- Email: [email protected]
Ceramic membrane elaboration using supercritical fluid
Materials Research Bulletin, Vol. 34, Nos. 12/13, pp. 2013–2025, 1999 Copyright © 2000 Elsevier Science Ltd Printed in the USA. All rights reserved 0025-5408/99/$–see front matter
PII S0025-5408(99)00225-1
CERAMIC MEMBRANE ELABORATION USING SUPERCRITICAL FLUID
J. Brasseur-Tilmant, K. Chhor, P. Jestin, and C. Pommier* Laboratoire d’Inge´nie´rie des Mate´riaux et des Hautes Pressions, CNRS LPR 1311, Institut Galile´e, Universite´ Paris Nord, Avenue Jean-Baptiste Cle´ment, 93430 Villetaneuse, France (Refereed) (Received September 8, 1998; Accepted February 15, 1999)
ABSTRACT Macroporous alumina supports were prepared from commercial ␦-Al2O3 and modified by hydrolytic decomposition of titanium isopropoxide in supercritical propan-2-ol. Various experimental parameters were studied in relation to the deposition of TiO2 particles on the substrate surface and inside the pores: temperature, precursor concentration, water-to-alkoxide ratio, solution flow rate, and deposition time. A set of these parameters was found to allow supported membranes with high nitrogen permeation (9.5 ⫻ 10⫺6 mol䡠m⫺2䡠 s⫺1䡠Pa⫺1 ) and low average pore radius (1.5 nm) to be obtained. © 2000 Elsevier Science Ltd
KEYWORDS: A. ceramics, A. thin films, B. sol-gel chemistry, C. electron microscopy INTRODUCTION The application of porous inorganic membranes in filtration and catalytic reactions has attracted much attention because of their high thermal, chemical, and mechanical stability. Supported membranes exhibit very low pore size (in the nanometer range) and high selectivity while keeping high permeability. They are obtained from a macroporous ceramic support treated in such a way that nanoparticles can be deposited either as a thin film on the surface or inside the pores in order to decrease their diameter. In the first case, sol-gel
*To whom correspondence should be addressed. Fax: ⫹33-1-49-40-34-14. E-mail: pommier@limhp. univ-paris13.fr. 2013
2014
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
techniques, either colloidal or polymeric, have been used widely [1–7]. In such processes, a sol must first be prepared under controlled conditions. After coating, drying has to be performed very slowly at room temperature before calcination, usually above 400°C, in order to obtain crystallized oxide. Numerous studies have been conducted by various laboratories to get a better understanding of the chemical and physical phenomena involved during the sol transformation and to develop specific strategies leading to controlled pore sizes for flaw-free supported ceramic membranes [2]. First, a variety of deposition techniques, in addition to the initial slip casting method, such as deep coating, filtration deposition, and permformation, has been proposed [6]. Second, many metal oxides used as membrane materials (alumina, titania, zirconia) exhibit a structural phase transformation between 450 and 900°C, which favors crystallite growth and sintering [5]. In order to obtain crack-free membranes and to improve their thermal stability, many precautions must be taken during the drying and calcination steps. Phase transformation must be retarded at higher temperature, for example, by using a doping oxide [5] or avoided by starting with a suitable sol (rutile sol in the case of TiO2 membrane) [7]. Such a route to modify supports to obtain membranes for ultra- or nano-filtration is very time-consuming, but a pore diameter in the deposit can be lowered to 3–5 nm. However, to prevent crack formation, the optimum thickness of the layer obtained under each coating treatment is claimed to be less than 0.2 m [4]. Another route to reduce pore size is the chemical vapor deposition (CVD) process. Reaction between two reactive gas flows usually leads to the formation of already crystallized metal oxide particles. In the “classical” method, streams are supplied on the same side of the support: reaction occurs at the corresponding surfaces and substrate pores are covered or plugged. In a “modified” CVD method, reactant flows counterdiffuse from both sides of the membrane and react when they contact each other on the inner surface of the pores. Comparing the two methods with the SiCl4/H2O system, it has been shown that a one-sided membrane has significantly higher selectivity for H2/N2 separation [8]. However, it has been noted that, in the counterdiffusion technique, silica formation in zeolite materials can be controlled by the molecule size of the reactant [9]. This last process has been used for TiO2 and ZrO2 deposition either inside the pores of the substrate (pore diameter ⬎ 100 nm) or in the supported top layers of the membrane with smaller pore size (⬍20 nm) and thickness (about 5 m) [10,11]. Reaction between the two gas flows (metal chloride in argon and water in air) was conducted between 700 and 1000°C during a relatively short time (about 20 min). Metal oxide deposition was found at the substrate surface exposed to the metal chloride vapor, and a kinetic study showed that it proceeded according to a heterogeneous reaction [10]. Pore reduction observed from gas permeation measurements was shown to strongly depend on the initial pore size distribution, but no final diameter values were reported [11]. Recently, a new method to modify the pore size distribution of ceramic membranes, using a supercritical fluid solution of a metal oxide precursor (such as metal alkoxide) infiltrated through the macropores was proposed [12]. The infiltrated samples are further immersed in water vapor for complete hydrolysis and finally dried and treated at 450°C in order to obtain the final products. Propane used as solvent is saturated with aluminum isopropoxide at 130°C under 30 MPa and infiltrated into ␣-Al2O3 membranes under the same pressure and temperature conditions. During the infiltration, the pressure of the supercritical solution falls off along the flow direction, and solute supersaturation is reached inside the pores. It has been shown that either homogeneous or heterogeneous nucleation can then occur. The first mechanism is observed when the surface energy of alumina support is lowered before
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2015
treatment by a first coating with methyl cellulose, which makes the surface more wettable. Without such a first coating, the porosity and the total area of open pores also decrease. However, heterogeneous nucleation leads to an excessive reduction of smaller pores and consequently to an increase in average size. Supercritical fluids effectively exhibit physical characteristics that may be of great potential interest in the processing of ceramic materials: low viscosity, high diffusivity, and high solvent power with strong dependence on pressure. In the last decade, various applications have been described for the use of submicronic powder or thin film synthesis. In our research group, we have developed new processes that start with a metal alkoxide as precursor and a supercritical alcohol as solvent. For example, we have reported that reaction in such a medium leads to MgAl2O4 or TiO2 powder crystallization, and that particles with diameters ranging from 20 to 100 nm are obtained [13,14]. Another advantage of such a process is the formation of TiO2 films on alumina substrate at a high deposition rate [15]. It has been shown that under experimental conditions (T ⫽ 300°C, P ⫽ 10 MPa), the precursor is hydrolyzed by water produced in the dehydration of propan-2-ol used as solvent [16]. In the present work, we studied the opportunity to prepare supported ceramic membranes by applying such a reaction to the formation of a thin microporous layer at the surface of a macroporous support. EXPERIMENTAL Support Preparation. Porous alumina discs (25 mm diameter, 1.6 mm thickness) were prepared from ␦-Al2O3 (Alumina C from Degussa, particle size 15 nm; ⬎99% purity). About 1500 ppm magnesium oxide (Fluka, ⬎98% purity) and 5% polyvinyl alcohol (M ⫽ 22000 g䡠mol⫺1) as organic binder were added, in order to favor sintering and to retain high porosity, respectively. The mixture was compacted under 30 MPa and heated up to 1400°C, with intermediate stages at 600 and 1100°C. The obtained discs had a porosity of about 60%, pore size of about 200 nm, and were crystallized under an ␣-alumina structure. Supported Membrane Preparation. Titania particles were deposited on the support, hydrolyzing titanium tetraisopropoxide (TTIP), Ti(O-iC3H7)4 (Aldrich, 98% purity) in supercritical propan-2-ol (Tc ⫽ 235°C, Pc ⫽ 4.76 MPa). Experimental device. The whole system used is schematized in Figure 1. A high-pressure pump (HPP) sent liquid alcohol or alkoxide solution flow inside the reactor (R) at temperature T (T ⬎ Tc). A regulating valve (Vr) allowed the pressure to be maintained at about 10 MPa inside the system. The solvent was cooled down and condensed. At the end of the experiment, an inert gas (N2) can be sent through the system, to evacuate alcohol and residual by-products. A closer view of the reactor is shown in Figure 2. The substrate to be treated was fastened between two PTFE flat rings, to prevent fluid leaks between upstream and downstream zones (A and B, Fig. 2). This system appeared efficient up to about 350°C. Glass wool allowed a temperature gradient to set up between the two sides of the substrate and acted as a catalyst for alcohol dehydration; water was produced in zone B of the reactor. Experimental procedure. The successive steps involved in the TiO2 deposition process are schematized in Figure 3. The closed reactor, containing about 100 cm3 of pure propan-2-ol (higher than the critical volume), was first heated up to temperature TB for time t1 (40 to 100
2016
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
FIG. 1 Experimental device for dynamic treatment of ceramic supports (R ⫽ reactor; HPP ⫽ high pressure pump; V1 to V7 ⫽ valves; Vr ⫽ regulating valve). min). During this step, water was formed, the amount depending on TB and t1. (This point is discussed in the next section.) Then, a pure, cold solvent flow was sent through the system for time (t2 ⫺ t1), i.e., 5 to 10 min, to evacuate water from zone A. During this step, temperature TA decreased and reached an equilibrium value of about 60 to 80°C lower than TB, depending on the introduced alcohol flow rate. At the same time, some of the water present in zone B was evacuated; thus, its concentration decreased. At t2, the pure alcohol flow was replaced by TTIP solution at the same flow rate, and a reaction took place for time (t3 ⫺ t2), usually 10 to 15 min. During the last step, pure solvent flow was sent through the system, to evacuate unreacted TTIP before decompressing (t4). Finally, inert gas (N2) was used to remove any remaining by-products. At time t1, the pure alcohol flow was sent through the system until thermal equilibrium was reached (t2). The time duration (t2 ⫺ t1) was higher than that necessary for complete water
FIG. 2 Closer view of the reactor, showing substrate fastening and glass wool location.
