Pergamon (heroical Enginee~inH Science, Vol.

52, Nt~ 9, pp, 142~,~ 14>8, 1997

199"7 Elsevier Science l a d All rights reseived
Prinled m Great Britain
Plh S0009-2509(96)00515-5 ooo9:51)9<~:*tv.l><l~ootl

Dynamic simulation of dispersed

gas-liquid two-phase flow using a discrete
bubble model
E. Delnoij,* F. A. Lammers, J. A. M. Kuipers and W. P. M. van Swaaij
Department of Chemical Engineering, Twente University, P.O. Box 217, 75(10 AE Enschede.
The Netherlands

(Received 4 September 1996; in revised form 3 December 1996~ accepted 9 December 19961

Abstract In this paper a detailed hydrodynamic model for gas liquid two-phase flow will be
presented. The model is based on a mixed Eulerian Lagrangian approach and describes the
time-dependent two-dimensional motion of small, spherical gas bubbles in a bubble column
operating in the homogeneous regime. The motion of these bubbles is calculated from a force
balance tbr each individual bubble, accounting for all relevant forces acting on them. Contribu-
tions from liquid-phase pressure gradient, drag, virtual mass, liquid-phase vorticity and gravity
are considered, whereas direct bubble-bubble interactions are accounted for via an interaction
model resembling the collision model developed by Hoomans et a/. (1996) to model gas-
fluidized beds. The liquid-phase hydrodynamics are described using the volume-averaged,
unsteady, Navier-Stokes equations. A preliminary model validation has been performed by
comparing the computational results with experimental observations published previously in
literature by various authors. The model is shown to predict correctly the motion of a bubble
plume in a pseudo-two-dimensional bubble column operated at different superficial gas
velocities, provided that a detailed description of the bubble dynamics is incorporated in thc
model. The effect of bubble column aspect ratio on the hydrodynamic behaviour of the column
has also been investigated. Our model predicts the effect of aspect ratio on the flow structure in
the bubble column. The importance of the various forces acting on the bubbles will also be
discussed and it will be shown that the added mass lbrce and the lift force cannot be neglected in
bubble column simulation. Finally, the model has been used to study the start-up behaviour of
a two-dimensional bubble column. It will be shown that the history of the gas-liquid two-phase
flow significantly affects the flow structure ultimately obtained in a bubble column. This finding
has, to our knowledge, not been reported before in literature..~, 1997 Elsevier Science Ltd. All
rights reserved

Kevwords: Bubble columns; two-phase flow; hydrodynamics.

1. INTRODUCTION There are two well-known approaches to model

Bubble column reactors are frequently employed in gas-liquid two-phase flows. One approach is based oil
the biological, chemical and petrochemical industry. the mixture formulation in which effective (transport)
Compared to other multiphase reactors, such as properties are used to describe these two-phase flows.
packed towers and trickle bed reactors, bubble col- This approach is relatively simple to use, bu! cannot
umns offer some distinct advantages, among which its account for any interaction li.e. slip, mass and heat
simple construction and excellent heat transfer char- transfer) between the phases. In the second approach
acteristics can be mentioned. The biggest disadvan- both phases are treated separately where transfer of
tage, on the other hand, is the limited knowledge mass, heat and momentum between the phases can be
about the fluid dynamics of these gas-liquid two- accounted for. Within the context of this approach
phase systems, despite the significant research efforts one can distinguish Eulerian/Eulerian ~or two fluid)
which have been made both in academic and indus- models which are based on the concept of interpenet-
trial research laboratories. Modelling the fluid dy- rating continua, and Eulerian/Lagrangian models
namics of gas-liquid bubble columns is therefore still which adopt a continuum description for the liquid
a challenging problem. phase and additionally track each individual bubble
using the Newtonian equations of motion.
In recent years a number of authors (Torvik and
*Corresponding author: E-mail: [email protected]. Svendsen, 1990: Ranade, 1992; Grienberger and
1429
1430 E. Delnoij et al.
Hofman, 1992; Hjertager and Morud, 1995; and intermediate gas velocities, a large number of different
Sokolichin and Eigenberger, 1994) have published bubbles with varying size and shape can be observed.
interesting results obtained with two fluid gas-liquid Bubble shapes range from spherical to spherical cap.
models. Among others Sokolichin and Eigenberger Slug flow occurs at high gas velocities in small dia-
have contributed to the development of this type of meter bubble columns.
model. Their calculations clearly demonstrate the As stated in the Introduction, the model developed
time-dependent behaviour of a gas-liquid bubble col- applies to the homogeneous or dispersed bubble re-
umn, and thus stress the importance of dynamic mod- gime. This implies that the bubbles can be considered
els. Long time averaging of their instantaneous as spherical, which simplifies the calculation of the
velocity distributions leads to the familiar, experi- force exerted by the liquid on the bubbles signifi-
mentally observed, liquid velocity profile with liquid cantly. We will assume that this force Ftotal acting on
upflow in the column centre and liquid downflow near a non-deformable spherical bubble rising in an un-
the walls of the column. steady, and non-uniform liquid flow is composed of
Due to advances in computer technology Euler- separate and uncoupled contributions from pressure
ian/Lagrangian models for gas-liquid bubble col- gradient, drag, virtual or added mass, vorticity and
umns have received increased attention in recent gravity (Auton, 1983):
years. Pioneering work in this field was conducted by
Ftotal = Fe + Fo + FVM + FL + F~. (1)
Trapp and Mortensen (1993), Lapin and Ltibbert
(1994) and Devanathan et aL (1995). The models de- The acceleration of a bubble can then be calculated
veloped differ in their description of the bubble dy- from a force balance over that bubble:
namics, and in the representation of the coupling
between the phases. In their recent paper Lapin and dv
mb~ -~ Ftota I (2)
Liibbert developed a two-dimensional Eulerian/Lag-
rangian model of a bubble column using a simple
where rnb represents the mass of a bubble. Once the
description of the bubble dynamics. Coupling be-
bubble acceleration is know, the new bubble velocity
tween the gas and the liquid phase was achieved
can be calculated using a simple, explicit integration
through the effective density of the mixture. No ex-
change of momentum between the phases was incorp- formula:
orated in their model. / / d v ~ n+l
In the present paper we will present a detailed v"+' = v" + ~-d-~) DT (3)
Eulerian/Lagrangian model for a gas-liquid bubble
column operating in the homogeneous regime. The which states that the new bubble velocity equals the
liquid phase will be described using the well-known sum of the bubble velocity at the previous time level
volume averaged Navier-Stokes equations whereas and the most recently obtained acceleration of the
the dispersed phase will be described by the equations bubble multiplied by the time step DT. The calcu-
of motion for each individual bubble. The exchange of lation of the new bubble positions, from their new
momentum between the gas and the liquid phase will velocities and previous time level positions, requires
also be accounted for. In addition, bubble dynamics to account for possible direct bubble-bubble interac-
will be described in detail, incorporating all relevant tion (collisions) as will be described in Section 2.5.
forces acting on a bubble rising in a liquid. In order to First the various forces acting on a bubble will be
prevent, physically impossible, bubble-bubble over- discussed in more detail.
lap, direct bubble-bubble interactions will be de-
scribed using a collision model. Finally a preliminary 2.1. Gravity and far-field pressure contribution
experimental validation of the model will be reported The force acting on a bubble due to the pressure
using experimental results obtained from the litera- gradient in the liquid incorporates contributions from
ture. the Archimedes displacement force, inertial forces and
viscous strain in the liquid. The sum of liquid-phase
pressure gradient and gravity equals:
2. BUBBLE D Y N A M I C S
Gas-liquid bubble columns can be operated in FG + Fp = pg Vbg -- VhVP. (4)
a number of different regimes. The flow regime en-
countered in the column depends on one hand on the 2.2. Drag force
superficial gas velocity and the physical properties of A bubble moving with a constant velocity through
the phases, and on the other hand, on the aspect ratio a uniform liquid flow field experiences a drag force,
of the column. One can distinguish the homogeneous which consists of a form drag and a friction drag
or dispersed bubble regime, the heterogeneous or exerted by the liquid on the moving bubble. The drag
churn turbulent regime and the slug flow regime force acting on a suspended sphere is given by Odar
(Deckwer and Schumpe, 1993). The homogeneous (1964), and is found to be proportional to the relative
flow regime is characterised by relatively low gas velocity between the phases as follows:
velocities and small, spherical bubbles. In a bubble
column operating in the heterogeneous regime, i.e. at FD = -- ½Co&~R~Iv - ul(v - u). (5)
 Simulation of dispersed gas liquid two-phase flow 1431

