Chemical Engineering Science 61 (2006) 665 – 675
Taha Taha, Z.F. Cui ^{∗}
Department of Engineering Science, Oxford University, Parks Road, Oxford OX1 3PJ, UK
Received15 June 2004; receivedin revisedform 5 July 2005; accepted6 July 2005 Available online 12 September 2005
Abstract
An attempt is made to model slug ﬂow inside square capillaries containing Newtonian liquids. In square micro-channels, slug ﬂows provedto promote high rates of heat andmass transfer between the solidboundaries andthe bulk ﬂow. This is due to the combination of bubble inducedsecondary ﬂows andthe creation of very thin liquidﬁlms. There are many industrial applications for which it is necessary to predict this rate of transfer. The volume of ﬂuid(VOF) methodimplementedin the commercial CFD package, FLUENT is used for this numerical study. A comprehensive description of such ﬂow is obtained and a comparative study is conducted on the hydrodynamics of slug ﬂow inside circular capillaries and their square counterparts. Computed values of the bubble velocity and diameter were in excellent agreement with publishedexperimental measurements. The detailedvelocity ﬁeldaroundthe bubble was also computed andcomparedfavourably with those experimental results reportedin literature. Basedon the hydrodynamic study explanation of mass and heat augmentation due to slug ﬂow is drawn. 2005 Elsevier Ltd. All rights reserved.
Keywords: Multiphase ﬂow; Numerical simulation; VOF; Square capillaries
1. Introduction
Slug ﬂow is the most important of the two-phase ﬂow patterns observedinside capillaries which attractedconsid- erable research interest because of numerous industrial and practical applications. These include enhanced oil recov- ery (Gauglitz et al., 1987; Hirasaki, 1989; Holm andGar- rison, 1988; Ransohoff et al., 1987; Ransohoff andRadke, 1987; Ransohoff andRadke, 1988; Ratulowski andChang, 1989), trickle- bedreactors (Charpentie andFavier, 1975), coating technology andpolymer processing (Huzyak and Koelling, 1997; Poslinski et al., 1995), in various micro- electromechanical systems (Ghiaasiaan andAbdel-Khalik, 2001) andmonolith froth reactors (Cybulski andMoulijn, 1994; Irandoust and Anderson, 1988). Slug ﬂow inside circular capillaries is characterised by the motion of a train of long capsule-shapedbubbles, frequently referredto as Taylor bubbles or gas slugs. These almost ﬁll the cross-sectional area of the capillary andare separated
^{∗} Corresponding author. Tel.: +44 1865 273118; fax: +44 1865 283273. E-mail address: zhanfeng.cui@eng.ox.ac.uk (Z.F. Cui).
0009-2509/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2005.07.023
from the wall by a liquidﬁlm andfrom each other by liquid plugs (Fig. 1). In a frame of reference attachedto the bubble andat low capillary numbers (Ca = U _{T} _{B} / ) strong circu- lation of the toroidal vortex inside the liquid slug ahead of the bubble exists. The bubble-ends are spheroidal and the shape of streamlines at both ends of the bubble are nearly identical. Close to the bubble-ends streamlines bow sharply to complete the vortex path (Fig. 1a). As capillary number is increased(Fig. 1b), the vortex in the liquidplug becomes smaller andthe radial position of the vortex centre shifts to- wards the capillary axis (Thulasidas et al., 1997; Taha and Cui, 2004). Also, the bubble nose becomes more slender and liquidﬁlm ﬂowing aroundthe bubble is thicker. Thulasdias et al. (1997) reportedthat the transition to complete bypass ﬂow occurs around Ca = 0.5. Slug ﬂow provides a useful andconvenient methodof augmenting radial heat and mass transfer. Slug ﬂow has been demonstrated to augment signiﬁcantly radial mass transfer in reactors with catalytically active walls (Horvath et al., 1973). Radial mixing was found to increase with Reynolds number, andrapidly increasedwith decreasing slug length. Enhancement of radial increase in heat transfer by slug ﬂow
666 T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
Fig. 1. Velocity ﬁeldin a bubble train inside circular capillaries with a frame of reference moving with the bubble: liquidusedis 50 centistokes silicone oil ( _{L} = 0.048 Pa s, _{L} = 957 kg m ^{3} , = 22.18 mN/m); capillary diameter = 2 mm; (a) Ca = 0.032; (b) Ca = 0.924.
has been demonstrated by Oliver andWright (1964) and Oliver andYoung Hoon (1968). Bellara et al. (1997) demon- stratedthat injecting gas bubble is a cheap andeffective way of reducing concentration polarisation and thus enhanc- ing the permeate ﬂux in hollow ﬁbre membrane modules.
