Finite-Element Analysis of Taylor Flow

Rolf K. Edvinsson and Said Irandoust

Dept. of Chemical Reaction Engineering, Chalmers University of Technology, S-412 96 Goteborg, Sweden

A finite-element analysis of Taylorflow in a cylindrical capillary was performed using

a commercial FEM program (FIDAP) to solve the fundamental fluid dynamics equa-
tions together with the capillary forces at the gas - liquid intetface. A moving-surface
fornulation was used to calculate the bubble shape, The thickness of the liquid film
surrounding the gas bubble, the degree of miwing in the liquid phase, and the slip veloc-
ity between the two phases were calculated. These parameters influence the pe$ormance
of monolith reactors operating in the Taylor flow regime. On comparison with expen-
mental results it was found that the FEM calculation generally predicts a thinner liquid
film, which can possibly be explained in terms of a peripheral variation in sutface ten-
sion. Moreover, the wavelength of the wiggles predicted in the liquid film near the tail
end of the bubble was compared to those arisingfrom a simplified mathematical analy-
sis available in the literature. Good agreement was found for Ca < 0.005, while for
higher Ca the FEM predicts significantly shorter wavelengths, indicating that the lubri-
cation theory is not valid here.

Introduction
Segmented gas-liquid flow, also known as Taylor flow, is can be modeled as a network of narrow, cylindrical channels
an important flow pattern in cocurrent gas-liquid flow in with a radius typically in the range 1-100 p m (Schwartz et
monolithic catalyst reactors. A monolith consists of many al., 1986). With low flow rates and narrow channels, capillary
narrow, parallel channels with the catalyst deposited on the forces (surface tensional forces) dominate. As the channel di-
walls. Monoliths are characterized by low-pressure drops and ameter and flow rate are increased, viscous, inertial, and
high surface-to-volume ratio. They are primarily used for gravitational forces become influential as well. Flow regimes
gas-solid processes, b u t a r e also attractive for in gas-liquid flow are discussed by, among others, Taitel and
gas-liquid-solid reactions. The topic has been reviewed by Dukler (1976), Troniewski and Spisak (1987), and Whalley
Irandoust and Anderson (1988) and Cybulski and Moulijn (1987).
(1994). Fairbrother and Stubbs (1935) pointed out that the liquid
Taylor flow exhibits some special advantages for chemical film surrounding the gas bubble caused the two phases to
processes. The gas bubble, having an equivalent diameter ex- travel with slightly different speeds. This had implications for
ceeding that of the capillary, is deformed into a long plug the practice of measuring liquid flow velocities by introducing
separated from the wall by only a very thin liquid film that an indicator bubble and measuring the speed of the latter.
permits high gas-to-wall mass-transfer rates. A recirculation Their experiments led them to recommend the following
pattern is induced in the liquid plug trapped between two gas correlation correcting for this difference: 1 - UL/lJc =
plugs, which increases radial mass transfer and reduces axial ( p U c / y ) v , where U, and U, denote the mean velocities of
dispersion. the liquid and gas phases. The group pU/y is a capillary
Another example of two-phase flow of a wall-wetting liquid number, Ca, and is the ratio between the viscous and the
and a dispersed fluid in a capillary is found in enhanced oil capillary forces ( p is the viscosity of the continuous, liquid
recovery. Oil yield is increased by driving the oil from the phase). If nothing else is stated, the reference velocity, U, is
porous rock by another fluid (a liquid or a foam). The rock the linear speed of the bubble. Their expression was valid in
the range lop3< Cu < Later, Taylor (1961) was able to
Correspondence concerning this article should be addressed to S. Irandoust.
Current address of R. K. Edvinsson: Dept. of Chemical Engineering, Technical
extend this range to < Cu < lo-'. If the film is thin, it is
University of Delft, Julianalaan 136, 2628 BL Detft, The Netherlands. reasonable to assume that it is at rest with respect to the tube

