Alumina and aluminate ultrafiltration membranes derived from alumina nanoparticles

Journal of Membrane Science 224 (2003) 11–28

Alumina and aluminate ultrafiltration membranes derived from alumina nanoparticles Kimberly A. DeFriend a,c , Mark R. Wiesner b,c , Andrew R. Barron a,c,∗ a

Department of Chemistry, Rice University, MS-60, 6100 Main Street, Houston, TX 77005, USA Department of Civil and Environmental Engineering, Rice University, Houston, TX 77005, USA Center for Biological and Environmental Nanotechnology, Rice University, Houston, TX 77005, USA b

c

Received 20 December 2002; received in revised form 2 June 2003; accepted 16 June 2003

Abstract The fabrication of alumina ultrafiltration membranes using acetic acid surface stabilized alumina nanoparticles (A-alumoxanes) has been investigated. The pore size, pore size distribution, and molecular weight cut-off (MWCO) parameters of the resulting membranes are highly dependant on the uniformity of the nanoparticle precursor, which is a function of the reaction time and reaction pH during their synthesis. By the control over the alumina nanoparticles, a significant improvement of the membrane performance is observed over our prior results. The new alumoxane-derived membranes have a molecular weight cut-off in the range of 15

13

growth and NaCl (0.05 M) was added to control the ionic strength. Ultrapure water was used in all the experiments (10 M). Samples were prepared using the same procedure for the permeability measurements and placed in a stirred ultrafiltration cell (Amicon, Model 8200). The cell was modified to allow for recirculation of the feed. A hole was drilled in the top of the cell to insert a 3.125 mm (1/8 in.) stainless steel tube, used as inlet for the feed. The existing fitting was used as exit of flow. The feed was pumped at a rate of 100–110 ml min−1 . A valve and pressure gage at the flow exit of the cell allowed for transmembrane pressure control. The pressure in the regulator was set at 48 ± 3 × 103 Pa (7.0 ± 0.5 psi). The permeate samples were collected after allowing the system to run for 30 min. The velocity of the flow across the membrane was kept below 0.15 ml min−1 to avoid deformation effects of the macromolecules with a peristaltic pump at the permeate line. A sample of the feed was taken at the end of each filtration. Feed and permeate samples were analyzed by gel permeation chromatography (GPC). A HPLC system (Waters 717+ Autosampler, Waters 600E System Controller) was used with a GPC column (TosoHaas G4000PWXL). Calibration curves (MW versus elution time) were obtained running each dextran fraction separately. The peak was assumed to correspond to the average molecular weight given by the manufacturer. Elute was collected at every ml with a fraction collector (Waters). The concentration of solute in each fraction was determined measuring organic carbon concentration, using a Total Organic Carbon Analyzer (Shimatzu, TOC 5050A). Three HPLC runs were performed with each sample and each fraction was analyzed three times by the TOC analyzer. The MWCO measurements can be correlated to the average pore size for retention of coefficients greater than 80% Eq. (1) [9]. Determining the %R of the solute of a known molecular weight, the solute radius is known, rsolute , then the average pore radius, rpore , may be approximated as: rsolute %R = × 100 (1) rpore In order to calculate the true radius of dextran, an average needs to be taken between the radius of gyration of a real chain, ar , and a geometric chain, ag Eqs. (2) and (3), respectively [9]. Where ar and ag are expressed

14

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

in Angstroms and M is the molecular weight of the dextran: ar = (0.096)M 0.59

(2)

ag = (0.128)M 0.5

(3)

2.5. Retention studies by pressurized filtration The ␣-alumina membrane/support was placed in a Nalgene filtration cell. About 50 ml of the desired solution was poured on top of the mounted support. The cell was closed, and a pressure of 5.5 × 104 Pa (8 psi) was applied to the top of the cell. Aliquots of the permeate were collected and pipetted into a UV cuvette for analysis. The concentration of the initial feed was compared to the concentration of the permeate over time, by UV analysis, to determine the percentage of retention. This methodology was used for all solution studies presented herein. 2.6. Surface modification of α-alumina membranes The ␣-alumina support and ␣-alumina support/ membrane was placed in 20 ml of hexanoic acid or p-hydroxybenzoic acid solution. The solution was gently refluxed for 24 h. After cooling the membrane was removed and allowed to dry thoroughly before use.

3. Results and discussion 3.1. Membrane formation Fabrication of a membrane is achieved by bringing the surface of the support into contact with a solution of alumoxane. The solution is drawn into the surface pores of the support by capillary forces. The membrane deposited onto the surface of the support should be uniformly thin throughout in order to maximize the flux, which is important when the pore size of the membrane is significantly smaller than the pore size of the support [10–13]. Our initial studies used commercially available ␣-alumina (Refractron Technologies Corp.) as supports for casting the alumoxane-derived membranes. The surface area (4.1 m2 g−1 ) and average pore size