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2017
FIG. 3 Successive steps in the substrate treatment process: (a) reactor thermal equilibrium setting, (b) ROH flow, (c) TTIP solution flow, (d) ROH flow, and (e) decompressing, N2 flow.
evacuation from zone A, as estimated from a plug flow model. However, under experimental conditions, water concentration in zone B did not reach a constant value (even without hydrolysis reaction), because the amount formed in the alcohol dehydration was lower than that evacuated by the supercritical flow. A decrease in water concentration between t2 and t3 is evidenced from gas chromatographic analysis (⬍25%). In the following section (h calculation, Table 1), mean values are used. Characterization Methods. On all modified membranes, the crystal structure of the deposited layer was studied by X-ray diffraction (Philips PW1729 diffractometer) and Raman spectroscopy (XY Dilor). Thickness, particle size, and shape in both the deposited layer and
TABLE 1 Characteristics of Alumina Substrate and Some Modified Membranes Obtained Under Various Experimental Conditions (TB ⫽ 330°C, f ⫽ 5 cm3䡠min⫺1) Run 0
Run 1
Run 2
Run 3
Run 4
Run 5
CTi (mol䡠L⫺1) h
— —
0.025 15
0.025 15
0.025 25
0.0125 30
0.0125 30
e (m) d (m) F0
— — 18
3 50 3.5
— — 3.7
4 25 8.3
1 15 9.3
— — 9.5
␣  /␣ r (nm)
12.8 347 267 114
3.46 5.38 15.5 6.5
3.66 5.77 15.8 6.5
8.23 9.21 11.2 4.5
9.17 2.58 2.81 1.3
9.46 2.71 2.86 1.2
⫺2
␣ ⫽ coefficient related to Knudsen contribution (10⫺6 mol䡠m 䡠s⫺1䡠Pa⫺1).  ⫽ coefficient related to Poiseuille contribution (10⫺13 mol䡠m⫺2䡠s⫺1䡠Pa⫺2). F0 ⫽ gas permeation (10⫺6 mol䡠m⫺2䡠s⫺1䡠Pa⫺2l) obtained from N2 permeation.
2018
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
the infiltrated zone were investigated by scanning electron microscopy (Leica S440). For some samples, properties related to their porous structure were studied in more detail. Gas permeation measurements were performed on samples before and after deposition treatment, using a laboratory-made permeation apparatus at room temperature and helium, nitrogen, or carbon dioxide as the permeating gas. Gas flow at constant flow rate was passed through the membrane. Higher upstream pressure Ph and lower downstream pressure Pl were measured with two precision gauges. The applied mean pressure, P¯ ⫽ 1/2 (Ph ⫹ Pl), was from 0.1 to 0.2 MPa. The transmembrane pressure ranged from 100 to 200 kPa, and the flux of permeating gas was measured with a soap bubble flowmeter. Similarly, water permeability and rejection properties were studied. These measurements were performed using polyethyleneglycol (PEG) solutions (Aldrich-Chimie, Mw ⫽ 400, 600, 1000, and 1500 g䡠mol⫺1). Their concentrations in feed solution and in permeate were obtained from size exclusion chromatographic analyses (column, Shodex SB-802 HQ; detector, Gilson 132 RI). RESULTS AND DISCUSSION It is well known that alcohols can be dehydrated over metal oxide based catalysts such as alumina and that the facility of the reaction increases from primary to tertiary derivatives [17]. We have observed that such a dehydration occurred with propan-2-ol in our experimental device (described above), at temperatures higher than 300°C. At 330°C, a rather high water concentration (⬇0.6 mol䡠L⫺1) could be reached after 1 h under static conditions. Above 300°C, with an excess amount of water, the TTIP hydrolysis reaction was complete, and the resulting titanium tetrahydroxide was transformed into TiO2. The overall reaction can be written Ti(OR) 4 ⫹ 2H 2O 3 TiO 2 ⫹ 4ROH
(1)
The ratio between water and alkoxide mole numbers, defined as h ⫽ nH2O/nTi ⫽ cH2O/2Ti, is equal to 2 under stoichiometric conditions. However, it has been shown that complete elimination of organic groups only occurs for h ⬎ 2 [18,19]. For lower h values, the oxo-alkoxide compounds are formed according to the following reaction: Ti(OR) 4 ⫹ xH 2O 3 Ti(OR) 4 –x(OH) x ⫹ xROH
(2)
Supported Membranes Synthesis. In a preliminary study, the objective was to determine experimental conditions allowing the formation of water in an amount sufficient to lead to a complete reaction of the introduced alkoxide precursor, and the deposition of the formed titanium dioxide located only on one side of the porous support. When TB was regulated at 330 or 350°C, temperature gradients between the two opposite faces of the substrate were about 80 and 60°C for liquid flow rates of 5 and 3 cm3䡠min⫺1, respectively. With TB ⫽ 350°C, t1 ⫽ 50 min, and f ⫽ 5 cm3䡠min⫺1, the water concentration in zone B increased up to [H2O]B max ⫽ 0.6 mol䡠L⫺1 and its mean value during the deposition step (c, Figure 3) was [H2O]B ave ⫽ 0.4 mol䡠L⫺1. Under another set of experimental conditions (TB ⫽ 330°C, t1 ⫽ 75 min, f ⫽ 5 cm3䡠min⫺1), these values were [H2O]B max ⫽ 0.5 mol䡠L⫺1 and [H2O]B ave ⫽ 0.35 mol䡠L⫺1. In all cases, the quantity of formed water was highly excessive compared to the amount of titanium alkoxide introduced into the system, even for the highest investigated
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2019
values of solution flow rate and concentration, e.g., f ⫽ 5 cm3䡠min⫺1 and CTi ⫽ 0.05 mol䡠 L⫺1. At high titanium alkoxide concentration (CTi ⫽ 0.05 mol䡠L⫺1) and lower solution flow rates (f ⱕ 3 cm3䡠min⫺1), TiO2 particles were formed both on the upper and lower faces of the substrate, as well as inside its pores. Furthermore, the amount of solid deposited in zone A was less influential on increasing flow rate, because increase in the amount deposited induced a temperature decrease inside this area. No more particles were formed on the substrate face A when TA was equal to or less than 275°C. Starting with lower alkoxide concentration (CTi ⱕ 0.025 mol䡠L⫺1), the reaction occurred only on the lower face of the substrate when the flow rate was between 3 and 5 cm3䡠min⫺1 and TB between 330 and 350°C. The localization of the TiO2 deposit on face B of the substrate could be obtained from a conjunction of suitable values for the three parameters: fluid flow rate, temperature, and alkoxide concentration. The temperature gradient between the two faces had to be sufficiently high so that the reaction occurred only on face B. At a low fluid flow rate, water formed in zone B could diffuse more easily through the substrate and react inside its pores. A higher flow rate limited such a phenomenon and increased the temperature gradient. A low TTIP concentration