AIChE Journal July 1996 Vol. 42, No. 7 1815

 Table 1. Correlations for Estimation of Liquid-Film Thickness
hdR Range Method Continuous Phase Dispersed Phase Reference
0.5 caw 5X < Ca < 3 X 10V4 Conductimetry Water Air Chen (1986)
1.337 Ca2P 2x < Ca < 2 x Conductimetry Water Oil Chen (1986)
[0.89-0.05AU,)1p] Ca'b 7X < Ca < 2 x Conductimetry Water, aqueous glycerol Air, benzene Marchessault and
U, in cm/s Mason (1960)
1.337 Ca2P 1 0 - ~< cu < Volumetry Aniline, benzene Air Bretherton (1961)
0.36(1-exp(-3.08 C U ' . ~ ~ } ) <Cu<1.9 Colorimetry Water, ethanol, Air Irandoust and
glycerol (aq) Andersson (1989b)
0.5 Ca'p 0.013 < Cu < 0.09 Microscopy Various oils Various liquids Goldsmith and
Mason (1963)
0.5 Ca'b 10-3 < ca< 10-2 Volumetry Aniline, ethanol, Air Fairbrother and
benzene, water Stubbs (1935)

wall, and then a simple mass-conservation argument leads to to a monolithic catalyst reactor. The objectives are (1) to adapt
the following relationship: a method that allows an accurate calculation of the flow pat-
tern; (2) to derive performance characteristics from this solu-
tion; and (3) to establish, by a parameter study, the essential
features of the relationship between, primarily, the film
thickness and the various flow parameters.

Here h, is the film thickness, measured far enough behind Theory

the bubble head to allow a film of constant thickness to de- Model assumptions
velop, and R is the capillary radius. The relation between The model of Taylor flow, on which the subsequent devel-
film thickness and capillary number has been investigated ex- opment is built, is based on the following assumptions:
perimentally by several workers, as summarized in Table 1. 1. Equal sue and spacing of the gas bubbles.
The mathematical analysis of Taylor flow is complicated by 2. Periodicity of the velocity field in the axial direction.
the fact that the bubble shape is part of the problem. At low 3. The flow is steady in a coordinate system moving with a
Reynolds numbers it is justified to linearize the Navier-Stokes gas bubble. In a fixed coordinate system the flow would be
equations by the creeping flow assumption. It is, however, nonstationary. In particular the boundary conditions at the
not sufficient to solve the equations of flow; the location of channel inlet/outlet would be time-dependent.
the boundary, or the shape of the domain, must be deter- 4. The channel orientation is vertical with flow in either
mined as well. Hence the overall problem remains nonlinear. direction.
By using additional simplifications, valid in the thin-film re- 5. The flow is axially symmetric and nonswirling.
gion, the problem can be reduced to a nonlinear ordinary 6. The flow is laminar, that is, the full Navier-Stokes equa-
differential equation (Bretherton, 1961). For high flow rates tions are solved and no turbulence model is used.
the assumption of potential flow has been used as a starting 7. Constant pressure within a gas plug. Also, the gas is as-
point for analyzing slug flow but, obviously, the problem of sumed to have negligible density. This implies that the gas
determining the bubble shape remains. For further details exerts a pressure on the liquid, but carries no momentum.
see Bendiksen (1985), Collins et al. (1978), Fabre and Link This assumption allows us to remove the gas phase from the
(1992), Ginely and Radke (1989), Goldsmith and Mason explicit calculation and greatly improves computational speed.
(1963), Ratulowski and Chang (1990). This approach is also supported by work of Schwartz et al.
As an alternative to analytical methods, it is also possible (1986) and Westborg and Hassanger (1989).
to perform computational fluid dynamics simulations. A sim- 8. All physical properties are considered to be constants,
ple strategy is to solve the equations of flow for a fixed geom- though the qualitative effect of a suspected variation in sur-
etry and then adjust the shape on the basis of the solution in face tension is given some attention.
a second separate step. These two steps are then repeated 9. The liquid plugs are long enough for a fully parabolic
until convergence is obtained. Irandoust and Andersson velocity profile to develop in the middle of them. This is not
(1989a) and Mao and Dukler (1990) used the finite-dif- a necessary, but a convenient, assumption simplifying the for-
ference method (FDM), while Shen and Udell (1985) and mulation of the boundary conditions.
Westborg and Hassager (1989) used the finite-element
method (FEM).
This decoupling of flow and shape, though simple, can Governing equations
cause convergence problems. A more sophisticated approach The equations of flow and the necessary boundary condi-
is to incorporate the location of the interface directly into the tions given in the following sections are all presented in the
system of nonlinear equations arising from the equations of dimensionless form. Details are given in the Notation section
flow. One way to achieve this is to use the boundary integral and Figure 1. Conservation of momentum in the radial and
equation method (Martinez and Udell, 1990; Stone and Leal, axial directions:
1990).
In this work we examine the use of a FEM code to simu- du, au, 1 d 1 durz
late Taylor flow over a wide range of flow conditions relevant
ur-+uZ-=--(rurr)--u
dr dz r dr r
+- d z (2)

