Chemical Engineering Science 60 (2005) 1117 1126

www.elsevier.com/locate/ces

CFD simulations of wall mass transfer for Taylor ow in circular capillaries

J.M. van Baten, R. Krishna
Van t Hoff Institute for Molecular Sciences, University of Amsterdam, Nieuwe Achtergracht 166, 1018 WV Amsterdam, The Netherlands

Abstract
Computational uid dynamics is used to investigate the mass transfer from the liquid phase to the channel wall for Taylor ow of
bubbles rising in circular capillaries. The separate inuences of the Taylor bubble rise velocity, unit cell length, gas holdup, and liquid
diffusivity on mass transfer were investigated for capillaries of 1.5, 2 and 3 mm diameter. A correlation is proposed for estimation of the
wall mass transfer coefcient and this correlation has been tested against published experimental data.
2004 Elsevier Ltd. All rights reserved.

Keywords: Taylor bubbles; Mass transfer; Capillary; CFD; Monolith reactors; Wall; Film

1. Introduction
transfer from Taylor bubbles to the surrounding liquid phase
and used extensive computational uid dynamics (CFD)
Multiphase monolith reactors offer many potential advan-
simulations in order to develop a model, based on the pene-
tages over trickle beds, slurry bubble columns and airlifts
tration theory, to allow the prediction of the volumetric mass
that include low-pressure drop, high mass transfer rates, and
transfer coefcient kL a from information on the various hy-
ease of scale up (Edvinsson and Cybulski, 1995; Kreutzer,
drodynamic and system parameters. The focus of the current
2003; Nijhuis et al., 2001; Stankiewicz, 2001). Monolith
work is on mass transfer from the liquid phase to the walls
reactors are being applied in laboratory studies and in
of the circular capillary channels of monoliths.
commercial practice for carrying out reactions such as
Fig. 1b shows the liquid phase velocity proles within a
hydrogenations. Provided the gas and liquid phases are uni-
liquid slug separating two Taylor bubbles (these proles
formly distributed over the various channels of the monolith,
were obtained from CFD simulations to be described later).
commercial reactors of large dimensions can, in principle,
The recirculation of the liquid phase causes a signicant
be scaled up from information on the hydrodynamics, mass
enhancement to radial mass transfer within the slug. In
transfer, and mixing within a single channel that has dimen-
Fig. 2a we illustrate this by replotting the experimental data
sions typically in the 13 mm range (Boger et al., 2003; Roy
of Bercic and Pintar (1997) in terms of the Sherwood num-
et al., 2004). Inside each capillary, we usually have Taylor
ber for wall mass transfer, Sh kw dc /D, as a function of the
ow of gas bubbles, as shown schematically in Fig. 1a. In the
Graetz number Gztube Ltube (1
G )D/dc2 Vb . Note that
development and design of monolith loop reactors for fast
we include the liquid holdup in the denition of Gztube ; the
reactions, there are two mass transfer processes to reckon
rationale will be clear in the discussions below. The experi-
with: (a) transfer from the Taylor gas bubbles to the sur-
mental data for Taylor ow mass transfer are about one order
rounding liquid phase, and (b) transfer from the liquid phase
of magnitude higher than the corresponding value for single-
to the walls of the monolith channels. In recent work (van
phase liquid ow (
G = 0) in circular tubes, for which Sh =
Baten and Krishna, 2004b) we had focussed on the mass 1/3
1.62Gztube , provided Gztube < 0.05 (Shah and London,
1978). The experimental data of Horvath et al. (1973) on wall
Corresponding author. Tel.: +31 20 525 7007; fax: +31 20 525 5604. Sherwood number also show a dependence on the ratio of
E-mail address: R.Krishna@uva.nl (R. Krishna). the slug length to capillary diameter, Lslug /dc (see Fig. 2b).

0009-2509/$ - see front matter 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2004.10.001
1118 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126

CFD, this CFD, 3 mm

(b) 10 3 work CFD, 2 mm
CFD, 1.5 mm
Bercic expt

Sherwood number, Sh
Taylor
z 10 2 flow
single
phase
flow

L tube
dc
CFD: dc = 1.5, 2, 3 mm
Bercic expt: dc = 2.4-3.2 mm
10 1
10 -6 10 -5 10 -4 10 -3 10 -2
L film
(a) Graetz number, Gz tube

CFD, this
103 work Horvarth, Lslug/dc= 1.6
Horvarth, Lslug/dc= 5
Profiles within Horvarth, Lslug/dc= 16
liquid slug
Sherwood number, Sh