allowed a complete reaction on the substrate face B. Further experiments were conducted with TB ⫽ 330°C and liquid flow rate f ⫽ 5 cm3䡠 min⫺1 for reactive solutions with TTIP concentration CTi ⫽ 0.025 mol䡠L⫺1. Study of the Deposited Layers. In Table 1 are reported some results obtained under various synthesis conditions. Experiments conducted at TB ⫽ 350°C were not taken into account: A rather poor reproducibility was found, related to technical problems due to the behavior of PTFE rings at this temperature. Furthermore, compared to that obtained at 330°C, the homogeneity of the deposits was not improved. Many of the modified membranes exhibited a gray coloring, probably due to the presence of carbon particles formed during alkoxide decomposition. However, these particles were easily eliminated, and perfectly white samples were recovered after thermal air treatment at about 350°C for 2 h. It was shown that such treatments did not modify significantly the microstructure and filtration properties of the deposited layer. As expected from a previous study [14], the solid particles formed under the above conditions were crystallized in the anatase structure. This can be evidenced by the presence of its strong line at 2 ⫽ 25.4° on the X-ray diffraction diagram. As shown in Figures 4a and 4b, the pattern obtained from face A of the treated substrate is that of the original alumina support. Conversely, in the diagram recorded from face B (Fig. 4c), the intensity of the Al2O3 peaks is much lower, and lines associated with the presence of TiO2 anatase deposit appear. Figure 5 compares Raman spectra obtained from both sides of a modified membrane. In the face A spectrum (Fig. 5a), only alumina Raman lines are observed. In the face B spectrum (Fig. 5b), only lines related to the anatase-deposited layer are present: The underlying alumina support was not reached by the incident laser light. Scanning electron micrographs (Fig. 6) show that three zones can be distinguished in the cross section of the treated alumina discs. First, according to our experiments, TiO2 particles were formed on face B as a layer with a thickness e ranging from 1 to 4 m (Figs. 6a and c). Second, particles were observed to have a somewhat lower density inside the alumina support macropores, because of water counterflow diffusion (Fig. 6d). The infiltrated zone depth d ranged from 15 to 50 m according to our experiments. The third zone was an unmodified alumina support, in which only a few isolated nanometric TiO2 particles were observed in some macropores (Fig. 6b).
2020
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
FIG. 4 X-ray diffraction patterns of alumina support (a) and modified membranes face A (b) and face B (c). Starred (夽) lines correspond to TiO2 anatase.)
The precursor concentration and the amount of water inside the reactor mainly influenced the surface morphology and thickness of the deposited layer, as well as the permeation properties of the modified membranes. For h ⫽ 15 and CTi ⫽ 0.025 mol䡠L⫺1 (run 1, Table 1), angular particles below 1 m in size were observed and e ⬇ 3 m (Fig. 6a). For a higher h value (h ⫽ 30) and CTi ⫽ 0.0125 mol䡠L⫺1 (run 4, Table 1), rather spherical agglomerates of particles less than 30 nm in size formed a dense thin layer (e ⬇ 1 m) (Figs. 6c and d). In these two samples, the depth d of the infiltrated zone was about 50 and 15 m, respectively. Results on permeation measurements and rejection properties of the membranes are reported in Figures 7 to 10. Figure 7 compares the nitrogen permeation of substrate and modified membranes. After various applied treatments, permeation was divided by a factor of 2 to 5, according to the amount of titania particles both inside and on the surface of the alumina support. The results obtained with various gases are reported in Figure 8. As expected, permeability decreased with an increase of the gas kinetic radius. For example,
FIG. 5 Raman spectra of modified membrane obtained in run 3: (a) face A and (b) face B.
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2021
FIG. 6 SEM micrographs on modified membranes obtained in run 1: (a) face B, (b) cross section (50 m from surface); and in run 4: (c) face B, (d) cross section. when comparing He and N2 curves, a large difference is observed in their ordinate at origin, which is related to a difference in Knudsen diffusion contribution [20,21]. For treated and untreated membranes, water flux as a function of pressure drop, ranging from 0.1 to 1 MPa, is compared in Figure 9. In the investigated range, a linear variation was observed, as predicted by the Kozony-Carman equation [22,23].
FIG. 7 N2 permeation as function of mean pressure for the substrate (a) and modified membranes obtained in (b) run 1, (c) run 3, and (d) run 4.
2022
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
FIG. 8 He, N2, and CO2 permeation as function of mean pressure for modified membrane obtained in run 4.
In order to estimate the mean pore size of the deposited layer, rejection properties of modified membranes were studied, and the retention ratio for various PEG samples are reported in Figure 10. is defined as ⫽ 1 ⫺ (Cp/Cf), where Cp and Cf are solute concentrations in permeate and feed solutions, respectively. The cut-off value is usually taken for a 90% solute retention. In the present case, the corresponding solute molar weight was estimated to be about 2000, associated with a gyration radius of about 1.5 nm. In the theoretical treatment of gas permeation measurements, simplifying assumptions are usually introduced. Both the deposited layer and support are considered to have cylindrical pores with average pore radii of rl and rs, respectively. Furthermore, the gas flow through the membrane is assumed to be the sum of Knudsen diffusion and Poiseuille laminar flow contributions [24,25]. Then, by comparing results obtained on membranes before and after treatment, the average pore size change under treatment can be checked. Measuring the gas flow rate Q passing through a cross-sectional area A of the membrane, the gas permeation F0 can be calculated according to its definition:
FIG. 9 Dependency of water flux with pressure drop for (a) the substrate and (b) modified membrane obtained in run 4.
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2023
FIG. 10 Retention behavior of PEG polymers on modified membrane obtained in run 4.
