ow variables is
P P done in a staggered sense. Scalar quantities are dis-
qR 1#C cretised using TVD-schemes to improve accuracy (see
g "  , b"  . (10)
P q$ o /o #C e.g. Lathouwers, 1999). Cartesian co-ordinates have been
   
used in the present simulations, in view of the rectangular
3.1.3. Solids distribution airlift loop. Due to the presence of volume fractions and
Once a steady-state of the pseudo-two-phase #ow is large inter-phase coupling terms, a `standarda explicit or
obtained, the distribution of the solids can be obtained implicit treatment of the discretized balance equations
by solving a mass balance equation (see Table 1). Instead will lead to instability or extremely slow convergence,
of solving the momentum equation, the velocity di!er- making a more elaborate treatment necessary. After hav-
ence between the solid and liquid phases is assumed ing de"ned proper initial conditions (e.g. velocity "eld is
equal to a gravitational slip velocity plus a drift velocity. solenoidal by virtue of mass conservation), the following
Essentially, this is equivalent to neglecting all terms in the steps are undertaken to solve them:
momentum equation for the solid phase, except the inter- E The `guesseda velocities are computed using a pre-
facial drag and the gravitational forces; in non-dimen- viously obtained pressure "eld.
sional form the equation becomes E Correction to the `guesseda velocities is applied, which
1 g is closely related to a pressure correction step in
I "!a , (11) single-phase #ow.
  Fr "g"
E Updating pressure and the velocity "eld of both
where the Froude number, Fr, represents the reciprocal phases.
of the non-dimensional gravitational force. (Fr";/g¸ , E The volume fractions are calculated using the dis-
A A
where ; and ¸ are the characteristic velocity and cretized mass balance equation of the dispersed phase.
A A
length scale, respectively.) The interfacial drag is due to E Determination of the turbulent kinetic energy and the
the relative velocity between the particles and the liquid. dissipation rate of both phases.
Similarly, to Eq. (4), the relative velocity is decomposed E Optional step: determination of the solid volume frac-
into a mean velocity di!erence between the solid and tion a , which is only done when steady-state solutions

liquid phases and a drift velocity. Analogously to the of the `pseudoa-two-phase #ow calculations are
#uid-bubble treatment, the drift velocity is modelled as obtained.
being proportional to the gradient of the particle concen-
tration (Viollet & Simonin, 1994). 3.2.1. Boundary condition
The internal-loop airlift column has three distinct

 

a
(1!a )
v "!DR !  , (12) boundaries: walls, inlet and a free surface at the top (see
   a 1!a Fig. 1). The carrier-#ow is assumed turbulent and to
 
where the dispersion tensor DR is formulated in a similar determine the wall boundary conditions the boundary
 layer is assumed to be in equilibrium for which a logar-
way as done for the #uid-bubble dispersion tensor (see
Eq. (5)). For #ows where the gravitational forces are ithmic velocity pro"le holds. This approach has the ad-
much larger than the inertial forces (Fr;1, in our case vantage that no excessive grid re"nement has to be done
FrK0.1), the convective #ux of momentum and the near the walls. It is standard for single-phase #ows and
particle kinetic stresses can be neglected with respect to here is assumed to be valid for the carrier phase. The
the gravitational forces. Eq. (11) is a good approximation following relations hold in the equilibrium log-layer re-
except in regions with large gradients in the mean velo- gion:
city and particle kinetic stresses. For instance, near the u u
k " O , e " O , P "e, (13)
side walls, the gradient in the particle kinetic-stresses,  (C u  iy I
associated with the gradient in the liquid Reynolds stres- K
ses can be very high, inducing a `turbophoretic forcea. where u is the friction velocity. The inlet conditions for
P
This leads to a particle #ux towards the wall and/or the two-#uid model are based on the speci"cation of the
 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681 677

#uxes. At the free surface the bubbles are allowed to exit

the domain, while for the liquid a zero stress condition is
imposed. The conditions for the solid phase are reduced
to zero-#ux conditions at the boundaries, which requires
that the total velocity of the particles (i.e. the sum of the
liquid, slip and drift velocity) equals zero. Care has to be
taken while dealing with these conditions numerically in
order to obtain steady state and physically realistic solu-
tions. Improper handling may lead to nonconservation of
solid-phase mass. For example, if the liquid velocity is
zero the semi-discretized transport equation for a