(94 nm) of these supports were quite low and the pore size distribution was very broad; the maximum pore size was over 180 nm. Subsequent batches of the Refractron supports had a smaller average pore size (80 nm) and a lower surface area (3.5 m2 g−1 ). These improvements in the commercial supports result in a higher resistance of the support. The membrane thickness is controlled by the concentration of the A-alumoxane precursor. A concentration of 1 wt.% A-alumoxane, produces the thinnest homogeneous coverage (Fig. 2a), with a thickness of ca. 1 ␮m (Fig. 3). Using higher concentrations (i.e. 6 wt.% A-alumoxane) increased the membrane thickness, but also increases the degree of surface cracking (Fig. 2b). If the alumoxane solution is too dilute (i.e. 0.75 wt.% A-alumoxane) incomplete coverage of the support is seen (Fig. 2c). Controlling and limiting the amount of surface defects is crucial when forming a membrane that does not leak or foul. These results mimic our previous studies and indicated that the choice of a 1 wt.% A-alumoxane solution provides a uniform defect free membrane. Allowing the dip-coated membrane to dry completely before sintering, will also reduce the extent of cracking by prohibiting premature drying. Although A-alumoxane ceramic yield is high (75%) some shrinkage must occur during sintering. The shrinkage may cause stress cracks in the membranes. Samples of the membranes were sintered to final temperatures of 600 and 1000 ◦ C. Calcining the Aalumoxane membrane to 600 ◦ C over 6 h with a dwell time of 5 h achieved a smooth membrane (Fig. 2a) with an average pore size of 11 nm and narrow pore size distribution (±4 nm). AFM data are consistent with the N2 adsorption measurements. The surface roughness, calculated from the AFM Nanoscope software, was determined to be 30 nm. If the final sinter temperature is raised to 1000 ◦ C, grain growth (48 nm from 42 nm for 600 ◦ C) and phase transformation (␣-Al2 O3 from ␥- and ␦-Al2 O3 at 600 ◦ C) occurs resulting in an increased pore size (20 nm) and a broader pore size distribution (±8 nm). Calcining to 1000 ◦ C also increases the surface roughness (36 nm). These results suggest that the calcination/sinter process per se is not responsible for the bleed-through of our original test membranes that were sintered to 600 ◦ C. In our previous work we showed that the size of the membrane’s pores are related to the size of the

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

15

Fig. 3. SEM images of cross-sections of an A-alumoxane (1 wt.%) derived alumina membrane (top layer) cast on an ␣-alumina support (lower layer) after firing to 600 ◦ C.

Fig. 2. SEM images of the surface of alumina membranes formed by dip-coating a support with (a) 1 wt.% A-alumoxne, (b) 6 wt.% A-alumoxane, and (c) 0.75 wt.% A-alumoxane, and firing to 600 ◦ C.

alumoxane nanoparticles [1]; if the particle size distribution is broad, then the pore size distribution would be expected to be large. The pore size analysis and PCS of previously synthesized A-alumoxane showed a broad range of pores, 7–25 nm and particle sizes between 30 and 40 nm. Closer inspection of the particle sizes present in the

samples of A-alumoxane used for the test membranes (“original” A-alumoxane) showed the presence of a significant number of particles over 300 nm in addition to the expected 20 nm particles. These were determined to consist mainly of unreacted boehmite. Sintering this sample of A-alumoxane resulted in a bi-modal distribution of pores. Thus, in addition to the 12 nm pores previously observed, the membranes contained a number of pores over 20 nm in diameter. The presence of these larger pores is consistent with the poor rejection of viruses observed for the test membranes. These results indicate that the pore size and distribution is highly dependent on the size and uniformity of the nanoparticle precursors. This is consistent with experience with sol-gel derived membranes that are frequently formed from precipitates with a broad size range and ordinarily produce a broad pore size distribution, requiring multiple dip/fire sequences to assure membrane integrity. The average particle size of boehmite starting material is nearly two orders of magnitude larger than the A-alumoxane particles (3000 nm versus 30 nm). Thus, the presence of unreacted boehmite can alter the resulting pore size formation and increase the variance of the pore size distribution. Removal of unreacted boehmite from the A-alumoxane solution by centrifugation reduced the average particle size. The mean particle size before and after centrifuging was 120 and 18 nm, respectively. Subsequent calcination and sintering to 600 ◦ C yielded alumina with uniform 12 nm average pore size and no large pores (Fig. 4).

16

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

25

% volume

20

15

10

5

0 0

10

20

40 30 50 60 Average pore diameter (nm)

70

80

Fig. 4. The effect on pore size distribution of ceramic made from alumoxanes with narrow or broad particle size distribution.

While centrifugation is a possible purification route, we investigated alternative methods to ensure the removal of any unreacted boehmite. The pH of the reaction solution for A-alumoxane prepared by the previously reported method [8] was measured to be 4.5. At this pH, the average particle size was determined to be 37 nm. Decreasing the reaction pH by increasing the concentration of acetic acid is expected to produce smaller particles (see Fig. 5). As an alternative to using an aqueous solution, the reaction of

boehmite was carried out in glacial acetic acid (see Section 2). In addition to ensuring complete conversion of boehmite to A-alumoxane, this methodology results in smaller average particle size (18 nm) and a narrower particle size distribution (±7 nm) than that obtained from the aqueous synthesis (37, ±15 nm) [8]. We have previously reported a correlation between reaction time and particle size for a range of alumoxanes [8]. In the present case, ceasing reaction after 30–45 min was determined to be the optimal reaction

50 45

Particle size (nm)

40 35 30 25 20 15 10 5 0 1

1.5

2

2.5

3 pH

3.5

4

4.5

5

Fig. 5. Particle size (nm) dependence of synthesized A-alumoxane as a function of solution pH.

