Fo

Ba

Sb

a

ARRAA

KCPFV

1

titHswiuoso

taaoaT

0d

Journal of Membrane Science 405– 406 (2012) 275– 283

Contents lists available at SciVerse ScienceDirect

Journal of Membrane Science

jo u rn al hom epa ge: www.elsev ier .com/ locate /memsci

inger-like voids induced by viscous fingering during phase inversionf alumina/PES/NMP suspensions

o Wanga,b, Zhiping Laia,∗

Advanced Membranes and Porous Materials Center, Division of Chemical and Life Science and Engineering, King Abdullah University of Science and Technology, Thuwal 23955-6900,audi ArabiaSchool of Bioengineering and Chemical Engineering, Anhui Polytechnic University, Wuhu 241000, PR China

r t i c l e i n f o

rticle history:eceived 6 December 2011eceived in revised form 4 March 2012ccepted 8 March 2012vailable online 17 March 2012

eywords:eramic hollow fibre membranes

a b s t r a c t

The formation mechanism of phase-inversion ceramic hollow fibre membranes has not been well under-stood. In this paper, we report on the formation of finger-like macrovoids during non-solvent-inducedphase inversion of alumina/PES/NMP suspensions. A membrane structure without such finger-likemacrovoids was observed when the suspension was slowly immersed into pure ethanol or a mixtureof 70 wt% NMP and 30 wt% water, whereas finger-like macrovoids occurred when the suspension wasslid into the non-solvents at higher speeds. We found that the formation process of finger-like macrovoidscould be fully or partially reversed when nascent membranes were taken out from water shortly after

hase inversioninger-like voidsiscous fingering

immersion, depending on the duration of the immersion. Splitting of the fingers during the formationof the macrovoids was also observed during the phase inversion of two alumina/PES/NMP suspensions.These experimental observations were not predicted by current theories of finger-like macrovoid forma-tion in polymer membranes, but appear to mimic the well-known viscous fingering phenomenon. Wetherefore propose that in the phase inversion of ceramic suspensions, the viscous fingering phenomenonis an important mechanism in the formation of finger-like voids.

. Introduction

Ceramic hollow fibre (HF) membranes fabricated throughhe combined phase-inversion/sintering method have attractedncreasing attention over the past 10 years. By combining the fea-ures of ceramic materials with small tubular geometries, ceramicF membrane modules with high packing densities and high

urface-to-volume ratios are able to work under harsh conditionshereas polymeric membranes fail under such conditions. Phase-

nversion ceramic HF membranes have been employed for variousses, such as dense perovskite HF membranes for high-temperaturexygen production [1–4], porous alumina HF membranes as sub-trates for functional layers [5–8], and HF-membrane-based solidxide fuel cells (SOFC) [9–12].

Phase-inversion ceramic HFs normally contain two regions inheir macrostructures. One is a region of finger-like macrovoids,nd the other is the macrovoid-free part that sometimes is referreds the “sponge-like structure” [13]. The existence and arrangement

f finger-like macrovoids may significantly affect the permeabilitynd the mechanical strength of a ceramic HF membrane [12,13].o achieve desired properties, tailoring the macrostructure and the

∗ Corresponding author. Tel.: +966 2 8082408.E-mail address: zhiping.lai@kaust.edu.sa (Z. Lai).

376-7388/$ – see front matter © 2012 Elsevier B.V. All rights reserved.oi:10.1016/j.memsci.2012.03.020

© 2012 Elsevier B.V. All rights reserved.

microstructure of ceramic HF membranes is necessary. For exam-ple, an asymmetric perovskite HF membrane for oxygen separationrequires a thin, dense skin layer and a thick, highly porous sup-port layer dominated by finger-like macrovoids [3,14]; while inan anode-supported micro-tube SOFC, an optimized combinationof a finger-macrovoid layer and a sponge-like layer is required toachieve both high mechanical strength and good electro-chemicalperformance [12].

To control the macrostructure of ceramic HF membranes, theoccurrence and growth of finger-like macrovoids must be manip-ulated during the phase-inversion process. Such manipulationrequires understanding about the formation mechanism of themacrovoids. The formation of the finger-like macrovoids in poly-mer membranes has been studied for a long time, and severalmechanisms have been proposed to interpret empirical observa-tions of the morphologies of polymer membranes [15,16]. Thesemechanisms can be roughly classified into two categories. The firstcategory is based on the assertion that the formation of finger-likemacrovoids is induced by incoming non-solvent flows [17–21], andthe other is based on the assertion that their formation is caused bydiffusional flows, i.e., the growth of nucleated polymer-lean phases

in polymer solutions [22]. The latter category seems more likelyin polymer membranes, though evidence has shown that bothmechanisms are possible, depending on the polymer/solvent/non-solvent system involved [23]. The diffusional-flow mechanism from

2 rane Science 405– 406 (2012) 275– 283

pfita

agfitopaitflfsbc

2p

floptmflnoiiorao[

iitpritnicicpiv

so(aibtt

�

76 B. Wang, Z. Lai / Journal of Memb

olymer membranes was also used to interpret the formation ofnger-like macrovoids in ceramic HF membranes [14], althoughhere is a big difference between dual-phase ceramic suspensionsnd single-phase polymer solutions.