Prothero andBurton (1961) pointedout that strong circula- tion aheadof the redcells similar to those shown in Fig. 1b was very effective for nutrient distribution. Early experimental investigation on the motion of bub- ble trains in capillaries dates back to Fairbrother andStubbs (1935). Their experiments yielded an empirical correlation for the fraction of the cross section of the tube occupiedby a wetting liquid. This fraction, W , was foundto be propor- tional to the square root of a dimensionless capillary num- ber. This result was foundto be satisfactory for capillary numbers up to Ca = 10 ^{−}^{2} . Taylor (1961) rediscovered the problem andconﬁrmedthe results of Fairbrother andStubbs (1935). He also extended its validity up to Ca = 10 ^{−}^{1} and foundthat, for high capillary numbers, the fraction of the
depositedﬁlm no longer followedthe
1
_{2} -power law but in-
steadasymptotically approacheda constant value of 0.56. Later work due to Cox (1964) indicted that this asymptotic value was about 0.6. Most recent experimental work reportedby Thulasidas et al. (1995) covering upward, downward and horizontal ﬂow indicted that in circular capillaries for Ca > 0.3, the differ- ences in ﬁlm thickness for upwardanddownwardﬂow are negligible. At large Ca , viscous forces are dominant and the effect of gravity is minimised. In the lower Ca range, how- ever, the liquidﬁlm thickness foundto be almost twice that for horizontal ﬂow whereas for downwardﬂow andhorizon- tal ﬂow the bubble diameters are almost undistinguishable (Thulasidas et al., 1995). In a ﬂow-up paper (Thulasidas et al., 1997) the authors conducted experimental work to in- vestigate the ﬂow patterns in the liquidplug. Their ﬁndings were similar to those suggestedbyTaylor (1961). In the past, theoretical studies were limited to the use of lubrication approximation. Bretherton (1961) pioneered such a theory for vanishingly small capillary numbers by assuming that the bubble acts as a tight-ﬁtting piston with a thin liquid ﬁlm lubricating it. The domain is divided into two regions: the constant-ﬁlm thickness region andthe bub- ble nose region. In the ﬁlm region, the problem is solved by the methodof regular perturbation theory (Mazouchi and Homsy, 2000). In the bubble nose region, the interface is as- sumedto be almost hydrostatic andits shape nearly circular. Interfacial tension is important in this region. Since this re- gion cannot be smoothly matchedto the constant-thickness region, there exists a transition region in which the shape is deformedby viscous traction andthus viscous forces also become important. The technique of matchedasymptotic ex- pansions is usedbetween the two sub regions: the capillary- static region where the shape of the interface takes on a nearly spherical cap andthe transition region where the lu- brication approximation can be applied. Bretherton (1961) foundthat the fraction of the liquiddepositedon the walls by a wetting bubble follows the well-known ^{2} -power law:
3
W = 1.29(3Ca) ^{2}^{/}^{3} . Bretherton introduced a correction to this correlation using a factor of 1 ± ^{2} Bo for upwardanddown- wardﬂow, respectively. He also performedexperiments to verify his theory. However, the agreement between theory
3
T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
667
andexperiment was less than satisfactory. For Ca > 10 ^{−}^{4} the predicted ^{2} -power law was approximately obeyed. How- ever at slower bubble speeds, the measured value of ﬁlm thickness greatly exceeded the theoretical value. This inabil- ity of the Bretherton theory to correctly predict the fraction of the wetting layer is surprising since the theory is basedon
3
the assumption of vanishingly small capillary number. Chen (1986) performedexperiments to measure the ﬁlm thickness of long andshort bubbles andreporteda signiﬁcant devia- tion from the Bretherton theory. Chen attributedthis effect to surface roughness of the very small capillaries being used. Ratulowski andChang (1990) suggestedan explanation of the experimental discrepancies found by Bretherton (1961) basedon the effect of impurities on the surface tension of newly createdsurfaces. Recently, there have been several numerical studies on the motion of long bubbles through two-dimensional channels andcircular capillaries. Reinlet andSaffman (1985) useda ﬁnite difference method whilst most other researches have adopted either boundary integral methods (Martinez and Undell, 1989) or ﬁnite element methods (Shen andUndell, 1985; Giavedoni and Saita, 1997). Both methods by Reinlet andSaffman (1985) and Martinez andUndell (1989) showed goodagreement with experiments if 10 ^{−}^{2} Ca 2 and 10 ^{−}^{2} Ca 0.2, respectively. Neither numerical method is able, apparently, to treat the low capillary number re- gion because of the difﬁculty in adequately resolving the thin ﬁlm region. At Ca = 10 ^{−}^{2} , both solutions predict a thinner wetting ﬁlm than that given by Bretherton (1961). For capillary numbers greater than 0.2, the ﬁnite element solutions due to Shen andUndell (1985) showedunrealistic oscillations in the proﬁle of the bubble nose. Martinez and Undell (1989) usedthe boundary integral methodandtheir results are in goodagreement with experimental results in the range of 10 ^{−}^{2} Ca 10. The solution obtainedby Giavedoni and Saita (1997) agreedwell with experimental results as well as with numerical solutions by other re- searchers. Their solutions match the theoretical expression due to Bretheton (1961) up to Ca = 10 ^{−}^{1} . They computed asymptotic values of the ﬁlm fraction, 0.559 at Ca = 0.2 and0.592 at Ca = 10, in agreement with the experimental values reportedby Taylor (1961) and Cox (1964). Martinez andUndell (1989) computeda value of 0.59 at Ca = 10. While the motion of a single Taylor bubble inside circular capillaries has been studied extensively, both experimentally andtheoretically their square counterpart has been overshad- owedexcept of a few studies. Only recently, researchers have started to conduct experimental studies in square capillar- ies (Kolb andCerro, 1991; Ratulowski andChang, 1990). Thulasidas et al. (1995, 1997) performedextensive experi- mental measurements in both circular andsquare capillar- ies. These included the measurement of the bubble velocity, ﬁlm thickness andvelocity ﬁeldaroundthe bubble. The motion of Taylor bubble inside circular capillaries differs from those inside square ones because circular cap- illaries lack the corner ﬂow of the liquidﬁlm. In circular
capillaries, the bubble is axisymmetric acting like a tight-
ﬁt piston with a liquidﬁlm lubricating the bubbles. The ﬂow inside square capillaries, however, is essentially three- dimensional and the bubble acts like a leaky piston. At low Ca the bubble is not axisymmetric, it ﬂattens out against the walls andthe liquidlenses at the leaky corners. At mod-
erate andhigh Ca the bubble body is cylindrical and the bubble is axisymmetric (Wong et al., 1995). This transition is reportedby Kolb andCerro (1991) to occur at Ca = 0.1. Thulasidas et al. (1995) reporteda higher value of Ca = 0.4 which is consistent with the numerical value calculatedby Ratulowski andChang (1989). The leaky corners—which are main features of square capillaries—are characteristics of porous media, monolith froth reactors andmicroelectromechanical systems. Thus, the intention of this work is twofold: ﬁrstly, to provide in- sight into the small-scale hydrodynamics of slug ﬂow with net liquidﬂow inside square capillaries andcompare it to our previous study of their circular counterparts (Taha and Cui, 2004). Secondly, to quantify and understand the result- ing mass andheat transfer enhancement due to two-phase slug ﬂow. A 2 mm capillary are usedfor all simulations. The liquidandgas feedsystem are identical to those used by Thulasidas et al. (1995). Silicone oil was usedas liquid and air as gas. Two different oils were used: 1000 centis- tokes silicone oil and50 centistokes silicone oil with vis- cosities 0.971 and0.048 Pa s, respectively. The densities of
these oils are 971 and957 kg/m ^{3} , respectively. Typical sur- face tension for silicone oil is 22.18 mN/m.