F 0 ⫽ Q/( A ⫻ ⌬P)
(3)
where ⌬P ⫽ Ph ⫺ Pl is the pressure drop between upstream (h) and dowstream (l) sides. It has been shown [24] that, even for composite membranes, gas permeation usually follows a linear variation vs. the average applied pressure P¯ ⫽ 1/2 (Ph ⫹ Pl): F 0 ⫽ ␣ ⫹ P¯ F 0
(4)
In this relation, ␣ and  are coefficients related to the Knudsen and Poiseuille contributions, respectively. As noted above, Figures 7 and 8 show a linear variation of permeation F0 vs. the applied mean pressure. Corresponding ␣ and  parameters are reported in Table 1. Under treatment, the membrane permeability decreased as the result of pore narrowing. Accordingly, the ␣ parameter, which is related to the Knudsen diffusion contribution, also decreased. Furthermore, a very important decrease in  values is noted. Such a behavior is commonly observed when pore size is drastically reduced [20], i.e., the laminar flow contribution is negligible. In terms of resistance to gas flow (R), eq. 4 is usually written as F 0 ⫽ 1/R ⫽ 1/R k ⫹ (1/R p)P¯
(5)
where Rk and Rp are the Knudsen and Poiseuille resistances. Knudsen and Poiseuille resistances vs. membrane characteristics (thickness L, tortuosity q, porosity ⑀, mean pore radius r) and gas properties (viscosity , molar weight M) can be expresssed as [21,22]:
Rk ⫽
1 ⫽ ␣
Rp ⫽
3Lq
冉
R gTM 8 2rε
1 8LqR gT ⫽  r 2ε
冊
1/2
(6)
(7)
where Rg and T are gas constant and temperature, respectively. Rearrangement of this set of equations leads to the mean pore radius being expressed as
2024
J. BRASSEUR-TILMANT et al.
16 r⫽
冉 冊 8R gT M 3
or r ⫽ 8.51
1/2
冉 冊 R gT M
Vol. 34, Nos. 12/13
䡠
Rk Rp
(8)
 ␣
(9)
1/2
䡠
These expressions are related to homogeneous membranes with mean pore size r. In the present case, resistances Rk and Rp would have to be written both for the support and modified zones (deposited and infiltrated layers). However, in a recent study on pore modification of alumina substrates using the infiltration method, Wang et al. [12] estimated the pore radius of treated membranes from the /␣ ratio,  and ␣ values being obtained experimentally. Such calculated mean pore sizes were found to be in close agreement with those determined by a gas bubble point method. Similarly, in the present study, eq. 9 has been applied to calculate the mean pore radii reported in Table 1. In the case of run 4, we verified that the calculated value (1.2 nm) was close to that obtained experimentally from rejection ratio measurements (1.5 nm). Such values are averaged over both the deposited layer and infiltrated zone. Their closeness validates the use of eq. 9 and can be tentatively explained by a major contribution of these modified zones to the flow resistance of the final membrane. Eq. 4 and Table 1 data allowed to calculate the relative importance of Poiseuille flow in the membrane permeation: It was about 29% of the total flow for the untreated support and fell below 0.4% after treatment. Thus, only the Knudsen diffusion term can be considered in eq. 5, for the modified samples. We can now tentatively explain the influence of experimental conditions, applied in the substrate treatment, on the membrane characteristics as reported in Table 1. For the lower h values (runs 1 and 2), the measured low permeation can be related to the rather thick infiltrated zone (d ⫽ 50 m), with much narrower pores than in the original alumina substrate. For h ⫽ 30 (runs 4 and 5), the membrane permeation is higher because of the lower thickness of both the infiltrated zone and deposited layer, about 15 and 1 m, respectively. The calculated mean pore radius was significantly smaller than in the previous case, probably due to a higher density of this thin layer related to its different morphology, as seen in SEM micrographs (Fig. 6c). It was shown that, in the present experimental conditions (e.g., h ⬎ 10, T ⱖ 300°C), the dominant mechanism governing particles formation was nucleation rather than crystal growth [14]. In addition, with higher h values (h ⫽ 30), a complete alkoxide hydrolysis occurred, leading to the elimination of all organic groups and, consequently, to a more compact deposit. When h ⫽ 25 (run 3), both reported F0 and r values are between those found in the above two cases. CONCLUSIONS Supported membranes have been elaborated from macroporous alumina substrates treated with supercritical solutions of titanium isopropoxide in propan-2-ol. This precursor was hydrolyzed by water produced from alcohol dehydration inside the reactor, leading to the formation of titania in the anatase structure. Under experimental conditions, TiO2 particles
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2025
were deposited both as a thin layer on the substrate surface and, near this surface, inside the macropores. Membranes obtained starting with a low precursor concentration (CTi ⫽ 0.0125 mol䡠L⫺1), a high water-to-alkoxide ratio (h ⫽ 30), liquid flow rate f ⫽ 5 cm3䡠min⫺1, and deposition time t ⫽ 10 min exhibit the best characteristics. The thickness of the deposited layer and infiltrated zone is about 1 and 15 m, respectively, so that N2 permeation was only reduced by a factor less than 2 after membrane treatment (F0 ⫽ 9.5 ⫻ 10⫺6 mol䡠m⫺2䡠s⫺1䡠Pa⫺1). The mean pore radius in the external layers was then about 1.5 nm, i.e., reduced by a factor of 60. REFERENCES 1. A. Julbe, C. Guizard, A. Larbot, L. Cot, and A. Giroir-Fendler, J. Membrane Sci. 77, 137 (1993). 2. C.J. Brinker, R. Sehgal, S.L. Hietala, R. Deshpande, D.M. Smith, D. Loyd, and C. S. Ashley, J. Membrane Sci. 94, 85 (1994). 3. K.N. Kumar, K. Keizer, and A.J. Burggraaf, J. Mater. Chem. 3, 1141 (1993). 4. J. Etienne, A. Larbot, C. Guizard, and L. Cot, J. Membrane Sci. 86, 95 (1994). 5. Y.S. Lin, C.H. Chang, and R. Gopalan, Ind. Eng. Chem. Res. 33, 860 (1994). 6. R.A. Peterson, E.T. Webster, G.M. Niezyniecki, M.A. Anderson, and C.G. Hill, Sep. Sci. Technol. 30, 1689 (1995). 7. K.N. Kumar, K. Keizer, A.J. Burggraaf, T. Okubo, and H. Nagamoto, J. Mater. Chem. 3, 923 (1993). 8. M. Tsapatsis, S. Kim, S. Nam, and G. Gavalas, Ind. Eng. Chem. Res. 30, 2152 (1991). 9. M. Nomura, T. Yamaguchi, and S.I. Nakao, Ind. Eng. Chem. Res. 36, 4217 (1997). 10. G.Z. Cao, H. Brinkman, J. Meijerink, K. Vries, and A.J. Burggraaf, J. Mater. Chem. 3, 1307 (1993). 11. Y.S. Lin and A.J Burggraaf, AIChE J. 38, 445 (1992). 12. Z. Wang, J. Dong, N. Xy, and J. Shi, AIChE J. 43, 2359 (1997). 13. M. Barj, J.F. Bocquet, K. Chhor, and C. Pommier, J. Mater. Sci. 27, 2187 (1992). 14. K. Chhor, J.F. Bocquet, and C. Pommier, Mater. Chem. Phys. 32, 249 (1992). 15. J.F. Bocquet, K. Chhor, and C. Pommier, Surf. Coat. Tech. 70, 73 (1994). 16. V.G. Courtecuisse, K. Chhor, J.F. Bocquet, and C. Pommier, Ind. Eng. Chem. Res. 35, 2539 (1996). 17. H. Kno¨zinger and R. Ko¨hne, J. Catal. 5, 264 (1966). 18. D.C. Hague and M.J. Mayo, J. Am. Chem. Soc. 77, 1957 (1994). 19. D.C. Bradley, R.C. Mehrotra, and D.C. Gaur, Metal Alkoxide, Academic Press, San Diego (1990). 20. R.J.R. Uhlhorn, M.H.B. J. Huis in’t Veld, and K. Keizer, J. Mater. Sci. 27, 527 (1992). 21. R.S.A. de Lange, J.H.A. Hekkink, K. Keizer, and A.J. Burggraaf, Microporous Mater. 4, 169 (1995). 22. P.C. Carman, Flow of Gases Through Porous Media, Butterworths, London (1956). 23. A.F. Leenaars and A.J. Burgraaf, J. Membrane Sci. 24, 245 (1985). 24. Y.S. Lin and A. J. Burggraaf, J. Membrane Sci. 79, 65 (1993). 25. P. Uchytil, J. Membrane Sci. 97, 139 (1994).