Kingsbury et al. investigated the formation of finger-like voids inlumina HF membranes [13,24,25], and they proposed that the fin-ers are formed in the spinning process by the well-known “viscousngering” phenomenon. They were able to qualitatively interpretheir results using the “viscous fingering” hypothesis by focusingn the effects of the suspension’s viscosity. However, their inter-retation did not exclude the possibility that other mechanismsre at work, and stronger evidence distinguishing finger formationn ceramic suspensions from that in polymer solutions is neededo verify the “viscous fingering” hypothesis. In this study, we usedat-sheet alumina membranes as a benchmark to verify the unique

eatures of the viscous fingering phenomenon in alumina/PES/NMPuspensions and to demonstrate that viscous fingering is responsi-le for the formation of finger-like voids in the phase inversion oferamic suspensions.

. Distinguishing characteristics of the viscous fingeringhenomenon

Viscous fingering is commonly observed when a less viscousuid replaces a more viscous fluid, during which uneven frontsf the replacing fluid invade the replaced fluid and form com-lex patterns, often taking finger-like shapes. Early studies ofhis phenomenon originated from observations of fluid replace-

ent in porous media [26]. Typical examples of replacing/replaceduid systems are slurry/water systems [27], which are similar toon-solvent/ceramic suspension systems in which phase inversionccurs. Most investigations on viscous fingering were carried outn a two-dimensional shallow geometry called a Hele-Shaw cell,n which viscous fingering is analogous to the fluid replacementccurring in a porous medium. The onset and growth of fingersely on the instability of the interface between the replaced fluidnd the replacing fluid. These issues have been discussed both the-retically and experimentally in the literature on fluid mechanics28–30].

There are two main types of viscous fingering processes. Ones the immiscible viscous fingering in which there is a distinctnterface between the replacing fluid and the replaced fluid, withhe interfacial tension playing an important role in the fingeringrocess. The other is the miscible viscous fingering in which theeplacing fluid and the replaced fluid are miscible and there is nonterfacial tension, but the dispersion of the finger front affectshe fingering process [30]. In a system consisting of a displacingon-solvent and a displaced ceramic suspension, both of which are

nvolved in phase-inversion ceramic membrane fabrication pro-esses, the interface between the suspension and the non-solvents always distinct although the exchange between the solvent of theeramic suspension and the non-solvent occurs during the wholehase-inversion process. Such a system can thus be treated as an

mmiscible system with varying interfacial tensions and varyingiscosities.

In the immiscible interface between a non-solvent and a ceramicuspension, when it moves towards the suspension side and a peri-dic disturbance with a wavelength � is applied on the interfaceillustrated in Fig. 1), the disturbance can be either stabilized ormplified depending on the pressure jump across the perturbednterface. The pressure jump comes from the viscosity differenceetween the non-solvent and the suspension, which can be quan-

ified by assuming that the flux of the non-solvent passing throughhe interface follows Darcy’s law [26,30]:

pV = k(�1 − �2)Uıx (1)

Fig. 1. Illustration of a moving interface between two fluids.

where �pV is the pressure jump across the interface due to theviscosity difference, k is the resistance constant of the suspensionto the permeation of the non-solvent, �1 and �2 are the viscosity ofthe suspension and the non-solvent, respectively, U is the velocityof the moving interface, and ıx is the variation of the interface.Because the viscosity of the suspension is usually much larger thanthe viscosity of the non-solvent, we can reduce Eq. (1) to

�pV = k�Uıx (2)

Here, we replace �1 with � for simplicity. This pressure jump tendsto amplify the disturbance towards the suspension and thus initiatethe fingering process.

In the case of flat-sheet membranes where the non-solvent/suspension interface is horizontal, the influence of gravitycannot be neglected. Then, at the disturbed interface, a pressurejump applies due to gravity:

�pG = (�1 − �2)gıx (3)

Here, �1 and �2 are the density of the suspension and the non-solvent, respectively. In the case of ceramic suspensions, normallywe have �1 > �2. Then, the direction of �pG is against �pV and tendsto flatten the interface.

Another force that produces a pressure jump against �pV is theinterfacial tension. The pressure jump, �pT, produced by the inter-facial tension is the product of the interfacial tension, T, and thecurvature, �, of the perturbed interface; that is,

�pT = T� (4)

In the onset and growth of fingers, �pV must be larger thanthe sum of �pG and �pT. Hence, a critical moving velocity of theinterface (or finger fronts) is defined by

Uc = �pG + �pT

k�ıx(5)

The interfacial moving velocity must be higher than the criticalvelocity to initiate or continue fingering processes [28].