The drag coefficient C'D depends on the flow regime 2.4. The virtual muss ./orce
and the liquid properties. For rigid spheres the drag The total force exerted by the liquid on the bubble
coefficient is usually approximated by the standard is in part independent of changes in the slip velocity
drag curve (Cliff et al., 1978): between the phases and in part dependent on changes
in the slip velocity. This latter part can be seen as
24 a resistance to acceleration. This resistance to acceler-
( ' t ~ - - I Re < 1000 ~ RT(1 + 0 . 1 5 R e °'6s;) ation is due to an 'added mass" of liquid that has to be
(6)
( Re'> 1000 ~ 0.44. accelerated when a bubble accelerates. According to
Auton (1983) this added or virtual mass force can be
modelled using eq. ( I 0):
However, because of internal gas circulation and de-
formation, bubbles do not necessarily behave as rigid
spheres; their drag coefficient can therefore differ from !%.u = -
(°'
, i)t- + I V u
)
that predicted by the standard drag curve. According ( 101
to Clift et al., the drag on bubbles in pure water is less I = C'v.upt I/h(v - u).
than the drag predicted by the standard drag curve. In
The material derivative in this equation for the virtual
contaminated systems, on the other hand, the surfac- mass force should be the derivative pertaining to the
rants tend to collect at the rear of the bubble whereby gas bubble. Van Wijngaarden (1976)argued that the
the slip along the surface of the bubble is reduced. In
influence of neighbourmg bubbles on the virtual mass
contaminated systems, therefore, bubbles behave coefficient of the bubble under consideration can be
more like rigid particles (Auton, 1983L As tap water is expressed in terms of the average void fraction in the
used m most experiments reported in the literature, it
vicinity of that bubble as follows:
is decided to use the standard drag curve equation as
an estimate for the drag coefficient of a spherical C'~,u = C , M [ I + 2.78(1 - el)]. (I 1)
bubble,
For all computations reported in this paper the value
for the virtual mass coefficient CvM has been taken lu
2.3. The lit? ./orce equal 0.5.
Bubbles rising in a non-uniform liquid flow field,
experience a lift force due to vorticity or shear in the 2.5. Direct Bubble huhble interaction
liquid flow field. Auton (1983) calculated the lift force When two bubbles approach each other, the liquid
exerted by an inviscid liquid on a bubble in a vertical, between the bubbles will resist this relative motion.
linear shear flow represented by: Due to this motion the pressure in lhe liquid film
between the bubbles increases and as a consequence
u~. = (t,x + u~iJ. (71
the liquid is forced to move out of this fihn. The final
He derived that the lift force depends on the vector outcome of this process depends on the magnitude of
product of the slip velocity and the curl of the liquid the pressure in the liquid film between these two
velocity, resulting in a lift force that acts in a direction bubbles. If the pressure is high enough to stop the
perpendicular to both the direction of the slip velocity relative motion of the bubbles, the bubbles will
and the direction of the curl of the liquid velocity field. bounce; otherwise bubble coalescence will result.
If the bubble velocity exceeds the liquid velocity the Duineveld (1994) showed that the Weber number
lift force is directed towards a region with a lower based on either the rise velocity of the bubbles or the
liquid velocity. If the bubble moves with a lower approach velocity of the bubbles determines whether
velocity the lift force is directed towards the high two bubbles will coalesce, bounce and then coalesce
velocity region: or bounce and separate. He determined a minimum
bubble size for bouncing and separation of air bubbles
FL = -- ('~Pt Vh(v - u) x in pure water of 1.72 ram. Assuming a bubble size
(8) exceeding this minimum bubble size, but small
~:=Vxu. enough to assume a spherical bubble shape, a colli-
sion model can be used to describe the bouncing and
Expression (8) for the lift force is valid under the
separation of two bubbles.
assumption of local homogeneity of the flow, or:
The processing of a sequence of collisions is based
ghK on the method developed by Hoomans eta/. (1996). In
-- <{ 1. (9) this method a constant time step DT is used to ac-
IJrisc
count for the forces acting on a bubble. Within this
This assumption is satisfied throughout the homo- time step, the velocity of the bubbles is assumed to
geneous regime where both bubble radius and liquid change onl~ due to binary collisions between bubbles:
velocity gradients are rather small; therefore, Auton's a sequence of collisions is then processed one collision
lift force equation has been implemented in our at a time.
model. The value of the lift coefficient has been cal- In this computation a collision between two
culated by Auton (1983) and was found to be equal to bubbles is supposed to occur when the distance be-
0.53. tween these two bubbles equals the sum of the two
1432 E. Delnoij et al.
radii of the respective bubbles plus an added volume updated using this time interval and the bubble vel-
equivalent to the volume occupied by the virtual mass ocities [obtained from eq. (3)] as follows:
of liquid:
ro(t + t,,b) = r~(t) + v~t.b. (15)
[ro -- rbl = (R. + Rb)~/1 + C v M . (12) However, within this time interval tab there are at least
In order to process a sequence of collisions, the time two bubbles that bounce. Their velocities change due
to collision for each bubble with another bubble or to this collision as depicted in Fig. 1. In the first step
one of the containing walls has to be determined. This the velocities of the individual bubbles are split into
two components, one along the line connecting the
time step tab can be calculated from the initial position
of the bubbles and their velocities. This yields a quad- centres of mass of both bubbles and one perpendicu-
ratic equation in t~b, in which the smallest positive lar to that line. Three simplifying assumptions are
root corresponds to the collision time: made to calculate the new velocities; the collisions are

- b.b -- x / b ~ 2 - Iv. - Vbl2(lr. -- rbl 2 -- (R. + Rb)2(l + C v M ) 2/3)

~ab "~ (13)
iv, - vbl2

where instantaneous, completely elastic and the rotation of

the bubbles remains unaffected by the collision.
bab = (r~x - rbx)(Vax -- Vbx) + (r,r - r b y ) ( U a y - - Uby). Therefore, the velocity component perpendicular to
(14) the centreline does not change, whereas the velocity
component along the centreline can be computed
As the time interval to the next collision has been from:
calculated for each bubble, the smallest collision time
of all bubble pairs contained in the column can be - 2 ma Vra -~- IHb 1)rb
(Vra)af. . . . . l.sio. V,,. (16)
determined. The bubble positions are subsequently m a -4- m b

[] []
Vb Vsb
Va Vb Vsa

)
"-.~ . . . . . . .°°"
Y -- , ~ . , s

g r

[] []
Vsb
Va ---'vsa t- .... !

