Journal of Membrane Science 167 (2000) 253261

An improved synthesis of ultraltration zirconia membranes via the solgel route using alkoxide precursor
Jeffrey Chi-Sheng Wu , Li-Chuen Cheng
Department of Chemical Engineering, National Taiwan University, Taipei 10617, Taiwan Received 22 June 1999; received in revised form 23 August 1999; accepted 27 August 1999

Abstract Crack-free ultraltration zirconia membranes were synthesized using zirconium butoxide via the solgel route. Unlike the conventional technique, the esterication of anhydrous butanol and glacial acetic acid provided the hydrolyzing water. In addition, acetic acid also served as a chelating ligand to stabilize the hydrolysiscondensation process and minimize the agglomeration of zirconia. Esterication was veried by its product, butyl acetate, as identied by a GC/Mass. FTIR spectrum conrmed the presence of bidental ligand. HNO3 was used to perform the peptization, creating a stable particulate sol with pH near 1. The sol was spin-coated on anodic alumina discs, dried under programmed temperature and humidity, then calcined at 500 C to remove organics. Grazing incident XRD veried that the zirconia membrane was partially transformed to the tetragonal phase. SEM showed a uniform layer 0.15 m thick on top of the support. Based on nitrogen permeation, the ow-average pore diameter of the ZrO2 membrane was estimated near 3.6 nm. The pore size distribution of unsupported zirconia was measured from nitrogen adsorption/desorption indicating two modes, centered at 1.5 and 3.6 nm, respectively. TEM revealed that the particle size of zirconia ranged from 5 to 10 nm. Our results illustrate that ZrO2 layer is the closed-packing of nano-size spherical particles. Supported zirconia membrane was tested by a cross-ow ultraltration. The rejection of soybean oil/hexane solution achieved 37% which was much better than that of bare disc. 2000 Elsevier Science B.V. All rights reserved.
Keywords: Zirconia; Porous membrane; Solgel; Gas permeation; Ultraltration

1. Introduction Ceramic membranes are highly promising for use in current separation technology, particularly at high temperatures and in chemically harsh conditions. Ceramic membranes not only have excellent thermal and mechanical properties, but are also resistant to organic solvent and microbiological attack. The unique prop Corresponding author. Tel.: +886-2-23631994; fax: +886-2-23623040. E-mail address: cswu@ccms.ntu.edu.tw (J.C.-S. Wu).

erties of ceramics make them appropriate for many more applications in chemical plants than their organic polymer counterparts. An important application is membrane reactors, capable of enhancing conversion and selectivity by removing product(s) during reactions simultaneously [1]. Zirconia, a widely used ceramic membrane, can be used in aqueous ltration with much better alkali durability than other ceramic membranes, such as alumina and silica [2]. In addition, zirconia membrane allows steam sterilization and cleaning with caustic solution [3]. Porous zirconia membrane is stacked with spherical particles resulting

0376-7388/00/$ see front matter 2000 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 6 - 7 3 8 8 ( 9 9 ) 0 0 2 9 4 - X

254

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261

in higher liquid permeation ux. Studies showed that the drinking water ux of ZrO2 membrane was 510 times higher than that of Al2 O3 under the same pore size [4]. Solgel technique does not require complicated instruments like chemical vapor deposition in membrane preparation. It provides a simple and easy means of synthesizing nano-size particles, which is essential for ultraltration membrane. Because of this simplicity, the synthesis of ceramic membranes via solgel has received considerable attention [47]. Their studies have indicated that stable ZrO2 colloid could be obtained at a low pH after hydrolyzing zirconium alkoxide. The hydrolysis of mixed metal alkoxides could be controlled through the esterication of acetic acid and alcohols [810]. A porous ZrO2 membrane prepared from zirconyl oxalate had a pore diameter near 4 nm with thickness ranging from 0.1 to 0.3 m [11]. The stability of particulate sol, and its drying and calcination procedures are crucial in obtaining crack-free membranes. The pore size of a porous membrane is determined by the particle size because the pore is the space of particle packing. One of the difculties in the solgel method is that the hydrolysis of zirconium alkoxide is highly sensitive to water. Even if the solution of zirconium alkoxide is stirred vigorously, the rate of hydrolysis is so quick that large agglomerated zirconia particles will be precipitated once water is added. This problem makes the production of an industrial batch impossible. This study presents an improved method of preparing ZrO2 sol by controlling the hydrolysis of zirconium alkoxide through the slowly-released esterication water. In addition, the acetic acid also serves as a chelating ligand to reduce the rate o f hydrolysis so that very ne particles can be obtained [12,13]. This concept can also be applied to industrial scale solgel processes to prepare crack-free supported zirconia membranes. Such membranes have the feature of a thin active membrane with good mechanical support.