The stability of the interface can be assessed by a dimensionlessparameter, the capillary number Ca, which measures the viscousforce relative to the interfacial tension by

Ca = k�UW

T(6)

where W is a dimension-related parameter. In the Hele-Shaw cell,this parameter is the width of the cell, but it needs to be clarified inreal membrane fabrication cases. The fingering process will occurat a high Ca number when the interface becomes less stable, whileit stops or reverses when the Ca is lower than a certain value.

Finger splitting is a commonly observed phenomenon in viscousfingering studies when the capillary number, Ca, is very high. Thisphenomenon is based on the fact that disturbances also exist onthe interface surrounding the moving fingers. It has been foundthat the minimum required wavelength, l, and the amplitude, A, of

a disturbance for initiating fingers are determined by the capillarynumber, Ca:

A∼ 1

e√

Ca(7)

B. Wang, Z. Lai / Journal of Membrane Science 405– 406 (2012) 275– 283 277

Table 1Composition of the suspensions.

Suspension-A (wt%) Suspension-B (wt%)

NMP 36.5 NMP 34.5

l

owfibpsot

cfiafifsnpTmbti

3

3

6(PPe3wt

3

wotdtwf

3m

sc

Al2O3 56.7 Al2O3 57.3PVP 1.0 PD 3315 2.4PES 5.8 PES 5.8

>W√Ca

(8)

When fingers develop, finger fronts are widened and the flatnessf the finger fronts increases. For a large Ca, the required minimumavelength, l, of the disturbance could be small enough so that thenger fronts can be deemed as essentially flat, and thus the distur-ance may initialize secondary fingers on the front interface of therimary fingers, i.e., to split the primary fingers [28]. Splitting on aide interface other than the front is less favored, mainly becausef the much lower interfacial moving velocity of the side interfacehan that of the front and thus the lower Ca.

In ceramic membranes fabricated by the phase-inversion pro-ess, if finger-like voids were formed in the manner of viscousngering, three distinguishing characteristics would be expectedccording to the rules of the viscous fingering theory. Therst is the effect of the interfacial moving velocity on finger

ormation, especially the existence of a critical velocity; theecond is the reversibility of fingers when the stability of theon-solvent/suspension interface increases; and the third is thehenomenon of finger splitting when the instability increases.hese characteristics are not predicted by the theories of finger for-ation in polymeric membranes. Thus, if these characteristics can

e experimentally verified during ceramic membrane fabrication,hat would suggest the validity of the viscous fingering mechanismn ceramic suspensions during phase inversion.

. Experimental

.1. Materials

Baikalox® �-alumina powder (d50 = 0.6 �m, surface area m2/g) was purchased from Baikowski. 1-Methyl-2-pyrrolidinoneNMP, HPLC grade, 99.5%) was purchased from Alfa Asea.olyether Sulfone (PES, Ultrason® E6020P) was provided by BASF.olyvinylpyrrolidone (PVP, Fluka K90, molar weight ∼36,000) andthanol absolute were purchased from Sigma–Aldrich. Zephrym PD315 was provided by CRODA. The alumina powder and the PESere stored in 120 ◦C before use to remove adsorbed moisture, and

he other materials were used as received.

.2. Preparation of alumina suspensions

Alumina powder and PVP (or Zephrym PD 3315) were mixedith NMP and then ball milled at 400 rpm for 3 h. After addition

f PES, the suspension was further ball milled for another 15 ho achieve uniform dispersion. The prepared suspension was thenegassed in a vacuum with gentle stirring for 24 h. The composi-ions of the suspensions used are listed in Table 1. Suspension-Aas used throughout the study, and membranes were fabricated

rom suspension-A unless otherwise specified.

.3. Fabrication of flat-sheet membranes with varied immersionethods and immersion durations

To fabricate flat-sheet membranes with different immersionpeeds, a Teflon O-ring with an inner diameter of 17 mm and aross-section diameter of 3 mm was placed on a glass microscope

Fig. 2. Image of membranes fabricated by using water as the non-solvent.

slide, and the suspension was poured into the inner side of theO-ring. The slide with the suspension was then immersed into anon-solvent bath by one of two means. The first slide was hori-zontally immersed into the non-solvent bath slowly. The secondslide was slid into the non-solvent bath from a height of 4 or 10 cmon a smooth leading rail that was inserted into the non-solventbath. The wall and the bottom of the container of the non-solventformed an angle of 26.5◦ between the leading rail and the surfaceof the non-solvent, guaranteeing a high enough sliding speed whileensuring that the suspension would not spill out of the O-ring dur-ing sliding. The immersed slide was then placed horizontally inthe non-solvent bath for 24 h and then dried at room tempera-ture. Two non-solvents were used; one was pure ethanol and theother was a mixture of 70 wt% NMP and 30 wt% water. An extrastep of water–NMP exchange was carried out with the mixturenon-solvent in a water bath to remove NMP prior to drying.