(Table 1) at a horizontal wall, becomes after rearranging:



*a v DH v DH
 "   ! a #   # a , (14)
*t 2 *y  2 *y 

where DH is the di!usion coe$cient, a and a repres-

 
ent the solid volume fraction at the nearest and second
nearest grid cell to the wall, respectively. Clearly, in order
to reach a steady state one must have Fig. 2. Di!erent regimes in an internal-loop airlift reactor.


v DH
  ! )0. (15)
2 *y
Eq. (15) imposes a reverse restriction on the grid spacing
at the bottom wall. In the current simulation, we opted to
circumvent this di$culty by imposing a value for DH (or
v ) at the grid cell closest to the bottom wall, such that
 
Eq. (15) equals zero. This can be seen as a very simple
wall function for DH (or v ).
 

4. Results and discussion

4.1. Two-phase yow xeld

The calculations were done in two dimensions in Car-

tesian co-ordinates on a 66;82 grid, with constant grid Fig. 3. Liquid velocity in the riser: simulations and mechanical energy
size. The dimensions of the internal-loop airlift column balance.
are shown in Fig. 1. The two-phase #ow in the air lift can
be categorized into three regimes, following Heijnen et al.
(1997). At low gas #ow rates, the liquid circulation is not
strong enough to drag the bubbles into the downcomer. boundary of the second and third regime. In the last
At intermediate gas #ow rates, the bubbles are dragged picture clearly gas is recirculating. Notice that according
into the downcomer but do not enter the riser again to the simulations, the gas re-entering the riser has the
(regime II). At high gas #ow rates, the circulation is so tendency to move along the riser wall. The airlift can be
strong that the bubbles are #owing from the downcomer modelled using a simple mechanical energy balance, as-
back into the riser. This is illustrated in Fig. 2. The suming that the separation of the gas at the top is com-
simulation reveals the same trends, as is shown in Fig. 3. plete. The riser liquid velocity, v , follows from
0
Here the gas fraction distribution in the reactor is shown
for four di!erent super"cial gas velocities: ; " 2g¸ ;
 v " 0  (16)
0.75, 1.25, 1.75, 2.25 cm/s, respectively. The "rst two are in 0 K v #v
0  
the "rst regime: no bubbles are found in the downcomer,
apart from the gas that is sucked into the two vortices with g the acceleration of gravity, ¸ the riser length,
0
that are formed at the entrance of each downcomer. At K the e!ective friction coe$cient and v the slip velo-
 
; "1.75 cm/s gas is drawn into the downer, move all city of the gas bubbles. The latter is assumed to be

the way down and re-enter the riser. This picture is at the constant (25 cm/s). The friction coe$cient is "tted on the
678 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681

Fig. 4. Total gas hold-up versus super"cial gas velocity: ;"riser; #"downcomer.

Fig. 5 . Liquid velocity: (left-top) ; "0.75 cm/s, (right-top) ; "1.25 cm/s, (left-bottom) ; "1.75 cm/s, (right-bottom) ; "2.25 cm/s.
   
 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681 679

lowest super"cial gas velocity, where the gas separation

at the top is complete. The resulting prediction of the
riser liquid velocity is compared with the data from the
simulation in Fig. 4. The agreement is good, but the data
start to deviate at a super"cial gas velocity of 1 cm/s, i.e.
at a riser gas fraction of about 1.5%. This coincides with
entering the second regime: bubbles are dragged into the
downcomer thereby decreasing the driving force for the
circulation (the averaged liquid velocity in the down-
comer is then 24 cm/s almost equal to the slip velocity).
The averaged gas fraction in the riser and in the down-
comer are given in Fig. 5. Also from this plot the change
from the "rst regime to the second can be seen: at
; "1.25 cm/s the gas fraction in the downcomer

starts to increase. The liquid velocity "eld is shown in
Fig. 6. The presence of the vortices is seen in all cases,
especially at the higher super"cial gas velocities. The
liquid #owing around the corner at the top of the riser
has a relatively high velocity as only part of the cross-
sectional area of the downcomer is available. Conse-
quently, bubbles can be trapped in this #ow and dragged
into the downcomer. Below the vortices the liquid velo-
city relaxes as the total area is now available (see Fig. 6).
Then the drag force on the bubbles is no longer su$cient
to drag them downwards to the entrance of the riser. This Fig. 6. Gas fraction distribution (in %) at four di!erent super"cial gas
explains the existence of regime II. The turbulent kinetic velocities: (left-top) ; "0.75 cm/s, (right-top) ; "1.25 cm/s,
 
energy distribution is plotted in Fig. 7. With increasing (left-bottom) ; "1.75 cm/s, (right-bottom) ; "2.25 cm/s.
 