1816 July 1996 Vol. 42, No. 7 AIChE Journal

 ~ ~ ~~ ~ ~ ~~ ~

Figure 1. Geometry of the computational domain.

1 d considered to be a mapping of the interval [0,1] onto the r - z

u,-
du,
dr
+ u,- du,
dz
= --(rq.,)+
r dr
d%
-dz
+ -.Fr1 (3) plane:

This can be made to be applied to both upward and down- r5=ws) 0 1 ~ ~ 1 . (10)
ward flow if we assume that the direction of flow is the posi-
tive z-direction and use the convention that Fr > 0 corre- At the gas-liquid interface
sponds to downward flow and Fr < 0 signifies upward flow.
The various stresses that appear in Eqs. 2 and 3 are given by
the following expressions: = -vy, (11)

2 du, 2 ur
a,,= - p +- -;
Re d r
a,, = - p +- -;
Re r
and the righthand side of this expression reduces to zero when
variations in surface tension are neglected. A stable gas plug
moving with the same speed as the coordinate system, u , must
obey the following condition in order to retain its shape and
position:
The continuity equation: Unlr, = 0. (12)

1 d du, The endpoints of the curve defining the interface must, due
--(rur)+ - =O.
r dr dz to symmetry considerations, intersect the axis of symmetry at
a right angle.
Boundary conditions Since we are considering incompressible fluids the pres-
sure level can be selected arbitrarily. With the gas side pres-
The equations are solved with reference to a translating sure set to zero, the stress balance at the curved interface
coordinate system. The constant rate of translation is v > 0. relates the curvature of the interface to the liquid side pres-
A parabolic velocity profile is imposed at the end of the com- sure:
putational system, that is, on the boundary segment r, (see
Figure 1) 2 H + - -2. d u n
p'r5=-z
Re dn
(13)

The boundary conditions on the liquid-phase velocities are

and the periodic velocity condition is such that liquid volume is conserved. Since the system vol-
ume is constant, the gas volume is also conserved.

At the wall: Nondimensionul parameters

The linear velocity of the bubble is frequently used as a
U , k * = 0; u,lrz = - v. (8) reference velocity but, since it is treated as a dependent vari-
able in this work, it is more convenient to use the mean ve-
Axial symmetry: locity of the two phases as the reference velocity:

4 Qc+QL
(U)=- ~ (14)
rr( D2 ) *

Since we assume that the bubble possesses rotational symme- With the channel diameter, D, as the characteristic length,
try, it suffices to use a single curve to represent its shape. the unit of time becomes D / ( U ) , the time required to move
The interface between the gas and the liquid, r,, can thus be a distance D at the average velocity. To completely charac-

AIChE Journal July 1996 Vol. 42, NO. 7 1817

 V recirculation near-wail
region \ region
\

fixed elements flexible elements fixed elements

Figure 2. Final mesh and the corresponding streamlines.

terize the system, five dimensionless groups are needed. They requirement that the bubble should remain at the same posi-
are the Reynolds, Froude, and capillary groups tion. This leads to the following nonstationary problem, rep-
resented by a function f 2 :

( v” ,Sf) = f ( Re,Fr ,Cu ,E ,a ,v r) (18)

as well as the void fraction, E, and the aspect ratio, a : Here v r is the currently best estimate of the bubble velocity,
which is identical to the translational speed of the coordinate
system. The bubble is now free to move, and the net drift
(16) velocity is the difference between the actual velocity and the
translation velocity of the coordinate system:

Finite-Element Simulation (19)