To fabricate flat-sheet membranes with varied immersion dura-tions, the slide carrying the suspension was horizontally immersedinto a water bath slowly and then taken out after a certain dura-tion. As-prepared slides were then put in a sealed container. A cupof water (∼50 ml) was placed in the same container to generatewater vapor. The slides were stored inside the container for 24 hat room temperature so that the membranes were fully solidified.After that, the membranes were immersed in a water bath againovernight to remove NMP and then dried at room temperature. Forcomparison, a flat-sheet membrane with an immersion durationof 24 h in water but without the water-vapor treatment and thesecondary solvent exchange was also fabricated.

3.4. Observation of macrovoid formation by microscopy

A drop of suspension was dropped on a slide and then coveredby another slide. The slides were spaced by two pieces of weighingpaper to create a gap of ∼30 �m. A few drops of water were thendelivered at the edge of the slides and water was adsorbed intothe gap due to capillary force, which led to phase inversion of thesuspension. An inverted optical microscope was used to record thephase-inversion process.

3.5. Characterization of the alumina membranes

The macrostructure and microstructure of the green body of themembranes were observed by a Quanta 200 field-emission scan-ning electron microscope (SEM). The green bodies were treated byliquid nitrogen to obtain undeformed fractures on the cross section.All specimens were coated with gold before examination.

4. Results and discussion

All flat-sheet membranes prepared in this study were deemedto have good integrity. Fig. 2 is an image of samples fabricated by

278 B. Wang, Z. Lai / Journal of Membrane Science 405– 406 (2012) 275– 283

F n and

a nd 30

m

hawd

4fl

enstn

ig. 3. The macrostructure of flat-sheet membranes fabricated by (a) slow immersiond (e) sliding from a height of 4 cm or (f) 10 cm into a mixture of 70 wt% NMP aembrane surface.

orizontal and slow immersion into water. This image shows thatlthough the membranes were as thick as ∼3 mm, their integrityas not damaged by the difference of shrinking rates at differentepths when the phase inversion occurred.

.1. The effects of immersion methods on the macrostructure ofat-sheet membranes

Flat-sheet membranes were fabricated by using either purethanol or a mixture of 70 wt% NMP and 30 wt% water as the

on-solvent. The macrostructures of both kinds of membraneshowed a dependence on the method of immersion (Fig. 3). Whenhe suspension was horizontally and slowly immersed into eitheron-solvent, the membranes had symmetric structures free of

(b) sliding from a height of 4 cm or (c) 10 cm into ethanol; and by (d) slow immersionwt% water. In image (a), the membrane was vertically placed and the left edge is

finger-like macrovoids (Fig. 3a and d). However, when the suspen-sion was immersed by sliding, an asymmetric structure with longfingers was obtained when ethanol was the non-solvent (Fig. 3band c), while small and short fingers were observed when the mix-ture of 70 wt% NMP and 30 wt% water was the non-solvent (Fig.3e and f). We note that in the membranes obtained by sliding intoethanol, a large number of small fingers aligned in the top layer andevolved into few large fingers towards the bottom. This is identicalto the commonly observed “shielding” phenomenon in viscous fin-gering processes, where fingers moving faster outrun and prevent

the growth of slower fingers [30]. The suspension was slid fromtwo heights, i.e., 4 cm and 10 cm, but no notable difference in fin-ger length or width was observed in either non-solvents from theseheights.

rane S

onbpimtiifnrniitestimveTss

4fl

pTmfi(beot(loiitbtbo

cabsffmfiibIboc

B. Wang, Z. Lai / Journal of Memb

The big macrostructural difference between the membranesbtained by slow immersion and sliding can be attributed to theon-Newtonian fluid nature of the ceramic suspension, which haseen verified by Kingsbury’s rheology study of similar alumina sus-ensions [25]. When the suspension was immersed slowly, the

nterface between the non-solvent bath and the suspension barelyoved (relative to the glass plate) during the immersion, and

he interfacial moving velocity was lower than the critical veloc-ty for initiating viscous fingering. The surface of the suspensionmmersed by sliding underwent a shear rate parallel to the inter-ace during the sliding process because of the friction between theon-solvent and the suspension. With a Newtonian fluid, the shearate would not produce a movement normal to the interface, but aon-Newtonian fluid would produce a movement normal to the

nterface because of the first normal stress difference [31]. Thiss identical to the expansion when a polymer solution is ejectedhrough a bore. It has been found that the first normal stress differ-nce of a non-Newtonian fluid is proportional to the square of thehear rate when the shear rate is slow [31]. Thus, in the sliding case,he movement normal to the interface would be much faster thann the case of slow immersion, which might have led the interfacial

oving velocity to exceed the critical velocity, and thus finger-likeoids were initiated in the membranes. This result indicates thexistence of a critical velocity during the finger formation process.he reason why the finger length was identical regardless of theliding height is probably because the sliding process ended in ahort time, and thus it only affected the initiation stage of the voids.