LUC
db
Taylor
102
flow

film single
phase
flow

(a) (b)
Fig. 1. (a) Schematic of Taylor bubble rise in a capillary and (b)
liquid velocity vectors obtained from CFD simulations for dc =
3 mm, Vb = 0.45 m/s, LU C = 0.025 m, lm = 48
m,
G = 0.272,
Llm = 5.321 mm, D = 1 109 m2 /s.
Bercic and Pintar (1997) and Irandoust and Andersson
(1988) have both put forward empirical correlations for es-
timation of the wall mass transfer coefcient kw . Kreutzer
(2003) and Kreutzer et al. (2001) have used axi-symmetric
CFD simulations to determine the wall mass transfer coef-
cient, without explicitly accounting for the presence of the
Taylor bubble and the resulting ow of liquid in the lm sur-
rounding the bubble. As we shall demonstrate in this paper,
such estimations are rather optimistic for high gas holdups,
long Taylor bubbles and long contact with the liquid lm.
The major objective of the present work is to use CFD in
order to investigate wall mass transfer in the real situation
including the presence of the Taylor bubble. Furthermore,
we study the separate inuences on Sh of various hydrody-
namic and system parameters such as bubble rise velocity
CFD: dc = 1.5, 2, 3 mm
Horvarth expt: dc = 2.4 mm
101
10-6 10-5 10-4 10-3 10-2
(b) Graetz number, Gz tube
Fig. 2. Sherwood number for wall-mass transfer plotted against the Graetz
number dened as Gztube Ltube (1
G )D/dc2 Vb for single phase
ow and for Taylor ow in circular capillaries. (a) The symbols show
the experimental data of Bercic and Pintar (1997) obtained in capillaries
of diameters dc = 2.4, 2.5 and 3.2 mm. The 2.5 mm capillary had a
length Ltube = 0.35 m, the 3.2 mm capillary had lengths of 0.250.27 m.
(b) The symbols show the experimental data of Horvath et al. (1973)
obtained in a capillary of dc = 2.4 and three different slug lengths. Also
plotted in (a) and (b) are all the CFD simulations carried out in this
work for capillaries of diameters dc = 1.5, 2 and 3 mm for a variety of
hydrodynamic parameters as specied in Table 1.
Vb , unit cell length LU C , gas holdup
G , liquid phase dif-
fusivity D, thickness of liquid lm surrounding the Taylor
bubble lm , and capillary diameter dc . Such a study will
provide insights into the mass-transfer mechanisms and al-
low more reliable scale up of monolith reactors. The second
objective is to suggest a practical, engineering, correlation
for estimation of Sh.
 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126 1119

2. CFD model development

We consider an idealized geometry for the Taylor bub-

bles, consisting of two hemispherical caps and a cylindrical
body. The Taylor bubble is considered as a void, acting
as a free surface with the surrounding liquid phase; see the
computational domain shown in Fig. 3a. The inuence of
the assumption of hemispherical caps is of secondary impor-
tance to the mass transfer contributions from the wall. The
distribution of velocities in the system is more important,
and will not be inuenced signicantly by the precise shape
of the interface, since the interface does not exert drag on
the liquid. The volume-averaged mass and momentum con-
servation equations in the Eulerian framework are given by

QuL = 0, (1)

*uL

L + Q L uL uL L (uL + (uL )T )
*t Fig. 4. (a) Schematic of coarse grid and (b) schematic of ne grid. The
= p + L g, (2) ne grid details are specied in Table 1.

where L , uL and L represent, respectively, the macro-

scopic density, velocity and viscosity of the liquid phase, p
is the pressure and g is the gravitational acceleration. Lam-
inar ow conditions are assumed to prevail.
A commercial CFD package CFX, version 4.4, of AN-
SYS Inc., Canonsburg, USA was used to solve the equa-
tions of continuity and momentum. This package is a nite
Symmetry
volume solver, using body-tted grids. The grids are non-
staggered and all variables are evaluated at the cell centres.
An improved version of the RhieChow algorithm (Rhie
and Chow, 1983) is used to calculate the velocity at the cell
faces. The pressurevelocity coupling is obtained using the
SIMPLEC algorithm (van Doormal and Raithby, 1984). For
the convective terms in Eqs. (1) and (2), the SUPERBEE-
MUSCL differencing scheme was used (higher order upwind

scheme with ux delimiters). A fully implicit backward dif-

axis

ferencing scheme was used for the time integration.

The boundary condition was periodic in the vertical direc-
tion (utop = ubottom , ptop = pbottom ). The hydrostatic pressure
Periodic drop of gL is added as a source term to the momentum bal-
Boundary: Top
ance Eq. (2) to correct for the periodic boundary conditions.
Simulations were performed in a reference frame in which
Wall slug

the bubble is stationary and the system moves up with the

bubble rise velocity Vb . At the outer wall, the boundary
condition was set to uz = Vwall = Vb , ur = 0, where r and
L UC z are the radial and axial coordinates. The bubble surface is
specied as free-slip: du /dn=0 with u being the velocity
su rfac e
Bu bble

Wall film

component in the direction of the bubble surface, and n is

the normal direction to the bubble surface. At the axis of
symmetry, we have: duz /dr = 0.
The simulations were carried out using axi-symmetric 2D
Wall slug

grids using cylindrical coordinates. The calculation of the

wall

wall mass-transfer coefcient using CFD simulations was

carried out in three consecutive campaigns.
Periodic In the rst campaign, the mass and momentum transfer
Boundary, Bottom equations are solved for a coarse grid (typical coarse grid
(a) (b) (c) shown in Fig. 4a) to obtain the steady-state hydrodynamics.
Using this velocity eld solution as initial guess, the hydro-
Fig. 3. (a) Schematic overview of the computational domain of unit cell
with periodic boundary conditions at inlet and outlet, (b) highlights the dynamic steady state is re-calculated in a second campaign
wallslug contribution to mass transfer and (c) highlights the walllm using a ne grid that has much smaller cells near the mass-
contribution to mass transfer. transfer surfaces, i.e., the wall (see Fig. 4b). The steady-state
1120 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126