Discretization Now v is determined iteratively by finding the value of v that
The meshing of the domain is also illustrated in Figure 2. makes the drift velocity zero, that is, v is selected so that the
Nine node quadrilateral elements have been used. Two types boundary condition (Eq. 12) is satisfied. In practice, a tran-
of elements have been employed; those filling the region sur- sient solution is obtained for a short time period based on a
rounding the bubble are flexible and may change their shapes first guess of the drift velocity. The observed drift velocity is
and sizes in order to accommodate the motion of the inter- then used to adjust this value. This procedure normally con-
face, while the rest are fixed in space. The interface must lie verges rapidly, after which it is possible to perfom a long
inside the region of flexible elements. Moreover, each node is transient simulation with virtually no change in bubble shape
constrained to a prespecified line, denoted a spine. For fur- and position. In most runs, this technique was used to pin
ther details see FIDAP, Theoretical Manual (1991). down the velocity to four or five decimal places. It should be
pointed out that if a steady-state simulation is attempted, even
the slightest error in the estimate of v would cause a net
Solution strategy
drift. As the interface reaches the limit of the flexible ele-
The mathematical problem is composed of several parts. ments, mesh destruction is inevitable.
The conservation equations apply in the fluid phase, a two- Another problem, related to the formulation of the bound-
dimensional domain. The bubble shape is described by a curve ary conditions, is how to drive the liquid through the channel.
and thus has a one-dimensional domain. Finally, the (scalar) It would be desirable to impose a finite pressure drop over a
speed of the gas plug is a dependent variable. To illustrate long section encompassing many bubbles. This would be
this we may represent the problem that we set out to solve by computationally demanding, and in the present work we have
a function fi: restricted ourselves to one repetitive unit. Three formula-
tions of the boundary conditions have been tried: (1) impos-
ing a pressure drop across one gas/liquid unit; (2) imposing a
parabolic; and (3) a flat axial velocity profile on rl and r,.
Unfortunately this problem cannot easily be solved in a sin- All formulations gave very similar results. The Parabolic
gle-step simulation with the present simulation. This is be- profile is used henceforth since it is both convenient and
cause the boundary conditions are functions of the bubble realistic.
speed, which we do not know beforehand. Two approaches
are possible. First we may vary the translational speed of our
coordinate system so that it matches the bubble speed. This Results and Discussion
will require inclusion of acceleration terms, since the coordi- A typical shape of a Taylor bubble is shown in Figure 3a.
nate system is not moving with constant speed. The alterna- The plot of streamlines indicates the surrounding flow field.
tive approach, used in this work, is to temporarily drop the It can be seen that the liquid phase can be divided into two

1818 July 1996 Vol. 42, No. 7 AIChE Journal

 At both ends of the computational system, approximately
in the center of the liquid plug, a parabolic velocity profile
was prescribed. As a test, a flat velocity profile, giving the
same volumetric flow rate as the parabolic one, was imposed
instead. The effect on the film thickness, and thus V , was
minor, even when the liquid plug length was about one capil-
lary diameter. Similarly, the length of the gas plug had no
effect as long as it was large enough to form a plug.
The proper convergence of the solution was tested in sev-
eral ways. The nodal density of the mesh was varied in order
to establish grid independence. Since a transient solution
method was used only as a means to obtain a steady-state
solution, a relatively high tolerance was used. As a steady-
state solution appeared to have been reached, the tolerance
could be reduced by five orders of magnitude without signifi-
cantly altering the solution. Before the bubble had relaxed to
its final shape, the time increment was limited by stability
considerations; as the final shape was reached this limit in-
creased, sometimes permitting a single time step of 20 to be
taken. This corresponds to the bubble moving a length of ap-
proximately 20 diameters in one step without changing shape,
and implies that a steady-state solution of high quality was
obtained. In most cases, a solution could be obtained within
5 to 30 CPU minutes using a HP720 workstation, provided
that a good starting guess was supplied. In many cases the
procedure had to be repeated a few times to allow revision of
meshing.
The calculation of wiggles, especially when the inertial
forces were significant, was more demanding. In order to re-
solve the wiggles a finer mesh was employed; moreover, a
small time increment was required to maintain stability. As a
result, some of these calculations required 10 CPU hours or
a more.