2. Dimensional analysis
Using Pi-theorem, the variables that can affect the motion of bubbles train in capillaries with net ﬂow are:
L
,
G
L
,
G
_{L} U SL D t
L
,
L T B
D
t
,
L
LP
D
t
,
U SL _{L}
and
g( _{L} − _{G} )D
2
t
^{U} ^{T} ^{B} ^{} ^{L}
,
where D _{t} is the diameter of the tube, U _{T} _{B} is Taylor bubble velocity, _{L} and _{G} are the density of the liquidandgas, respectively, _{L} and _{G} are the viscosity of the liquidand gas respectively, is the surface tension, L _{T} _{B} is the length of Taylor bubble, L _{L}_{P} is the length of liquidplug and U _{S}_{L} is the liquidsuperﬁcial velocity. The relative bubble velocity andthe ﬁlm thickness shouldbe a function of these parame- ters. Under the assumption that inertial forces in the gas are far smaller than the inertial forces in the liquid ( _{L} / _{G} 1), _{L} / _{G} can be eliminated. In their experiments, Suo and Grifﬁth (1964) foundthat the effect of the length of gas slug andliquidplugs are unimportant. They also foundthat the motion of the bubble is independent of _{L} / _{G} as long as _{L} / _{G} > 25. Thus, in the absence of the effects due to gas viscosity andinertia, andfor bubbles which move
668 T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
independently, the factors affecting the bubble relative ve- locity andﬁlm thickness are:
Eo =
2
g( _{L} − _{G} )D
t
,
U SL _{L}
and
Ca =
Re =
_{L} U SL D t
L
,
U T B _{L}
.
The ﬁrst group (Eo) is the ratio between buoyancy andsur- face tension forces. Bretherton (1961) usedBondnumber, B _{o} = Eo/4. The secondgroup is the liquidReynolds num- ber; the thirdgroup represents the relative importance of viscous andsurface tension effects, andthe ﬁnal group is the capillary number which is the ratio of viscous to sur- face tension forces. The ﬁrst three groups describe the bal- ance between the retarding forces, namely inertia, viscosity, surface tension, andbuoyancy. The secondandthe thirddi- mensionless groups can be combinedto yielda dimension- less group which is independent of liquid velocity and is a constant for a particular ﬂuidin a particular pipe; Suo and Grifﬁth (1964) referredto it as = _{L} ^{2} /D _{t} _{L} = Ca/Re . Other andthese widely useddimensionless groups can all be derived by manipulating and/or combining two or more of above three groups. In circular capillaries, the values of Eo are small (Eo < 1); in other words, the predominance of surface tension force leads to insensitivity to ﬂow direc- tion andcapillary orientation. Suo andGrifﬁth (1964) de- rivedthe following criterion for neglecting buoyancy effect in capillary slug ﬂow: Eo< 0.22.
3. CFD model development
The CFD software FLUENT (Release 5.4.8, 1998) was usedto simulate the motion of a single Taylor bub- ble rising in a ﬂowing liquidthrough a capillary with a square cross section. In FLUENT, the control vol- ume method—sometimes referred to as the ﬁnite volume method—is used to discretize the transport equations. The movement of the gas–liquidinterface is trackedbasedon the distribution of _{G} , the volume fraction of gas in a com- putational cell, where _{G} = 0 in the liquidphase and _{G} = 1 in the gas phase (Hirt andNichols, 1981). Therefore, the gas–liquidinterface exists in the cell where _{G} lies between 0 and1. The geometric reconstruction scheme that is based on the piece linear interface calculation (PLIC) methodof Youngs (1982) is appliedto reconstruct the bubble free sur- face. The surface tension is approximatedby the continuum surface force model of Brackbill et al. (1992). The numerical results through this work are basedon a three-dimensional grid. The length of the domain is 11L, where L is the square capillary side length. The grid is made up of quadrilateral control volumes. The control volume are smaller in the vicinity of the wall in order to improve res- olution of the liquidﬁlm in this crorical region. The main gridcontains 52 × 52 × 560 volumes. Near the walls, the last row of cells near the walls is sub-divided 3 times. The result
Fig. 2. Initial and boundary conditions for a Taylor bubble rising inside a square capillary in a moving coordinate moving with the bubble.
is a 59 × 59 × 560 volumes in the domain. This reﬁnement method, ensuring ﬁlm region grid independence, does, how- ever, not always guarantee full grid independence in regions where the air–liquid interface is highly curved. Conﬁdence of grid independence results is gained by selecting simula- tions that were run with the gridcells number doubled(i.e., choosing results from a 52 × 52 × 560 gridrather than a 26 × 26 × 280).