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

17

45

Particle size (nm)

40 35 30 25 20 15 10 5 0 0

50

300 150 200 250 Reaction time (min)

100

350

400

Fig. 6. Average particle size (nm) for A-alumoxane as a function of reaction time (min).

time for producing the smallest particle sizes (Fig. 6). If the reaction is halted prematurely, the conversion of boehmite to A-alumoxane is not complete resulting in a wide range of particle sizes. However, excessive reaction times yield agglomerates that do not remain in solution. Table 2 summarizes the particle size, pore volume, and specific surface area for the two different reactions for A-alumoxane. The specific surface area of the membrane is the ratio of the membrane pore surface area to the volume of the pores. A high specific surface area may be a desirable characteristic of a reactive membrane or catalyst support. 3.2. Performance characteristics of the membrane Given the improved synthesis of small, highly uniform, A-alumoxane particles and the subsequent formation of alumina membranes with small pore size and uniform pore size distribution, we have prepared

a “second generation” ultrafiltration membrane using the new A-alumoxane. The fabrication of these membranes is described in Section 2. Permeate flux, permeability and molecular weight cut-off measurements were performed and a comparison was made with our previously reported alumoxane-derived membranes (Table 3). The permeability of the commercial ␣-alumina support was found to be 36.7 (±0.5) nm2 . Deposition of an alumina membrane on this support, using A-alumoxane with a broad particle size, results in a decrease in the overall permeability to 21 (±1) nm2 . The smaller more uniform particle size A-alumoxane produced a nearly identical permeability, 22 (±4) nm2 . The rejection characteristics of the membrane were investigated using molecular weight cut-off measurements [14–18]. The rejection of a membrane can range from 100% (complete retention) to 0% (no retention). An MWCO of a filter is defined as the molecular weight for which a given percent of the molecules in

Table 2 Comparison of physical properties of the two syntheses of A-alumoxane Synthesis

Particle size distribution (nm)

Surface areaa (m2 g−1 )

Pore volumea (ml g−1 )

Specific surface area (m2 ml−1 )

Originalb Newc

30–40 7–25

115 111

0.34 0.32

338 348

Value for alumina body formed by thermolysis at 600 ◦ C. Reaction of boehmite and acetic acid at pH 4.5 in aqueous solution. c Reaction of boehmite in acetic acid. a

b

18

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

Table 3 Comparison of the performance and rejection characteristics of the support and the alumoxane-derived membranes Performance measurements

␣-Alumina support

Original A-alumoxane membranesa

New A-alumoxane membranesb

Particle size (nm) Surface area (m2 g−1 ) Average pore size (nm) Surface roughness (nm) Permeate flux (10−6 m s−1 )c Permeability (nm2 )c MWCO (g mol−1 ) MWCO pore size (nm)

100–500 3.85 50 67 1.42 (2) 36.7 (5) >500000 31

30–40 115 12 16 0.81 (5) 21 (2) 40000–200000 9–18

7–25 111 7 9 0.85 (5) 22 (4) >1000 >4

a

Alumina membrane derived from A-alumoxane prepared by the reaction of boehmite and acetic acid at pH 4.5 in aqueous solution. Alumina membrane derived from A-alumoxane prepared by the reaction of boehmite in acetic acid. c Errors given in parenthesis. b

the feed are rejected. The rejection coefficient (%R) is defined by Eq. (4), where cp is the concentration of the permeate, and cf is the concentration of the feed [14–18]. In this work we adopt a criterion of 80% rejection for the purpose of defining MWCO:
 cp %R = 1 − × 100 (4) cf Different molecules can be used to determine the MWCO [19], however, dextrans are commonly chosen because they are readily available in a variety of molecular weights, are inexpensive, and they are highly branched and rigid. Because the dextran compounds studied are not completely spherical, there is some error associated with the interpretation of MWCO data in terms of pore size. The average molecular weight and approximate diameters of the dextrans used as the feed components are shown in Table 1. It was determined that the Refractron supports allowed for all of the dextran samples studied to pass through, meaning that the MWCO for the supports is >500,000 g mol−1 . This corresponds to a pore size greater than 40 nm, and is in agreement with the average pore size of 80 nm determined by BET measurements. Since the support fails to retain dextrans, all dextran retention can be attributed to the A-alumoxane-derived membrane. MWCO measurements of a membrane synthesized using boehmite contaminated A-alumoxane (37 nm mean particle size and 16 nm pore size) were compared to membrane synthesized with “refined A-alumoxane particles” described above (18 nm mean

particle size and 7 nm pore size). The original test membrane with the larger pore size gave a cut-off between 40,000 and 200,000 g mol−1 , at 80% rejection. Using Eqs. (2) and (3), this corresponds to a pore diameter between 8 and 23 nm. The MWCO of the filter prepared using the refined nanoparticles gave an 80% rejection of molecular weights between 9000 and 10,000 g mol−1 , corresponding to a pore diameter of ca. 4 nm. The lowest molecular weight available for a dextran (T-10) has a nominal value of 10,500 g mol−1 . MALDI mass spectroscopy was performed on T-10 dextran solution and showed intensity peaks at 1000 g mol−1 and lower. This indicates that T-10 actually contains a broad range of dextrans with MW from 1000 to 10,000 g mol−1 . Therefore, the lower bound on the MWCO determined by this method can only be estimated as falling within a range of 1000 and 10,000 g mol−1 for these membranes. There is a remarkable difference in the performance and rejection characteristics of a membrane derived from different methods of A-alumoxane syntheses. This result indicates that the range of the initial alumoxane particle sizes has a direct effect on the membrane’s pore size and the pore size distribution (Table 2), underscoring the importance of controlling the particle size of the ceramic precursor when trying to obtain a high performance filter. 3.3. Separation of dye molecules As noted above, MWCO calculated using dextrans can only be estimated as being between 1000 and

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

(a)

N N

NaO3S

N N

19

SO3Na

HO

O HN C

(b)

SO3Na

O H2N

C HN

(c)

SO3Na

O HN C

CH CH

NH2 NaO3S N N

O HN C

SO3Na H2N O NH C NH

N N

OH

HO

NaO3S (d)

SO3Na

SO3Na N N

O NH C NH

H3C

SO3Na N N CH3

SO3Na (e)

NH2

SO3Na OH

SO3Na N N

N N

N N NaO3S

SO3Na

NH2

SO3Na

Fig. 7. Molecular representation of (a) Direct Red 81, (b) Direct Yellow 62, (c) Direct Red 75, (d) Direct Yellow 50, and (e) Direct Blue 71.