Figure 3. Effect of viscosity ( pb = 5 p a ) on the stream-

Influence of the capillary number
line pattern.
(a) Re = 100, Fr = 1, Ca = 0.04, a = 8, E = 0.50.(b) Re = 20, Figure 4a-4d constitutes a sequence illustrating what hap-
Fr = 1, Ca = 0.04, a = 8, E = 0.50. pens when the viscous effects become more dominant. The
parameters of Figure 4 are given in Table 2. The convexity of
the rear meniscus is first lost and then finally inverted. If the
regions. The closed streamlines within the liquid plug form a diameter is sufficiently large and the liquid viscous enough, it
recirculation region, and the open-ended streamlines passing is possible for the inverted rear meniscus to penetrate all the
through the liquid film mark liquid elements that will remain way through the bubble. This can under some circumstances
in the vicinity of the wall. Increasing the viscosity by a factor result in a stable toroidal bubble. This peculiar flow pattern
of 5 results in the situation illustrated in Figure 3b. The bub- is called stalactite flow (Edvinsson et al., 1992).
ble head is more pointed and the tail almost flat. In addition,
a much thicker film is formed and the recirculation is signifi- The eflect of inertial forces
cantly reduced.
In reaction engineering applications, we can expect to en-
counter higher flow rates than those typical of capillary flows.
As the Re number increases, the almost perfectly spherical
Solution accuracy ends typical of the flow dominated by capillary forces are lost.
When assessing the quality of the numerical solution, two The sequence in Figures 4e-4g, corresponding to Reynolds
kinds of questions need to be addressed: (1) Is this an accept- numbers 500, 1,000, and 2,000, illustrates this. In the film
ably converged solution to the set of equations we wanted to near the rear meniscus, a multiple wavelet has formed and its
solve? and (2) Are the assumptions made in order to obtain amplitude increases with Re. The film is also predicted to
this set of equations acceptable? The latter question can be thicken slightly, though a precise measure is difficult since
represented by the artificial boundary conditions that must there no longer appears to be a region of constant thickness.
be applied in order to limit the size of the computational This sequence also represents increasing computational diffi-
domain. Ideally we would like these assumptions to be neu- culties; as the critical Re number is approached, one needs
tral in the sense that they do not significantly affect the im- to shorten the time steps in order to achieve convergence at
portant aspect of the resulting solution. all. Therefore the cautious reader should consider the shapes

AIChE Journal July 1996 Vol. 42, NO.7 1819

 .:
C d

-l

Figure 4. Calculated bubble shapes.

The parameters are summarized in Table 2.

in Figures 4e-4g as “snapshots” rather than steady-state so-

lutions, since it cannot be guaranteed that the solution is fully
converged. Indeed, an oscillating solution is encountered.
P1,2 =
Caby
2 ~ /w t
27r,
- ; Ca,=Cav;
~

Undulations in the liquid film

Wiggles, very similar to the ones calculated here, have been
observed experimentally (Figure 5). As their amplitudes are
increased still further, a snap-off process can occur in which
the bubble splits into two smaller ones in close proximity.
Ratulowski and Chang (1989) used an arclength formula-
tion of the lubrication equations to calculate the bubble pro-
files. After some simplifications, their analysis led to an ana-
lytical expression for the characteristic wavelength, which in
our notation becomes

where

Table 2. Parameters Used in the Simulations in Figure 4

Label CI E Re 1/Fr Ca
a 5 0.2 20 1.o 0.20
b 5 0.2 20 1.0 0.38
C 5 0.2 20 1.0 0.50
d 5 0.2 20 1.0 1.0
e 8 0.5 500 0 0.01 Figure 5. Undulations in the liquid film.
f 8 0.5 1,000 0 0.01 Petroleum ether and air at 20°C in a glass capillary of inter-
8 0.5 2,000 0 0.01 nal diameter 4.3 mm. Flow rates: petroleum ether 0.523
g mL/s, air 0.023 mL/s (upflow).

1820 July 1996 Vol. 42, No. 7 AIChE Journal

 -2.5 -2.0 -1.5 -1.0
1.0 u -2.5 -2.0 -1.5 -1.0
log10 Ca loglo Ca
Figure 6. Shapes of Taylor bubbles as a function of the - Irandoust and Andersson (1989b) - - - Bretherton (1961)
capillary number.
Ca is (a) 0.001; (b) 0.0032; (c) 0.01;(d) 0.03;and (e) 0.06. ..... Fairbrother and Stubbs (1935) rn Thisstudy
Taylor (1961)
Figure 8. Influence of Ca on (a) S,; (b) Y.