gas #ow rate, obviously the turbulent kinetic energy
increases. The #uid clearly is set in motion more and
more in the lower parts of the downcomers. Note that the
increasing region of turbulent kinetic energy in the down-
comer.

4.2. Solids transport

Using the previous obtained "elds of the liquid phase

(phase 1), the transport equation for a (i.e. the solid

particles) can now be solved. A solid-particle diameter of
0.1 mm is taken; the solid density is twice that of the
liquid phase. As an initial condition a constant percent-
age (i.e. 1%) of liquid volume fraction was taken for the
solids volume fraction. It was thus assumed that the
particles were uniformly distributed in the liquid. Simula-
tion were done for a super"cial gas velocity of 1.25 cm/s.
The steady-state solid fraction distribution obtained, is
given in Fig. 8. The solids fraction is relatively constant,
but is higher in the lower part of the downcomers. This is
a consequence of the low velocity here compared to both
the riser and the upper part of the downcomers, where
the two vortices force the liquid to #ow through a small
cross section. The solids motion is governed by the liquid,
gravity (resulting in slip velocity of solid between the
liquid) and the turbulent dispersion mechanism. The
dispersion is high where the turbulent kinetic energy is Fig. 7. Turbulent kinetic energy (in m/s): (left-top) ; "0.75 cm/s,
high. When the dispersion mechanism is turned o! the 
(right-top) ; "1.25 cm/s, (left-bottom) ; "1.75 cm/s, (right-bot-
 
solids start to settle under the action of gravity. However, tom) ; "2.25 cm/s.

680 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681

tered the riser. As soon as the gas fraction in the down-

comer starts to rise, the simulation and the simple model
deviate, as they should. The solid transport appears to
depend on the turbulence in the liquid phase. If no
turbulent dispersion is taken into account, the solids
"nally settle in the dead corners. If turbulent dispersion is
taken into account, an equilibrium between settling and
dispersion is found in which the solids fraction is highest
in the `deada corners and lowest in the entrance of the
riser where the transport velocity is highest.