L. o° , ",. .,-° , . I
,o'1 *, / "., I
• *, d k I
," ", -' :1

$ " " ~ . ° . . . . . . . . . °o'°° "-. ....... °°"° y ....... \, /"

r x

Fig. 1. New velocities of two bouncing bubbles are calculated using a collision sub model: (A) bubbles in
Cartesian grid; (B) colliding bubbles in coordinate system used to describe the collisions IC) new velocities
from collision dynamics; (D) new Cartesian velocities.
 Simulation of dispersed gas--liquid two-phase flow 1433
The final step consists of calculating the Cartesian erties are required, which have to be derived fiom
velocities of both bubbles from the velocity compo- volume averaged liquid properties.
nents used in the collision coordinate system. The In addition to the m o m e n t u m transfer between
interested reader is referred to the paper of H o o m a n s both phases, the liquid volume fraction ~:t of a com-
et al. for additional details. putational cell has to be calculated from the volume
occupied by the bubbles present in the cell under
3. LIQUID-PHASE HYDRODYNAMICS consideration and the volume of the computational
cell:
3.1. Gorerning equations
The liquid-phase hydrodynamics are described us- V. V,,i
~:1 = 1 - " " ~ (20)
ing the volume-averaged mass and m o m e n t u m con- V~H
serwttion equations: In calculating this liquid volume fraction it is very
d(c,p~) important to account for bubbles overlapping with
- - + V'c,ptu = 0 (17) more than one computational cell. The volume occu-
?t
pied by these bubbles has to be divided over the
8tc~ptu) respective computational cells.
- - + V'~tpluu = - ~:IVP - V'clzz + rlplg + q). The m o m e n t u m transfer from the bubbles to the
?t
liquid per unit volume G L dispersion, q~, is just the
(18) opposite of the m o m e n t u m transfer rate due to drag.
lift and virtual mass forces exerted by the liquid on the
The spatial resolution with which the liquid velocity bubbles present in the computational cell under con-
field is resolved is small compared to the size of the sideration. Thus, qb can be calculated from
bubbles; the liquid-bubble interaction is therefore
superimposed on the liquid velocity field. This (l)r~ll = ~',.(FOi + Fv,qi + F/,) 1211
liquid bubble interaction is modelled using a source where the magnitude of the various tbrces is obtained
term q) which accounts for the m o m e n t u m exchange from the discrete bubble submodel.
between the bubbles and the liquid. In order to calculate the force acting on a bubble
In the present study two-dimensional, isothermal from eq. (1) local values of the pressure, of the liquid
motion of both the gas and the liquid phase is as- velocity, of the partial and substantial derivatives of
sumed. The basic variables that have to be calculated the pressure and of the partial and substantial deriva-
from the model are pressure, liquid velocity and the tives of the liquid velocity have to be available tit the
velocity and position of each individual bubble. All position of the bubble. However, these Eulerian vari-
other variables in the balance equations must be spe- ables and their derivatives are only known at discrete
cified in terms of these basic variables. nodes in the computational domain. Therefore, an
area-weighted averaging technique is used to obtain
3.2. Constitutive equations these local values from the values of the Eulerian
The liquid-phase viscous stress tensor rt is modelled variables at the four nodes surrounding the bubble.
assuming general Newtonian behaviour of the liquid: With reference to Fig. 2, the local value of a quantity
f can be calculated using
% .... [(2, - ~I~,)(V- u)E + IL,((Vu) + (Vu)T)]. (19)

The bulk viscosity ),~ is set to zero in all simulations f (r ) -- d \ d.l' ,,~1 ,4. Ii, (221
presented in this paper, in most simulations water is
used as the liquid phase which corresponds to a shear with f. being some Eulerian quantity at node n. and
viscosity of 1.0x 1 0 3 kg m ~ s- A,, representing an area equal to
A I = tdx- axl{dy - ,Sy)
3.3. Couplin.q between bubbles and liquid and vice versa
In our model the coupling between the gas and the .4_, = avldy ,Sy)
1231
liquid phase appears through the liquid volume frac-
,4.~ -- (dx &v)6y
tion ~:, and the source term 0p in the liquid-phase
m o m e n t u m equation which represents m o m e n t u m A 4- = ~'~',-~iv.
exchange between the gas phase and the liquid phase.
As the system of partial differential equations govern- 3.4. Boundary conditions
ing the liquid-phase hydrodynamics is solved using The boundary conditions required to solve the
a finite differencing technique this source term has to model equations are incorporated into the model us-
be calculated in accordance with the number of ing the flag matrix concept. This concept, also used by
bubbles present in a differential volume element and Kuipers et al. (1993), allows boundary conditions to
the individual velocities of these bubbles. This pro- be set for each individual computational cell. Using
vides the m o m e n t u m transfer from the bubbles to the this flag matrix concept a variety of boundary condi-
liquid; the liquid on the other hand, exerts a force on tions can be set by specifying the value of the cell flag
the bubbles. To calculate this force local liquid prop- fll i, j). In Table 1 an overview of the various cell flags
1434 E. Delnoij et al.
dx
y

A A

'f3 A

dy

Bubble
Eulerianquantity ~y
at node in compu-
tational grid

fl f2 T T
~L
V

~x

Fig. 2. Area weighting of the four values at the Eulerian nodes surrounding a bubble to obtain the value of
the quantity under consideration at the position of the bubble in the computational domain.

Table 1. Cell flags and corresponding cell types used in defining boundary
conditions

fl (i, j) Physical meaning of cell (i, j)

Interior cell, no boundary conditions have to be specified

Impermeable wall, free slip boundary
Impermeable wall, no slip boundary
Fluid-phase influx cell, normal velocity has to be specified
Prescribed pressure cell, free slip boundary
Continuous outflow cell, free slip boundary
Impermeable floor, free slip boundary
Impermeable floor, no slip boundary
Corner Cell, no boundary conditions have to be specified

and the corresponding b o u n d a r y conditions is given. This model has been implemented in a computer
Figure 3 depicts a typical grid layout used in many of code, called LeBuc, written in C. A simplified flow-
the simulations presented in this paper. sheet of the computer code is presented in Fig. 4. The
The b o u n d a r y conditions for the discrete part of the computer code takes consecutive steps in time, during
model are essentially provided by the collision routine each of which the code calculates the forces acting on
which accounts for wall-bubble interactions. the bubbles present in the system, the new bubble
velocities and their new positions, taking into account
4. N U M E R I C A L S O L U T I O N possible collisions between bubbles. Finally, the code
In the second and third section of this paper the calculates the new liquid velocity field using an algo-
physics of our model have been described in detail. rithm also used by Kuipers et al. (1993). At the end of
 Simulation of dispersed gas-liquid two-phase flow I435

liiiiiliiiiii li]iiiiii
i iiiii i ]i iiii i liiiiiiiiiiii
ii::i::ili::i::ii!::!iiiiiiiiii::
lii
:::::::::::::::::::::::::::::::::::: 1 1 1 1 ::::::::::::::::::::::::::::::::
iiiiiiiil;i;iiiiiii!iiiii!i ::iiiiiiiiiiiiiiiiiiiiiiiii
!!!!!i::ii[iiii!i::!!iiiiiiil
iiii!!!iiii~iiii!ii;ii!i2 1 1 1 I
• • .,.. • ............, ,...

ili!i
• -.-.- -.,.... - ./......
• .............../,....
.......-....... ,......

:i:i:i:i:i:i:i!!:i:i!i!i!i:
. •. ,...- ..., ..,....,

i!iiiii!i!i~i!i!i!i!iiiii
........... • ..-,.,.,...
1 1 1 1
. ..-.,/....... ,......
.......,......,.........
........,.............. •
...,../.. • ...........,

1 1 1 1 ::)ii!i!!!~)!::::i::::ili::
ii

'>>2">2":'.'>:'2".':':'
ii::ii!i::iii::::i::iiiii!::i::il
2">>>:-'_,32,',~.>',',
• .. • ......, ......-.,.,
• • -.... ,... - • ..-...

• ..............-.... • •

i:i:i:i:i:i~li:i:i:i:i:i:i 1 1 1 1 • ....,..........., / ....

.:.:->:.:.>>>'.1.%.1,
. ............,.....,.,, .,,.,.,...............-.-.

• .....,. • ..,....,..... . . . . . . . . . . . . . .

. ....-.....-....,......-., :.:.:.:.:.>:.>> > >:.:.

i:i:!:i:i:i3i:!:i:i:i:i:i 1 1 1 1
iiiiiliiiiiiiiiiiiiil;iill
• ,.,..../...., • • .....
- ... • • ./ ,..........
• ........ • ,...,...,.,...
........ • ...........,. •
. . . . . . . . . . . . .

;.~ ).i.i.;.;.~).;.2.;.;.: ...................,......

• ................ -.....
-. -.-.../ .......... •
• .... • ......,. ,.......
.......... •...../.,

!:!:i:i:i:i3!:[:i:i:i:i:! 1 1 1 1 iiiiiii!i:~iiii!i!iii!i!
":':+:+:+: :':':': :
•,, .............-..
• • • -.......- ,.,.-...-.
•.
iiiiiiiiiiiiiiiiiiiiiiiiii
• .....-...,...........
:,1,:,:,; :,>:,J.>>2.:,i .,,.....,.,., .....,.....