Fig. 1. The solgel procedure of ZrO2 membrane synthesis.

2. Experimental The hydrolysis processes were performed in a glove box maintained at a relative humidity near 25% by purging with tank nitrogen. Zirconium(IV) butoxide (Zr(OC4 H9 )4 ,7680 wt.% in n-butanol, Strem Chem-

icals) was the starting material. To avoid fast precipitation during polycondensation or the formation of unstable colloidal sols, the hydrolyzing water was homogeneously released by the esterication of butanol and acetic acid. Fig. 1 illustrates the procedure of membrane synthesis. Basically, the intention was to provide the stoichiometric amount of water to hydroloyze zirconium butoxide in the hydrolysis process. A typical batch contained 0.01 mol zirconium butoxide, 0.04 mol anhydrous butanol (min. 99.8%) and 0.04 mol glacial acetic acid (min. 99.7%). The clear solution was stirred for 23 h then became cloudy, indicating the zirconium butoxide was hydrolyzed. The pH value of the solution was measured to be near 5. To rule out the inuence of moisture in the glove box during hydrolysis, a test was performed without adding glacier acetic acid into the zirconia butoxide/butanol mixture. In this test the mixture remained clear even after 24 h of stirring, indicating negligible hydrolysis. The analysis of hydrolyzed sol by a GC/Mass

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261

255

(HP GCD plus) identied the product of esterication, butyl acetate. Hydrolyzed zirconia solution was then peptized by 0.5 ml HNO3 (70%). The sol became transparent again and the pH value was near 1. The HNO3 peptized sol formed a clear stable particulate sol that remained stable at least for 1 month. Synrex (Taiwan) spinner was used for spin-coating zirconia particulate sol on a Whatman asymmetrical anodic disc with 4.7 cm OD. The alumina disc comprises of 0.02 m pores on the top and a bottom structure of 0.2 m straight pores. The total thickness of the disc is near 60 m. The disc was initially spun-up to 1500 rpm. Approximately 2 ml of zirconia sol was dropped in the center of the disc and spun off for 20 s. The spinning speed was then increased to 2000 rpm and another 2 ml of sol was dropped and spun off for additional 25 s. A second spin-coating could be applied after drying, and higher spinning speeds of 2000 and 3000 rpm were used (Fig. 1). The thickness of the coating was only inuenced by the characteristics of the solution, the spinning speed and its duration [14]. The amount of zirconia sol was not critical since oversupply was simply spun off. The membrane was dried in a programmable temperature and humidity chamber (Yih-Der Company, Taiwan). The temperature was raised from 30 C to 70 C, increasing at a rate of 10 C/h, and the relative humidity was reduced from 80% to 40%, decreasing at a rate 10%/h, simultaneously. The temperature and the relative humidity were then maintained at nal condition for 30 min. The dried membrane was calcined under air circulation at 500 C for 60 min, with an increase of 0.5 C/min from ambient to 200 C, and then 1.5 C/min to 500 C. Finally the furnace was turned off and allowed to cool to ambient temperature. Grazing incident X-ray diffractometer (GIXRD) identied the phase of supported ZrO2 membrane. Incidental X-ray with grazing angle (3 ) was applied to measure the thin layer of ZrO2 on the disc. The morphology and structure were examined by scanning electron microscopy (SEM). SEM samples were sputtered with Pt-carbon lm on the supported ZrO2 membrane surface to increase the electrical conductivity. A batch of un-supported zirconia akes was prepared using the same procedures as the membrane. ZrO2 particles were observed using transmission electron microscopy (TEM). TEM samples were prepared by suspending un-supported zirconia in solvent then