T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
669
The frame of reference, initial and the boundary condi- tions are similar to those adopted in axisymmetric simula- tions presentedin our previous study for circular capillaries
(Taha andCui, 2004). The top andthe bottom of the initial bubble are two hemispheres andare connectedto a cylin- der of the same radius (Fig. 2). If other shapes were used (e.g., only a cylinder), the ﬁnal shape of the bubble is found to be similar except the convergence is slower in the latter case. Thus, for the simulations the former initial shape is adopted. The no-slip wall condition is applied to the walls. The ﬂuidmass ﬂux at the inlet is speciﬁedusing a proﬁle for a fully developed ﬂow through a pipe. The governing equa- tions are solved for a domain surrounding a Taylor bubble in a frame of reference attachedto the rising Taylor bubble. With these coordinates, the bubble becomes stationary and the pipe walls move with a velocity U _{w}_{a}_{l}_{l} , numerically equal to that of the Taylor bubble rise velocity, U _{T} _{B} . The liquid is fedat the inlet with an average velocity U _{i}_{n}_{l}_{e}_{t} , which is equal to U _{T} _{B} − U _{S}_{L} . A fully developed velocity proﬁle is imposedat the inlet. The value of U _{T} _{B} is adjusted after the initial guess until the nose of the bubble ceasedto move in the axial direction. Trial simulations were conducted to ex- amine the effect of using a ﬁxedframe of reference; they run longer with the same ﬁnal result as with that of a moving frame of reference.
4. Results and discussion
4.1. Bubble shape
The propagation of long bubbles inside a square capillary is different from those observed in circular capillaries espe- cially at low capillary numbers. In circular capillaries, the ﬂow assumes a cylindrical symmetry. However, for capillar- ies with square cross section the ﬂow is symmetric (but not axisymmetric). This results from the corners being affected by viscous forces to a greater extent than regions near the walls. A three-dimensional view of three bubbles is depicted in Fig. 3 andthe evolution of the bubble shape with increas- ing Ca is shown in Figs. 4 and 5. At low Ca , the bubbles have spherical ends andtendto ﬂatten out against the walls. With increasing Ca , the bubble ends loses its sphericity and a small indentation appears at the rear of the bubble before it start to penetrate the bubble along its axis (Figs. 3 and 5). Other researchers made similar observations (Goldsmith andMason, 1963; Martinez andUndell, 1989; Olbricht and Kung, 1992; Tsai andMiksis, 1994; Giavedoni and Saita, 1999). Fig. 5 shows that, with increasing Ca , the bubble become longer andcylindrical in shape; the liquidﬁlm is thicker andthe nose is sharper resembling those seen in- side circular capillaries. In Fig. 4, the bubble radius in the diagonal direction is found to increase with decreasing Ca , while the liquidﬁlm thickness along the walls decreases and eventually approaches an asymptotic value. At high Ca the bubble radius in the side plane gradually decreases until it
usedis 1000 centistokes silicone oil ( _{} _{L} = 0.971 Pa s, _{} _{L} = 971 kg m ^{3} ,
= 22.18 mN/m); side length = 2 mm; = 21.89; (a) Ca = 0.009; (b) Ca = 1.35; (c) Ca = 3.04.
is equal to that in the diagonal direction and the bubble con- forms to an axisymmetric shape. Thulasidas et al. (1995) foundthat this transition takes place at Ca = 0.4 conﬁrm- ing the numerical computations of Ratulowski andChang (1989). Kolb andCerro, 1991 reporteda higher value of Ca = 0.1. Wong et al. (1995) studied the structure of the ﬂuid ﬁlms deposited by a bubble in the limit of vanishingly small Ca . They foundthe depositedﬁlm is non-uniform in the cross-stream direction and rearranges in downstream. For the Ca limit we studied here it was not feasible to capture this trendbecause of the extreme thinning of the ﬁlm and the very long axial length-scale requiredfor signiﬁcant rear- rangement. The results reportedhere however were similar to those reportedby Hazel andHeil (2002). At Ca = 0.009 the thickness of the liquidﬁlm depositedat the wall for up- wardﬂow is foundto be thicker than that for downwardﬂow due to gravity forces, in accordance with the experimental ﬁndings by Thulasidas et al. (1995). This is shown in Fig. 6.
4.2. Bubble velocity
The dimensionless bubble velocity for square capillaries deﬁned as = U _{T} _{B} / U _{S}_{L} is plottedagainst capillary num- ber in Fig. 7. The overall behaviour of is essentially sim- ilar to the one foundfor circular capillaries, except at low
670 T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
= 0.054; ( ) Ca = 0.054, = 21.89; ( ) Ca = 0.21, = 21.89; (◦)
Ca = 0.4, = 21.89; (×) Ca = 1.35, = 21.89; (+) Ca = 3.04,
= 21.89.
Fig. 5. Bubble shape proﬁle in the side plane for square capillary: liquid usedis 1000 centistokes silicone oil ( _{} _{L} = 0.971 Pa s, _{} _{L} = 971 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; V b = 0.01 ml; = 21.89; ( ) Ca = 0.001; (◦) Ca = 0.026; ( ) Ca = 0.067; ( ) Ca = 0.21; (•) Ca = 0.40; ( ) Ca = 1.35.
Ca : it increases with Ca . The transition to complete bypass, that is when the bubble velocity is equal to the maximum liquidvelocity ( = 2) occurs aroundthe value of 0.4 in agreement with Thulasidas et al. (1997).
4.3. Velocity ﬁeld
Streamlines patterns aroundthe bubble rising inside a square capillary are presentedin Fig. 8. The frame of ref-
Fig. 6. Cross-sectional view of the bubble proﬁle at Ca = 0.009: liq- uidusedis 50 centistokes silicone oil ( _{} _{L} = 0.048 Pa s, _{} _{L} = 957 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; = 0.054 ( ) Upward; (•) Down- ward.