PII S0025-5408(99)00225-1
CERAMIC MEMBRANE ELABORATION USING SUPERCRITICAL FLUID
J. Brasseur-Tilmant, K. Chhor, P. Jestin, and C. Pommier* Laboratoire d’Inge´nie´rie des Mate´riaux et des Hautes Pressions, CNRS LPR 1311, Institut Galile´e, Universite´ Paris Nord, Avenue Jean-Baptiste Cle´ment, 93430 Villetaneuse, France (Refereed) (Received September 8, 1998; Accepted February 15, 1999)
ABSTRACT Macroporous alumina supports were prepared from commercial ␦-Al2O3 and modified by hydrolytic decomposition of titanium isopropoxide in supercritical propan-2-ol. Various experimental parameters were studied in relation to the deposition of TiO2 particles on the substrate surface and inside the pores: temperature, precursor concentration, water-to-alkoxide ratio, solution flow rate, and deposition time. A set of these parameters was found to allow supported membranes with high nitrogen permeation (9.5 ⫻ 10⫺6 mol䡠m⫺2䡠 s⫺1䡠Pa⫺1 ) and low average pore radius (1.5 nm) to be obtained. © 2000 Elsevier Science Ltd
KEYWORDS: A. ceramics, A. thin films, B. sol-gel chemistry, C. electron microscopy INTRODUCTION The application of porous inorganic membranes in filtration and catalytic reactions has attracted much attention because of their high thermal, chemical, and mechanical stability. Supported membranes exhibit very low pore size (in the nanometer range) and high selectivity while keeping high permeability. They are obtained from a macroporous ceramic support treated in such a way that nanoparticles can be deposited either as a thin film on the surface or inside the pores in order to decrease their diameter. In the first case, sol-gel
*To whom correspondence should be addressed. Fax: ⫹33-1-49-40-34-14. E-mail: pommier@limhp. univ-paris13.fr. 2013
2014
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
techniques, either colloidal or polymeric, have been used widely [1–7]. In such processes, a sol must first be prepared under controlled conditions. After coating, drying has to be performed very slowly at room temperature before calcination, usually above 400°C, in order to obtain crystallized oxide. Numerous studies have been conducted by various laboratories to get a better understanding of the chemical and physical phenomena involved during the sol transformation and to develop specific strategies leading to controlled pore sizes for flaw-free supported ceramic membranes [2]. First, a variety of deposition techniques, in addition to the initial slip casting method, such as deep coating, filtration deposition, and permformation, has been proposed [6]. Second, many metal oxides used as membrane materials (alumina, titania, zirconia) exhibit a structural phase transformation between 450 and 900°C, which favors crystallite growth and sintering [5]. In order to obtain crack-free membranes and to improve their thermal stability, many precautions must be taken during the drying and calcination steps. Phase transformation must be retarded at higher temperature, for example, by using a doping oxide [5] or avoided by starting with a suitable sol (rutile sol in the case of TiO2 membrane) [7]. Such a route to modify supports to obtain membranes for ultra- or nano-filtration is very time-consuming, but a pore diameter in the deposit can be lowered to 3–5 nm. However, to prevent crack formation, the optimum thickness of the layer obtained under each coating treatment is claimed to be less than 0.2 m [4]. Another route to reduce pore size is the chemical vapor deposition (CVD) process. Reaction between two reactive gas flows usually leads to the formation of already crystallized metal oxide particles. In the “classical” method, streams are supplied on the same side of the support: reaction occurs at the corresponding surfaces and substrate pores are covered or plugged. In a “modified” CVD method, reactant flows counterdiffuse from both sides of the membrane and react when they contact each other on the inner surface of the pores. Comparing the two methods with the SiCl4/H2O system, it has been shown that a one-sided membrane has significantly higher selectivity for H2/N2 separation [8]. However, it has been noted that, in the counterdiffusion technique, silica formation in zeolite materials can be controlled by the molecule size of the reactant [9]. This last process has been used for TiO2 and ZrO2 deposition either inside the pores of the substrate (pore diameter ⬎ 100 nm) or in the supported top layers of the membrane with smaller pore size (⬍20 nm) and thickness (about 5 m) [10,11]. Reaction between the two gas flows (metal chloride in argon and water in air) was conducted between 700 and 1000°C during a relatively short time (about 20 min). Metal oxide deposition was found at the substrate surface exposed to the metal chloride vapor, and a kinetic study showed that it proceeded according to a heterogeneous reaction [10]. Pore reduction observed from gas permeation measurements was shown to strongly depend on the initial pore size distribution, but no final diameter values were reported [11]. Recently, a new method to modify the pore size distribution of ceramic membranes, using a supercritical fluid solution of a metal oxide precursor (such as metal alkoxide) infiltrated through the macropores was proposed [12]. The infiltrated samples are further immersed in water vapor for complete hydrolysis and finally dried and treated at 450°C in order to obtain the final products. Propane used as solvent is saturated with aluminum isopropoxide at 130°C under 30 MPa and infiltrated into ␣-Al2O3 membranes under the same pressure and temperature conditions. During the infiltration, the pressure of the supercritical solution falls off along the flow direction, and solute supersaturation is reached inside the pores. It has been shown that either homogeneous or heterogeneous nucleation can then occur. The first mechanism is observed when the surface energy of alumina support is lowered before
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2015
treatment by a first coating with methyl cellulose, which makes the surface more wettable. Without such a first coating, the porosity and the total area of open pores also decrease. However, heterogeneous nucleation leads to an excessive reduction of smaller pores and consequently to an increase in average size. Supercritical fluids effectively exhibit physical characteristics that may be of great potential interest in the processing of ceramic materials: low viscosity, high diffusivity, and high solvent power with strong dependence on pressure. In the last decade, various applications have been described for the use of submicronic powder or thin film synthesis. In our research group, we have developed new processes that start with a metal alkoxide as precursor and a supercritical alcohol as solvent. For example, we have reported that reaction in such a medium leads to MgAl2O4 or TiO2 powder crystallization, and that particles with diameters ranging from 20 to 100 nm are obtained [13,14]. Another advantage of such a process is the formation of TiO2 films on alumina substrate at a high deposition rate [15]. It has been shown that under experimental conditions (T ⫽ 300°C, P ⫽ 10 MPa), the precursor is hydrolyzed by water produced in the dehydration of propan-2-ol used as solvent [16]. In the present work, we studied the opportunity to prepare supported ceramic membranes by applying such a reaction to the formation of a thin microporous layer at the surface of a macroporous support. EXPERIMENTAL Support Preparation. Porous alumina discs (25 mm diameter, 1.6 mm thickness) were prepared from ␦-Al2O3 (Alumina C from Degussa, particle size 15 nm; ⬎99% purity). About 1500 ppm magnesium oxide (Fluka, ⬎98% purity) and 5% polyvinyl alcohol (M ⫽ 22000 g䡠mol⫺1) as organic binder were added, in order to favor sintering and to retain high porosity, respectively. The mixture was compacted under 30 MPa and heated up to 1400°C, with intermediate stages at 600 and 1100°C. The obtained discs had a porosity of about 60%, pore size of about 200 nm, and were crystallized under an ␣-alumina structure. Supported Membrane Preparation. Titania particles were deposited on the support, hydrolyzing titanium tetraisopropoxide (TTIP), Ti(O-iC3H7)4 (Aldrich, 98% purity) in supercritical propan-2-ol (Tc ⫽ 235°C, Pc ⫽ 4.76 MPa). Experimental device. The whole system used is schematized in Figure 1. A high-pressure pump (HPP) sent liquid alcohol or alkoxide solution flow inside the reactor (R) at temperature T (T ⬎ Tc). A regulating valve (Vr) allowed the pressure to be maintained at about 10 MPa inside the system. The solvent was cooled down and condensed. At the end of the experiment, an inert gas (N2) can be sent through the system, to evacuate alcohol and residual by-products. A closer view of the reactor is shown in Figure 2. The substrate to be treated was fastened between two PTFE flat rings, to prevent fluid leaks between upstream and downstream zones (A and B, Fig. 2). This system appeared efficient up to about 350°C. Glass wool allowed a temperature gradient to set up between the two sides of the substrate and acted as a catalyst for alcohol dehydration; water was produced in zone B of the reactor. Experimental procedure. The successive steps involved in the TiO2 deposition process are schematized in Figure 3. The closed reactor, containing about 100 cm3 of pure propan-2-ol (higher than the critical volume), was first heated up to temperature TB for time t1 (40 to 100
2016
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
FIG. 1 Experimental device for dynamic treatment of ceramic supports (R ⫽ reactor; HPP ⫽ high pressure pump; V1 to V7 ⫽ valves; Vr ⫽ regulating valve). min). During this step, water was formed, the amount depending on TB and t1. (This point is discussed in the next section.) Then, a pure, cold solvent flow was sent through the system for time (t2 ⫺ t1), i.e., 5 to 10 min, to evacuate water from zone A. During this step, temperature TA decreased and reached an equilibrium value of about 60 to 80°C lower than TB, depending on the introduced alcohol flow rate. At the same time, some of the water present in zone B was evacuated; thus, its concentration decreased. At t2, the pure alcohol flow was replaced by TTIP solution at the same flow rate, and a reaction took place for time (t3 ⫺ t2), usually 10 to 15 min. During the last step, pure solvent flow was sent through the system, to evacuate unreacted TTIP before decompressing (t4). Finally, inert gas (N2) was used to remove any remaining by-products. At time t1, the pure alcohol flow was sent through the system until thermal equilibrium was reached (t2). The time duration (t2 ⫺ t1) was higher than that necessary for complete water
FIG. 2 Closer view of the reactor, showing substrate fastening and glass wool location.