dropping solution on a copper-mesh holder. Sample was allowed to dry before TEM analysis. A batch of ZrO2 xerogel was obtained from particulate sol by drying under the same conditions as those of the ZrO2 membrane. Fourier transform infrared spectroscopy (FTIR) was used to identify the chelating ligand. Samples were prepared by pressing pellets of ZrO2 xerogel mixed with KBr. Thermogravimetric analysis (TGA) and differential thermal analysis (DTA) were performed on ZrO2 xerogel to study the weight loss and thermal reactions during calcination. The pore size distribution of un-supported zirconia was calculated from the N2 adsorption/desorption isotherms at 77 K. The nitrogen permeation of supported ZrO2 membrane was carried out in a disc module at room temperature. The feed-side pressures varied between 101 and 202 kPa, and the permeate side was open to the atmosphere. Gas permeate ow ranged from 3507000 ml/min, depending on membranes. The liquid-phase separation performance of ZrO2 membrane was evaluated by the cross-ow ultraltration of soybean oil/hexane solution. The solution, as obtained from Jia-Shin Oil and Grease Company (Taiwan), was directly collected from the solvent extractor in the plant without rening. Detailed procedures of cross-ow ultraltration can be found elsewhere [15].

3. Results and discussion 3.1. Hydrolysis and chelation Zirconium alkoxide is hydrolyzed and condensed to form polymeric species composed of MOM bonds in butanol solution. The process is generally described in Eqs. (1)(3), where R represents C4 H9 . Zr (OR)4 + H2 O Zr (OR)3 (OH) + ROH Zr (OR)3 (OH) + Zr (OR)3 (OH) (RO)3 ZrOZr (OR)3 + H2 O Zr (OR)3 (OH) + Zr (OR)4 (RO)3 ZrOZr (OR)3 + ROH (3) (2) (1)

256

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261

Fig. 3. TGA and DTA of ZrO2 xerogel Fig. 2. FTIR spectrum of ZrO2 xerogel.

peptization, due to the electrostatic repulsion between highly charged particles under low pH condition [16]. ROH + CH3 COOH H2 O + RCOOC2 H5 (4)

Hydrolyzing water can be supplied via two different mechanisms. The rst mechanism is the esterication of butanol and acetic in situ, as shown Eq. (4). The second one is the oxolation, as shown in Eq. (2). The esterication product, butyl acetate was found and identied by a GC/Mass. The esterication becomes much more favorable when the ratio of acetic acid/metal alkoxide is low [10]. Thus, hydrolysis rate is primarily controlled by esterication of water excluding the air moisture. Acetic acid also serves as a chelating ligand and can change the alkoxide precursor at the molecular level, thus modifying the hydrolysis process [12]. Fig. 2 presents the FTIR spectrum of zirconia xerogel. The adsorption peaks at 1453 and 1561 cm1 clearly indicate the bidental ligand CH3 COO . They are very different from acetic acid. Chelation can be described in Eq. (5) [13]. A fast temperature increase 10 C in 10 s) also occurred when acetic acid was added to Zr butoxide/butanol mixture during our experiment. Such a phenomenon is due to the replace with CH3 COO . The hydrolysis of ment of O-nC4 H9 chelated zirconium complex is deferred so that bidental ligand may still remain in the xerogel. Upon further hydrolysis, bidental ligand is broken off and gives to ZrOH bond. This mechanism decreases the rate of hydrolysis, meaning very ne particles of zirconium hydroxide will be formed and suspended in solution. The ne particulate sol is very stable after HNO3

(5) Fig. 3 displays the TGA/DTA results of zirconia xerogel. The weight loss before 120 C indicates the vaporization of physically adsorbed water and butanol (b.p. 117.7 C). As the temperature rises, a signicant exothermic peak accompanied with weight loss is observed near 160 C. Possibly, the oxidation of nitro-compounds is the cause of this huge exothermic reaction. The nitro-compounds may be formed by nitration of butyl acetate and HNO3 . Exothermic peaks between 360 C and 500 C imply the combustion of organics, something further suggested by the gradual decrease in weight over the same temperature range. The weight decrease stops above 500 C indicating that the loss of organics is completed, as reported by Muoz-Aguado et al. [9]. Thus, the calcined temperature of ZrO2 membrane is determined to be 500 C.

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261

257

Fig. 4. SEM cross-section of double-coated supported ZrO2 membrane.