::::::::':::::'
- .................-......
- • .......,.....
• • ... -. ,................
-...,,, i ii!!ii!!!!ii
; ~ ; 1 1 : ; ; i ; i ~ i
• /........ •, ,.... • ..

!:i:i:i:!:!3i:i:i:!:!:!:i
,:.>>:.>>>>>-.>>'.,
1 1 1 1 iiiiiiiii~iiiiiiiiiiiii
-...'.'., ..-........ ,..-.

-......./.,......-.- -.. ..........,.......-.-.. •

• ,-.-.....-.-..,....,....

• .... ,........ • ..... :::!:i. ::::~:;i:};::::::: :i~:~:::::i:i::':::::':::. ";:~:':::::::::~::::::::::: :~:::~::::::~:;::'::k~:: . . . . . .

ililil)ii!
iiil)ii!i;i!
)iil;i;!;!
;iiiiiil;!;iiiiiiiiil;i
iiil;iiiiiill
il;i;i}iii
;i)iiiil;i;i)

Fig. 3. T~pical grid klyotlt for a leBuc simulation. The cell flags depicted in 1his figure are explained in
Table I.

each timestep various user specified data in several file 5. RESULTS ANI) DIS('USSION
formats can be saved.
On average calculating I rain of the time-depen- In the previous sections we have presented a dis-
dent behaviour of a typical bubble column containing crete bubble model for a gas-liquid bubble column.
several thousand bubbles requires two and a half to The model incorporates both a bubble-bubble inter-
3 h dedicated CPU time on a Silicon Graphics In- action model and a fundamental description of the
digo 2 workstation. The CPU time depends on the size bubble dynamics. In this section the model will be
of the column and the number of bubbles present in compared to experimental data reported in the litera-
the bubble column. A twofold increase in the number ture by various workers. In addition, the model will be
of bubbles present in the bubble column (which can be used to simulate the time-dependent behaviour of
achieved by doubling the superficial gas velocity) a bubble column. Specifically, both start up behaviour
roughly doubles the CPU time required to calculate and the effect of column aspect ratio on the two-phase
one minute of real-time operation of the bubble flow will be studied. As the model incorporates a de-
column. tailed description of the bubble dynamics, the relative
1436 E. Delnoij et al.

Start
Time_liquid= 0
Initialisafion [

*,,.._[ Time~_gas=0
---I I
[ Calc'uYlatederivatives

] Calculateforceson bubbles

Calculatenew bubblevelocities

I M°ve]~[bubbles I

Time_gas = Time.gas + DTgas

...I..

Time'liquid = Time_liquid+ DT

Write'output to Hard Disk

End ~]

Fig. 4. Flowchart of the leBuc computer code.

importance of the various forces acting on a bubble eq. (24) yields an implicit equation in Vrisewhich can
will also be investigated. be solved numerically. This yields a terminal rise
velocity of 20.8 cm s - 1 for a 2 mm air bubble in water.
5.1. Terminal rise velocity of a single bubble Our two-dimensional gas-liquid model was also
As a first test case for our model the terminal rise used to determine the terminal velocity of a single
velocity of a single bubble as calculated from a force bubble rising in a liquid. We therefore simulated the
balance will be compared to the velocity predicted by rise of a single 2 mm air bubble in a bubble column of
our model. For a bubble rising with a steady velocity 1.5 m in height and 0.50 m in width (computational
through a quiescent liquid the buoyancy force just conditions are listed in Table 2). At a height of 1 m
equals the sum of gravity force and drag force. From above the gas distributor the bubble appeared to have
this force balance the bubble's steady rise velocity can a steady rise velocity of 20.9 cm s - 1, which agrees well
be derived: with the theoretically calculated bubble rise velocity.
~/~(Po -- Pl)Rbg
Vrise = C-'-D~/ " (24) 5.2. Comparison with experimental data reported by
Becker et al. (1995)
The drag coefficient Co is calculated using the stan- Becker et al. conducted experiments in a pseudo-
dard drag curve [eq. (6)]; including this equation into two-dimensional bubble column; the results of two of
 Simulation of dispersed gas-liquid two-phase flo~ 1437
Table 2. Parameters used to calculate terminal rise velocity of a single air bubble in
water using the Eulerian/Lagrangian model

Bubbles and liquid Column

Bubble diameter 2 (mm) Height 1.50 (m}

Gas flow rate Width 0.50 (m]
Density gas 1.2 (kg m 3t Number x-cells 50
Density liquid 1000 (kg m 3) Number )'-cells 100
Liquid shear viscosity 1 x 10 -3 (Pa s~ DT 5.0 × l0 3 (s)
Gas Air
Liquid Tap water

2.0 m

Liquid level
1.5m

0.08 m

0.50 m

Fig. 5. Experimental set up used by Becker et al. {1995). Liquid level is 1.5 m.

their experiments, i.e. with a small and a large superfi- Our model was used to simulate a two-dimensional
cial gas velocity, will be compared with computa- cross section of the bubble column under considera-
tional results obtained from our model. The tion. The gas bubbles, with a diameter of 2 mm, were
experimental setup used by Becker et al. is depicted in generated in a regular pattern at the gas distributor
Fig. 5- consists of a two-dimensional bubble column where the rate of bubble generation was calculated
(width: 50 cm, height: 2 m and depth: 8 cm). The col- from the specified gas flow rate. Experimental condi-
umn is equipped with a gas distributor consisting of tions and numerical parameters are listed in Table 3.
five, individually controllable porous plates.
Becker et al. fed gas (air) to the column (containing 5.2.1. Comparison between model and experiment
watert through only one of the five porous plates ./or large supetJicial,qas velocity. Becker et al. describe
during their experiments. This plate is located 15 cm in detail the flow structure which they have observed
from the left wall of the column and consists of in their bubble column. They reported the existence of
a 40 mm plastic disc with an active pore size of 15 pm. a gross circulation flow over the whole height of the
Becker et al. reported data on the liquid velocity field column which firmly pushes the bubble swarm against
obtained for two different superficial gas velocities. the left wall of the column. As the bubble swarm
Liquid velocities were measured using Laser Doppler reaches the free surface, it is influenced by a secondary
Anemometry and microturbine anemometry tech- vortex which develops at the upper left corner. This
niques. causes part of the swarm to flow downwards near the
1438 E. Delnoij et al.
Table 3. Parameters used in model simulation of experiments reported by Becker
et al.

Bubbles and liquid Column

Bubble diameter 2 (mm) Height 1.50 tm}

Gas flow rate lhight 8.0 (1min ~) Width 0.50 (ml
Gas flow rate (low) 1.6 (1 min 11 Number x-cells 50
Density gas 1.2 (kg m- 3) Number ),-cells 100
Density liquid 1000 (kg m 3) Orifice width 4.0 x 10 -2 (m)
Liquid shear viscosity 1 x 10 3 (Pa s) DT 5.0 x 10- 3 (s)
Gas Air
Liquid Water