Fig. 8. Solid volume fraction at ; "1.25 cm/s. References


Bel F'Dhila, R., & Simonin, O. (1992). Eulerian predictions of a turbu-
it takes quite some time before the solids reach the dead lent bubbly #ow down-stream a sudden pipe expansion. In M.
zones in the lower corners of the airlift loop. They tend to Sommerfeld (Ed.), Sixth Workshop on Two-Phase Flow Predictions,
follow the stream lines pattern for a long time: at the pp. 264}273.
Van Benthum, W. A. J. (1998). Integrated nitrixcation and denitrixcation
bottom part of the air lift gravity moves them to an in bioxlm airlift reactors: Bioxlm development, process design and
`outer stream linea, but at the top of the part gravity hydrodynamics. Ph.D. thesis, Delft University of Technology, Delft,
moves them inwards again. Once in a dead zone, they The Netherlands.
cannot be redispersed. In the steady state, most of the Van Benthum, W. A. J., van der Lans, R. G. J. M., van Loosdrecht, M.
particles are found in the dead corners, but still a signi"- C. M., & Heijnen, J. J. (1999). Bubble recirculation regimes in an
internal-loop airlift reactor. Chemical Engineering Science, 54,
cant portion is trapped in the circulation. When disper- 3995}4006.
sion is turned on the particles tend to deviate from these Chisti, M. Y. (1989). Airlift bioreactors. London: Elsevier Applied
streamlines of the liquid. The time for the particles to Science.
reach the other streamlines decreases and it will become Cockx, A. L. A., Roustan, M., Do-Quang, Z., & Lazarova, V. (1997).
much easier for them to reach the zones where there are Numerical simulation and physical modelling of the hydrodynamics
in an air-lift loop reactor. Chemical Engineering Science, 52,
low velocities. As in these zones the turbulent kinetic 3387}3793.
energy is also low, redispersion is small. Finally, equilib- Delhaye, J. M. (1974). Jump conditions and entropy sources in
rium is reached between incoming and out #owing par- two-phase systems. International Journal of Multiphase Flow, 1,
ticles at a certain solid fraction gradient that counter 395}409.
balances the settling via gravity. Drew, D. A. (1983). Mathematical modelling of two-phase #ow. Annual
Review of Fluid Mechanics, 15, 261}291.
Elgobashi, S., & Abou-Arab, T. (1983). A two-equation turbulence
model for two-phase #ows. Physics of Fluid, 26(4), 931}938.
5. Concluding remarks Heijnen, J. J., Hols, J., van der Lans, R. G. M., van Leeuwen, H. L. J. M.,
Mulder, A., & Weltevrede, R. (1997). A simple hydrodynamic model
In this paper, a model and method to calculate two- for the liquid circulation velocity in a full scale three phase air lift
rector. Chemical Engineering Science, 52, 2527}2539.
and three-phase #ow in internal-loop airlift column, us- Lahey, R. T. Jr., & Drew, D. A. (1992). On the development of multi-
ing the two-#uid method is presented. The three-phase dimensional two-#uid models for vapor/liquid two-phase #ows.
#ow is treated as a pseudo-two-phase mixture in which Chemical Engineering Committee, 118, 125}139.
the liquid phase consists of the real liquid phase in which Lain, S., Broder, S. D., & Sommerfeld, M. (1999). Experimental and
the solid particles are dispersed. The bubbly #ow is numerical studies of the hydrodynamics in a bubble column. Chem-
ical Engineering Science, 54, 4913}4920.
simulated using a full two-#uid model. From the #ow Lathouwers, D. (1996a). An evaluation of two-#uid models for two-
"eld the solid distribution in the liquid is obtained by phase turbulent bubbly #ows. In M. Sommerfeld, (Ed.), Eighth
solving a mass balance in which the solid velocity is the Workshop on Two-Phase Flow Predictions.
sum of the liquid velocity, a gravitational slip velocity Lathouwers, D. (1996b). A numerical method for the solution of two-
and a dispersive term due to the turbulence in the liquid #uid model equations. In Numerical Methods for Multiphase Flows,
FED vol. 236, pp. 121}126. ASME.
phase. The pseudo-two-phase simulations are done by Lathouwers, D. (1999). Modelling and simulation of turbulent bubbly yow.
varying the gas #ow rate, keeping all other parameters Ph.D. thesis, Delft University of Technology, Delft, The Nether-
"xed. It is found that at low gas #ow rates no gas enters lands.
the downcomer, whereas at higher gas #ow rates it does Simonin, O. (1990). Eulerian formulation for particle dispersion in
and eventually re-enters the riser. This observation is in turbulent two-phase #ows. In M. Sommerfeld (Ed.), Fifth Workshop
on Two-Phase Flow Predictions, pp. 156}166.
accordance with literature. Predictions from a simple Sokolichin, A., & Eigenberger, G. (1999). Application of the standard
energy balance are in agreement with the results from the k}e model to the dynamic solution of bubble columns. Chemical
simulations at low gas #ow rates, when no gas has en- Engineering Science, 54, 2273}2284.
 R. S. Oey et al. / Chemical Engineering Science 56 (2001) 673}681 681

Thai Van, D., Minier, J., Simonin, O., Freydier, P., & Olive, J. (1994). Viollet, P. L., & Simonin, O. (1994). Modelling dispersed two-phase
Multidimensional two-#uid model computations of turbulent #ows: Closure, validation and software development. Applied Mech-
dispersed two-phase #ows. ASME, Numerical Methods in Multiphase anics Reviews, 47 (6, Part 2), S80}S84.
Flows, FED-185, 277}291. Wells, M. R., & Stock, D. E. (1983). The e!ect of crossing trajectories on
Tchen, C. M. (1947). Mean value and correlation problems connected the dispersion of particles in turbulent #ow. Journal of Fluid Mech-
with the motion of small particles suspended in a turbulent yuid. Ph.D. anics, 136, 31}62.
thesis, Delft University of Technology, Delft, The Netherlands.