3.2. Characteristics of zirconia membrane Fig. 4 shows the SEM cross-section of a ZrO2 membrane prepared with double coatings of ZrO2 sol. A uniform zirconia layer appears on top of the disc. The thickness of the ZrO2 layer is approximately 0.15 m. The ZrO2 particles are too small to be distinguished under 50 000 magnication by SEM. Other SEM micrographs of single-coated ZrO2 membrane show that no separate top layer can be observed. This implies that most ZrO2 sol was trapped in the 0.02 m pores of disc during the rst coating, and might reduce the pore size. Therefore, the ZrO2 sol could stand on the top of disc during the second coating. Fig. 5 displays the TEM micrograph of unsupported ZrO2 . Most of the ZrO2 particles are spherical and their diameter is in the range of 510 nm. The particles of ZrO2 membrane should be similar to the unsupported ZrO2 since they are prepared by the same procedures. Fig. 6 presents the GIXRD spectrum of the ZrO2 membrane with a broad base line. Peak positions and relative intensities indicate the tetragonal phase of ZrO2 [11,17]. The broad base line of spectrum implies the existence of amorphous zirconia, which is a common result in the solgel process. In sum, the ZrO2 membrane is a packing of both amorphous and nano-crystal zirconia.

Fig. 5. TEM micrograph of unsupported ZrO2 .

Fig. 6. Grazing angle (3 ) XRD of double-coated ZrO2 membrane.

258

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261

Fig. 7. Nitrogen permeation of calcined membranes, 0.02 m disc, single-coated ZrO2 membrane with disc, double-coated ZrO2 membrane with disc, calculated single-coated ZrO2 layer, - - - calculated double-coated ZrO2 layer. Table 1 Calculated N2 permeances and estimated pore diameters of membranes Knudsen permeance (mole/(m2 Pa s)) 0.02 m disc ZrO2 membranea ZrO2 membraneb
a Single b Double

3.3. Pore size and pore size distribution Fig. 7 depicts the permeation of nitrogen versus various mean transmembrane pressures. In general, porous membrane comprises of two kinds of pores. One is continuous-network pore which can let uid pass through the membrane. These pores determine the separation performance and permeation ux of a membrane. The other kind are dead-end pores which can be considered as the void of membrane structure and make no contribution to permeation. The gas permeation is primarily contributed by Knudsen and Poiseuille ows in a porous membrane under our permeation condition. The average radius of continuous-network pores can be estimated from the gas permeation data in Eqs (6) and (7) [18,19]. The total resistance of gas permeation is the linear combination of resistance of support and membrane for a supported-membrane. Thus, the permeation of a single ZrO2 layer (also shown in Fig. 7) can be calculated based on the resistance-in-series model [20]. The ow-average pore diameters of the disc and a single ZrO2 layer can be calculated from their slopes ( ) and intercepts ( ), which are obtained by linear regression from Fig. 7. Table 1 lists the calculated Knudsen permeances and ow-average pore diameters of disc

Pore diameter (nm) 17.6 15.5 3.6

5.67 105 1.23 104 3.04 105

coated. coated.

and sole ZrO2 membranes, respectively. Knudsen permeance is determined by extrapolating permeance to zero pressure, that excludes the Poiseuille ows. The ow-average pore diameter of the disc is 18 nm. This value is close to the 0.02 m pore disc supplied by manufacturer. The average pore diameters are reduced to 15 and 3.6 nm for single and double coating membranes, respectively. The former pore is reduced by the entrapped ZrO2 particles within pores, while the latter is the pores of the top ZrO2 layer (Fig. 4). F = + pavg L r = 3.162 RT M (6)

(7)

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261 Table 2 Cross-ow ultraltration of soybean oil/hexane solutiona Soybean oil/hexane solution Feed (wt.%) Permeate (wt%) Rejection (%)e Stage cut Supportb 34.7 28.7 17 0.033 ZrO2 c 30.3 23.8 21 0.0052

259

ZrO2 d 31.3 19.6 37 0.009

a Cross-ow ltration condition: transmembrane pressure 505606 kPa, paddle agitation 120 rpm, Soybean oil concentration measured by UV adsorption at 458 nm. b 0.02 m anodic disc, Wu and Lee [15]. c Single-coated supported ZrO membrane. 2 d Double-coated supported ZrO membrane. 2 e Rejection = (oil conc. of feed oil conc. of perm)/(oil conc. of feed) 100%.