The value of ro can be evaluated by using Bretherton's limit Re = 200, (Y = 8, E = 0.5, Fr-' = 0.

10 = 1 - 1.3375 C U ; ~ . (22)
For small capillaries it is sufficient to correlate the film
thickness and the relative velocity with the Cu number. As
As an illustration consider the bubble profiles shown in Fig- flow rates and diameters are increased, the influences of in-
ure 6. The capillary number, and thus the film thickness, in- ertial forces as well as gravitational forces become more pro-
crease in the sequence a-e (note that the aspect ratio has nounced. The dependence of 6f and u on Cu is illustrated in
been changed to highlight the wiggles). Figure 8. Gravitational forces have been omitted here. Super-
It is possible to estimate the wavelengths from Figure 6 imposed are the results of some other workers. The overall
even though the undulations are not perfect sine waves. We behavior is the same, though the FEM solution systematically
see that initially both the amplitude and the wavelength go predicts thinner liquid films than those reported experimen-
through a maximum on increasing Cu. The estimated wave- tally (see Table 1).
lengths are plotted in Figure 7. In addition the analytical pre- The influence of gravity is illustrated in Figure 9. The re-
diction (Eqs. 20-22) is plotted. The middle curve was ob- ciprocal of the Froude group is proportional to the gravita-
tained by using the ro from the FEM calculation rather than tional acceleration. A negative sign represents upflow. For
the Bretherton limit of Eq. 22. The discrepancy between the low Cu numbers, the film is thin and only a small velocity
two predictions is reduced, though the maximum found in gradient develops in the film. As the film becomes thicker,
the FEM calculation is not predicted. This suggests that the substantial deviations from the assumption of a stagnant film
lubrication theory is acceptable up to a Cu of about 0.005. A occur. The strongest effect is found in upward flow.
possible explanation for the different behavior at higher Cu Based on the sequence in Figures 4e-4g, we note that there
is that the bubble becomes unsteady. appears to be a dependence of the film thickness on the Re
number as well. As Re increases the film becomes thicker,
Parameter study and the slip velocity between the two phases increases. As
The dependence of the film thickness, Sf, and the relative long as the flow is laminar, the effect is rather small and, due
velocity of the gas plug, v, on a number of parameters has to the problems of ensuring that this is a converged solution,
been studied. The focus has been on the Re, Fr, and Cu no attempt was made to quantify it.
groups since the aspect ratio, a,and the void fraction, E , are There are a few possible explanations for the discrepancy
found to be relatively unimportant in the region of interest. between the film thicknesses predicted by the simulation and
As long as the gas bubble is large enough to allow a distinct those measured experimentally. The correlation of Irandoust
film region to develop, any change in its volume will affect and Andersson (1989b) was based on measurements of light
only the bubble length, not the film thickness and the flow absorption. In order to minimize the distance the light has to
rate. travel through the liquid film, it must follow a line that passes
through the centerline of the capillary. If the focus is slightly

nn,
0.""

1 1 \ca= bl
1.4 0.04
I

-3 -2.0 -1.0 0 1.0 2.0 -2.0 -1.0 0 1.0 2.0

FI-' Fr.'
Figure 7. Wavelengths near tail.
(a) Predicted by Eqs. 20-22; (b) as case a , but ro is taken Figure 9. Influence of f r -' on (a) 6,; (b) V.
from the FEM simulation; (c) FEM simulation. Re = 200, (Y = 8 , c = 0.5.

