3238
Ind. Eng. Chem. Res. 2005, 44, 3238-3248
Influence of Cross-Linking and Process Parameters on the Separation Performance of Poly(dimethylsiloxane) Nanofiltration Membranes John P. Robinson,† E. Steve Tarleton,*,† Katrin Ebert,‡ Chris R. Millington,§ and Arian Nijmeijer| Advanced Separation Technologies Group, Department of Chemical Engineering, Loughborough University, Loughborough LE11 3TU, U.K., GKSS Research Centre Geesthacht GmbH, Institute of Chemistry, Max-Planck-Strasse 1, 21502 Geesthacht, Germany, Shell Global Solutions, Cheshire Innovation Park, P.O. Box 1, Chester CH1 3SH, U.K., and Shell Global Solutions International BV, P.O. Box 38000, 1030 BN Amsterdam, The Netherlands
The separation of organic solutes in organic solvents was assessed using dense poly(dimethylsiloxane) (PDMS) membranes with different degrees of cross-linking and varying thickness of the dense PDMS layer. The predominant rejection mechanism for low-polarity organic solutes is shown to occur via size exclusion, with the rejection also being dependent on the degree of membrane cross-linking, the swelling propensity of the membrane-feed stream, and the transmembrane pressure. It is postulated that the size-exclusion mechanism arises as a consequence of the relatively large degree of swelling of the PDMS material (up to 300%), which induces appreciable regions between the polymer chains for solvent and solute transport to take place. The degree of swelling governs the relative size of the transport regions within the membrane and, hence, the overall solvent flux and solute rejection characteristics. It is shown that solvent-solute coupling plays a major role in solute transport, with the convective element of solute flow increasing as the degree of swelling increases and the solute size decreases. Despite the existence of a size-exclusion mechanism, it is difficult to rule out the solution-diffusion model as an interpretation of the data; however, it is also demonstrated that models based on pore flow can adequately define the experimental data. The similarities between the two approaches are discussed, and potential evidence of a transition between solution diffusion and pore flow is introduced. Introduction In recent years, the potential of using membrane processes to perform organic-organic separations has been studied by many researchers, and poly(dimethylsiloxane) (PDMS) has been identified as a candidate polymer in the cases of nanofiltration,1 pervaporation,2 and vapor permeation.3 Because of its hydrophobic, nonpolar nature, PDMS is prone to swelling in the presence of alkane and aromatic solvents. The dimethylsiloxane oligomer is highly soluble in nonpolar solvents so a cross-linking process must be performed if the membrane is to remain chemically stable in their presence. Cross-linking is achieved by producing covalent bonds between the polymer chains. The energy required to allow their formation can be provided in the form of heat (e.g., as with a thermosetting polymer) or energy irradiation (e.g., light, electron, UV, and γ radiation). The extent of polymer swelling is dependent on the degree of cross-linking and the swelling propensity of the solvent; a relatively high cross-linking density will restrict the physical expansion of the polymer network and result in a lower degree of swelling.4 Swelling has * To whom correspondence should be addressed. E-mail:
[email protected]. † Loughborough University. ‡ Institute of Chemistry. § Shell Global Solutions. | Shell Global Solutions International BV.