Fig. 7. Dimensionless bubble diameter as a function of capillary num- ber for square capillaries: liquids used are 1000 centistokes silicone
oil ( _{L} = 0.971 Pa s, _{L} = 971 kg m ^{3} , = 22.18 mN/m, = 21.89)
and50 centistokes
silicone
oil
( _{L}
= 0.048 Pa s,
_{L}
= 957 kg m ^{3} ,
= 22.18 mN/m, = 0.054); side length = 2 mm; (• ) numerical results; (◦)
Thulasidas et al. (1995).
erence is attachedto the bubble andthe wall moves down
with a relative velocity equal to the bubble. Fig. 8 shows the streamlines patterns obtainedfor ﬁve different Ca values; these streamlines resemble the patterns observedin circular capillaries. Fig. 8a andb show clearly two vortexes aheadof the bubble. The vortices behindthe bubble are nearly iden- tical to those aheadof the bubble with the eye of the vor- tex easily identiﬁable. As the capillary number is increased the eye of the recirculating ring move away from walls andshift towards the symmetry line; this phenomenon is
T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
671
Fig. 8. Flow velocity ﬁeldarounda gas slug rising inside square capillary with a frame of reference moving with the bubble : liquidusedis 1000 cen- tistokes silicone oil ( _{L} = 0.971 Pa s, _{L} = 971 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; = 21.89; (a) Ca = 0.04; (b) Ca = 0.067; (c) Ca = 0.4; (d) Ca = 1.35; (e) Ca = 0.06 (side direction); (f) Ca = 0.06 (diagonal direction).
observedin both circular andsquare capillaries. Augmenta- tion in Ca results in a detachment of the back ﬂow from the interface producing a new recirculating ﬂow pattern where the recirculating rigs shrink towardthe symmetry line. Fi- nally, the ﬂow recirculation vanishes producing a complete
side view: liquidusedis 1000 centistokes silicone oil ( _{L} = 0.971 Pa s,
_{L}
= 971 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; _{} = 21.89; (1)
Ca = 0.005; (2) Ca = 0.21; (3) Ca = 0.40; (4) Ca = 1.35.
Fig. 10. Velocity proﬁle inside a liquid plug in square capillary along diag- onal direction: liquidusedis 1000 centistokes silicone oil ( _{L} = 0.971 Pa s,
_{L}
= 971 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; = 21.89; (1)
Ca = 0.005; (2) Ca = 0.21; (3) Ca = 0.40; (4) Ca = 1.35.
bypass when Ca is increasedfurther. Apart from the stag- nation point at the vertex of the meniscus andthe above two
stagnation rings, there are two additional stagnation points locatedin the ﬁlm region before the streamlines bow back (e.g., Fig. 8a). In this ﬁgure, the effect on the curvature of the streamlines near the bubble nose can be clearly seen. Fig. 8 indicates that the stagnation points in the ﬁlm region tend to recede toward the nose of the bubble as Ca increases. Side anddiagonal views of the streamlines at Ca = 0.036 are shown in Fig. 8e andf, respectively. The two views are es- sentially similar, except that the liquidﬁlm close to the wall surrounding the vortex andthe bubble is thicker andmore discernable from the diagonal view. The vortex in square capillaries is not toroidal as it lacks cylindrical symmetry and its diagonal view is wider than its side view. Dimension- less velocity proﬁles inside the liquid plugs for four values of capillary number in a frame of reference moving with the bubble are shown in Fig. 9 (side view) and Fig. 10 (diagonal view). Velocity proﬁles in the liquidplug inside square cap- illaries are qualitatively different from those inside circular
672 T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
Fig. 11. Bubble proﬁle and wall shear stress distribution in square capillaries along the side plane at Ca = 0.009: liquidusedis 50 centistokes silicone
oil ( _{L} = 0.048 Pa s, _{L} = 957 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; = 0.054; (•) bubble shape, upwardﬂow; (◦) bubble shape, downward
ﬂow; (—) wall shear stress, upwardﬂow;
(. . .
.) wall shear stress, downward ﬂow.
capillaries. For the lowest Ca scenario, the dimensionless liquidvelocity at the capillary centre moves down with a velocity about 50% faster than the bubble ( = 1.4) while the liquid in the wall vicinity moves downward; hence the vortex in the liquidplug. With increasing the Capillary num- ber, the bubble velocity increases but the relative velocity at the capillary centre decreases. The onset of a complete bypass scenario isreachedwhen the bubble moves at a ve- locity twice that of the average liquidvelocity. At Ca = 0.4 the bubble velocity is nearly twice that of the average liq- uidvelocity indicating that the bubble velocity is approach- ing the maximum liquidvelocity at the capillary axis. In this condition most of the liquidmoves towardthe bubble nose. Another important trendthat can be clearly observed is the position of the vortex eye where the liquidaheadof the bubble becomes motionless. It is clearly evident that the radial position of the vortex eye is displaced toward the capillary wall in accordance with experimental ﬁndings by Thulasidas et al. (1997). The deviation from perfect toroidal vortex can be seen by comparing the position of the vortex eye from the side and diagonal views. For the lowest Ca shown in Figs. 9 and 10, the axial distance from the capil- lary axis is 0.61 from the side view compared to 0.67 from the diagonal view hence the imperfect symmetric (not ax- isymmetric) toroidal vortex. The degree of deviation from perfect toroidal vortex diminishes with increasing capillary numbers. At low capillary numbers, the dimensionless bub- ble velocity in square capillaries is higher than those of their circular counterparts. This indicates that for square capillar- ies the liquidon the capillary axis moves up with a slower velocity. The speedat which liquidin the vortex inside the liquidplug is very important to mass transfer andmixing inside the plugs (Thulasidas et al., 1997).