right wall of the column. The liquid is observed to gas velocities Becker et al. observed a remarkably
flow upwards in the vicinity of the bubble swarm and different flow pattern in their bubble column. The
downwards along the right wall of the column. bubble swarm now moves upwards in a meandering
A number of snapshots visualising the developing manner. Several liquid circulation cells were reported
flow as calculated by our model are shown in Fig. 6.* which change their location and size continuously.
It can be seen from these figures that the calculated The flow was observed to be highly dynamic. Long-
flow structure is similar to that reported by Becker et time measurements of the vertical liquid velocity at a
al. Our model predicts the development of a powerful point 900 mm above the distributor and 35 mm from
liquid circulation which pushes the bubble swarm the left wall of the column, revealed a period of
firmly towards the left wall of the column. This strong oscillation of the vertical velocity of approximately
liquid circulation is induced by the large n u m b e r of 41s.
bubbles which rise through the column. At the upper Figure 8 shows snapshots of the behaviour of the
part of the left wall a secondary vortex develops which bubble plume at these lower gas velocities as cal-
changes the path of the bubble swarm and causes culated from our model. The same meandering behav-
a part of the bubbles to flow downwards near the right iour as was reported by Becker et al. is clearly
side of the column, as reported by Becket et al. These reproduced by our model. From video representation
bubbles are more or less trapped in the downward of the computational results it can clearly be seen that
liquid flow and form a rotating bubble-cloud which the meandering behaviour of the plume is caused by
does not change in size considerably. a vortex that develops at the upper left corner of the
Becker et al. reported an essentially steady-state column. Subsequently, this vortex moves downwards
flow structure. In this respect the results obtained with along the left wall of the column and disappears at the
our model differ from those obtained by Becker et al. bottom of the column. This process continues with the
The calculated flow structure appears to be more development of a new vortex at the upper left corner
oscillatory due to the interaction of the bubble plume of the column. This continuous vortex development
with vortices that develop near the upper left corner. and disappearance together with the weak circula-
These vortices move up and down along the upper ting liquid flow allows for the bubble plume to
part of the column's left wall. The aforementioned meander.
difference between model and experiment is most Figure 8(f) depicts the vertical liquid velocity as
likely due to the two-dimensional nature of our a function of time at the point chosen by Becker et al.
model. for their measurements of the vertical liquid velocity.
The computed time-averaged liquid velocity field is The mean vertical liquid velocity predicted by the
depicted in Fig. 7, it clearly indicates the existence of a model is - 0.077 m s ~ which differs from the result
gross liquid circulation with liquid upflow near the obtained by Becket et al., who measured a mean verti-
left wall of the column and downflow near the opposite cal liquid velocity of - 0 . 0 3 8 m s-1. The period of
wall of the column. The velocity field depicted in Fig. 7 oscillation of the vertical liquid velocity calculated by
was averaged over the first 300 s of simulation time. the model is approximately 30 s. The smaller period of
oscillation found in our computation and the difference
5.2.2. Comparison between model and experiment between the calculated and the measured mean vertical
.for small superficial 9as velocity. At lower superficial liquid velocity is most likely due to the two-dimen-
sional nature of our model which does not account
for the front and back wall of the 'real-life' pseudo-
*The bubbles indicated in the figures presented in this two-dimensional bed used by Becker et al. To study
paper have been exaggerated in size in order to make them this effect the extension of the present two-dimen-
visible to the reader. sional model to a full three-dimensional model is
 Simulation of dispersed gas--liquid two-phase flow 1439

t = 10,0000 Is]

.~ .-) --) ~ ½

.a .a --) -) ~,

oo ~

o~OOO ,

,® ~'t,

~oo °

od,
oo

o
o
o
o

o
o

~- <-- e- e" ~/
f.. f-- 6- (-- ~ v

's ~ /' v- (-- (-- ~- # ~:

(a)

Fig. 6. Bubble positions and instantaneous liquid velocities at (a) 10 s; (b) 30 s: and (c) 60 s after start of gas
feed. Simulation of experiment by Becket et al. (1995) for high superficial gas velocity. Gas flow
rate = 8.0 l m i n - i . Additional conditions are listed in Table 3.
1440 E. Delnoij et al.

t = ao.oooo [s]

o o

o o

°OoO

o° o

Oo
o o

oo (
o

o
o o
o o o
o
o
o

°o o

o o o

~Oo o o o
o o o
o oOoeg
oo o o° Oo

"ooo o oo

o o °° o o o o o ° oo
°°o o o
o o ° o o o O00Do
000 0 0° 0 0 0 o
O0 oo 0 o
o
oo °o o % o

o oo o o o
o o
o o oc~Oo o o
o oo o o
o Ooo o o o

°Oo °° o o o
o o o o o o o
o o o o ° °° °°
Oo ° o o o o
o o o oo
o oo° o o o o
oo o o o o
Oo o o o o
o o ° o

o° o o o

ooo o o °o o OoO

° o
o ° ° o o
o oo ° o

o
oOO
o oo
o o o

(b)

Fig. 6(b).
 Simulation of dispersed gas liquid two-phase flow 1441

t = 60.0000 [s]

o o oo
o
o o o

o
o,,,
o o

o o °
o o o

o
o
o
o ° o

o oO
o o o
o
o

o
o o
o o
o
o

o
° o
o
Oo
o
o
o
o
o o
o
o o
o
o
o

o
o
o
Oo

o
o
o
o

o o
o
~, o

o o o o o
o o
o o
~o
o o
o~,
o o

o
~o
O°o

°o ao

ooo o o °
o

Ic)

Fig. 6(c).
1442 E. Delnoij et al.
and an empirically determined slip velocity. The coup-
¢ ,,3 ._.--~_--~, ,~'4"'9 ling between the phases is usually achieved through the
effective density of the mixture.
In order to asses the validity of these simplifications
we have used a simplified version of our model to
simulate the behaviour of the bubble plume at the
$ small superficial gas velocity. The model incorporated
a very simple description of the bubble dynamics, i.e.
a bubble velocity which is the sum of the local liquid
velocity and a constant slip velocity. Momentum trans-
,I fer between the phases was neglected. Figure 9 depicts
the behaviour of the bubble plume at the low gas
velocity. From these snapshots, it is clear that a simpli-
fied description of the bubble dynamics and a neglect of
momentum transfer between the phases cannot ac-
count for the highly time-dependent meandering be-
haviour of the bubble plume as observed by Becker et
al. and as calculated from our model incorporating
? detailed bubble dynamics and momentum transfer
between the phases. These detailed bubble dynamics
and the momentum coupling between the phases are,
in our opinion, essential for an accurate description of
the time-dependent behaviour of a gas-liquid bubble
column.

1 5.3. Effect o]" aspect ratio o] a bubble column on its

hydrodynamic hehaviour
There is an extensive literature on fluid circulation
inside bubble columns and its relation to the aspect
ratio of the bubble column. Well-known models are
the Gulfstream model introduced by Freedman and
Davidson (1969) and the multiple circulation cell
model due to Joshi and Sharma (1979). The effect of
the aspect ratio of the bubble column on the liquid
circulation was studied experimentally by Chen et al.
(1989). They studied liquid circulation in bubble
columns with varying liquid depth by streak photo-
graphy. Chen et al. conducted experiments in two
different pseudo-two-dimensional beds at constant
superficial gas velocity.
Chen et al. observed the Gulfstream mode of circu-
lation in bubble columns with an aspect ratio smaller
than or equal to unity. This Gulfstream model con-
sists of two vortex cells opposite to each other, with
Fig. 7. Time-averaged liquid velocity field (0 300 s). liquid upflow in the middle of the column and liquid
Simulation of experiment by Becker et al. (1995) for high downflow near both walls. If the aspect ratio exceeds
superficial gas velocity. Additional conditions are listed in unity, the flow pattern observed is that of multiple,
Table 3. staggered circulation cells. The circulation pattern
proposed by Joshi and Sharma (1979) was not ob-
served.
required. This will be an important aspect of our future Our model was used to study the effect of the aspect
work. ratio of the bubble column on the liquid circulation
pattern. Bubble columns with two different aspect
5.2.3. Comparison with experiments at small superfi- ratios have been studied at a constant gas flow rate of
cial 9as velocity using a simplified model version. In 1.5 ml s-1. Figures 10 and 11 show the calculated in-
recent years a number of publications have appeared in stantaneous liquid velocity profiles and bubble posi-
the literature reporting Eulerian/Lagrangian models tions at different time levels in bubble columns with
for gas-liquid bubble columns. Some of these models aspect ratio's of 2 and 4, respectively. It can be seen
neglect the momentum transfer between the gas bub- that the flow structure clearly changes with the aspect
bles and the liquid phase and evaluate the bubble ratio of the column. For an aspect ratio equal to 2 the
velocity as a simple sum of the local liquid velocity Gulfstream type of flow pattern is observed. It
 Simulation of dispersed gas liquid two-phase flow' 1443

t = 50.0000 [s]

~, ~,,~o Oo
oe ° g ~ ° o o
oOO o o o
o o oe o
o o oo
o o~ooooo o°

/7!.,
o°o%°~ °
¢9 o °o o
o ~ooo o

po. %
0 % ¢P o

Ooo~O o
o ~ o
?,~ ~k' o

o o ~° o

o o o
°° o

%0
~o
g o
o ~ o
o

\
\

N, 6.,..6--~ e----~.-"
i

(a)

Fig. 8. Bubble positions and instantaneous liquid velocities at (a) 50 s; (b) 60 s; (c) 70 s; (d) 80 s; (e) 90 s after
start of gas supply. (f) Vertical liquid velocity at 900 m m above gas distributor and 35 m m from left wall as
function of time (0 300 s). Simulation of experiment by Becker e t al. (1995) for low superficial gas velocily.
Gas flow rate = 1.6 1/min. Additional conditions are listed in Table 3.
1444 E. Delnoij et al.

t = 60.0000 [s]

o~o o oo

o ~, ooO%0o
0o2 o%
o~O,

°o~;
O~o~

o O~o
o o

Oo~o
o ~o

~o
o~

~N

o /
o /

o ~

'x ---~.--#,, a ~" ~ - ~ < - - - e--

(b)

Fig. 8(b).
 Simulation of dispersed gas-liquid two-phase flow 1445

t = 70.0o00 Is]
k.