10,000 g mol−1 , corresponding to a molecular diameter of less than 4 nm. For a more precise measure of membrane performance an alternative reference must be used. A membrane with a pore size of ca. 4 nm is within the range associated with large organic molecules such as simple oligosaccharides, pharmaceuticals, and synthetic dyes. We have therefore investigated the separation and discrimination of various synthetic dyes with a range of molecular weights, sizes, and formal charge.

The synthetic dyes investigated were Direct Red 81, Direct Red 75, Direct Blue 71, Direct Yellow 62, and Direct Yellow 50 (Fig. 7). These dyes are polyaromatic organic salts absorbing in the visible region between 400 and 800 nm. The strong adsorption in the visible spectrum makes their detection and quantification easily accomplished by UV-Vis spectroscopy. Table 4 provides a summary of the characteristics of the dyes and the concentrations employed.

20

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

Table 4 Physical characteristics and solution concentrations of the synthetic dyes studied Synthetic dyes

Concentration (mol l−1 )

ε (l mol−1 cm−1 )

Diameter (Å)

Length (Å)

MW (g mol−1 )

Charge

Direct Direct Direct Direct Direct

5.5 × 10−5 1.0 × 10−4 5.5 × 10−5 5.5 × 10−5 5.5 × 10−5

33400 14400 21780 17600 24500

7.45 7.00 12.5 17.5 9.17

26.21 32.50 21.7 27.0 27.52

675 771 991 957 1030

−2 −2 −4 −4 −4

Red 81 (2610-11-9) Yellow 62 (6409-90-1) Red 75 (2829-43-8) Yellow 50 (4399-55-7) Blue 71 (3214-47-9)

1 Å = 1 × 10−10 m.

Standard solutions of each dye were prepared in aqueous solution. The concentrations were chosen to allow for UV-Vis spectroscopic analysis on the source solution. The dye solutions were passed through a dead-end filtration system of similar configuration to that used for evaluating MWCO with the difference that the rejectate was allowed to concentrate. Eq. (4) was used to calculate the percent retention (%R), where cf and cp are the concentration of the feed and permeate, respectively and may change over time. Initial studies were performed on solutions prepared by dissolution of the dye in DI water. The pH of these solutions was between 5.7 and 7.1 for all the dyes. The percentage retention of the asymmetric membrane (i.e. membrane and support) was compared to the sup-

port alone. A summary of the retention coefficients is available in Table 5. The highest retention for the 80 nm average pore size support is ca. 40% for Direct Blue 71, Direct Red 75 and Direct Yellow 50. A lower retention was observed for Direct Yellow 62 and essentially no retention was observed for Direct Red 81. These results are suggestive of surface adsorption of the dyes rather than membrane rejection, but they do provide a base-line comparison for the A-alumoxane-derived alumina membrane that showed a retention of >70%. The Direct Yellow 62 gave inconsistent results for multiple sample runs. The 7 nm A-alumoxane membrane shows a 40–90% improved retention over the support, without significant increase in the total surface area.

Table 5 Summary of the retention of selected dyes by A-alumoxane-derived alumina membranes at a specific pH and the resulting λmax Synthetic dyes

pH

λmax (nm)

ε (l mol−1 cm−1 )

Retention membrane (5 nm) (%)a

Direct Red 81

1.87 6.05 12.52

543 527 542

36000 33400 39200

100 90 94

0 0 0

Direct Yellow 62

1.63 6.45 12.56

350 341 343

6000 14400 13800

97 20 66

20 20 20

Direct Red 75

1.76 6.95 12.41

516 521 521

13000 21800 18000

100 90 95

38 38 38

Direct Yellow 50

1.82 7.07 12.44

400 395 397

17000 17500 16900

99 73 44

39 39 39

Direct Blue 71

1.62 5.73 12.57

576 585 581

24600 24500 31600

100 88 93

40 40 40

a b

Retention coefficient of the membrane. Retention coefficient of the support.

Retention support (80 nm) (%)b

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

1.1

permeate flux (10-6 m2.s -1)

During the process of dead-end filtration the retention characteristics may change with time. The rejection may increase if the membrane becomes blocked, or it may decrease due to an increase in concentration of the solution above the membrane. To preclude these effects multiple measurements were taken during the initial 50% of the dye solution. A timed collection of the filtrate for each dye was performed over a period of 5 h. After 1 h, the concentration of the filtrate for each dye had not increased significantly over time. In addition, a single filter was used for multiple separations of Direct Red 81 (showing no significant variance of the retention values) and then subjected to EDX analysis of a cross-sectioned sample. Although there was significant dye trapped on the surface of the membrane, there were only traces detected within the membrane and support. A used membrane was back-flushed to remove any dye present on the membrane surface, after which a fresh sample of Direct Red 81 was filtered, for which there was no change in the retention. As noted above, the pH of the dye solutions were close to neutral. Each of the dyes is a poly-sulphate salt and as such it is expected that the charge on the dye molecule is pH dependent. Furthermore, the surface charge on the alumina surface of the membrane will be pH dependent. The retention characteristics of the membranes under acidic (pH ≈ 1.5) and basic (pH ≈ 12.5) conditions were investigated. Alumina membranes have a high resistance to strong acids and bases [20]. Neither acidic or basic conditions had an effect on the retention characteristics of the support, but alteration of pH did show some improvement for the retention characteristics of the membranes (Table 5). The variation of retention with pH may be explained by a combination of alteration in the speciation of the dyes with pH and change in surface charge on the membrane. At pH values below the i.e.p. of the alumina membrane (pH < 8) the membrane carries a net positive charge. At higher pH values, the membrane is negatively charged [20]. Under acid conditions functional groups on the dyes will tend to be protonated, increasing the hydrophobicity of the molecules. In contrast, for basic conditions dye molecules may carry a negative charge and a repulsive interaction potential exists between the dye molecules and the membrane surface. Thus, adsorp-