Table 1
Details of variations of parameters and grid cells used in the simulations

dc (mm) LU C (mm) Vb (m/s) Llm (mm) lm (
m)
G () Total number of grid cells Smallest cell size (
m)

3 40 0.15, 0.3, 0.55 5.32 48 0.17 85,848 1.00

3 50 0.45 5.32 48 0.136 109,848 1.00
3 35 0.45 5.32 48 0.194 73,848 1.00
3 15, 20, 25, 30 0.45 5.32 48 0.2270.453 61,84825,848 1.00
3 40 0.45 6.60, 10.9, 15.1, 19.4 48 0.200.5 83,35258,776 1.00
3 100 0.05, 0.1, 0.2, 0.3 51.4 48 0.5 141,336 1.00
2 40 0.15, 0.3, 0.45 5.97 32 0.170 66,620 0.60
2 15, 25, 40 0.45 5.97 32 0.1700.453 66,62021,620 0.60
2 100 0.05, 0.1, 0.2, 0.3, 0.4 52.1 32 0.5 115,612 0.60
1.5 40 0.15, 0.3, 0.45 6.29 24 0.170 90,312 0.40
1.5 15, 25, 40 0.45 6.29 24 0.170.453 90,31228,812 0.40
1.5 100 0.05, 0.1, 0.2, 0.3, 0.4 52.4 24 0.5 149,400 0.40

solutions were typically obtained within 30,00040,000
iterations and it was veried that convergence was indeed
obtained.
From the converged hydrodynamics, the average liquid
velocity is calculated from a summation over the cells at the
top boundary of the computational domain:
 lm between the bubble surface and the capillary wall.
top voli ui
uL,domain =  . (3)
top voli
Here, voli is the volume of cell i and ui is the vertical velocity
in cell i. The summation in Eq. (3) is over the total number
of cells at the top boundary. The supercial liquid velocity
is now calculated from
UL = (uL,domain Vwall )(1
G ),
where the gas holdup
G is the bubble volume divided by
the unit cell volume. Note that uL,domain is non-zero in our
simulations,
bubble volume

G = . (5)
LU C dc2 /4
The supercial gas velocity is calculated from
UG =
G Vb .
The volumetric ow of liquid through the thin lm sur-
rounding the bubble, Qlm , is calculated from the following
relation derived by Thulasidas et al. (1995) using a mass
balance over a cross section through the bubble and a cross
section through the liquid slug:
Ab Vb = Ac ULS + Qlm ,
where ULS is the velocity of the liquid slug that is determined
from
ULS = uL,domain Vwall .
Rearranging Eq. (7) we get
 2
Qlm = (d Vb dc2 ULS ). (9)
4 b
The lm ow Qlm calculated from Eq. (9) agrees with the
value obtained by integration of the velocity prole in the
The converged hydrodynamics obtained were used in
a subsequent third campaign for wall mass transfer using
the ne grid, carried out in a transient manner, for which
300010,000 time steps of 0.01 s were used. To correctly
capture the steep concentration gradients near the wall,
we used a cell size smaller than 1
m close the wall, with
an exponential increase away from the wall, as shown
in Fig. 4. The grid details are specied in Table 1. The
(4) neness of the grid was chosen such that the calculated
mass transfer coefcient was unaffected by further grid
renement, as discussed in our earlier work (van Baten
and Krishna, 2004b). The tracer concentration throughout
the system was initially set to zero. At t = 0, the tracer
concentration at the wall was set to unity (a.u.) to deter-
mine the mass transfer from the solid phase to the liq-
uid phase. The following equation is solved for the mass
tracer:
*
( CL ) + (L uL CL DL CL ) = 0. (10)
(6) *t L
Here, CL is the concentration of mass-tracer in the liquid
(a.u.) and D is the diffusion coefcient of mass tracer in
the liquid. At the top and bottom, the periodic boundary
conditions were used: CL,top = CL,bottom . Though the peri-
odic boundary condition for the tracer is not strictly true, it
must be borne in mind that during one simulation cycle
(7) the amount of mass transferred is negligibly small. Essen-
tially we are assuming a well-mixed liquid phase during one
simulation cycle. For each run, the liquid circulates from
top to bottom about 1000 times. We have estimated that
the error in assuming CL,top = CL,bottom is at most 0.03%.
(8) Zero tracer ux was allowed through the bubble surface:
 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126 1121

1.0 The wall mass transfer coefcient kw , averaged over the

period t = 0 to t is then calculated from
Liquid concentration, CL,system / [a.u.]