been identified as a key parameter in the performance of PDMS membranes for organic-organic separations. Scarpello et al.5 showed how the separation of catalysts from organic media was dependent on the physiochemical interactions within the membrane-solvent system. Van der Bruggen et al.6 found that solute rejection was lowest with a hexane solvent and hydrophobic membrane when compared with separations involving alcohol and water and suggested that this was due to an increased mobility of the polymer chains because of contact with the organic solvent. Other workers have adopted the use of surface tension to correlate polymersolvent interactions with solvent flux,7,8 while a recent paper by the authors9 has shown how the chemical nature of different solvent groups can govern hydraulic transport through a swollen membrane. In a further study, significant evidence was presented to show how PDMS membranes exhibit the characteristics of a porous structure when swollen with solvent.10 The aim of the current study is to investigate the relationship between membrane swelling and the separation performance by measuring the rejection of organic solute molecules from organic solvents using membranes with different degrees of cross-linking. Two fundamental concepts exist for transport through dense membranes such as PDMS: the pore-flow model and the solution-diffusion (SD) model. The two approaches differ in the way pressure and concentration gradients within the dense layer are expressed. A summary of both models is given below; more detailed
10.1021/ie0496277 CCC: $30.25 © 2005 American Chemical Society Published on Web 03/19/2005
Ind. Eng. Chem. Res., Vol. 44, No. 9, 2005 3239
derivations and discussions of the SD model are given in refs 11-14 and various pore-flow models in refs 15, 16, and 18. SD Model. The SD model assumes a constant pressure within the selective layer, which is equal to the pressure of the feed liquid. Although the process is pressure-driven, unlike hydraulic permeation the pressure induces a concentration gradient within the dense layer and hence results in diffusive transport. An increase in pressure reduces the activity of each species at the permeate side of the membrane, whereas the activity at the feed side of the membrane is equal to that in the feed liquid. This has been demonstrated experimentally by Paul and Ebra-Lima.11 For nanofiltration or reverse-osmosis applications where the permeate pressure is atmospheric, the activity of a component, i, at the permeate side of the membrane can be related to its activity in the permeate liquid: m ) aiP exp aiP
(
)
(
)
-υi∆P -υi∆P m m f γiP CiP ) γiPCiP exp RGT RGT (1)
The activity at the feed side of the membrane is equal to the activity in the liquid feed mixture; hence m m m ) aiF f γiF CiF ) γiFCiF aiF
(2)
The ratio of activity coefficients in the liquid and membrane phases is the partition coefficient, Ki. The concentration gradient within the membrane can, therefore, be expressed as
dCi ) dx
[
Ki CiF - CiP exp
(
)]
-υi∆P RGT
x
(3)
For the case of Fickian diffusion, the flux of component i in an ideal mixture can be written as
Ji )
(
[
)]
DiKi -υi∆P CiF - CiP exp x RGT
(4)
The primary separation potential of a membrane system governed by the SD theory is due to a combination of the unique diffusivity and solubility of each component. The solubility is consistent with FloryHuggins sorption behavior for nonpolar solvents and PDMS.17 Pore-Flow Model. Although the term “pore flow” may be interpreted as one of several different mechanisms, the fundamental difference between these and the SD approach is that the membrane and permeants are treated as separate phases, and it is a pressuredriven force rather than a concentration-driven force that governs solvent and solute transport. One widely used interpretation of the pore-flow approach is the Spiegler-Kedem model,18 which assumes both convective and diffusive components of solute transport. In this case, the solute rejection can be written as
R) where
(1 - F)σ 1 - σF
(5)
[
F ) exp
]
-JV(1 - σ) PS
and σ is the “reflection coefficient”, JV the total solution flux (hydraulic), and PS the solute permeance. The reflection coefficient is a parameter used to describe the maximum possible retention of a solute. A simpler approach has been adopted by Gilron et al.,19 JagurGrodzinski and Kedem,20 and Burghoff et al.,21 where the solute flux (Ji) is expressed as a linear combination of convection and diffusion and can be written as
Ji ) Di
dC + JVC(1 - σ) dx
(6)
Although the two approaches are distinctly different, their respective predictions of solvent flux and solute rejection are very similar; both models predict a linear increase in flux with pressure (at pressures up to 10 bar) and an increase in solute rejection with increasing pressure. In many applications with dense membranes, mixtures of solvents of similar polarity have been shown not to separate when permeating the membrane.9,22 In this case, one may intuitively opt for a pore-flow approach because the nonseparation of mixtures can be easily rationalized by bulk convective transport of the solvent mixture because of a hydraulic driving force. Indeed, the same authors both cite the importance of solvent viscosity on the solvent flux obtained with dense membranes. It is perhaps more difficult to rationalize such a result using the SD mechanism as represented by eq 4 because each component in the mixture