4.4. Wall shear stress distribution
In the open literature, different mechanisms have been suggestedfor the back transport of particles from the mem-
brane wall to the bulk (Huisman et al., 1999). Shear- induced diffusion seems to be the main mechanism governing the permeate ﬂux. The wall shear stress is relatedto the mass transfer coefﬁcient and several ultraﬁltration models provide a relationship between permeate ﬂux andmass transfer co- efﬁcient (Porter, 1972). It is imperative, therefore, to study the wall shear stress proﬁle in capillaries in order to quantify the mass/heat transfer enhancement square capillaries. The wall shear stress distribution around an air slug rising inside a square capillary for various capillary numbers is shown in Figs. 11 and 12. These ﬁgures compare the wall shear stress distribution (taken along the centre of the side wall) around the bubble andthe bubble proﬁle in the side plane for the upwardanddownwardﬂow at the same capillary number. The side wall shear stress distribution is nearly identical for both cases. As the liquidis draggedaroundthe bubble, it experiences the largest shear stress just before entering the ﬁlm. The ﬂuctuations in shear stress proﬁle are only seen for cases where Ca is low correspondto the wake aheadof the bubble. Similar feature were observedfor circular cap- illaries at low Ca (Taha andCui, 2004). A similar mechanism at the back endof the ﬁlm generates another peak in the shear stress. The central region for both cases is characterisedby a near-zero wall shear stress. Sim- ilar features were observedfor circular capillaries. In up- wardﬂow the corner wall shear stress changes its sign twice in the bubble region. The ﬁrst takes place when the liquid ﬁlm reaches its terminal thickness andthe secondat the end of the ﬁlm. For the upwardﬂow, the negative shear stress,
T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
673
Fig. 12. Bubble proﬁle and wall shear stress distribution in square capillaries along the diagonal direction at Ca = 0.009: liquidusedis 50 centistokes
silicone oil ( _{L} = 0.048 Pa s, _{L} = 957 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; = 0.054; (•) bubble shape, upwardﬂow; (◦) bubble shape,
downwardﬂow; (—) wall shear stress, upwardﬂow;
(. . .
.) wall shear stress, downward ﬂow.
indicating upwardﬂow, aheadof the bubble persists beyond the tip of the bubble before becoming positive as the down- ﬂow is establishedin the liquidﬁlm aroundthe bubble. The shear stress attains its maximum positive value at the onset of the wave before it decreases rapidly to zero near the bub- ble tail. For the downwardﬂow, the liquidﬁlm in the corner always ﬂows downward with a peak in the wall shear stress
proﬁle at the endof the bubble nose. Fig. 13 depicts the cor- ner wall shear stress together with the bubble shape proﬁle for four values of Ca . The inﬂuence with increasing Ca on the proﬁle of the bubble shape is similar to those discussed above. The wall shear stress decreases to zero coinciding with the liquidﬁlm reaching its terminal thickness. The wall shear stress then attains its maximum positive value just be- fore the wave at the bubble tail. Still in the bubble region, it decreases rapidly to zero before its full recovery in the liquid plug. With increasing Ca , the bubble nose region become longer andaccordingly transition from upwardto downward ﬂow in the ﬁlm shifts its position to the right andthe portion of the downward ﬂow becomes shorter.
4.5. Enhancement in mass and heat transfer by slowﬂow
Kawakami et al. (1989) reportedthat radial mass transfer during bubble train ﬂow through square capillaries is 4–5 times higher for upwardﬂow than for downwardﬂow. On the other hand, Bellara (1998), who foundthat upwardﬂow has no discernable difference to downﬂow operation when they introduced gas bubbles to enhance the ultraﬁltration of large molecules in aqueous solutions using hollow ﬁbre membranes. These apparently contradictory ﬁndings can be explainedby the substantial difference between the ﬂows aroundthe bubble inside circular capillaries andtheir square counterparts. In circular capillaries motion of the bubble is
Fig. 13. Corner wall shear stress distribution in square capillaries : liquid usedis 1000 centistokes silicone oil ( _{L} = 0.971 Pa s, _{L} = 971 kg m ^{3} , = 22.18 mN/m); side length = 2 mm; = 21.89; (1) Ca = 0.026; (2) Ca = 0.038; (3) Ca = 0.054; (4) Ca = 0.067.
axisymmetric andthe walls shear stress distribution around the bubble is identical for both upwardanddownwardﬂow. However, the ﬂow in square capillaries is asymmetric. As a consequence of thicker ﬁlm for upwardﬂow, the corner wall shear stress for upwardﬂow in the ﬁlm region is generally higher than that of downwardﬂow andof the same mag-
nitude to the one near the bubble ends (Figs. 11 and 12). The change in sign of the corner wall shear stress, seen only for upwardﬂow, is another source of enhancement of radial mass transfer. Radial mixing in bubble train ﬂow in capillaries is also strongly inﬂuencedby the bubble length andfrequency. Many researchers have reportedthat the length andfre- quency of the bubble plays a pivotal role in heat andmass
enhancement. Horvath et al. (1973) showedthat radial mass
674 T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675
transfer in bubble train ﬂow inside circular capillaries in-
creases when the bubble length decreases. Kawakami et al. (1989) however, reportedan increase in radial mass transfer with increasing gas ﬂow rates. Explanations to what could be contradictory observations can be sought from the wall shear stress distributions. In circular capillaries, the radial mass transfer is mainly due to the vortices aheadandbehind the bubble. Thus, injecting smaller gas bubbles in circular capillaries wouldshorten the near-zero wall shear stress in the central region aroundthe bubble. The corner wall shear stress in square capillaries resemble those observedin large tubes; for ﬁxedliquidﬂow rates the wall shear stress in- creases with increasing the bubble length (Li et al., 1997). Another effect of the corner ﬂow is the augmentation of axial mixing—that is exchange of matter between liquid plugs. For circular capillaries, the liquidplugs, bridgedby extremely thin ﬁlms, are nearly perfectly isolated(Skeggs, 1957; Thulasidas et al., 1999). Thanks to the corner ﬂow in square capillaries, a better exchange between the liquid plugs is achievedandthis explains the perfect mixing in the liquidplugs observedby Kawakami et al. (1989) inside their monolith bioreactors.