0.8
1 (CL,w CL,system,t=0 )
dc = 1.5 mm; kw = ln
aw t (CL,w CL,system )
film = 24 m;
0.6 1 (CL,w )
= 1x10-9 m2 s-1; = ln
LUC = 0.04 m; aw t (CL,w CL,system )
Vb = 0.45 m/s; 1 1
0.4 Lfilm = 6.29 mm; = ln , (12)
G = 0.17;
aw t (1 CL,system )

0.2 bulk liquid where we introduce the initial condition that the concentra-
wall tion in the system is 0, and that the concentration at the wall
is maintained at CL,w = 1 during the entire mass-transfer
0.0
process. In Eq. (12) aw represents the specic interfacial
area for wall mass transfer:
(a) time / [s]
aw = 4/dc . (13)
10 -3
The Sherwood number can now be calculated from:
wall mass transfer coefficient, kw / [m/s]

dc = 1.5 mm;
10 3
film = 24 m;
kw dc
Sh = .
-9 2 -1
= 1x10 m s ; (14)
Wall Sherwood number, Sh

LUC = 0.04 m; D
Vb = 0.45 m/s;
10 -4 Lfilm = 6.29 mm; Both the time-averaged kw and Sh values decrease with in-
G = 0.17;
creasing contact time, as demonstrated by a typical example
10 2 in Fig. 5b.
Since we use periodic boundary conditions, the total
length of the capillary tube, Ltube , that is traversed in the
kw
simulation time, t, is given by
10 -5 Sh

Ltube
10 1 t= . (15)
0.01 0.1 1 10 100 Vb
(b) time / [s]
The wall mass transfer process consists of two separate
Fig. 5. (a) Concentration versus time prole and (b) time-averaged wall contributions: (1) wallslug contribution of the regions in
mass transfer coefcient and Sherwood numbers as a function of time. The contact with the liquid slug (see Fig. 3b), and (2) walllm
hydrodynamic parameters are dc =1.5 mm, Vb =0.45 m/s, LU C =0.04 m, contribution of the region in contact with the liquid lm sur-
lm = 24
m,
G = 0.17, Llm = 6.29 mm. Animations showing the rounding the bubble (see Fig. 3c). By additionally monitor-
dynamics of mass transfer can be viewed on our website (van Baten and
ing the mass transfer uxes of the wallslug and walllm
Krishna, 2004a).
regions at each instant of time, we can also separately
determine the Sherwood numbers for these two regions,
Shwall.slug , Shwall.lm , respectively, after averaging from
dCL /dn = 0. Symmetry conditions apply to the center axis: t = 0 to t.
dCL /dr = 0. At the outer wall, the concentration is specied A total of 46 simulations with varying parameter values
as CL,w = 1. dc , Vb , LU C ,
G , lm , and D were carried out on Linux PCs
The total concentration of tracer in the system at each with a single AMD XP processor. The details of parameter
time step is determined from values used in these simulations are listed in Table 1. The
 range of parameter values was chosen to correspond with
domain voli CL,i those obtained in experimental studies (Vandu et al., 2005).
CL,system =  , (11)
domain voli Each hydrodynamic simulation on the coarse grid was solved
in a matter of minutes, each hydrodynamic simulation on the
where the summation is carried out over all the vol- ne grid took several hours. Each dynamic mass tracer run
ume elements in the computational domain. A typical also took several hours. Animations showing the dynamics
concentrationtime prole for a 1.5 mm diameter capillary of wall mass transfer can be viewed on our website (van
is shown in Fig. 5a. Baten and Krishna, 2004a).
1122 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126

3. CFD simulation results with a maximum velocity at the centre of the channel that is
twice the value of the liquid slug velocity, ULS . The liquid
The ShGztube relationship for all these 46 simulations velocity proles within the lm are shown in Fig. 6b. The
are shown by the continuous solid (dc = 3 mm), dasheddot velocity of the liquid at the surface of the lm is one and a
(dc = 2 mm), and dashed (dc = 1.5 mm) lines in Figs. 2a half times the average liquid velocity, in keeping with the
and b. The CFD simulations are in reasonable agreement classical solution for the velocity elds in falling lms (see
with the experimental data of Bercic and Pintar (1997) and Sherwood et al., 1975, p. 78). In Fig. 6c, we see that the
Horvath et al. (1973). We now proceed to examine in some magnitude of the velocity of the liquid lm at the surface
detail the precise nature of the ShGztube dependence with next to the Taylor bubble, Vlm , increases linearly with Vb .
the aim of developing a usable correlation of the CFD data. Since the bubble rises upwards counter-current to the liq-
Consider rst the hydrodynamics in the 2 mm capillary uid lm the contact time, tlm , between the bubble and the
with a unit cell length LU C = 0.1 m, with varying values liquid lm for a lm of length Llm can be calculated using
of the Taylor bubble rise velocity, Vb . The liquid phase ve- tlm = Llm /(Vb + Vlm ) where we note that Vlm has a
locity proles at the top of the computational domain (liq- negative value, as shown in Fig. 6c. The value of (UG + UL )
uid outlet) are shown in Fig. 6a, in the reference velocity is slightly smaller than the value of the bubble rise velocity,
frame with a stationary wall (these values are obtained by Vb and this is due to the backow of liquid through the lm
adding the bubble rise velocity to the values obtained from as described by Eq. (7) (see Fig. 6d). With increasing Vb
the simulations. The velocity proles are parabolic in shape the ow of liquid through the lm surrounding the bubble,

dc = 2 mm; film = 32 m;
Axial liquid velocity distribution, (uL(r)-Vb )/ [m/s]