21

1.0 0.9 0.8

0

5

10

15

pH Fig. 8. Permeate flux as a function of the pH of the feed solution.

tion should be greater under acid conditions, but with fouling of the membrane surface. Under basic conditions, rejection should be greater, but with less contamination of the membrane surface. Changes in the membrane surface charge are also known to affect the permeate flux across ceramic membranes as a function of pH in the absence of foulants [20]. However, the trend observed for the permeate flux as a function of pH is not clear. A lower net charge at the i.e.p. of the membrane is postulated to lead to a higher flux as electroviscous effects should be reduced and some investigators have observed such a minimum [21]. However, a trend of monotonically increasing permeate flux with decreasing pH, above and below the i.e.p., has been observed in another previous study using commercially available alumina membranes [22]. In yet another study, permeate flux was at a minimum near pH values of 9 when filtering suspensions of SiO2 on an Al2 O3 membrane with the highest permeation rates occurring at pH values of 2 and pHs greater than 10 [23]. A trend similar to that observed in the latter study was observed in the current study when dyes were filtered on the alumoxane-derived alumina membranes. Permeate flux was higher at pH values both above and below the i.e.p. of the membrane (Fig. 8). With the exception of the Direct Yellow 50, the trends in dye retention with pH follow the trend in permeate flux with a minimum in apparent rejection occurring at pH values near the i.e.p. of the membrane (Table 5). We conclude that under acidic conditions there is significant absorption of the dyes to the alumina surface and hence a high apparent rejection. Conversely, at high pH values true rejection of the dyes occurred due in part to charge repulsion between dye and the membrane pore surfaces.

22

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

3.4. Fabrication of metal-doped membrane One method previously reported to alter pore size, flux, and retention of an alumina membrane is to dope the alumina with transition metals or form a binary oxide, Mx Aly Oz [24,25]. We have previously shown that the alumoxane nanoparticles readily react with metal acetylacetonate complexes to form metal-doped carboxylate-alumoxanes and aluminum acetylacetonate via a facile transmetalation reaction [26]. Thermolysis of the metal-doped carboxylate-alumoxanes yields the crystalline aluminates [27]. We have investigated the effects on pore size, pore distribution, and performance characteristics of iron(III), lanthanum(III) and manganese(III) doped alumina membranes. The choice of iron and lanthanum is based upon their ability to directly replace aluminum in the alumina lattice with the formation of FeAlO3 and LaAlO3 , respectively. We have previously reported that Mn-doped alumina is an effective oxidation catalyst [28] and therefore we are interested in this material as a possible catalytic membrane. Iron(III), manganese(III), and lanthanum(III) metal cations can be easily doped into alumoxanes by a transmetalation reaction between a metal salt and MEEAalumoxane (see Section 2). The membranes were fabricated in the manner described above. However, due to the low ceramic yield of MEEA-alumoxane (39%), a base layer, prepared from A-alumoxane, alumoxane, was deposited to ensure the integrity of the membrane (Fig. 9). SEM, AFM, and flow rate measurements were made on the aluminate membranes. Free-standing samples of the aluminates were prepared to allow for XRD and BET measurements.

aluminate membrane

Thermolysis of Fe-MEEA-alumoxane at 600 ◦ C results in the formation of FeAlO3 as confirmed by powder X-ray diffraction (JCPDS #30-0024). The FeAlO3 crystallite size was measured to be 20.9 nm, by XRD as compared to 17.0 nm for the undoped alumina prepared under the same conditions. As has been previously reported [24], thermolysis of La-doped MEEA-alumoxane to 600 ◦ C over 6 h and a 5 h dwell time, forms the LaAlO3 phase (JCPDS #31-0022) confirmed by powder X-ray diffraction. The LaAlO3 crystallite size for material made under these conditions was calculated from XRD to be 14.5 nm. The manganese doped alumoxane was synthesized with Mn(acac)3 , producing Mn-MEEA-A. Upon thermolysis to 600 ◦ C, spinel MnAl2 O4 (JCPDS #29-0880) and rutile MnO2 , (JCPDS #30-0820) are formed, as confirmed by powder X-ray diffraction. The presence of MnO2 is due to the disproportionation of Mn3+ to Mn2+ and Mn4+ in aqueous solution. The former presumably undergoes transmetalation with the MEEA-alumoxane (along with Mn3+ ), while the latter is inert and is oxidized upon thermolysis to MnO2 . The average pore size, pore size distribution, and surface area for the aluminate ceramics were determined on free-standing samples. A summary of the data is given in Table 6. The iron and manganese aluminates, derived from the appropriately doped MEEA-alumoxane, both exhibit a larger pore size and pore volume than the alumina analog (Table 6). In addition, both the iron and manganese aluminates have larger pore size distributions (see Fig. 10). The LaAlO3 material has a smaller pore diameter than the Al2 O3 membrane, determined by N2 absorption, where 75% of the pores of the

membrane layer

A-alumoxane derived base layer porous alumina substrate

Fig. 9. A schematic representation of the aluminate membrane prepared from metal-doped MEEA-alumoxane.