~o

o oo~ oo

4~

+
~og ~oo

'4

~o

o ~

' I

t \,

m I

ic)

Fig. 8 ci.
1446 E. Delnoij er al.

t = 8 0 . 0 0 0 0 [s]
O;o:~ -O~o:
0% ~--~o.
oo e ° n o ~ °
o° :8o~ o°

°:~&T'
o=~P
0ooXZ°~
oo,=
o@o
O.oV
o o

o o*e

2
(d)

Fig. 8(d).
 Simulation of dispersed gas--liquid two-phase flow 1447

t = 90.0000 [s]
o oooo o 1
oOOo o o o
%0~* ° ° , % , o
ooo o oo o o
o o o °°o° o

,o~q,~#o o oO o
I ~0~Oo Oooo
%, * o***oO o
,o% ~0**%%~o Oo ¢

• ~ ,,,oo,~eo**Oo
o o %11 o

: ~ N e oo

8 *k~****

/2;?
o o

• o

\
\
\
\
¢

%,

tel

Fig. 8(e).
1448 E. Delnoij et al.
0.4 ::::::::::::::::::::::::::

0.3 !iiiiiiiiiiiiiii!i!
0.2
0.1
0
"~ -0.1
;.= -0.2
-0.3
-0.4
-0.5
-0.6
-0.7
{f~ Time Is]

Fig. 8(f).

can be seen that both initially and after several min- presented in this paper, bubbles are assumed to grow
utes of simulation time the bubble plume rises in at the distributor until a specified radius has been
between two circulation cells. For an aspect ratio of 4, attained; at that point in time the bubbles detach from
the behaviour of the bubble column is quite different. the distributor. Near the distributor the bubbles accel-
The bubble plume rises through the liquid in a me- erate under the influence of buoyancy. The only force
andering manner. The plume is influenced by multiple restricting this acceleration to realistic values is the
vortices that develop at the free surface; these vor- added mass force; drag forces are not yet important
tices possess a staggered orientation with respect to because after detachment the bubble velocity is still
each other. In this respect our model compares well very small. Without added mass forces included in the
with Chen's experimental observations, although model, the bubble acceleration reaches an unrealistic
Chert et al. predict a transition from the Gulfstream value because of the relatively strong buoyant forces
flow pattern to the meandering time-dependent flow and the small mass of the bubbles. The omission of the
pattern at an aspect ratio of one. Our results also added mass force causes the model to become un-
indicate that the structure of the two-phase gas-liquid stable. From our model simulations and careful anal-
flow depends on the aspect ratio of the column. ysis of the computed data, we conclude that added
Figure 12 shows the time-averaged liquid velocity mass forces are essential for an accurate description of
profile in the bubble column with an aspect ratio of the behaviour of small, spherical gas bubbles in the
four. The velocities were averaged over the time inter- vicinity of the gas distributor region of the bubble
val from 0 to 60 s. Clearly, the well-known liquid column.
velocity profile with upflow in the column centre and The importance of the lift force acting on the
downflow near the column walls is obtained in this bubbles has also been investigated. The time-depen-
case. The velocity pattern is observed to be highly dent behaviour of the bubble column with a length-to-
symmetric. From Figs 11 and 12 it is clear that the diameter ratio of 4 as studied in Section 5.3 has also
time averaged liquid velocity profile does not capture been calculated without lift forces acting on the bub-
the time-dependent behaviour of the flow in a bubble bles. Figure 13 depicts the instantaneous bubble posi-
column with an aspect ratio equal to four. tions. From this figure and from Fig. l l(a) it can
clearly be seen that lift forces cause the bubble plume
5.4. The importance of the added mass force and the to expand, as the shear-induced lift force acting on the
lift force bubbles in the bubble plume is directed towards the
One of the key features of the model presented in walls of the column. The bubble plume seen in Fig. 13
this paper is the detail with which the bubble dynam- remains narrow, the bubbles do not spread out over
ics is described. In Section 2 it has been assumed that the cross section of the column. The behaviour seen in
the force acting on a spherical bubble rising in a liquid Fig. 13 is not realistic; lift forces acting on bubbles are
is composed of separate and uncoupled contributions therefore important, and must be included in an Eu-
from pressure gradient, gravity, drag force, added lerian/Lagrangian model of a bubble column in the
mass force and the lift force. These last two forces are homogeneous regime. However, the exact value of the
of particular importance despite the fact that they are lift coefficient is still to be investigated.
not usually incorporated in fluid dynamic descrip-
tions of gas liquid two-phase flow. 5.5. Start up of a bubble column
The added mass force appears to be very important A problem of interest for practical application of
in the vicinity of the gas distributor. In the model bubble columns is concerned with their start-up
 Simulation of dispersed gas liquid two-phase flow 1449

t = 10.0000 Es] t = 6 o . 0 0 0 0 [s]

W
W
v
v
V

V
V

V
V

w
w
w ~mmi,
w
w
w
w I
I
I
I I
I I
I
i i
i I
I I
i I
i I
i i
i I
i m
I m
I I
I m
i m
I m

Fig. 9. Instantaneous bubble positions at two different time levels. Simulation performed with a simplified
version of our model which does not incorporate a detailed description of the bubble dynamics and does
not account for m o m e n t u m exchange between both phases. Conditions used in the simulation are derived
from Becker et al.'s experiment for low superficial gas velocity (Table 3}.
 t = 2o.oooo Is]

oo
o ¢
o

% g
ol ,o
o %

,t "- "" I ,,
% o

~:o
o, ~
O,

0 ¢ ~0
g 'o

t"
co°,,,H,

t = 30.0000 [s]
/ e.---e-----4---- ~ ..-~ ----~. .----.-.-~--~
oo o % ~ 0oO~0 0
ooo ,cf e---,%.
oo ~b o o ~ ' o °
o ~ o ooo o
0 ~0000
0 0 0
o%oo~o
~o oo
o~ ~o
%
Oo
T
T
o
' T 1" v
, T

L, T

k,-
o a 1" k-

k.

e-
~H*ou,, ----> .2~ e-- e-" 4
----~ ----> ...# 4------ to---

(a)

Fig. 10 (a) Bubble positions and instantaneous liquid velocity field in a bubble column with an aspect ratio
of 2 at two different time levels. A i ~ w a t e r system. G a s flow rate = 6.0 ml s - '. Grid: 20 x 50 computational
cells. (b) Bubble positions in a bubble column with an aspect ratio of 2 at two different time levels.
A i r - w a t e r system. G a s flow rate = 6.0 ml s '. Grid: 20 x 50 computational cells.
 Simulation of dispersed gas liquid two-phase flow 1451
t = 300.0000 [s] t = 600.0000 [s]

8oooo ~ o ~ °°o °o ~ . . . .
~ o °o °~8 ° o
~o~oo} o°o° ,% o o ~ o o °
% oo~
%o~o ° o oo~ o o
o o o %oO~o
oO°~o o

gos g oO~ o
o°~O
00%°
;g
,g
oo
%
o

(b)

Fig. 10(b).