AIChE Journal July 1996 Vol. 42, No. 7 1821

 observed experimentally by Irandoust and Andersson (1989b).
The influence of the Reynolds number was found to be small
in the laminar regime. The influence of gravitational forces
was also studied and was found to be significant for high Ca
and upflow. For Ca < 0.01 and Fr < 1, this effect is small.
The size of the recirculation cell within the liquid plug was
found to decrease rapidly with increasing film thickness.
Predictions of the undulations occurring in the liquid film,
predominantly near the tail end, were compared to an analy-
Figure 10. Buildup of surface tension gradients through sis using lubrication theory by Ratulowski and Chang (1989).
the adsorption of surfactants. Agreement was good for Ca < 0.005. while the FEM pre-
dicted shorter wavelengths at higher Ca. The behavior of the
FEM solution indicates that the oscillations become unstable
off, or the light beam passes through an area rather than a for higher Ca, suggesting that the lubrication theory no longer
point, the result will be an overprediction of the film thick- applies.
ness. The calculated thickness, on the other hand, represents
the result of a perfect reading, and is thus the smallest value
that would be observable for real flow. Acknowledgment
The influence of a variable surface tension can also explain The financial support for this work provided by the Swedish Na-
a similar discrepancy. In a system with very pure liquid, the tional Board for Industrial and Technical Development is gratefully
assumption of constant surface tension is likely to be correct; acknowledged.
however, such systems are not likely to be found outside the
laboratory. As a small portion of liquid, containing trace Notation
amounts of surface-active impurities, is brought to the inter- g =gravitational acceleration, m/s2
face, these impurities start to accumulate at the interface. As H =mean Gaussian curvature
the surface concentration increases, the interfacial tension is L, = period length (gas +liquid plug), m
reduced. Hence the surface tension changes with age. A vari- n =normal direction
p =pressure
ation of the surface tension leads to a tangential force at the Q =flow rate, mys
interface in the direction from low to high surface tension. r = R/D, radial coordinate
This effect explains the significantly higher rise velocity of ro =bubble radius
an air bubble in very pure water compared to water of ordi- s =dummy variable introduced in Eq 10
t =tangential direction
nary purity (Clift et al., 1978). The fresh liquid is exposed to ( U ) =total mean linear velocity, m/s
the gas at the top of the bubble, and as it moves along the u =velocity component
periphery surfactants accumulate, surface tension drops, and V=volume of the gas phase in the computational domain, m3
a force, opposing the motion, results. z =axial coordinate
In the case of a Taylor bubble, the situation is a bit more
complicated. The fresh liquid now reaches the interface at Greek letters
the stagnation ring, rather than at the top, and is there di-
vided into two streams (see Figure 10). Hence the surface pi=parameter defined by Eq. 21
y =surface tension, N/m;parameter defined by Eq. 21
tension will have its maximum value at the stagnation ring, 0 =angular direction
and a gradient will form on both sides. In the film this gradi- A = characteristic wavelength
ent will give rise to a tangential force that tends to drain n =anchor point of interface (Figure 1)
liquid from the film, thus making it thinner. In front of the p =density of liquid, kg/m3
u =stress
stagnation ring, the gradient gives rise to a tangential force =function describing bubble shape
acting in the opposite direction and thus increasing the film
thickness. The rate at which the surface tension drops from
the value of the fresh liquid/gas interface is a complicated Literature Cited
function of the bulk concentration and distribution of impuri- Bendiksen, K. H., “On the Motion of Long Bubbles in Vertical
ties, their effective diffusivity, the adsorption isotherm at the Tubes,” fnt. J. Multiphase Ffow, 11(6), 797 (1985).
interface, and the relationship between surface concentration Bretherton, F. P., “The Motion of Long Bubbles in Tubes,” J. Fluid
Mech., 10, 166 (1961).
and surface tension. Chen, J.-D., “Measuring the Film Thickness Surrounding a Bubble
inside a Capillary,” J. Colloid Znterf: Sci., 109(2),341 (1986).
Conclusions Clift, R., J. R. Grace, and M. E. Weber, Bubbles, Drops, and Parti-
cles, Academic Press, New York (1978).
It has been shown that a commercial software package can Collins, R., F. F. de Moraes, J. F. Davidson, and D. Harrison, “The
Motion of a Large Gas Bubble Rising through Liquid Flowing in a
be used to analyze the flow of a Taylor bubble in a capillary Tube,” J. Fluid Mech., 89(3), 497 (1978).
under a wide range of conditions. The approach taken here Cybulski, A., and J. A. Moulijn, “Monoliths in Heterogeneous Catal-
allows the simultaneous inclusion of viscous, gravitational, in- ysis,” Cat. Reu. Sci. Eng., 36(2), 179 (1994).
ertial, and capillary forces into the model. The bubble shape Edvinsson, R., R. Sandboge, S. Irandoust, and W. Spisak, “The Sta-
lactite Flow of a Gas-Viscous Liquid in Vertical Tubes,” Adu. Fi-
and the velocity field in the liquid phase are obtained from nite Elem. Anal. Fluid e n . , FED, 137, 107 (1992).
the simulations. The predicted thicknesses of the liquid film Fabre, J., and A. Lin6, “Modeling of Two-Phase Slug Flow,” Ann.
surrounding the gas bubble were slightly lower than those Rev. Fluid Mech., 24, 21 (1992).