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

23

Table 6 Summary of the performance characteristics of the alumina and aluminate membranes Performance characteristics

Al2 O3

FeAlO3

LaAlO3

MnAl2 O4

Pore size (nm) Pore volume (ml g−1 ) Surface area (m2 g−1 ) Specific surface area (m2 ml−1 ) Surface roughness (nm) Flow (ml min−1 ) Permeate flux (106 m s−1 ) Permeability (nm2 )

7 0.32 111 347 35 0.070 0.85 22.2

12 0.44 210 477 127 0.091 1.10 28.4

4 0.45 174 387 42 0.100 1.21 32.0

16 0.49 162 330 99 0.066 0.80 20.5

70 60

% volume

50 40 30 20 10 0 0

10

20

50 30 40 60 70 80 Average pore diameter (nm)

90

100

Fig. 10. Nitrogen adsorption pore volume measurement for alumina (䊐), FeAlO3 ( ), LaAlO3 (䊏), and MnAl2 O4 (䊊) derived from alumoxanes.

lanthanum membrane are between 2 and 4 nm (Fig. 10) as compared to 40% of the alumina membrane pores are 6 nm in diameter (Fig. 10). The average pore size of the LaAlO3 as determined from BET measurements is confirmed by AFM to be 3–4 nm. AFM measurements also indicate that the surface roughness of the LaAO3 is slightly rougher than the alumina homologue. In contrast, the FeAlO3 and MnAl2 O4 membranes are significantly rougher, consistent with higher pore sizes. From SEM images it appears that the pores for the FeAlO3 membrane are diamond shaped (Fig. 11). The surface of the LaAlO3 membrane (Fig. 12) appears highly uniform and similar to that of the alumina membranes (Fig. 2a). The surfaces of both the LaAlO3 and FeAlO3 membranes are in contrast with that of the MnAl2 O4 membrane. SEM shows a

Fig. 11. SEM image of the surface of the FeAlO3 membrane.

24

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

Fig. 13. SEM image of the surface of the MnAl2 O4 membrane.

Fig. 12. Surface (a and b) and cross-sectional (c) SEM images of LaAlO3 membrane.

highly irregular shape of the pores (Fig. 13), crystallizing in a herringbone pattern. The irregularity of pore sizes also effects the fouling ability of the membrane [13,20]. A cross-section SEM of a fabricated membrane (Fig. 12c) confirms the presence of the multilayered structure. The surface area and the pore volume for the aluminates (Table 6) are all higher than the alumina ho-

mologue. In the case of the FeAlO3 and MnAl2 O4 materials this is in line with the increased pore sizes. However, the LaAlO3 samples have a higher surface area and pore volume despite the smaller pore size. The FeAlO3 and LaAlO3 membranes both showed an increased permeate flux and flow rate as compared to the alumina membrane. In contrast, the permeate flux and permeability for the MnAl2 O4 membrane are slightly lower than that of the alumina homologue. The increased permeate flux for LaAlO3 is despite the smaller pore size, while the lower flux for MnAl2 O4 is in spite of its higher pore size. A similar lack of correlation is observed for the flux with regard to porosity, surface area, or packing efficiency. We propose that this improvement is a function of the surface chemistry (acidity) of the membrane (see below). The retention of Direct Red 81 was investigated for each of the doped membranes (Table 6). In concert with its larger pore size (16 nm) and pore size distribution, the retention of Direct Red 81 by the MnAl2 O4 membrane was lower than the alumina membrane under neutral and basic conditions. However, under acidic conditions, the dye was retained in a comparable manner to alumina, suggesting the efficiency MnAl2 O4 membrane is also highly pH dependent. A slightly higher retention of Direct Red 81 is seen for the FeAlO3 membrane as compared to the alumina and manganese membranes under standard (neutral) conditions (90, 90, 88%), therefore, the selectivity of the iron doped membrane has increased. Since the lanthanum doped membrane produces pore sizes less than 4 nm, the retention coefficient

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

25

Table 7 Summary of the retention of the red and yellow dyes associated with metal membranes Synthetic dyes

pH

Al2 O3 (%)

FeAlO3 (%)

LaAlO3 (%)

MnAl2 O4 (%)

Direct Red 81

1.87 6.05 12.52

100 90 95

100 90 79

99 99 87

100 46 82

Direct Yellow 62

1.63 6.45 12.56

a

86 41 53

a

a

97 0–20 66

19 a

38 a

The poor results for Direct Yellow 62 under neutral conditions obviated further studies into pH effects.

should increase with Direct Red 81 compared to the retention coefficient of the alumina membrane, which is observed. Direct Yellow 62 gave an inconsistent retention coefficient when using the alumina membrane, between 0 and 20%, however, the lanthanum membrane has a narrower pore size and smaller pore size distribution, therefore the Direct Yellow 62 dye should have a higher retention coefficient when using the lanthanum doped membrane. Direct Red 81 and Direct Yellow 62 were filtered through the lanthanum membrane, separately, under acidic, neutral, and basic conditions. Table 7 provides a summary of the retention of the red and yellow dyes associated with metal membranes. The results obtained for the alumoxane-derived LaAlO3 membrane prompted further investigation to optimize the performance. In particular, the firing

temperature and dwell time should be important in determining the pore size and pore volume. Free standing samples of La-doped MEEAalumoxane were fired, in a single step over 3 h with a dwell time of 3 h, to form test samples typical of the membrane materials. As may be seen from Fig. 14 the average pore size increases with increased firing temperature, due to increased grain growth at higher temperatures. In concert with the increased pore size is an increase in surface roughness (Fig. 15a–c). Significant grain growth is seen for samples calcined to 1000 ◦ C. This is in agreement with the known crystallization of LaAlO3 at 1000 ◦ C [27]. The increase in crystallite size is confirmed by X-ray diffraction (5.87 × 10−10 m (5.87 Å) average crystallite size versus 4.71 × 10−10 m (4.71 Å) for samples sintered to 800 ◦ C). Samples heated with a longer ramp times