5. Conclusion
The characteristics of slug ﬂow inside circular capillar- ies andsquare micro-channels were revealedusing the VOF method, showing the fundamentally different hydrodynam- ics of such a ﬂow from that in larger tubes andthe effect of corner ﬂow in square capillaries. The velocity ﬁeldaround the bubble were shown in details for a wide range of Ca showing different ﬂow patterns inside liquid plugs in agree- ment with experimental ﬁndings. The shape and terminal velocity of the Taylor bubble were predicted by the VOF methodandagreedfavourably with the publishedexperi- mental ﬁndings. Based on hydrodynamics of slug ﬂow, a light was shedto explain andunderstandthe augmentation of mass andheat transfer in capillaries.
Notation
Bo
Ca
D _{t}
Eo
g
L
_{L}_{P}
L _{T} _{B}
r
R
R
_{t}
Bondnumber Bo = ^{4}^{g}^{(} ^{} ^{L} ^{−} ^{} ^{G} ^{)}^{D} capillary number =( U _{b} / ) diameter of capillary, m Eötvös number, dimensionless acceleration due to gravity, m /s ^{2} length of liquidplug, m length of Taylor bubble, m radial distance from the axis, m bubble radius, m capillary radius, m
t
2
, dimensionless
Re Reynolds number, dimensionless U _{i}_{n}_{l}_{e}_{t} inlet velocity, m/s U _{S}_{L} liquidsuperﬁcial velocity, m/s
U _{T} _{B} Taylor bubble velocity, m/s U _{w}_{a}_{l}_{l} wall velocity, m/s
v Radial liquid velocity, m/s
W velocity ratio for fraction liquiddepositedon
walls, dimensionless
x axial coordinate, m
Greek letters
_{G}
_{G}
_{L}
_{G}
_{L}
volume fraction of the gas phase in the compu- tational cell, dimensionless dimensionless number, dimensionless gas molecular viscosity, kg/m ^{3} liquidmolecular viscosity, kg/m ^{3} gas density, kg/m ^{3} liquiddensity, kg/m ^{3} surface tension, N/m bubble dimensionless velocity, dimensionless
Acknowledgements
T. Taha is grateful to Karim Rida Said Foundation for ﬁnancial support andto Dr DavidKenning for helpful suggestions. This work is partially sponsoredby EPSRC
(GR/66438).
References
Bellara, S.R., 1998. Novel developments in the application of ultraﬁltration for bioseparation. D.Phil. Thesis, University of Oxford. Bellara, S.R., Cui, Z.F., Pepper, D.S., 1997. Fractionation of BSA and lysozyme using gas spargedultraﬁltration in hollow ﬁbre membrane modules. Biotechnology Progress 13, 869–872. Brackbill, J.U., Kothe, D.B., Zemach, C., 1992. A continuum method for modeling surface tension. Journal of Computational Physics 100,
335–354.
Bretherton, F.P., 1961. The motion of long bubbles in tubes. Journal of FluidMechanics 10, 166–188. Charpentie, H.C., Favier, M., 1975. Some liquid holdup experimental data in trickle-bedreactors for foaming andnanofoaming hydrocarbons. A.I.Ch.E. Journal 21 (6), 1213–1218. Chen, J.D., 1986. Measuring the ﬁlm thickness surrounding a bubble
inside a capillary. Journal of ColloidandInterface Science 109,
341–349.
Cox, B.G., 1964. An experimental investigation of the streamlines in viscous ﬂuidexpelledfrom a tube. Journal of FluidMechanics 20,
193–200.
Cybulski, A., Moulijn, J.A., 1994. Monoliths in heterogeneous catalysis. Catalytic Review in Science andEngineering 36, 179–270. Fairbrother, F., Stubbs, A.E., 1935. Studies in electroendosmosis—VI. The “bubble tube” methodof measurement. Journal of Chemical Society 1, 527–529. Gauglitz, P.A., St. Slaurent, C.M., Radke, C.M., 1987. An experimental investigation of gas-bubble breakup in constrictedsquare capillaries. Journal of Petroleum Technology 39, 1137–1146.
T. Taha, Z.F. Cui / Chemical Engineering Science 61 (2006) 665 – 675 675
Ghiaasiaan, S.M., Abdel-Khalik, S.I., 2001. Two-phase ﬂow in microchannels. Advances in Heat Transfer 34, 145–254. Giavedoni, M.D., Saita, F.A., 1997. The axisymmetric and plane cases of a gas phase steadily displacing a Newtonian liquid—a simulation solution of the governing equations. Physics of Fluids 9, 2420–2428. Goldsmith, H.L., Mason, S.G., 1963. The ﬂow of suspensions through tubes—II. Single large bubbles. Journal of ColloidandInterface Science 18, 237–529. Hazel, A.L., Heil, M., 2002. The steady propagation of semi-inﬁnite bubble into a tube of elliptical or rectangular cross-section. Journal of FluidMechanics 470, 91–114. Hirasaki, G.J., 1989. The steam-foam process. Journal of Petroleum Technology 41, 449–456. Hirt, C.W., Nichols, B.D., 1981. Volume of ﬂuid(VOF) methodfor the dynamics of free boundaries. Journal of Computational Physics 39,
201–225.