is likely to exhibit a unique solubility and diffusivity. This is not necessarily the case, however, because the diffusivity of a component in a mixture is influenced by the other components within that mixture and can result in a coupling of diffusive fluxes. Alternatively, both Fick’s law and Maxwell-Stefan equations can contain convective terms that will account for flux coupling. The validity of the convective terms arises from considerations of the frame of reference for diffusion, which are discussed by Paul14 and Kamaruddin and Koros.23 The latter authors suggest that the convective term can often be neglected when the degree of swelling is low but is prevalent at high swelling. In the 1970s, the validity of the assumption of a constant pressure within the dense layer was questioned and led to something of a dispute between the pore-flow and SD mechanisms. Although the concentration gradient resulting from a SD mechanism was observed experimentally by Paul and Ebra-Lima,11 many studies of solvent-solute systems with dense membranes utilize models based on the pore flow. Jain and Gupta24 showed the Spiegler-Kedem model to apply to experimental reverse-osmosis systems, as did Gilron et al.19 for nanofiltration studies with several different membranes. Specific to PDMS-based membranes, Vankelecom et al.25 identified that convective flow played a major role in solvent transport, and Bhanushali et al.26 utilized the Spiegler-Kedem approach, among other pore-flow models, to describe experimental solute rejection data. Estimates of an effective pore size for dense membranes in nanofiltration applications have been obtained by Van der Bruggen et al.27 for a range of nanofiltration membranes and by Gibbins et al.28 for polyimide and PDMS-based membranes. It is suspected that, to some extent, confusion between the two approaches arises because of the presence of
3240
Ind. Eng. Chem. Res., Vol. 44, No. 9, 2005
convective flow. As discussed previously, both the SD and pore-flow models can account for convective flow of solvents and solutes in the dense membrane layer; however, the pore-flow models may be favored when the convective term is large. The aim of this work is to evaluate both models in terms of their applicability to the rejection of a range of organic solvents and membranes with varied cross-linking. Materials PDMS Membranes. The specially manufactured nanofiltration membranes were a composite of materials comprising a dense, selective PDMS layer and a porous support layer. The latter, made from poly(acrylonitrile) (PAN), was produced via a phase-inversion process. This support was subsequently coated with a solution containing the siloxane monomer and a catalyst and initially subjected to a thermal treatment process to induce cross-linking. Further cross-linking of the PDMS layer was performed via an irradiation process using a low-energy electron beam from a low-energy accelerator.29 The membranes were manufactured with selective layer thicknesses of 1, 2, or 10 µm, and these were verified experimentally by scanning electron microscopy imaging and nitrogen permeation tests. The coating procedure resulted in a well-defined PDMS layer with no evidence of pore intrusion. While some localized variation in the thickness was observed, gas permeation measurements showed that the average thickness was equal to the nominal thickness of the PDMS layer. Electron beam radiation generally has two opposing effects on PDMS composite membranes,30 namely, (i) cross-linking induced by the formation of radicals, which subsequently combine to form covalent cross-linking bonds, and (ii) degradation of the membrane material. An optimum radiation dose, therefore, exists that induces the maximum increase in the cross-linking density with only marginal degradation of the membrane material. For example, with composite membranes of PDMS on support layers of poly(vinylidene fluoride) and poly(ether imide), the optimum radiation dose was found to be 150 kGy, where 1 kGy ≡ 1 kJ/kg. When challenged by a process stream, membranes having undergone a higher radiation dose showed increased flux and decreased selectivity due to degradation of the membrane, and a possible effect of the radiation procedure on the support materials was discussed.31 In this case, the integrity of the membrane was maintained at a radiation dose of 200 kGy, and special care was taken to avoid the presence of electronbeam-induced free radicals in the finished membrane materials. A total of 14 different PAN-PDMS composite membranes were used in the current study. Of these, 13 samples were manufactured using the technique described above and included three samples with a 2-µm selective layer treated with radiation doses of 50, 100, and 200 kGy. Other radiation-cross-linked membranes were used with a nominal radiation dose of 80 kGy and a range of thickness of the PDMS layer. One sample was obtained that used an identical substrate material; however, the selective layer was formed by phase inversion and cross-linked using a thermal technique. A list of all of the membranes used is shown in Table 1. The irradiation dose gives a qualitative comparison of the degree of cross-linking for samples that were manufactured using this procedure. Nitrogen perme-
Table 1. Nominal Thickness of the PDMS Layer and Cross-Linking Procedures for All of the Membrane Samples Studied sample no.
dry thickness (µm)
crosslinking (kGy)
sample no.