0.8 -9 2 -1 0.0000
Vb=0.40 m/s = 1x10 m s ; LUC = 0.1 m; Vb=0.05 m/s
Lfilm = 5.207 mm; G = 0.5;
-0.0005 0.10
liquid velocity in film / [m/s]

0.6 0.15
0.30 -0.0010
0.20

0.4 -0.0015
dc = 2 mm;
0.20 0.30
film = 32 m;
0.15 -0.0020 = 1x10-9 m2 s-1;
0.2 0.40 LUC = 0.1 m;
0.10 Lfilm = 5.207 mm;
-0.0025
0.05 G = 0.5;
0.0 -0.0030
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) dimensionless radius, 2 r /dc (b) dimensionless distance along liquid film, r / film

0.0000 0.6
Liiquid velocity at surface of film, Vfilm / [m/s]

dc = 3 mm
dc = 3 mm
-0.0005 0.5 dc = 2 mm
dc = 2 mm
dc = 1.5 mm
dc = 1.5 mm
(UG+UL) / [m s ]

-0.0010 0.4
-1

-0.0015 0.3

-0.0020 0.2
G = 0.50;
-0.0025 LUC = 0.1 m; 0.1
Lslug = 4.6 mm;
-0.0030 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.6
-1
(c) Taylor bubble rise velocity, Vb / [m/s] (d) Taylor bubble rise velocity, V b / [m s ]

Fig. 6. Radial distribution of liquid velocity (a) at the outlet of the computational domain cell and (b) within the lm surrounding the bubble. In these
simulation campaigns dc = 2 mm, LU C =0.1 m, lm = 32
m,
G = 0.5, Llm = 5.207 mm, and D = 1 109 m2 /s. The velocity proles in (a) and (b)
are in the reference frame with respect to a stationary wall. (c) Dependence of surface velocity of the liquid lm on the bubble rise velocity Vb for different
channel diameters, dc = 1.5, 2 and 3 mm. (d) Dependence of (UG + UL ) on the bubble rise velocity Vb for channel diameters dc = 1.5, 2 and 3 mm.
 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126 1123

103 103 Vb = 0.05 m/s

Vb = 0.45 m/s Vb = 0.10 m/s
Vb = 0.30 m/s Vb = 0.15 m/s
Vb = 0.55 m/s Vb = 0.20 m/s
Sherwood number, Sh

Sherwood number, Sh
Vb = 0.30 m/s

102 dc = 3 mm; 102 dc = 3 mm;

film = 48 m; film = 48 m;
-9
= 1x10 m s ; 2 -1
= 1x10-9 m2 s-1;
LUC = 0.04 m; LUC = 0.10 m;
Lfilm = 5.32 mm; Lfilm = 5.14 mm;
G = 0.17; G = 0.5;
1 1
10 10
10-6 10-5 10-4 10-3 10-2 10-6 10-5 10-4 10-3 10-2

(a) Graetz number, Gztube (b) Graetz number, Gztube

103 G = 0.17 103

G = 0.2
= 1x10-9 m2 s-1
G = 0.3
= 5x10-9 m2 s-1
Sherwood number, Sh

Sherwood number, Sh
G = 0.4
G = 0.5
dc = 3 mm;
102 102
dc = 3 mm; film = 48 m;
film = 48 m; LUC = 0.04 m;
= 1x10-9 m2 s-1; Vb = 0.45 m/s;
Vb = 0.45 m/s; Lfilm = 5.32 mm;
LUC = 0.04 m;
G = 0.17;

101 101
10-6 10-5 10-4 10-3 10-2 10-6 10-5 10-4 10-3 10-2

(c) Graetz number, Gztube (d) Graetz number, Gztube

Fig. 7. The wall Sherwood number plotted against Gztube for 3 mm capillary. (a) Campaign for LU C =0.04 m, D = 1 109 m2 /s,
G = 0.17 with
varying Taylor bubble rise velocities, Vb . (b) Campaign for LU C =0.1 m, D = 1 109 m2 /s,
G = 0.5 with varying Taylor bubble rise velocities, Vb .
(c) Campaign for LU C = 0.04 m, Vb =0.45 m/s, D = 1 109 m2 /s,
G = 0.17 with varying gas holdup values
G . (d) Campaign for LU C =0.04 m,
Vb = 0.45 m/s,
G = 0.17 with varying liquid phase diffusivity values D.