behaviour. This problem has, to our kinowledge, not 'right-first" result, with both bubble plumes rising
been investigated theoretically yet. Our model has along the left wall of the column. It can therefore be
therefore been used to simulate the hydrodynamics concluded that the flow pattern obtained in a bubble
during start up of a bubble column. As a test problem column depends on the start up procedure of that
the behaviour of a bubble column aerated through bubble column; the history of the flow appears to have
two gas distributors was simulated; the total gas flow a profound effect on the flow pattern which is ulti-
rate used equalled 4.0 ml s-1. Two cases were con- mately established.
sidered; first gas was supplied to a gas distributor at Of course, the two scenarios considered here corres-
the right-hand side of the column centre and after 1 s pond to a rather extreme situation but these results
the gas supply to a distributor at the left-hand side indicate some of the complex features which can be
was started; this is termed the 'right-first' scenario. encountered in practice.
The second start up procedure (i.e. the 'left-first' scen-
ario) is just the mirror image of the first procedure; the
left-hand-side gas distributor was started 1 s before 6. CONCLUSIONS
the right-hand-side gas distributor was started. An Eulerian/Lagrangian model for a gas-liquid
Figure 14 compares the bubble positions for both bubble column has been developed which resolves the
cases 30 s after start up. Video representation of the two-dimensional, time-dependent motion of small
computational results indicated only minor changes spherical gas bubbles in a liquid. The model incorpor-
in the depicted gas liquid flow pattern after these ates all relevant forces acting on a bubble in a liquid,
initial 30 s. This can also be seen from Fig. 15 where such as contributions from the pressure gradient in
the flow pattern after 10 rain of simulation time has the liquid, drag, added or virtual mass, vorticity in the
been shown for the 'left-first' scenario. liquid phase and gravity. A direct bubble bubble in-
It is therefore concluded that Fig. 14 depicts the teraction model resembling Hoomans" collision
flow structure ultimately obtained in the bubble col- model for gas-fluidized beds has been incorporated in
umn. It can clearly be seen that there is a difference in our model. The liquid-phase hydrodynamics are de-
the flow structure obtained in either scenario. The scribed using the volume-averaged Navier Stokes
outcome of the 'right-first' simulation indicates that equations.
both bubble plumes rise along the right-side wall of Our model has been validated using experimental
the bubble column. A single circulation cell develops data from the literature obtained by various workers.
near the left wall of the column. The flow pattern for The model was shown to compare well to the experi-
the 'left-first' simulation is just the mirror image of the ments conducted by Becker et al. The behaviour of the
1452 E. Delnoij et al.

t = 25.0000 [s] t = 27.0000 [s] t = 29.0000 [s]

~oo~ o~oo~ ~o o o ~ =o 8 o~ o_o~
O oo; o ,-oo. o o. ~o o
o o 2 o_o_~o o
~o°~~ o

o 0%0, 0o0 8
o o ooO o o 0 or ~ 0 0
°~ooo o~
OooOO o o
o ~ Oo ° oO
oq~ o OoO o° o
~g og
~ oo
,go o OOoo o o %o Oo~ ~o
~, oo
o oo,
io f
o
o % ooo0 o°O° %: o % 0 0 0
ooo ~ o o o °o°
oOo
o o
o o
o Ooo o o ° ~
~
oOoi
OOOo
8 OoO o o 0% ~ o° oo o
o ooo o o
% Oo%O o o °o o oO O°o
o oooo~ ~ o, °o o o O O o
% oP o°o° °°° c% ° °Oo
o o o°oc o
°° oo
o o°°°
oo O O ° o
o
o o 0°o°° o o oO Oo oo o o o
o o o o oo o ( o oo
Oo oOO ;~ ) o ooooo o o o O o ° o ~o~O'°
o Oo o
oOO c
°oOo~
o o ,
~o
° o
08o
o °o~°o
o o 00000%
^

°%, o o°~ o
o o o~_oo
o o

o -o~
o °o og o o

o o o o oo oe o
o oo° o OooOe ° I
g ~I~
o o oc~ o % o 0OO6o o

° o o
o o
0 O0o°~ 0
oo°~°~°~° o
Oo o ~oOO
° o oool
o o 0%0 8~o Oo oO~ o0 o o
o o
v
o
0
oo 8 o ( ( Oo 008 e o o [

eT~
°o o°~ ~ o o
% oo°~o° % oo
° o cooo~ o o
o o ooooO o
OoO°gg CO !
0 o Oo % Oo~Oo~ oo
o°Ooo o~~~~ OoO°°o~
o oOo~OOo
%ooo~_.oo OOo
°~
o

go ~oo o
o
c
OoO( ooo
o ~%= o 0%00
%0
ooo Oo o Oo~
. °~ ~ooo
o o ~°
o~ %0
o 0 ° % oo oo o
o o~ ° o %0 o
ooo~,
~ ~o
0%% o
,~°o° oo o
oo o
8% i~ °°
~o oo
o
o(

o
%, o
o
o
c
o
OoO °o° ~ o
o ~o (9
o~
o o

o o °~ o, ~
o@
o~ ~
000tll~

(a)

Fig. 11 (a) Instantaneous bubble positions in a bubble column with an aspect ratio of 4 at three different
time levels. Bubbles are depicted somewhat larger as compared to the scale of the column. Air-water
system. G a s flow rate = 6.0 ml s - 1. Grid: 20 x 100 computational cells. (b) Corresponding instantaneous
liquid velocities in a bubble column with an aspect ratio of 4 at three different time levels.
 Simulation of dispersed gas--liquid two-phase flox~ 1453

J
• I!

") ;~ '3

~5! el
'N )
b ,f" t !ii • k

i '
I '
+ '~?2

I.
i~,J+ +

! +

(b)

Fig. ll(b).
1454 E. Delnoij et al.
found to be of the same order of magnitude as the one
observed by Becker et al.
The importance of incorporating a detailed descrip-
tion of bubble dynamics and of incorporating mo-
mentum transfer between the gas and the liquid phase
into the model was also shown. For the geometry
TT studied by Becker et al. it was shown that a simplified
version of our model, without detailed bubble dynam-
TT ics and momentum exchange, was not able to predict
the experimentally observed, time-dependent gas-
TT liquid flow pattern.

TT 1'^¢ The model was also used to investigate the effect of

the aspect ratio of the bubble column on the overall
TT liquid circulation pattern. This effect has been studied

Tt experimentally by Chen et al. Our computational

results indicate a transition in the gas liquid flow

tT pattern as the aspect ratio of the column changes from

two to four. For an aspect ratio equal to two the
1',,+ Gulfstream type of liquid circulation was observed. At
an aspect ratio of four however, a highly dynamic
liquid flow pattern with multiple vortices was ob-
served. These vortices were observed to be generated
T at the free surface and found to be positioned stag-

TT
gered with respect to each other. In part, these results
support Chen's findings. Chen et al. also observed
T a transition in the gas-liquid flow pattern from the
Gulfstream mode to the highly dynamic multiple vor-
tex mode. However, Chen et al. found this transition
to occur at an aspect ratio of one.
,r T One of the key features of the model presented in
this paper is the detailed bubble dynamics. The im-
portance of the added mass force and the lift force
acting on the bubbles was investigated theoretically.
It was found that the added mass force is of particular
importance near the gas distributor. Neglecting the
added mass force will (initially) cause the bubbles to
accelerate at an unrealistic high rate. The lift force
acting on the bubbles in the bubble plume is directed
towards the walls of the column, and causes the
~" ~ ¢ ~ bubble plume to diverge. Without lift forces, the bub-
bles do not spread out over the cross section of the
column, resulting in an unrealistic, narrow bubble
plume. It is therefore concluded that lift forces acting
on the bubbles must be included in any Euler-
ian/Lagrangian model of a gas-liquid bubble column
Fig. 12. Time-averaged liquid velocity field in the bubble operating in the homogeneous regime.
column with an aspect ratio of 4. Conditions as listed with Finally, our model was used to study effects en-
Fig. 11. Velocities averaged from 0 to 60 s after start up of countered in start up of bubble columns. A model
the bubble column. system was chosen to investigate the effect of start up
procedure on the steady-state hydrodynamics. In the
case one start up procedure the bubble column was
initially only aerated via a gas distributor at the right
side of the column. After 1 s gas was also fed to a gas
distributor at the left side of the column. The case two
bubble plume in Becker's two-dimensional bubble start-up procedure was just the mirror image of this
column was predicted accurately for both the high first start-up procedure. It was shown that the instan-
superficial gas velocity and the low superficial gas taneous liquid velocity fields ultimately obtained in
velocity case. The experimentally observed meander- these two different situations are mirror images of one
ing behaviour of the bubble plume at low superficial another. It was also found that the history of the flow
gas velocities was calculated where the period of oscil- had a marked influence on the prevailing flow pattern
lation of the bubble plume predicted by the model was under pseudo-steady-state conditions.
 Simulation of dispersed gas liquid two-phase ttow 1455

t = 2 7 . 0 0 0 0 [s]
o~
70 ¢-
1 9~'-;

o~

o ~
~o
o o
~g

¢, e U

r\ <,.s.-.e-- e-"