.2. The effects of immersion duration on the macrostructure ofat-sheet membranes

Fig. 4a shows the macrostructure of a flat-sheet membrane pre-ared by immersing the suspension into the water bath for 24 h.he figure shows that long fingers almost passed through the wholeembrane wall and stopped before the bottom. The surface of the

ngers exhibits an interesting structure with bumps and groovesFig. 4b). In some cases, the grooves went deeply into the wallsetween the fingers. Flat-sheet membranes fabricated with differ-nt immersion durations were used to examine the effects of timen finger formation. In the membrane not immersed in water butreated with water vapor, no finger-like macrovoids were foundFig. 4c). The membrane immersed in water for 30 s has a simi-ar macrostructure to that shown in Fig. 4c, and no fingers werebserved even in the very top layer (Fig. 4d). In the membranemmersed in water for 1 min, fingers were very short and almostnvisible in the cross-section view. They are clearly evident in theop view of the surface-peeled membrane (Fig. 4e). In the mem-rane immersed in water for 5 min (Fig. 4f), fingers are aboutwo-fifths as long as those shown in Fig. 4a. On the surface of mem-ranes immersed in water for up to 5 min, water accumulation wasbserved shortly after being taken out.

Finger formation in ceramic suspensions is a rather quick pro-ess and is normally completed within 30 s, as observed here as wells by Kingsbury [25]. We see that there were no fingers in the mem-rane immersed for 30 s (Fig. 4c). The lengths of the fingers werehort in membranes immersed for more than 1 min (Fig. 4e and). These membranes had remarkably different macrostructuresrom that immersed in water for 24 h, although a finger-dominant

acrostructure was expected in all these membranes given the fastngering speed. According to viscous fingering theory, if the viscos-

ty of a high-viscosity displaced fluid is unchanged, fingers woulde flattened if the capillary number Ca decreases to a low value.

n our case, when the membranes were removed from the waterath, the replacing water flow was cut off, and thus the velocityf the finger front U was reduced to zero, such that the interfa-ial tension together with the gravity of the suspension tended

cience 405– 406 (2012) 275– 283 279

to push the finger fronts back and to flatten the interface. Weobserved that liquid accumulated on the surface of those mem-branes after removed from water bath, which suggests that thewater was pushed out from the fingers. In the phase-inversion pro-cess of a ceramic suspension, solvent/non-solvent exchange startsfrom the suspension/non-solvent interface and then evolves to thebulk, which leads the polymer to precipitate and to solidify thesuspension. The shapes of the fingers can thus be captured by thepolymer precipitation. Polymer precipitation in a ceramic suspen-sion is a slower process compared with that of finger formation.If the membranes were taken out from the water bath shortlyafter immersion, polymer precipitation would have little time tooccur and the fingers would be completely flattened, as revealedby the membrane immersed for 30 s; by prolonging the immer-sion time, the precipitation process goes deeper into the membraneand the fingers are partially reversed, as revealed by the mem-branes immersed for 1–5 min; but if the immersion duration is longenough, precipitation continues and the fingers are entirely main-tained in the membranes, as revealed by the membrane immersedfor 24 h.

Fig. 5a shows the surface morphology of the membraneimmersed in water for 24 h. The surface is totally covered by poly-mer. A detailed microstructural examination at the top layer revealsthat this layer is full of sponge-like polymer (Fig. 5b), whereas thebottom part of the membrane is network-like polymer embeddedin the alumina particle bed (Fig. 5c). Visual observation suggeststhat the polymer content is higher in the top layer than in thebottom part. It is well known that in the non-solvent induced phase-inversion process of polymer solutions, the polymer solute diffusesto the surface due to the steep chemical potential gradient [18],which explains the polymer enrichment on the surface and in thetop layer of the current ceramic membrane’s green body. If the fin-gers are the consequence of viscous fingering, then the entrancesof the fingers should be seen on the membrane surface immersedfor 24 h. The entrances are however invisible as shown in Fig. 5a.The reason for this can be revealed by gradually evolved surfacemorphology of the membranes immersed in water bath for up to5 min. In the membrane immersed for 30 s, in which no fingerswere observed, the surface is composed of alumina particles withsmall primary packing pores between them, and no visible poly-mer blocks can be found (Fig. 5d); in the membrane immersed for1 min, in which short fingers remained in the bulk membrane, big-ger secondary pores can be found and polymer blocks are visibleon the surface (Fig. 5e). It is reasonable to believe that those sec-ondary pores were the entrances of the fingers. In the membraneimmersed for 5 min, where long fingers were observed, polymerblocks occupy a large proportion of the surface and fill the poresbetween the alumina particles, and the secondary pore structureis not visible (Fig. 5f). Fig. 5 demonstrated that although polymersdiffused to the surface and formed a polymer layer during the phaseinversion of the suspension, this was a slower process compared tofinger formation. After the completion of the fingering process, thefinger entrances were covered by the later-formed polymer layer.