Qlm also increases, thus increasing backow. This explains
the increasing deviation between (UG + UL ) and Vb with
increasing Vb .
We now consider the inuence of various parameters on
wall Sh, expressed as a function of Gztube Ltube (1

G )D/dc2 Vb t (1
G )D/dc2 . Typical simulation results for
the 3 mm capillary are shown in Fig. 7 for a variety of condi-
tions. Consider rst a campaign in which the unit cell length,
gas holdup, and liquid diffusivity are both held constant at
LU C = 0.04 m,
G = 0.17 and D = 1 109 m2 /s, respec-
tively. Varying the Taylor bubble rise velocity Vb in the range
0.30.55 m/s, yields result that are practically indistinguish-
able from one another (see Fig. 7a). This would lead us to
conclude that the ShGztube relation is not dependent on Vb .
This independence on Vb is further conrmed for simula-
tions in which we maintain LU C = 0.1 m,
G = 0.5 and D =
1 109 m2 /s and vary Vb in the range 0.050.30 m/s (see
Fig. 7b). In the campaign shown in Fig. 7c the Taylor bubble
velocity, unit cell length and liquid diffusivity are held con-
stant at Vb =0.45 m/s, LU C =0.04 m and D = 1109 m2 /s,
respectively, and the gas holdup
G is allowed to vary in
the range 0.170.5. We note that the ShGztube relationship
is only marginally inuenced by the gas holdup
G despite
the large variation in the gas holdup. This is because the
denition of Gztube includes the liquid holdup. There is a
modest, residual, inuence of gas holdup on the ShGztube
relationship because the liquid slug length varies when we
vary the gas holdup. A shorter liquid slug length yields a
higher value of Sh; this result is in agreement with the Hor-
vath data presented in Fig. 2b. In the campaign shown in
Fig. 7d the liquid diffusivity D is varied by a factor 5
from 1 109 to 5 109 m2 /s, keeping Vb = 0.45 m/s,
LU C =0.04 m and
G =0.17. We note that the ShGztube re-
lationship is dependent on the liquid-phase diffusivity D. A
higher liquid diffusivity leads to a lower Sh. The reason for
this observation is not intuitively obvious and an explanation
is given below. Exactly analogous results as shown in Fig. 7
were obtained with the 1.5 and 2.5 mm diameter capillaries.
1124 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126

To gain further insights into the reasons why ShGztube 103

relationship is inuenced by the gas holdup, slug length
and liquid diffusivity, we examine the contributions of the
wallslug and walllm regions separately. In Fig. 8a we

Sherwood number, Sh
wall, total
show the simulation results, obtained in the 1.5 mm diameter wall-film
capillary, for Sherwood numbers for the total wall region wall-slug
(as presented in the foregoing discussions), along with the 102
corresponding values for the wallslug and walllm regions
for a unit cell length LU C = 0.1 m, with
G = 0.5 and Vb =
-9 2 -1
0.05 m/s. The walllm mass transfer is signicantly poorer = 1x10 m s ;
LUC = 0.1 m; Vb = 0.05 m/s;
than that of the wallslug region. Since the walllm area
G = 0.5; dc = 1.5 mm;
is 50% of the total wall area, the poorer lm mass transfer
contribution causes the Sh for total wall to be signicantly 101
10-6 10-5 10-4 10-3 10-2
lower than that for the wallslug region alone. A further (a) Graetz number, Gztube
reason for poorer Sh for walllm is that for long lms, i.e.,
long contact time between bubble and lm (i.e., low Vb ), the
10 3
lm is most likely nearing saturation conditions. This point
has also been discussed by Kreutzer (2003) in some detail.
For shorter lms and shorter contact times between lm and

Sherwood number, Sh
bubble all three Sh numbers are close to one another, as
is evidenced by the simulation results shown in Fig. 8b for wall, total
wall-film
LU C =0.04 m, Vb =0.05 m/s and
G =0.17. The detrimental 10 2 wall-slug
effect of saturation in the liquid lm surrounding the Taylor
bubble is even more pronounced for long lms and high
liquid diffusivities. This is illustrated in Fig. 8c for LU C = -9
= 1x10 m s ;
2 -1

0.1 m,
G = 0.5, Vb = 0.10 m/s and D = 5 109 m2 /s. LUC = 0.04 m; Vb = 0.45 m/s;
The lm Sh is about one order of magnitude lower than the G = 0.17; dc = 1.5 mm;
wallslug value because the saturation of the lm is much 10 1
easier for higher diffusivity values. 10-6 10-5 10-4 10-3 10-2
For a given set of hydrodynamic parameters, the (b) Graetz number, Gztube
ShGztube relationship is additionally dependent on the
capillary diameter dc as illustrated in the simulation re- 10 3
sults shown in Figs. 9a and b for two sets of conditions as
specied in the legend. The reason for the dependence on
the capillary diameter is because the lm thickness lm is
Sherwood number, Sh

wall, total
smaller for the smaller diameter capillary. This smaller lm wall-film
thickness causes the lm contribution to be poorer because 10 2 wall-slug
of the approach to saturation within the lm.
On the basis of a systematic analysis of all the 46 simula-
tions, we derived the following correlation for the Sherwood
number, for the total wall region: -9
= 5x10 m s ;
2 -1

10 1 LUC = 0.1 m; Vb = 0.1 m/s;