1822 July 1996 Vol. 42, No. 7 AIChE Journal

Fairbrother, F., and A. E. Stubbs, “Studies in Electro-endosmosis. Ratulowski; J., and H.-C. Chang, “Marangoni Effects of Trace Impu-
Part VI: The ‘Bubble-tube’ Method of Measurement,” Chem. SOC., rities on the Motion of Long Gas Bubbles in Capillaries,” J. Fluid
1, 527 (1935). Mech., 210, 303 (1990).
FIDAP, Fluid Dynamics Analysis Package, Fluid Dynamics Interna- Schwartz, L. W., H. M. Princen, and A. D. Kiss, “On the Motion of
tional, Inc., Evanston, IL 60201 (1991). Bubbles in Capillary Tubes,” J. Fluid Mech., 172, 259 (1986).
Ginely, G. M., and C. J. Radke, “Influence of Soluble Surfactants on Shen, E. I., and K. S. Udell, “A Finite Element Study of Low
the Flow of Long Bubbles through a Cylindrical Capillary,” ACS Reynolds Number Two-Phase Flow in Cylindrical Tubes,” J. Appl.
Symp. Ser., 396, 480 (1989). Mech., 52, 253 (1985).
Goldsmith, H. L., and S. G. Mason, “The Flow of Suspensions Stone, H. A., and L. G. Leal, “The Effects of Surfactants on Drop
through Tubes. 11: Single Large Bubbles,” J. Colloid Sci., 18, 237 Deformation and Breakup,” J. Fluid Mech., 220, 161 (1990).
(1963). Taitel, Y., and A. E. Dukler, “A Model for Predicting Flow Regime
Irandoust, S., and B. Andersson, “Monolithic Catalysts for Nonauto- Transitions in Horizontal and Near Horizontal Gas-Liquid Flow,”
mobile Applications,” Cat. Rev. Sci. Eng., 30(3), 341 (1988). AZChE J., 22, 47 (1976).
Irandoust, S., and B. Andersson, “Simulation of Flow and Mass Taylor, G. I,, “Deposition of a Viscous Fluid on the Wall of a Tube,”
Transfer in Taylor Flow through a Capillary,” Computers Chem. J. Fluid Mech., 10, 161 (1961).
Eng., 13(4,5), 519 (1989a). Troniewski, L., and W. Spisak, “Flow Patterns in Two-Phase Down-
Irandoust, S., and B. Andersson, “Liquid Film in Taylor Flow through flow of Gas and Viscous Liquid,” Znt. J. Multiphase Flow, 13, 257
a Capillary,” 2nd. Eng. Chem. Rex, 28(11), 1684 (1989b). (1987).
Mao, Z.-S., and A. E. Dukler, “The Motion of Taylor Bubbles in Westborg, H., and 0. Hassager, “Creeping Motion of Long Bubbles
Vertical Tubes: I. A Numerical Simulation for the Shape and Rise and Drops in Capillary Tubes,” J. Colloid Znterf: Sci., 133(1), 135
Velocity of Taylor Bubbles in Stagnant and Flowing Liquid,” J. (1989).
Comput. Phys., 91, 132 (1990). Whalley, P. B., Boiling, Condensation and Gas-Liquid Flow, Claren-
Marchessault, R. N., and S. G. Mason, “Flow of Entrapped Bubbles don Press, Oxford (1987).
through a Capillary,” 2nd. Eng. Chem., 52(1), 79 (1960).
Martinez, M. J., and K S. Udell, “Axisymmetric Creeping Motion of
Drops Through Circular Tubes,” J. Fluid Mech., 210, 565 (1990).
Ratulowski, J., and H.-C. Chang, “Transport of Gas Bubbles in Cap-
illaries,’’ Phys. Fluids A l , 10, 1642 (1989). Manuscript received Apr. 19, 1995, and reuision receiued Oct. 26, 1995

AIChE Journal July 1996 Vol. 42, No. 7 1823