4.3. The effects of the non-solvent

In this study we found remarkable differences in the mem-brane macrostructures when different non-solvents were used. Inmembranes made by horizontal and slow immersion into the non-solvent, the macrostructure was symmetric and without finger-likevoids when the non-solvent was ethanol or the mixture of NMP andwater, whereas the macrostructure was dominated by finger-like

voids when the non-solvent was water. Furthermore, in mem-branes made by sliding into the non-solvent, the finger length inthe case of ethanol was much longer than that in the case of themixture of NMP and water.

280 B. Wang, Z. Lai / Journal of Membrane Science 405– 406 (2012) 275– 283

Fig. 4. Structure of flat-sheet membranes fabricated by using water as the non-solvent. (a) The macrostructure of the membrane immersed for 24 h; (b) the finger surfaceo embm hich t5

iiscfibatsoot

f the membrane immersed for 24 h; (c) the macrostructure of the non-immersed membrane immersed for 1 min; the insert shows a top view of the membrane on w

min.

Although in viscous fingering, the viscosity of the non-solvents a factor that may affect the formation of fingers as describedn Eq. (1), it was too small compared with the viscosity of theuspension in this study, and thus minor deviations in the vis-osity of the non-solvent produced no notable changes in thengers. We used various non-solvents to fabricate flat-sheet mem-ranes, and no correlation between the viscosity of the non-solventnd the dimensions of fingers could be established. We mayherefore conclude that the difference in the viscosity of non-

olvents did not lead to macrostructural change. A commonlybserved phenomenon in polymer membranes is that the tendencyf forming finger-like voids is closely related to the precipita-ion rate of the polymer solution. The macrostructure change

rane; (d) the top layer of the membrane immersed for 30 s; (e) the top layer of thehe skin layer was peeled off; (f) the macrostructure of the membrane immersed for

in the ceramic membranes studied here also obeyed the empir-ical rule in polymer membranes: the strongest precipitant –water – produced a macrostructure dominated by fingers; themedium precipitant ethanol produced fingers only when the slid-ing immersion technique was used; the weakest precipitant, themixture of NMP and water, produced fingers that were muchshorter than in the ethanol case. The connection between themacrostructure of the membranes and the precipitation rate canbe explained using the change in the viscosity of the suspen-

sion. When the suspension contacted the non-solvent, to enhancethe instability of the suspension/non-solvent interface and initi-ate fingering, the viscosity of the suspension near the interfaceneeded to be further increased, and this increase was achieved

B. Wang, Z. Lai / Journal of Membrane Science 405– 406 (2012) 275– 283 281

F tructuo

bietttocetvIts

ig. 5. (a) Surface morphology, (b) top-layer microstructure, and (c) bottom microsf a membrane immersed in water for (d) 30 s, (e) 1 min, (f) 5 min.

y the solvent/non-solvent exchange and the consequent precip-tation of the suspension. Below, we use the cases of water andthanol where the suspension was slowly immersed to discusshe effects of the non-solvent. To reach the critical viscosity ofhe interfacial suspension needed for fingering, water took a shortime, during which the solvent/non-solvent exchange occurrednly in a relatively thin surface layer, and in this layer the vis-osity was notably increased. To reach the same critical viscosity,thanol took more time, and thus the viscosity-increased layer washicker than in the case of water. The viscosity-increased layer con-

ersely hampered the growth of the fingers by consuming energy.n the case of water, it was easier to grow fingers because of thehinner viscosity-increased layer and thus the less energy was con-umed. Furthermore, due to the fast precipitation of the suspension

re of a flat-sheet membrane immersed in water for 24 h; and surface morphology

surface by water, the water/suspension interface may obtain exces-sive instability in a short time, leading to higher tendency for fingergrowth. In the case of ethanol, the formation of the fingers wasstopped at the very beginning in the thicker viscosity-increasedlayer because of the higher energy consumption. This principlealso applies to the cases of ethanol and the mixture of NMP andwater when the suspension was slid into the non-solvent, exceptthe needed critical viscosity of the interfacial suspension was lowerbecause of the increased interfacial moving velocity.

4.4. Finger splitting in alumina suspensions

In the flat-sheet membrane immersed in water for 24 h, the fin-ger surface exhibited a rough morphology with bumps and grooves

282 B. Wang, Z. Lai / Journal of Membrane Science 405– 406 (2012) 275– 283

F ck edgB fabric

tsimcosTegwpsit6itfpb

gssactdsbt

ig. 6. Patterns generated by two-dimensional phase inversion. (a) Suspension-A, ba, back edge of the suspension. (d) The macrostructure of the flat-sheet membrane

hat penetrated into the walls between the fingers (Fig. 4b). Theame suspension was used to study the phase-inversion processn two dimensions, i.e., in the narrow gap between two optical

icroscopy slides. The suspension film between the slides was cir-ular in shape, and water was supplied to the suspension film fromne side, with the assumption that the front edge of the suspen-ion underwent a remarkably higher water flow than the back edge.wo different patterns were obtained at the front edge and the backdge as depicted in Fig. 6. At the back edge of the suspension, fin-ers similar to those observed in the membranes described earlierere obtained (Fig. 6a). At the front edge, however, highly branchedatterns resulted from finger splitting were obtained (Fig. 6b). Inuspension-B with the dispersant PD 3315, which suggests reducednterfacial tension of the suspension/water interface, branched pat-erns were observed even at the back edge of the suspension (Fig.c). Finger splitting was confirmed with a real membrane by slowly

mmersing suspension-B into water. In the as-obtained membrane,he walls between the fingers collapsed entirely when the fingersormed, and only a highly porous structure was left at the lowerart of the membrane (Fig. 6d). It was so porous that the connectionetween the top layer and the bottom layer was almost lost.