 0.5 G = 0.5; dc = 1.5 mm
Sh = ,  = 0.61Gz0.025
slug , = ,
Gztube (Gzslug /
G )0.15
10-5 10-4 10-3 10-2
Lslug D
Gzslug = 2 . (16) (c) Graetz number, Gztube
dc Vb
Fig. 8. Comparison of the wall (total), walllm and wallslug Sherwood
This correlation reects that fact the  and  in the Sh = numbers as a function of Gztube for 1.5 mm capillary. (a) Campaign for
/Gztube function is dependent on
G , Lslug and D as em- LU C =0.1 m, D = 1109 m2 /s,
G = 0.5, Vb = 0.05 m/s. (b) Campaign
phasized in the discussions surrounding Fig. 7. Eq. (16) is for LU C =0.04 m, D = 1 109 m2 /s,
G = 0.17, Vb = 0.45 m/s. (c)
valid for Gztube values lower than 0.01. For values of Gztube Campaign for LU C =0.1 m, D = 5 109 m2 /s,
G = 0.5, Vb = 0.1 m/s.
higher than 1 the Sh number tends to reach asymptotic val-
ues. However, values Gztube > 0.01 are not often realized of the slug length. For a wide range of parameter values,
in practice because this would inordinately long tubes. A the variation in the value of  is in the range of 0.450.52
further point to note is that the dependence of  on Gzslug and so a constant value of 0.48 may also be used for rough
is only very weak and its calculation requires knowledge estimations.
 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126 1125

10 3 Parity

Wall Sherwood number calculated from Eq. (16)

= 1x10-9 m2 s-1;
LUC = 0.025 m;
Vb = 0.45 m/s;
10 2
G = 0.272;
Sherwood number, Sh

10 2
Bercic expt data
Horvath, Lslug/dc = 1.6
Horvath, Lslug/dc = 5
Horvath, Lslug/dc = 16
dc = 3 mm
dc = 2 mm
10 1
dc = 1.5 mm 10 1 10 2
Experimental value of wall Sherwood number
10 1
Fig. 10. Parity plot of experimentally determined values of Sherwood
10-6 10-5 10-4 10-3 10-2
number by Bercic and Pintar (1997) and Horvath et al. (1973) against
(a) Graetz number, Gztube the predictions using Eq. (16). In addition to the parameters specied in
the paper by Bercic and Pintar (1997), it should be noted that the 2.5 mm
capillary had a length Ltube = 0.35 m, the 3.2 mm capillary had lengths
10 3 of 0.250.27 m.

= 1x10-9 m2 s-1;
LUC = 0.04 m; 4. Conclusions
Vb = 0.45 m/s;
G = 0.17; We have used CFD simulations to investigate mass trans-
Sherwood number, Sh

fer from the liquid phase to the wall during rise of Taylor
bubbles in circular capillaries. The Sherwood number for
wall mass transfer decreases signicantly with increasing
10 2
values of Gztube . The precise dependence of Sh on Gztube
depends on a variety of hydrodynamic parameters such as
slug length, gas holdup, liquid diffusivity. By tting the CFD
dc = 3 mm simulations we have derived the empirical correlation (16)
dc = 2 mm for practical estimation of wall mass transfer from informa-
dc = 1.5 mm tion on Taylor ow hydrodynamics.

10 1
10-6 10-5 10-4 10-3 10-2 Notation
(b) Graetz number, Gz tube a gasliquid interfacial area per unit cell
Fig. 9. Comparison of Sherwood number versus Gztube relationships volume, m2 /m3
for 1.5, 2 and 3 mm capillaries. (a) Campaign for LU C =0.025 m, A cross sectional area, m2
9 2
D = 1 10 m /s,
G = 0.272, Vb =0.45 m/s. (b) Campaign for C concentration of tracer in liquid phase, a.u.
LU C =0.04 m, D = 1 109 m2 /s,
G = 0.17, Vb = 0.45 m/s. dc capillary diameter, m
db bubble diameter; see Fig. 1(a), m
D liquid phase diffusivity, m2 /s
Fig. 10 compares the experimentally determined values g gravitational vector, m/s2
of Sh by Bercic and Pintar (1997) and Horvath et al. (1973) Gztube Graetz number based on tube length
against the predictions using Eq. (16). We note that the agree- ( Ltube (1
G )D/dc2 Vb ), dimensionless
ment is reasonably good, when we consider the scatter in Gzslug Graetz number based on slug length
the experimental data. ( Lslug D/dc2 Vb ), dimensionless
The corresponding correlation for the wallslug region kw wall-mass transfer coefcient, m/s
alone is somewhat simpler: Llm length of liquid lm (see Fig. 1a), m
Lslug length of liquid slug, Lslug = LU C (1
G ), m
2.4 + 1.5/(Lslug /dc ) Ltube total length of capillary tube (see Fig. 1a), m
Shwall.slug = . (17)
Gz0.45 LU C unit cell length, m
tube
n vector normal to the surface, dimensionless
1126 J.M. van Baten, R. Krishna / Chemical Engineering Science 60 (2005) 1117 1126