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2017
FIG. 3 Successive steps in the substrate treatment process: (a) reactor thermal equilibrium setting, (b) ROH flow, (c) TTIP solution flow, (d) ROH flow, and (e) decompressing, N2 flow.
evacuation from zone A, as estimated from a plug flow model. However, under experimental conditions, water concentration in zone B did not reach a constant value (even without hydrolysis reaction), because the amount formed in the alcohol dehydration was lower than that evacuated by the supercritical flow. A decrease in water concentration between t2 and t3 is evidenced from gas chromatographic analysis (⬍25%). In the following section (h calculation, Table 1), mean values are used. Characterization Methods. On all modified membranes, the crystal structure of the deposited layer was studied by X-ray diffraction (Philips PW1729 diffractometer) and Raman spectroscopy (XY Dilor). Thickness, particle size, and shape in both the deposited layer and
TABLE 1 Characteristics of Alumina Substrate and Some Modified Membranes Obtained Under Various Experimental Conditions (TB ⫽ 330°C, f ⫽ 5 cm3䡠min⫺1) Run 0
Run 1
Run 2
Run 3
Run 4
Run 5
CTi (mol䡠L⫺1) h
— —
0.025 15
0.025 15
0.025 25
0.0125 30
0.0125 30
e (m) d (m) F0
— — 18
3 50 3.5
— — 3.7
4 25 8.3
1 15 9.3
— — 9.5
␣  /␣ r (nm)
12.8 347 267 114
3.46 5.38 15.5 6.5
3.66 5.77 15.8 6.5
8.23 9.21 11.2 4.5
9.17 2.58 2.81 1.3
9.46 2.71 2.86 1.2
⫺2
␣ ⫽ coefficient related to Knudsen contribution (10⫺6 mol䡠m 䡠s⫺1䡠Pa⫺1).  ⫽ coefficient related to Poiseuille contribution (10⫺13 mol䡠m⫺2䡠s⫺1䡠Pa⫺2). F0 ⫽ gas permeation (10⫺6 mol䡠m⫺2䡠s⫺1䡠Pa⫺2l) obtained from N2 permeation.
2018
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
the infiltrated zone were investigated by scanning electron microscopy (Leica S440). For some samples, properties related to their porous structure were studied in more detail. Gas permeation measurements were performed on samples before and after deposition treatment, using a laboratory-made permeation apparatus at room temperature and helium, nitrogen, or carbon dioxide as the permeating gas. Gas flow at constant flow rate was passed through the membrane. Higher upstream pressure Ph and lower downstream pressure Pl were measured with two precision gauges. The applied mean pressure, P¯ ⫽ 1/2 (Ph ⫹ Pl), was from 0.1 to 0.2 MPa. The transmembrane pressure ranged from 100 to 200 kPa, and the flux of permeating gas was measured with a soap bubble flowmeter. Similarly, water permeability and rejection properties were studied. These measurements were performed using polyethyleneglycol (PEG) solutions (Aldrich-Chimie, Mw ⫽ 400, 600, 1000, and 1500 g䡠mol⫺1). Their concentrations in feed solution and in permeate were obtained from size exclusion chromatographic analyses (column, Shodex SB-802 HQ; detector, Gilson 132 RI). RESULTS AND DISCUSSION It is well known that alcohols can be dehydrated over metal oxide based catalysts such as alumina and that the facility of the reaction increases from primary to tertiary derivatives [17]. We have observed that such a dehydration occurred with propan-2-ol in our experimental device (described above), at temperatures higher than 300°C. At 330°C, a rather high water concentration (⬇0.6 mol䡠L⫺1) could be reached after 1 h under static conditions. Above 300°C, with an excess amount of water, the TTIP hydrolysis reaction was complete, and the resulting titanium tetrahydroxide was transformed into TiO2. The overall reaction can be written Ti(OR) 4 ⫹ 2H 2O 3 TiO 2 ⫹ 4ROH
(1)
The ratio between water and alkoxide mole numbers, defined as h ⫽ nH2O/nTi ⫽ cH2O/2Ti, is equal to 2 under stoichiometric conditions. However, it has been shown that complete elimination of organic groups only occurs for h ⬎ 2 [18,19]. For lower h values, the oxo-alkoxide compounds are formed according to the following reaction: Ti(OR) 4 ⫹ xH 2O 3 Ti(OR) 4 –x(OH) x ⫹ xROH
(2)
Supported Membranes Synthesis. In a preliminary study, the objective was to determine experimental conditions allowing the formation of water in an amount sufficient to lead to a complete reaction of the introduced alkoxide precursor, and the deposition of the formed titanium dioxide located only on one side of the porous support. When TB was regulated at 330 or 350°C, temperature gradients between the two opposite faces of the substrate were about 80 and 60°C for liquid flow rates of 5 and 3 cm3䡠min⫺1, respectively. With TB ⫽ 350°C, t1 ⫽ 50 min, and f ⫽ 5 cm3䡠min⫺1, the water concentration in zone B increased up to [H2O]B max ⫽ 0.6 mol䡠L⫺1 and its mean value during the deposition step (c, Figure 3) was [H2O]B ave ⫽ 0.4 mol䡠L⫺1. Under another set of experimental conditions (TB ⫽ 330°C, t1 ⫽ 75 min, f ⫽ 5 cm3䡠min⫺1), these values were [H2O]B max ⫽ 0.5 mol䡠L⫺1 and [H2O]B ave ⫽ 0.35 mol䡠L⫺1. In all cases, the quantity of formed water was highly excessive compared to the amount of titanium alkoxide introduced into the system, even for the highest investigated
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2019
values of solution flow rate and concentration, e.g., f ⫽ 5 cm3䡠min⫺1 and CTi ⫽ 0.05 mol䡠 L⫺1. At high titanium alkoxide concentration (CTi ⫽ 0.05 mol䡠L⫺1) and lower solution flow rates (f ⱕ 3 cm3䡠min⫺1), TiO2 particles were formed both on the upper and lower faces of the substrate, as well as inside its pores. Furthermore, the amount of solid deposited in zone A was less influential on increasing flow rate, because increase in the amount deposited induced a temperature decrease inside this area. No more particles were formed on the substrate face A when TA was equal to or less than 275°C. Starting with lower alkoxide concentration (CTi ⱕ 0.025 mol䡠L⫺1), the reaction occurred only on the lower face of the substrate when the flow rate was between 3 and 5 cm3䡠min⫺1 and TB between 330 and 350°C. The localization of the TiO2 deposit on face B of the substrate could be obtained from a conjunction of suitable values for the three parameters: fluid flow rate, temperature, and alkoxide concentration. The temperature gradient between the two faces had to be sufficiently high so that the reaction occurred only on face B. At a low fluid flow rate, water formed in zone B could diffuse more easily through the substrate and react inside its pores. A higher flow rate limited such a phenomenon and increased the temperature gradient. A low TTIP concentration