This phenomenon is consistent with the widely observed fin-er splitting in viscous fingering studies. The grooves on the fingerurface as shown in Fig. 4b are the traces of uncompleted initial-tage secondary fingers, which were interrupted by being pushedside from the finger front. As discussed above, finger splittingan be strengthened by increasing the capillary number Ca, andhis has been achieved by increasing U, as shown in Fig. 6b, or by

ecreasing the interfacial tension, T, as shown in Fig. 6c. Fig. 6dhows three-dimensional finger splitting in a real ceramic mem-rane fabrication, confirming that the finger splitting observed inwo dimensions was not an artefact due to the capillary effect

e of the suspension; (b) suspension-A, front edge of the suspension; (c) suspension-ated from suspension-B.

between the two slides. Finger splitting produces highly porousstructures and severely reduces the mechanical strength of mem-branes. It should be avoided in membrane fabrication. In somecircumstances, however, it might be useful. For example, in densemembranes used in high-temperature oxygen separation, the sur-face exchange of oxygen plays an important role in improvingoxygen permeation [3,14], while the requirement of the mem-brane’s mechanical strength is less rigorous. In this situation, ahighly porous structure produced by finger splitting would bea benefit because of high surface area and hence fast surfaceexchange.

5. Conclusions

By using alumina/PES/NMP suspensions as examples, we haveinvestigated non-solvent-induced phase inversion of the suspen-sions and discussed the formation of finger-like voids during thephase-inversion process. This investigation verified the followingthree distinguishing characteristics of finger formation in ceramicsuspensions:

(1) The structural transition of the membranes (with fingers orwithout fingers) when different immersion methods are usedis dramatic. We attribute this transition to the different movingvelocities of the non-solvent/suspension interface.

(2) Finger formation in the phase-inversion process of a ceramic

suspension is reversible, as observed by taking out nascentmembranes from the non-solvent bath shortly after immersion.

(3) Finger splitting occurs during finger formation in ceramic sus-pensions.

rane S

ftfv

A

Ae2P

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[[

[

[

B. Wang, Z. Lai / Journal of Memb

These characteristics of the finger formation process are uniqueeatures of the well-known viscous fingering phenomenon. It isherefore reasonable to suggest that the driving force of fingerormation during phase inversion of a ceramic suspension is theiscosity difference between the suspension and the non-solvent.

cknowledgements

This work was supported by baselines funds to Z.P. Lai from Kingbdullah University of Science and Technology. B. Wang acknowl-dges support from the Ministry of Education of China (Contract No.10093). We thank Dr. S.H. Choi for stimulating discussions and Dr.. Wang for providing access to the microscope.

eferences

[1] X. Tan, Z. Pang, K. Li, Oxygen production using La0.6Sr0.4Co0.2Fe0.8O3−ı (LSCF)perovskite hollow fibre membrane modules, J. Membr. Sci. 310 (2008) 550–556.

[2] T. Schiestel, M. Kilgus, S. Peter, K. Caspary, H. Wang, J. Caro, Hollow fibre per-ovskite membranes for oxygen separation, J. Membr. Sci. 258 (2005) 1–4.

[3] B. Zydorczak, Z. Wu, K. Li, Fabrication of ultrathin La0.6Sr0.4Co0.2Fe0.8O3−ı hollowfibre membranes for oxygen permeation, Chem. Eng. Sci. 64 (2009) 4383–4388.

[4] J. Sunarso, S. Liu, Y.S. Lin, J.C.D. da Costa, High performance BaBiScCo hol-low fibre membranes for oxygen transport, Energy Environ. Sci. 4 (2011)2516–2519.

[5] M. Kilgus, V. Gepert, N. Dinges, C. Merten, G. Eigenberger, T. Schiestel, Palladiumcoated ceramic hollow fibre membranes for hydrogen separation, Desalination200 (2006) 95–96.

[6] M.P. Gimeno, Z.T. Wu, J. Soler, J. Herguido, K. Li, M. Menendez, Combination ofa two-zone fluidized bed reactor with a pd hollow fibre membrane for catalyticalkane dehydrogenation, Chem. Eng. J. 155 (2009) 298–303.