Holm, L.W., Garrison, W.H., 1988. CO _{2} diversion with foam in an immiscible CO _{2} ﬁeldproject. SPE Reservoir Engineering 3, 112–118. Horvath, C., Solomon, B.A., Engasser, J.M., 1973. Measurement of radial transport in slug ﬂow using enzyme tubes. Industrial and Engineering Chemistry Fundamentals 12, 431–439. Huisman, I.H., Tragardh, G., Tragardh, C., 1999. Particle transport in cross microﬁltration—II. Effect of particle–particle interactions. Chemical Engineering Science 54, 281–289. Huzyak, P.C., Koelling, K.W., 1997. The penetration of a long bubble through a viscoelestic ﬂuidin a tube. Journal of Non-Newtonian Fluid Mechanics 71, 73–88. Irandoust, S., Anderson, B., 1988. Monolithic catalysts for nanoautomobile applications. Catalytic Reviews in Science andEngineering 30,
341–392.
Kawakami, K., Kawasaki, K., Shiraishi, F., Kusunoki, K., 1989. Performance of a honeycomb monolith bioreactor in a gas–liquid–solid three-phase system. Industrial and Engineering Chemistry Research 28, 394–400. Kolb, W.B., Cerro, R.L., 1991. Coating the inside of a capillary of square cross section. Chemical Engineering Science 46, 2181–2195. Li, Q.Y., Cui, Z.F., Pepper, D.S., 1997. Effect of bubble size and frequency on the permeate ﬂux of gas spargedultraﬁltration with tubular membranes. The Chemical Engineering Journal 67, 71–75. Martinez, M.J., Undell, K.S., 1989. Boundary integral analysis of the creeping ﬂow of long bubbles in capillaries. Transactions of ASME Journal of AppliedMechanics 56, 211–217. Mazouchi, A., Homsy, G.M., 2000. Thermocapillary migration of long bubbles in cylindrical tubes. Physics of Fluids 12, 542–549. Olbricht, W.L., Kung, D.M., 1992. The deformation and breakup of liquid drops in low Reynolds number ﬂow through a capillary. Physics of Fluids A 4, 1347–1354. Oliver, D.R., Wright, S.J., 1964. Pressure drop and heat transfer in gas–liquidslug ﬂow in horizontal tubes. British Chemical Engineering 9, 590–596. Oliver, D.R., Young Hoon, A., 1968. Two-phase non-Newtonian ﬂow—II. Heat transfer. Transactions of the Institution of Chemical Engineers 46, 116–122.
Porter, M.C., 1972. Concentration polarization with membrane ultraﬁltration. Industrial and Engineering, Product Research and Development 11, 234–248. Poslinski, A.J., Oehler, P.R., Stokes, V.K., 1995. Isothermal gas-assisted displacement of viscoplastic liquids in tubes. Polymer Engineering Science 35, 877–892.
Prothero, J., Burton, A.C., 1961. The physics of bloodﬂow in capillaries—I. The nature of the motion. Biophysical Journal 1, 565–
575.
Ransohoff, T.C., Radke, C.M., 1987. Weeping ﬂow around nonwetting bubbles in smoothly constrictednoncircular pores. Physics and Chemistry Hydrodynamics 8, 255–263. Ransohoff, T.C., Radke, C.M., 1988. Laminar ﬂow of a wetting liquid along the corners of a predominantly gas-occupied noncircular pore. Journal of ColloidandInterface Science 121, 392–401. Ransohoff, T.C., Gauglitz, P.A., Radke, C.M., 1987. Snap-off of gas bubbles in smoothly constrictednoncircular capillaries. A.I.Ch.E. Journal 33, 753–765. Ratulowski, J., Chang, H.C., 1989. Transport of gas bubble in capillaries. Physics of Fluids A 1, 1642–1655. Ratulowski, J., Chang, H.C., 1990. Marangoni effects of trace impurities on the motion of long bubbles in capillaries. Journal of FluidMechanics 210, 303–328. Reinlet, D.A., Saffman, P.G., 1985. The penetration of a ﬁnger into a viscous ﬂuidin a channel andtube. SIAM Journal of Science and Statistical Computation 6, 542–561. Shen, E.I., Undell, K.S., 1985. A ﬁnite element study of low Reynolds number two-phase ﬂow in cylindrical tubes. Transanctions of ASME, Journal of AppliedMechanics 52, 253–256. Skeggs, L.J., 1957. An automatic methodfor colorimetric analysis. American Journal of Clinical Pathology 28, 311–322. Suo, M., Grifﬁth, P., 1964. Two-phase ﬂow in capillary tubes. Journal of Basic Engineering 86, 576–582. Taha, T., Cui, Z.F., 2004. Hydrodynamics of slug ﬂow inside capillaries. Chemical Engineering Science 59, 1181–1190. Taylor, G.I., 1961. Deposition of a viscous ﬂuidon the wall of a tube. Journal of FluidMechanics 10, 161–165. Thulasidas, T.C., Abraham, M.A., Cerro, R.L., 1995. Bubble-train ﬂow in capillaries of circular andsquare cross section. Chemical Engineering Science 50 (2), 183–199. Thulasidas, T.C., Abraham, M.A., Cerro, R.L., 1997. Flow patterns in liquid slugs during bubble-train ﬂow inside capillaries. Chemical Engineering Science 52, 2947–2962. Thulasidas, T.C., Abraham, M.A., Cerro, R.L., 1999. Dispersion during bubble-train in capillaries. Chemical Engineering Science 54, 61–76. Tsai, T.M., Miksis, M.J., 1994. Dynamics of a drop in a constricted capillary tube. Journal of FluidMechanics 274, 197–217. Wong, H., Radke, C.J., Morris, S., 1995. The motion of long bubbles in polygonal capillaries. Part 1. Thin ﬁlms. Journal of FluidMechanics 292, 71–94. Youngs, D.L., 1982. Time-dependent multi-material ﬂow with large ﬂuid distortion. In: Morton, K.W., Baibnes, M.J. (Eds.), Numerical Methods for FluidDynamics. Academic Press, New York, p. 273.