allowed a complete reaction on the substrate face B. Further experiments were conducted with TB ⫽ 330°C and liquid flow rate f ⫽ 5 cm3䡠 min⫺1 for reactive solutions with TTIP concentration CTi ⫽ 0.025 mol䡠L⫺1. Study of the Deposited Layers. In Table 1 are reported some results obtained under various synthesis conditions. Experiments conducted at TB ⫽ 350°C were not taken into account: A rather poor reproducibility was found, related to technical problems due to the behavior of PTFE rings at this temperature. Furthermore, compared to that obtained at 330°C, the homogeneity of the deposits was not improved. Many of the modified membranes exhibited a gray coloring, probably due to the presence of carbon particles formed during alkoxide decomposition. However, these particles were easily eliminated, and perfectly white samples were recovered after thermal air treatment at about 350°C for 2 h. It was shown that such treatments did not modify significantly the microstructure and filtration properties of the deposited layer. As expected from a previous study [14], the solid particles formed under the above conditions were crystallized in the anatase structure. This can be evidenced by the presence of its strong line at 2 ⫽ 25.4° on the X-ray diffraction diagram. As shown in Figures 4a and 4b, the pattern obtained from face A of the treated substrate is that of the original alumina support. Conversely, in the diagram recorded from face B (Fig. 4c), the intensity of the Al2O3 peaks is much lower, and lines associated with the presence of TiO2 anatase deposit appear. Figure 5 compares Raman spectra obtained from both sides of a modified membrane. In the face A spectrum (Fig. 5a), only alumina Raman lines are observed. In the face B spectrum (Fig. 5b), only lines related to the anatase-deposited layer are present: The underlying alumina support was not reached by the incident laser light. Scanning electron micrographs (Fig. 6) show that three zones can be distinguished in the cross section of the treated alumina discs. First, according to our experiments, TiO2 particles were formed on face B as a layer with a thickness e ranging from 1 to 4 m (Figs. 6a and c). Second, particles were observed to have a somewhat lower density inside the alumina support macropores, because of water counterflow diffusion (Fig. 6d). The infiltrated zone depth d ranged from 15 to 50 m according to our experiments. The third zone was an unmodified alumina support, in which only a few isolated nanometric TiO2 particles were observed in some macropores (Fig. 6b).
2020
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
FIG. 4 X-ray diffraction patterns of alumina support (a) and modified membranes face A (b) and face B (c). Starred (夽) lines correspond to TiO2 anatase.)
The precursor concentration and the amount of water inside the reactor mainly influenced the surface morphology and thickness of the deposited layer, as well as the permeation properties of the modified membranes. For h ⫽ 15 and CTi ⫽ 0.025 mol䡠L⫺1 (run 1, Table 1), angular particles below 1 m in size were observed and e ⬇ 3 m (Fig. 6a). For a higher h value (h ⫽ 30) and CTi ⫽ 0.0125 mol䡠L⫺1 (run 4, Table 1), rather spherical agglomerates of particles less than 30 nm in size formed a dense thin layer (e ⬇ 1 m) (Figs. 6c and d). In these two samples, the depth d of the infiltrated zone was about 50 and 15 m, respectively. Results on permeation measurements and rejection properties of the membranes are reported in Figures 7 to 10. Figure 7 compares the nitrogen permeation of substrate and modified membranes. After various applied treatments, permeation was divided by a factor of 2 to 5, according to the amount of titania particles both inside and on the surface of the alumina support. The results obtained with various gases are reported in Figure 8. As expected, permeability decreased with an increase of the gas kinetic radius. For example,
FIG. 5 Raman spectra of modified membrane obtained in run 3: (a) face A and (b) face B.
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2021
FIG. 6 SEM micrographs on modified membranes obtained in run 1: (a) face B, (b) cross section (50 m from surface); and in run 4: (c) face B, (d) cross section. when comparing He and N2 curves, a large difference is observed in their ordinate at origin, which is related to a difference in Knudsen diffusion contribution [20,21]. For treated and untreated membranes, water flux as a function of pressure drop, ranging from 0.1 to 1 MPa, is compared in Figure 9. In the investigated range, a linear variation was observed, as predicted by the Kozony-Carman equation [22,23].
FIG. 7 N2 permeation as function of mean pressure for the substrate (a) and modified membranes obtained in (b) run 1, (c) run 3, and (d) run 4.
2022
J. BRASSEUR-TILMANT et al.
Vol. 34, Nos. 12/13
FIG. 8 He, N2, and CO2 permeation as function of mean pressure for modified membrane obtained in run 4.
In order to estimate the mean pore size of the deposited layer, rejection properties of modified membranes were studied, and the retention ratio for various PEG samples are reported in Figure 10. is defined as ⫽ 1 ⫺ (Cp/Cf), where Cp and Cf are solute concentrations in permeate and feed solutions, respectively. The cut-off value is usually taken for a 90% solute retention. In the present case, the corresponding solute molar weight was estimated to be about 2000, associated with a gyration radius of about 1.5 nm. In the theoretical treatment of gas permeation measurements, simplifying assumptions are usually introduced. Both the deposited layer and support are considered to have cylindrical pores with average pore radii of rl and rs, respectively. Furthermore, the gas flow through the membrane is assumed to be the sum of Knudsen diffusion and Poiseuille laminar flow contributions [24,25]. Then, by comparing results obtained on membranes before and after treatment, the average pore size change under treatment can be checked. Measuring the gas flow rate Q passing through a cross-sectional area A of the membrane, the gas permeation F0 can be calculated according to its definition:
FIG. 9 Dependency of water flux with pressure drop for (a) the substrate and (b) modified membrane obtained in run 4.
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2023
FIG. 10 Retention behavior of PEG polymers on modified membrane obtained in run 4.