Fig. 8. Pore size distribution of unsupported ZrO2 by nitrogen adsorption/desorption.

Fig. 8 displays the pore size distribution of unsupported zirconia calculated from nitrogen desorption isotherm. Pore size distribution is split into two modes, and the average pore diameters are at 1.5 and 3.6 nm, respectively. Total porosity is approximately 10% estimated from the nitrogen adsorption volume. Such low porosity could be due to the sintering of very ne ZrO2 particles. The specic surface area is approximately 4 m2 /g calculated by BET equation. The drying of stable sol can give denser material because particle repulsion avoids aggregation, and the particles move around until the most compact packing is formed [5]. Because of this behavior, calcined ZrO2 membrane could form a closed-packed solid sphere. The pore volume is created from the space of ZrO2 particle packing. From above results, the dead-end pores may possess an average diameter of approximately 1.5 nm, and contribute to the microporosity of the membrane. The continuous-network pores may contribute to the mesoporosity, which has an average diameter near 3.6 nm. This diameter is consistent with the ow-average diameter from gas permeation (Table 1). If we assume that the mesoporosity contributes the continuous-network pores, the tortuosity ( ) of the ZrO2 layer can be estimated by using Eq. (8). Tortuosity can only be calculated if mesoporosity and permeate rates are known. The

mesoporosity of membrane is estimated to be near 4% (Fig. 8). The Knudsen permeance of top ZrO2 layer is 3.04 105 mol/(m2 Pa s) (Table 1). The thickness of the ZrO2 layer is near 0.15 m (Fig. 4). Thus, the tortuosity is estimated to be 1.9. This value is lower than that of traditional alumina membrane, which is normally 3.9. This supplementary evidence indicates that the zirconia layer is constructed of spherical particles while the particles in the alumina layer are plate-shaped [6]. F L = = 31.62
0

1 L

1 RT M

(8)

3.4. Cross-ow ultraltration A cross-ow ultraltration of soybean oil/hexane solution was used to evaluate the separation performance of supported ZrO2 membrane. Specically, this work was done to show the defect free nature of the ZrO2 membrane studied here. Table 2 summarizes the separation results. The molecular weight of soybean oil is estimated to be approximately 900 g/mole. A previous separation study reported that 17% rejection could be achieved using a bare 0.02 m disc [15]. Under the same ltration conditions of previous study, the rejection increases to 21% and 31% on single and double coating ZrO2 membranes, respectively. The separation performance improves signicantly because the pore size is reduced. Compared with bare disc, the stage cut reduction of ZrO2 membranes is due to the decrease of permeation as expected. The stage cut of double coated membrane is slightly higher than that

260

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261

of single coated membrane. It is possibly due to less in-depth fouling because of smaller pores on top layer. Separation and gas permeation results illustrate that the supported ZrO2 membrane is crack and defect free. 4. Conclusion

under the project number NSC-86-2214-E-002-009, and the partial support from the Tzong Jwo-Jang Educational Foundation, contract no. 87-S-029, under the auspices of Tjing-Ling Industrial Research Institute, National Taiwan University. References

A crack-free ultraltration zirconia membrane is successfully synthesized on the top of a disc by solgel method using zirconium butoxide. The esterication of acetic acid and alcohol is conrmed and is a feasible means of providing the water necessary for hydrolysis. The water is released homogeneously, resulting in a controllable hydrolysis. Acetic acid also plays an important role in the hydrolysis process through the chelation, thereby stabilizing zirconium alkoxide. Acid peptization can give a stable sol, which makes the spin-coating on the disc reproducible. The results of gas permeation and ultraltration indicate that the zirconia membrane is defect-free. This improved solgel method is highly promising for use in large-scale manufacturing of membranes in industry.