70 60

% volume

50 40 30 20 10 0

0

10

20

30 40 50 60 70 80 Average pore diameter (nm)

90

100

Fig. 14. Nitrogen adsorption pore volume measurements of LaAlO3 membrane sintered to 600 ◦ C (䊏), 800 ◦ C ( ), and 1000 ◦ C (䊊) over 3 h each, and 600 ◦ C (䊐) over 6 h.

26

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

3.5. Chemically functionalized membrane The surface of the A-alumoxane-derived membranes is moderately hydrophilic as measured by a contact angle of 40–50◦ [29]. Previous research within the Barron group has shown that the surface of alumina may be readily functionalized by the reaction with a carboxylic acid [29]. The resulting surface can be made to be either hydrophobic (contact angle > 70◦ ) or more hydrophilic (contact angle < 35◦ ) than the untreated surface. We have investigated the effects of surface modifications of the A-alumoxane-derived membranes on their flux and rejection characteristics for Direct Red 81. Hexanoic acid (hydrophobic) and p-hydroxybenzoic acid (hydrophilic) were used to provide hydrophobic and hydrophilic functionalization, respectively, to the interior and surface of the alumina filter (see Section 2). As may be expected, the flux for water measured for the hexanoic acid functionalized alumina membrane is significantly lower (3.14 × 10−7 m s−1 ) than for the untreated membrane (0.85×10−6 m s−1 ). Conversely, the flux for the p-hydroxybenzoic acid treated membrane (1.27 × 10−6 m s−1 ) increases over the untreated membrane. Given that the lengths of the two carboxylic acids are comparable, this result confirms that the flux is highly dependent on the hydrophilicity (or wettability) of the membrane surface. A plot of flux versus the surface contact angle for each surface shows a good correlation (Fig. 16). Furthermore, the data suggest that an upper limit of the flux of water of ca. 2 × 14

Fig. 15. SEM images of LaAlO3 formed from La-doped alumoxane sintered to (a) 600 ◦ C, (b) 800 ◦ C, and (c) 1000 ◦ C for 3 h each.

(i.e. 6 h) showed smaller average pore diameters (Fig. 14) and less surface defects (Fig. 15). This result is presumably due to the slower expulsion of pyrolysis products (CO2 and H2 O), and therefore a lower proportion of defects. For the present system the optimum process conditions involve a sinter temperature of 600 ◦ C with a ramp time of 6 h.

Permeate flux (107 m.s-1)

12 10 8 6 4 2 30

40

50 Contact angle (˚)

60

70

Fig. 16. Plot of flux vs. surface contact angle for chemically functionalized alumina membranes (R = 0.969).

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

10−6 m s−1 should be attained for a 5 nm membrane assuming 100% wetting of the surface, i.e. zero contact angle. The p-hydroxybenzoic acid functionalized alumina membrane shows a small but significant increase in the retention of Direct Red 81 (96%) as compared to the untreated membrane (90%) at the same pH. Further studies on chemically functionalized membranes are required to better understand the relationship of the surface chemistry and rejection of organic molecules. 4. Conclusions Carboxylate-alumoxane nanoparticles are an alternative to the sol-gels traditionally used for forming ultrafiltration alumina membranes. A study of the alumoxane synthesis, dip-coat protocol, calcination and sintering conditions have demonstrated the importance of uniformly small particle size in the fabrication of defect free ceramic membranes. Altering the reaction pH and controlling the reaction time of the A-alumoxane narrowed the particle size distribution of the nanoparticles. The consequential decrease in the pore size and the pore size distribution of the membrane improved the membrane’s rejection characteristics. An increase in flux can be accomplished by using doped aluminate membranes. For example, lanthanum-doped membranes show an increased retention coefficient of the dyes. Retention coefficients and permeate flux values may also be altered by the chemical functionalization of the interior surface of the membranes. Acknowledgements Financial support for this work was provided jointly by the Environmental Protection Agency (EPA), the National Science Foundation (NSF) under the Technology for a Sustainable Environment Program, and the Center for Biological and Environmental Nanotechnology. References [1] R.L. Callender, A.R. Barron, Chemical control over ceramic porosity using alumoxane precursors, Adv. Mater. 12 (2000) 734–738.