[7] Z. Wang, Q. Ge, J. Shao, Y. Yan, High performance zeolite LTA pervaporationmembranes on ceramic hollow fibers by dipcoating-wiping seed deposition, J.Am. Chem. Soc. 131 (2009) 6910–6911.

[8] L. Shan, J. Shao, Z. Wang, Preparation of zeolite MFI membranes on aluminahollow fibers with high flux for pervaporation, J. Membr. Sci. 378 (2011)319–329.

[9] N. Droushiotis, U. Doraswami, D. Ivey, M.H.D. Othman, K. Li, G. Kelsall,Fabrication by co-extrusion and electrochemical characterization of micro-

tubular hollow fibre solid oxide fuel cells, Electrochem. Commun. 12 (2010)792–795.

10] M.H.D. Othman, N. Droushiotis, Z. Wu, G. Kelsall, K. Li, High-performance,anode-supported, microtubular SOFC prepared from single-step-fabricateddual-layer hollow fibers, Adv. Mater. 23 (2011) 2480–2483.

[

[

cience 405– 406 (2012) 275– 283 283

11] M.H.D. Othman, N. Droushiotis, Z. Wu, G. Kelsall, K. Li, Novel fabrication tech-nique of hollow fibre support for micro-tubular solid oxide fuel cells, J. PowerSources 196 (2011) 5035–5044.

12] M.H.D. Othman, Z. Wu, N. Droushiotis, G. Kelsall, K. Li, Morphological studies ofmacrostructure of Ni-CGO anode hollow fibres for intermediate temperaturesolid oxide fuel cells, J. Membr. Sci. 360 (2010) 410–417.

13] B.F.K. Kingsbury, K. Li, A morphological study of ceramic hollow fibre mem-branes, J. Membr. Sci. 328 (2009) 134–140.

14] X. Tan, N. Liu, B. Meng, S. Liu, Morphology control of the perovskite hollowfibre membranes for oxygen separation using different bore fluids, J. Membr.Sci. 378 (2011) 308–318.

15] P. van de Witte, P. Dijkstra, J. vandenBerg, J. Feijen, Phase separation processesin polymer solutions in relation to membrane formation, J. Membr. Sci. 117(1996) 1–31.

16] G.R. Guillen, Y. Pan, M. Li, E.M.V. Hoek, Preparation and characterization ofmembranes formed by nonsolvent induced phase separation: a review, Ind.Eng. Chem. Res. 50 (2011) 3798–3817.

17] M. Frommer, R. Messalem, Mechanism of membrane formation. 6. Convectiveflows and large void formation during membrane precipitation, Ind. Eng. Chem.Prod. Res. Dev. 12 (1973) 328–333.

18] H. Strathmann, K. Kock, Formation mechanism of phase inversion membranes,Desalination 21 (1977) 241–255.

19] R. Ray, W. Krantz, R. Sani, Linear-stability theory model for finger formation inasymmetric membranes, J. Membr. Sci. 23 (1985) 155–182.

20] F. Paulsen, S. Shojaie, W. Krantz, Effect of evaporation step on macrovoid for-mation in wet-cast polymeric membranes, J. Membr. Sci. 91 (1994) 265–282.

21] R. Matz, Structure of cellulose acetate membranes. 1. Development of porousstructures in anisotropic membranes, Desalination 10 (1972) 1–15.

22] C. Smolders, A. Reuvers, R. Boom, I. Wienk, Microstructures in phase-inversionmembranes. Part 1. Formation of macrovoids, J. Membr. Sci. 73 (1992) 259–275.

23] J. Lai, F. Lin, T. Wu, D. Wang, On the formation of macrovoids in PMMA mem-branes, J. Membr. Sci. 155 (1999) 31–43.

24] B.F.K. Kingsbury, Z. Wu, K. Li, A morphological study of ceramic hollow fibremembranes: a perspective on multifunctional catalytic membrane reactors,Catal. Today 156 (2010) 306–315.

25] B.F.K. Kingsbury, A morphological study of ceramic hollow fibre membranes: aperspective on multifunctional catalytic membrane reactors, PhD thesis, Impe-rial College London, London, UK, 2010.

26] S. Hill, Channelling in packed columns, Chem. Eng. Sci. 1 (1952) 247–253.27] H. Vandamme, F. Obrecht, P. Levitz, L. Gatineau, C. Laroche, Fractal viscous

fingering in clay slurries, Nature 320 (1986) 731–733.28] D. Bensimon, L. Kadanoff, S. Liang, B. Shraiman, C. Tang, Viscous flows in 2

dimensions, Rev. Mod. Phys. 58 (1986) 977–999.29] P. Saffman, Viscous fingering in Hele-Shaw cells, J. Fluid Mech. 173 (1986)

73–94.30] G. Homsy, Viscous fingering in porous media, Annu. Rev. Fluid Mech. 19 (1987)

271–311.31] R.P. Chhabra, J.F. Richardson, Non-Newtonian Flow in the Process Industries:

Fundamentals and Engineering Applications, Butterworth-Heinemann, 1999.