5. Nomenclature F/L L r M pavg R T permeance [mole/(m2 Pa s)] membrane thickness [m] pore radius [m] gas molecular weight [g/mol] average of feed and permeate pressure [Pa] gas constant, [8.314 j/(mol K)] temperature [K]

Greek Symbols intercept, in Eq (6) and (7). slope, in Eq (6) and (7). viscosity of gas [poise] tortuosity porosity

Acknowledgements The authors are grateful for the nancial support of National Science Council, Taiwan, Republic of China

[1] H.P. Hsieh, Inorganic Membrane Reactors A Review, in: R. Govind, N. Itoh (Eds.), Membrane Reactor Technology, AIChE symposium series 268, vol. 85, 1989, p. 53. [2] T. Yazawa, H. Tanaka, H. Nakamichi, T. Yokoyama, Preparation of water and alkali durable porous galss membrane coated on porous alumina tubing by solgel method, J. Membr. Sci. 60 (1991) 307. [3] A.J. Burggraaf, K. Keizer, Synthesis of inorganic membranes, in: R.R. Bhave (Ed.), Inorganic Membranes, Synthesis, Characterization and Applications, Van Nostrand Reinhold, New York ,1991, chap. 2. [4] L. Cot, Ch. Guizard, A. Larbot, Novel ceramic material for liquid separation process: present and prospective applications in microltration and ultraltration, Ind. Ceramic 8 (3) (1988) 143. [5] A. Larbot, J.P. Fabre, C. Guizard, L. Cot, Inorganic membranes obtained by solgel technique, J. Membr. Sci. 39 (1988) 203. [6] A. Julbe, C. Guizard, A. Larbot, A.L. Cot, A. Giroir-Fendler, The solgel approach to prepare candidate microporous inorganic membranes for membrane reactors, J. Membr. Sci. 77 (1993) 137. [7] C.J. Brinker, R. Sehgal, S.L. Hietala, R. Deshpande, D.M. Smith, D. Loy, C.S. Ashley, Solgel strategies for controlled porosity inorganic materials, J. Membr. Sci. 94 (1994) 85. [8] C. Guizard, N. Cygankiewicz, A. Larbot, L. Cot, Solgel transition in zirconia systems using physical and chemical processes, J. Non-Crystalline Solids 82 (1986) 86. [9] M.J. Muoz-Aguado, M. Gregorkiewitz, A. Larbot, Solgel synthesis of the binary oxide (Zr, Ti)O2 from the alkoxides and acetic acid in alcoholic medium, Res. Bull. 17 (1992) 87. [10] S. Doeuff, M. Henry, C. Sanchez, Solgel synthesis and characterization of titanium oxo-acetate polymers, Mater. Res. Bull. 25 (12) (1990) 1519. [11] J. Etienne, A. Larbot, A. Julbe, L. Cot, A microporous zirconia membrane prepared by the solgel process from zirconyl oxalate, J. Membr. Sci. 86 (1994) 95. [12] S. Doeuff, M. Henry, C. Sanchez, J. Livage, Hydrolysis of the titanium alkoxides: modication of the molecular precusor by acetic acid, J. Non-Crystalline Solids 89 (1987) 206. [13] C. Wolf, C. Rssel, Solgel formation of zirconia: preparation, structure and rheology of sols, J. Mater. Sci. 27 (1992) 3749. [14] J.D. Wang, The Study of Spin Coating, Ph. D. Thesis, National Taiwan University, 1995. [15] J.C.-S. Wu, E.H. Lee, Ultraltration of soybean oil/hexane extract by porous ceramic membrane, J. Membr. Sci. 154 (1999) 251.

J.C.-S. Wu, L.-C. Cheng / Journal of Membrane Science 167 (2000) 253261 [16] A. Larbot, J.-P. Fabre, C. Guizard, L. Cot, J. Gillot, New inorganic ultraltration membranes: titania and zirconia membranes, J. Am. Ceramic Soc. 72 (2) (1989) 257. [17] Q. Xu, M.A. Anderson, Solgel route to synthesis of microporous ceramic membranes: thermal stability of TiO2 ZrO2 mixed oxides, J. Am. Ceramic Soc. 76 (8) (1993) 2093. [18] Y.S. Lin, A.J. Burggraaf, Preparation and characterization

261

of high-temperature thermally stable alumina composite membrane, J. Am. Ceramic Soc. 74 (1) (1991) 219. [19] Y.S. Lin, A.J. Burggraaf, Experimental studies on pore size change of porous ceramic membranes after modication, J. Membr. Sci. 79 (1) (1993) 65. [20] M. Mulder, Basic Principles of Membrane Technology, Kluwer Academic Publishers, Dordrecht, 1991, p. 204 and 230231.