Fig. 13. Instantaneous bubble positions and liquid velocity field at 27 s after slart up in a bubble coluxnn
with an aspect ratio of 4. Bubbles are depicted somewhat larger as compared to the scale of the bubble
column. Air water system. Gas flow rate = 6.0 ml s ~. Grid: 20 × 100 computational cells. Lift forces are
not included in the model.
1456 E. Delnoij et al.

t = 3 0 . 0 0 0 0 [s]

t = 3 0 . 0 0 0 0 Is]

4.
"i' 1" ^ o., ~, ¥ ~ 1" 4, +
,L
J,

$
$

L//'-~.~\ L

I
b
L
;\ \ ~ . < - - ~ Z i i

Fig. 14. Bubble positions and liquid velocities at 30 s after start up for both start up procedures. Top:
"right-first' senario. Bottom: 'left-first" senario. Gas flow rate = 4.0 ml s - ~. Grid: 20 × 50 computational
cells.
 Simulation of dispersed gas-liquid two-phase flow 1457

t = 570.0000 Is]

O o ® o~ ~°o~ o

I7~o
i®~
OooOOo
~o o o
oo-O
o oo Oo°
P.ooo ....
o o
2°o

^ o o Oo
I o ~ o ~ , ~ o o o °o
o
~
o~
oo
Oo
oi Oo-O
< ~_
,
,0 oo
o 10o
~-o
oo o

! >0%0
I ~ °° 2 o o o~ ~
o oS
I'<~ I ~ <~
o o oo
~> ~e 0%0
|o~o o ~ ~o
• o ~ OOoooo~ ~ ~o
] ° o< #0°%° 0 o~ o oo 8 ~o~ o
...%-~
°

Ooo
oO.... ~ o

~, ~ ~/ ~/

Fig. 15. Bubble positions and liquid velocities at 570 s after start up for "left-first' scenario. Gas flow
rate = 4.0 ml s ~. Grid: 20 x 50 computational cells.

NOTATION Greek letters

A area, ITI 2 el volume fraction liquid in computational cell,
CD drag coefficient, dimensionless dimensionless
CI. lift force coefficient, dimensionless ~,- velocity gradient in calculating lift force, s
Cv,~ virtual mass coefficient, dimensionless )-1 bulk viscosity liquid, kg m 1 s i
DT time step solver liquid flow field, s h shear viscosity liquid, kg m ~s
E unit tensor, dimensionless p~ density gas phase, kg m 3
Fo drag force on a bubble, N Pl density liquid phase, kg m s
F¢; force on bubble due to gravity, N rl stress tensor liquid, N m - - '
Fz. lift force on bubble, N 4~ typical Eulerian variable, dimensionless
Fp force on bubble due to pressure gradient, N m o m e n t u m exchange gas to liquid, N m .t
~total total force on a bubble, N f~ vorticity in liquid phase, s 1
Fv.w virtual mass force on bubble, N
REFERENCES
g acceleration due to gravity, m s 2
Auton, T. R. (1983) The dynamics of bubbles, drops
I kelvin impulse deformable body, Ns and particles in motion in liquids. Ph.D. thesis,
D~b mass of a bubble, kg University of Cambridge, Cambridge, U.K.
P pressure. N m 2 Becker, S., Sokolichin, A. and Eigenberger, G. (19951
r position of bubble, m Gas-liquid flow in bubble columns and loop reac-
ra position of bubble of bubble a, m tors. Part II. Comparison of detailed experiments
rh position of bubble of bubble b, m and flow simulations. Chem. Engng Sci. 49, 5747.
Rb radius of a bubble, m Chen, J. J. J., Jamialahmadi, M. and Li, S. M. (1989)
Re~ Reynolds number for flow around a bubble, Effect of liquid depth on circulation in bubble col-
umns: a visual study. Chem. Engng Res. Des. 67, 203.
dimensionless
Chesters, A. K., van Doorn, M. and Goossens, L. H. J.
tab time until collision between bubbles a and
(1980) A general model for unconfined bubble
h.s plumes from extended sources. Int. J. Multiphase
U liquid velocity, m s Flow 6, 499.
/'/rise rise velocity of a bubble, m s Clift, R., Grace, J. R. and Weber, M. E. (1978) Bubbles
¥ velocity of bubble, m s Drops, and Particles. Academic Press, New York,
v~ volume of a bubble, m 3 U.S.A.
1458 E. Delnoij et al.
Deckwer, W. D. and Schumpe, A. (1993) Improved simulation of the hydrodynamics of a two dimen-
tools for bubble column reactor design and scale sional gas-fluidized bed. Comput. Chem. Engng 17,
up. Chem. Engng Sci. 48, 889. 839.
Devanathan, N., Dudukovic, M. P., Lapin, A., Liib- Lapin, A. and Liibbert, A. (1994) Numerical simula-
bert, A. (1995) Chaotic flow in bubble column reac- tion of the dynamics of two phase gas-liquid flows
tors. Chem. Engn9 Sci. 50, 2661. in bubble columns. Chem. Engng Sci. 49, 3661.
Duineveld, P. G. (1994) Bouncing and coalescence of Odar, F. and Hamilton, W. S. (1964) Forces on
two bubbles in water. Ph.D. thesis, Twente Univer- a sphere accelerating in a viscous liquid. J. Fluid
sity, Enschede, The Netherlands. Mech. 18, 302.
Freedman, W. and Davidson, J. F. (1969) Holdup and Ranade, V. V. (1992) Flow in bubble column: some
liquid circulation in bubble columns. Trans. lnstn numerical experiments. Chem. Engng Sci. 47, 1857.
Chem. Engng 47, T251. Sokolichin, A. and Eigenberger, G. (1994) Gas liquid
Grienberger, J. and Hofman, H. (1992) Investigations flow in bubble columns and loop reactors. Part I.
and modelling of bubble columns. Chem. Engng Sci. Detailed modelling and numerical simulation.
47, 2215. Chem. Engng Sci. 49, 5735.
Hjertager, B. H. and Morud, K. (1995) Computational Svendsen, H. F., Jakobsen, H. A. and Torvik, R. (1992)
fluid dynamics of bioreactors. Mod. Indent. Control. Local flow structures in internal loop and bubble
16, 177. column reactors. Chem. Engng Sci. 47, 3297.
Hoomans, B. P. B., Kuipers, J. A. M., Briels, W. J. and Torvik, R. and Svendsen, H. F. (1990) Modelling of
van Swaaij, W. P. M. (1996) Discrete particle simu- slurry reactors, a fundamental approach. Chem.
lation of bubble and slug formation in a two dimen- Engng Sci. 45, 2325.
sional gas-fluidised bed: a hard sphere approach. Trapp, J. A. and Mortensen, G. A. (1993) A discrete
Chem. Engng Sci. 51, 99. particle model for bubble slug two phase flow. J.
Joshi, J. B. and Sharma, M. M. (1979) Some design Comp. Phys. 107, 367.
features of radial baffles in sectionalised bubble Wijngaarden, L. van (1976) Some problems in the
columns. Can. J. Chem. Engng 57, 375. formulation of the equations for gas/liquid flows. In
Kuipers, J. A. M., van Duin, K. J., van Beckum, Theoretical and Applied Mechanics, ed. W. T. Koiter.
F. P. H. and van Swaaij, W. P. M. (1993) Computer North-Holland, Amsterdam, The Netherlands.