F 0 ⫽ Q/( A ⫻ ⌬P)
(3)
where ⌬P ⫽ Ph ⫺ Pl is the pressure drop between upstream (h) and dowstream (l) sides. It has been shown [24] that, even for composite membranes, gas permeation usually follows a linear variation vs. the average applied pressure P¯ ⫽ 1/2 (Ph ⫹ Pl): F 0 ⫽ ␣ ⫹ P¯ F 0
(4)
In this relation, ␣ and  are coefficients related to the Knudsen and Poiseuille contributions, respectively. As noted above, Figures 7 and 8 show a linear variation of permeation F0 vs. the applied mean pressure. Corresponding ␣ and  parameters are reported in Table 1. Under treatment, the membrane permeability decreased as the result of pore narrowing. Accordingly, the ␣ parameter, which is related to the Knudsen diffusion contribution, also decreased. Furthermore, a very important decrease in  values is noted. Such a behavior is commonly observed when pore size is drastically reduced [20], i.e., the laminar flow contribution is negligible. In terms of resistance to gas flow (R), eq. 4 is usually written as F 0 ⫽ 1/R ⫽ 1/R k ⫹ (1/R p)P¯
(5)
where Rk and Rp are the Knudsen and Poiseuille resistances. Knudsen and Poiseuille resistances vs. membrane characteristics (thickness L, tortuosity q, porosity ⑀, mean pore radius r) and gas properties (viscosity , molar weight M) can be expresssed as [21,22]:
Rk ⫽
1 ⫽ ␣
Rp ⫽
3Lq
冉
R gTM 8 2rε
1 8LqR gT ⫽  r 2ε
冊
1/2
(6)
(7)
where Rg and T are gas constant and temperature, respectively. Rearrangement of this set of equations leads to the mean pore radius being expressed as
2024
J. BRASSEUR-TILMANT et al.
16 r⫽
冉 冊 8R gT M 3
or r ⫽ 8.51
1/2
冉 冊 R gT M
Vol. 34, Nos. 12/13
䡠
Rk Rp
(8)
 ␣
(9)
1/2
䡠
These expressions are related to homogeneous membranes with mean pore size r. In the present case, resistances Rk and Rp would have to be written both for the support and modified zones (deposited and infiltrated layers). However, in a recent study on pore modification of alumina substrates using the infiltration method, Wang et al. [12] estimated the pore radius of treated membranes from the /␣ ratio,  and ␣ values being obtained experimentally. Such calculated mean pore sizes were found to be in close agreement with those determined by a gas bubble point method. Similarly, in the present study, eq. 9 has been applied to calculate the mean pore radii reported in Table 1. In the case of run 4, we verified that the calculated value (1.2 nm) was close to that obtained experimentally from rejection ratio measurements (1.5 nm). Such values are averaged over both the deposited layer and infiltrated zone. Their closeness validates the use of eq. 9 and can be tentatively explained by a major contribution of these modified zones to the flow resistance of the final membrane. Eq. 4 and Table 1 data allowed to calculate the relative importance of Poiseuille flow in the membrane permeation: It was about 29% of the total flow for the untreated support and fell below 0.4% after treatment. Thus, only the Knudsen diffusion term can be considered in eq. 5, for the modified samples. We can now tentatively explain the influence of experimental conditions, applied in the substrate treatment, on the membrane characteristics as reported in Table 1. For the lower h values (runs 1 and 2), the measured low permeation can be related to the rather thick infiltrated zone (d ⫽ 50 m), with much narrower pores than in the original alumina substrate. For h ⫽ 30 (runs 4 and 5), the membrane permeation is higher because of the lower thickness of both the infiltrated zone and deposited layer, about 15 and 1 m, respectively. The calculated mean pore radius was significantly smaller than in the previous case, probably due to a higher density of this thin layer related to its different morphology, as seen in SEM micrographs (Fig. 6c). It was shown that, in the present experimental conditions (e.g., h ⬎ 10, T ⱖ 300°C), the dominant mechanism governing particles formation was nucleation rather than crystal growth [14]. In addition, with higher h values (h ⫽ 30), a complete alkoxide hydrolysis occurred, leading to the elimination of all organic groups and, consequently, to a more compact deposit. When h ⫽ 25 (run 3), both reported F0 and r values are between those found in the above two cases. CONCLUSIONS Supported membranes have been elaborated from macroporous alumina substrates treated with supercritical solutions of titanium isopropoxide in propan-2-ol. This precursor was hydrolyzed by water produced from alcohol dehydration inside the reactor, leading to the formation of titania in the anatase structure. Under experimental conditions, TiO2 particles
Vol. 34, Nos. 12/13
CERAMIC MEMBRANE
2025
were deposited both as a thin layer on the substrate surface and, near this surface, inside the macropores. Membranes obtained starting with a low precursor concentration (CTi ⫽ 0.0125 mol䡠L⫺1), a high water-to-alkoxide ratio (h ⫽ 30), liquid flow rate f ⫽ 5 cm3䡠min⫺1, and deposition time t ⫽ 10 min exhibit the best characteristics. The thickness of the deposited layer and infiltrated zone is about 1 and 15 m, respectively, so that N2 permeation was only reduced by a factor less than 2 after membrane treatment (F0 ⫽ 9.5 ⫻ 10⫺6 mol䡠m⫺2䡠s⫺1䡠Pa⫺1). The mean pore radius in the external layers was then about 1.5 nm, i.e., reduced by a factor of 60. REFERENCES 1. A. Julbe, C. Guizard, A. Larbot, L. Cot, and A. Giroir-Fendler, J. Membrane Sci. 77, 137 (1993). 2. C.J. Brinker, R. Sehgal, S.L. Hietala, R. Deshpande, D.M. Smith, D. Loyd, and C. S. Ashley, J. Membrane Sci. 94, 85 (1994). 3. K.N. Kumar, K. Keizer, and A.J. Burggraaf, J. Mater. Chem. 3, 1141 (1993). 4. J. Etienne, A. Larbot, C. Guizard, and L. Cot, J. Membrane Sci. 86, 95 (1994). 5. Y.S. Lin, C.H. Chang, and R. Gopalan, Ind. Eng. Chem. Res. 33, 860 (1994). 6. R.A. Peterson, E.T. Webster, G.M. Niezyniecki, M.A. Anderson, and C.G. Hill, Sep. Sci. Technol. 30, 1689 (1995). 7. K.N. Kumar, K. Keizer, A.J. Burggraaf, T. Okubo, and H. Nagamoto, J. Mater. Chem. 3, 923 (1993). 8. M. Tsapatsis, S. Kim, S. Nam, and G. Gavalas, Ind. Eng. Chem. Res. 30, 2152 (1991). 9. M. Nomura, T. Yamaguchi, and S.I. Nakao, Ind. Eng. Chem. Res. 36, 4217 (1997). 10. G.Z. Cao, H. Brinkman, J. Meijerink, K. Vries, and A.J. Burggraaf, J. Mater. Chem. 3, 1307 (1993). 11. Y.S. Lin and A.J Burggraaf, AIChE J. 38, 445 (1992). 12. Z. Wang, J. Dong, N. Xy, and J. Shi, AIChE J. 43, 2359 (1997). 13. M. Barj, J.F. Bocquet, K. Chhor, and C. Pommier, J. Mater. Sci. 27, 2187 (1992). 14. K. Chhor, J.F. Bocquet, and C. Pommier, Mater. Chem. Phys. 32, 249 (1992). 15. J.F. Bocquet, K. Chhor, and C. Pommier, Surf. Coat. Tech. 70, 73 (1994). 16. V.G. Courtecuisse, K. Chhor, J.F. Bocquet, and C. Pommier, Ind. Eng. Chem. Res. 35, 2539 (1996). 17. H. Kno¨zinger and R. Ko¨hne, J. Catal. 5, 264 (1966). 18. D.C. Hague and M.J. Mayo, J. Am. Chem. Soc. 77, 1957 (1994). 19. D.C. Bradley, R.C. Mehrotra, and D.C. Gaur, Metal Alkoxide, Academic Press, San Diego (1990). 20. R.J.R. Uhlhorn, M.H.B. J. Huis in’t Veld, and K. Keizer, J. Mater. Sci. 27, 527 (1992). 21. R.S.A. de Lange, J.H.A. Hekkink, K. Keizer, and A.J. Burggraaf, Microporous Mater. 4, 169 (1995). 22. P.C. Carman, Flow of Gases Through Porous Media, Butterworths, London (1956). 23. A.F. Leenaars and A.J. Burgraaf, J. Membrane Sci. 24, 245 (1985). 24. Y.S. Lin and A. J. Burggraaf, J. Membrane Sci. 79, 65 (1993). 25. P. Uchytil, J. Membrane Sci. 97, 139 (1994).