27

[2] R.L. Callender, M.R. Wiesner, A.R. Barron, Aqueous synthesis of water soluble alumoxanes: environmentally benign precursors to alumina and aluminum based ceramic materials, in: Proceedings of the 9th CIMTEC—World Ceramics Congress, Ceramics Getting into the 2000, Part C, 1999, pp. 135–142. [3] C.D. Jones, D.A. Bailey, M.R. Wiesner, A.R. Barron, Carboxylate-alumoxanes: environmentally benign precursors for developing aluminum based ceramic membranes and filter, in: Proceedings of the 9th CIMTEC—World Ceramics Congress, Ceramics Getting into the 2000, Part D, 1999, pp. 413–420. [4] R.L. Callender, A.R. Barron, A novel route to alumina and aluminate coatings on SiC, carbon and Kevlar® fiberreinforced ceramic matrix composites using carboxylatealumoxane nanoparticles, J. Mater. Res. 15 (2000) 2228–2237. [5] R.L. Callender, A.R. Barron, Formation and evaluation of highly uniform aluminate interface coatings for sapphire fiber reinforced ceramic matrix composites (FRCMCs) using carboxylate-alumoxane nanoparticles, J. Mater. Sci. 36 (2001) 4977–4987. [6] D.A. Bailey, C.D. Jones, A.R. Barron, M.R. Wiesner, Characterization of alumoxane-derived ceramic membranes, J. Membr. Sci. 176 (2000) 1–9. [7] C.D. Jones, M. Fidalgo, M.R. Wiesner, A.R. Barron, Alumina ultrafiltration membranes derived from carboxylatealumoxane nanoparticles, J. Membr. Sci. 193 (2001) 175–184. [8] R.L. Callender, C.J. Harlan, N.M. Shapiro, C.D. Jones, D.L. Callahan, M.R. Wiesner, R. Cook, A.R. Barron, Aqueous synthesis of water-soluble alumoxanes: environmentally benign precursors to alumina and aluminum-based ceramics, Chem. Mater. 9 (1997) 2421–2433. [9] M.N. Sarbolouki, A general diagram for estimating pore size of ultrafiltraion and reverse osmosis membranes, Sep. Sci. Technol. 17 (1982) 381–386. [10] T. Matsuura, Synthetic Membranes and Membrane Separation Processes, CRC Press, Ann Arbor, 1994. [11] R.W. Baker, Membrane Separation Systems: Recent Developments and Future Directions, Noyes Data Corp., Park Ridge, NJ, 1991. [12] H.D. Hsieh, K.K. Sirkar, D.R. Lloyd (Eds.), in: New Membrane Materials and Processes for Separation, vol. 84, American Institute of Chemical Engineering, New York, 1988. [13] A.J. Burggraaf, Membrane science and technology series 4, in: A.J. Burggraaf, L. Cot (Eds.), Fundamentals of Inorganic Membrane Science and Technology, Elsevier, New York, 1996. [14] American Society for Testing and Materials, Publication E, pp. 1343–1390. [15] H. de Balmann, R. Norbrega, The deformation of dextran molecules: causes and consequences in ultrafiltration, J. Membr. Sci. 40 (1989) 311–327. [16] M. Meirles, A. Bessieres, I. Rogissart, P. Aimar, V. Sanchez, An appropriate molecular size parameter for porous membranes calibration, J. Membr. Sci. 103 (1995) 105–115. [17] S. Kulkarni, E. Funk, N. Li, in: W.S.W. Ho, K.K. Sikar (Eds.), Membrane Handbook, Van Nostrand Reinhold, New York, 1992, p. 393.

28

K.A. DeFriend et al. / Journal of Membrane Science 224 (2003) 11–28

[18] A. Julbe, J.D.F. Ramsay, Membrane science and technology series 4, in: A.J. Burggraaf, L. Cot (Eds.), Fundamentals of Inorganic Membrane Science and Technology, Elsevier, New York, 1996, p. 99. [19] R. Nobrega, H. de Balmann, P. Aimar, V. Sanchez, Transfer of dextran through ultrafiltration membranes: a study of rejection data analysed by gel permeation chromatography, J. Membr. Sci. 45 (1989) 17–36. [20] R.R. Bhave, Inorganic Membranes: Synthesis, Characteristics, and Applications, Van Nostrand Reinhold, New York, 1991. [21] M.D. Moosemiller, C.G. Hill Jr., Physicochemical properties of supported ␥-Al2 O3 and TiO2 ceramic membranes, Sep. Sci. Technol. 24 (1989) 641–657. [22] F. Nazzal, M.R. Wiesner, pH and ionic strength effects on ceramic membranes, J. Membr. Sci. 93 (1994) 91– 103. [23] M.R. Hoogland, The effect of pH on the crossflow filtration of mineral slurries using ceramic membranes, in: Proceedings of the 1st International Conference on Ceramic Membranes, Montpellier, 1990. [24] B.A. McCool, G. Xomeritakis, Y.S. Lin, Composition control and hydrogen permeation characteristics of sputter deposited

[25]

[26]

[27]

[28]

[29]

palladium–silver membranes, J. Membr. Sci. 161 (1999) 67– 76. B.A. McCool, Y.S. Lin, Nanostructured thin palladium-silver membranes: effects of grain size on gas permeation properties, J. Mater. Sci. 36 (2001) 3221–3227. A. Kareiva, C.J. Harlan, D.B. MacQueen, R. Cook, A.R. Barron, Carboxylate substituted alumoxanes as processable precursors to transition metal-aluminum and lanthanidealuminum mixed metal oxides: atomic scale mixing via a new transmetalation reaction, Chem. Mater. 8 (1996) 2331–2340. R.L. Callender, A.R. Barron, Facile synthesis of aluminum containing mixed metal oxides using doped carboxylatealumoxane nanoparticles, J. Am. Ceram. Soc. 83 (2000) 1777–1789. R.L. Cook, C. Wong, C.J. Harlan, A. Kareiva, A.R. Barron, Alumoxane precursors to designer catalysts and catalyst supports: catalytic oxidation of dichloromethane, Mater. Res. Soc., Symp. Proc. 454 (1997) 169–176. C.T. Vogelson, A. Keys, C.L. Edwards, A.R. Barron, Molecular coupling layers formed by reactions of epoxy resins with self-assembled carboxylate monolayers grown on the native oxide of aluminum, J. Mater. Chem. 13 (2003) 291–296.