p system pressure, Pa format for reanalysis and replotting purposes. The criti-
Qlm volumetric liquid lm ow; m3 /s cal comments of Dr. Bercic on the draft manuscript are
r radial coordinate, m most appreciated. The Netherlands Organisation for Scien-
Sh Sherwood number for wall mass transfer tic Research (NWO) is gratefully acknowledged for pro-
( kw dc /D),dimensionless viding nancial assistance in the form of two grants: Pro-
t time, s grammasubsidie and TOP subsidie for development of novel
tlm contact time of liquid lm with Taylor gas bubble, concepts in reactive separations technology and for reactor
m/s intensication.
u liquid velocity vector, m/s
u velocity in z-direction, m/s
UG supercial gas velocity, m/s References
UL supercial liquid velocity, m/s
Bercic, G., Pintar, A., 1997. The role of gas bubbles and liquid slug lengths
ULS mean velocity of liquid slug, m/s
on mass transport in the Taylor ow through capillaries. Chemical
voli volume of cell i, m3 Engineering Science 52, 37093719.
Vb Taylor bubble rise velocity, m/s Boger, T., Roy, S., Heibel, A.K., Borchers, O., 2003. A monolith loop
Vlm velocity at surface of liquid lm, next to reactor as an attractive alternative to slurry reactors. Catalysis Today
bubble, m/s 79, 441451.
Edvinsson, R.K., Cybulski, A., 1995. A comparison between the
Vwall velocity of wall boundary condition in
monolithic reactor and the trickle-bed reactor for liquid-phase
simulations, equal to Vb , m/s hydrogenations. Catalysis Today 24, 173179.
z axial coordinate, m Horvath, C., Solomon, B.A., Engasser, J.M., 1973. Measurement of slug
ow using radial transport in enzyme tubes. Industrial and Engineering
Chemistry Fundamentals 12, 431439.
Greek letters Irandoust, S., Andersson, B., 1988. Mass-transfer and liquid-phase
reactions in a segmented two-phase ow monolithic catalyst reactor.
Chemical Engineering Science 43, 19831988.
lm thickness of lm surrounding bubble, m
Kreutzer, M.T., 2003. Hydrodynamics of Taylor ow in capillaries and

G gas holdup, dimensionless monolith reactors. Ph.D. Thesis, Delft University of Technology, Delft,
L liquid viscosity, Pa s The Netherlands.
L density of liquid phase, kg/m3 Kreutzer, M.T., Du, P., Heiszwolf, J.J., Kapteijn, F., Moulijn, J.A.,
2001. Mass transfer characteristics of three-phase monolith reactors.
Chemical Engineering Science 56, 60156023.
Subscripts Nijhuis, T.A., Kreutzer, M.T., Romijn, A.C.J., Kapteijn, F., Moulijn,
J.A., 2001. Monolithic catalysts as more efcient three-phase reactors.
Catalysis Today 66, 157165.
b refers to Taylor bubble Rhie, C.M., Chow, W.L., 1983. Numerical study of the turbulent ow
bottom bottom boundary of computational domain past an airfoil with trailing edge separation. AIAA Journal 21,
c refers to capillary 15251532.
domain in computational space Roy, S., Heibel, A.K., Liu, W., Boger, T., 2004. Design of monolithic
catalysts for multiphase reactions. Chemical Engineering Science 59,
lm refers to liquid lm
957966.
G refers to gas phase Shah, R.K., London, A.L., 1978. Laminar ow forced convection in ducts:
L refers to liquid phase a source book for compact heat exchanger analytical data. In: Shah,
i referring to cell number i R.K., London, A.L. (Eds.), Advances in Heat transfer: Supplement 1.
r in radial direction Academic Press, New York.
Sherwood, T.K., Pigford, R.L., Wilke, C.R., 1975. Mass Transfer,
s at the bubble surface
McGraw-Hill, New York, USA.
slug referring to liquid slug Stankiewicz, A., 2001. Process intensication in in-line monolithic reactor.
system referring to the computational system Chemical Engineering Science 56, 359364.
top top boundary of computational domain Thulasidas, T.C., Abraham, M.A., Cerro, R.L., 1995. Bubble-train ow in
tube referring to capillary tube capillaries of circular and square cross-section. Chemical Engineering
Science 50, 183199.
UC refers to unit cell
van Baten, J.M., Krishna, R., 2004a. Mass Transfer for Taylor ow in
wall refers to wall Capillaries. University of Amsterdam, Amsterdam, The Netherlands,
z in axial direction http://ct-cr4.chem.uva.nl/taylor_cfd, 12 March.
perpendicular van Baten, J.M., Krishna, R., 2004b. CFD simulations of mass transfer
from Taylor bubbles rising in circular capillaries. Chemical Engineering
Science 59, 25352545.
van Doormal, J., Raithby, G.D., 1984. Enhancement of the SIMPLE
method for predicting incompressible ows. Numerical Heat Transfer
Acknowledgements 7, 147163.
Vandu, C.O., Ellenberger, J., Krishna, R., 2005. Hydrodynamics and mass
We are grateful to Dr. G. Bercic for providing us the transfer in an upow monolith loop reactor. Chemical Engineering and
experimental data reported in his 1997 paper in electronic Processing 44, 363374.