dry thickness (µm)
crosslinking (kGy)
1 2 3 4 5 6 7
2.0 2.0 2.0 2.0 2.0 2.0 2.0
80 80 80 80 80 80 80
8 9 10 11 12 13 14
2.0 1.0 10.0 1.5 2.0 2.0 2.0
80 80 80 thermal 50 100 200
Table 2. Molecular Weight and Size of the Solute Compounds Used in This Study
compound
molecular weight (g/mol)
minimum size (nm)
maximum size (nm)
thiophene 1-butanethiol acenaphthene anthracene phenanthrene ferrocene pyrene coronene 9,10-diphenylanthracene 1,1,2,2-tetraphenylethylene iron(III) acetylacetonate iron naphthenate rubrene copper naphthenate
84 90 154 178 178 186 202 300 330 332 353 373 532 611
0.295 0.331 0.670 0.561 0.771 0.522 0.771 0.968 0.968 0.708 1.291 1.818 1.210 0.916
0.516 0.828 0.727 0.968 0.848 0.538 0.848 0.981 1.274 0.995 1.291 1.818 1.274 3.516
abilities for all of the samples were indistinguishable, which is consistent with the observations of Dudley et al.32 Further aspects of this work involve characterization of the degree of cross-linking such that all 14 samples can be compared, which is discussed in later sections. Solvents and Solutes. The low-polarity solvent and solutes were obtained from Sigma-Aldrich Ltd. The solvent used was xylene (mixture of isomers), to which were added a range of low-polarity organic solute compounds as shown in Table 2. All of the solutes were evaluated in xylene at concentrations of 20-50 ppm. The sizes of the solutes were estimated using bond lengths (resolving to a single plane using bond angles as necessary) and covalent radii from the solute molecular structures. The low-polarity nature of the solvent and solutes infers that any solvent-solute interactions are minimal; hence, the solute dimensions in the solvent are equivalent to those of pure substances. Experimental Section The separation characteristics of the solvent-solute systems were studied in the cross-flow membrane filtration apparatus shown schematically in Figure 1. The solvent-solute mixture was added to the 2.5-L capacity reservoir (A), from which an air-driven pump (B) delivered the fluid to the membrane module (C) via a variable-area flowmeter (F), a flow control valve (V6), and a 15-µm rated prefilter (D). The permeate could be either circulated back to the reservoir or collected separately for subsequent sample analysis. The retentate stream returned to the reservoir through a cooler (E), which employed the exhaust air stream from the pump to maintain the temperature of the circulating fluid. The transmembrane pressure and cross-flow rate were controlled primarily by the backpressure regulator
Ind. Eng. Chem. Res., Vol. 44, No. 9, 2005 3241
Figure 2. Flux-pressure relationship for xylene with samples 12 (50 kGy), 13 (100 kGy), and 14 (200 kGy). Table 3. Xylene Permeability and SD Model Parameters from Equation 7 for the 14 Membrane Samples Studied Figure 1. Schematic of the cross-flow nanofiltration apparatus: (A) reservoir; (B) air-driven pump; (C) membrane module; (D) prefilter; (E) cooler; (F) flowmeter; (P) pressure gauge; (V1) drain valve; (V2) air regulator; (V3) sampling valve; (V4) permeate recycle valve; (V5) permeate valve; (V6) flow control valve; (V7) backpressure regulator.
(V7) and the air regulator to the pump (V2); the flow control valve (V6) was sometimes used to make minor adjustments to process conditions. The circular, flat sheet membrane was mounted in an Osmonics DESAL membrane cell to give a wetted surface area of 75 cm2. The membranes were mounted dry, with 100 mL of the process fluid being used to flush away any residual solvent from the manufacturing process. When the desired process conditions were set, the permeate was returned to the reservoir to allow a steady state to be achieved. The permeate was then diverted to a separate collecting vessel for a set time, which allowed flux measurements to be obtained. Samples of this permeate were subsequently used to determine the concentration of the desired solute. The amount of permeate collected was such that the recovery was of the order of 10%. In all cases, a solute mass balance was obtained to within 1% based on concentrations and masses of the feed, permeate, and retentate. No degradation in the membrane performance was observed for the duration of the testing period. Experimental Results Solvent Flux. The xylene flux was measured at pressures of 2-9 bar and at 1 bar increments for each of the 14 membrane samples studied. In all cases, the flux-pressure relationships were linear over this pressure range, an example of which is shown in Figure 2. The xylene permeability (rationalized using the dry thickness of the PDMS layer) for each membrane is also shown in Table 3. For those samples whose relative degrees of crosslinking can be compared, a higher degree of crosslinking results in a lower xylene permeability and vice versa. To our knowledge, the dependence of the solvent flux in nanofiltration on the degree of membrane crosslinking has not previously been explicitly stated; however, this result may be inferred from discussions of the
sample no.
dry thickness (µm)
crosslinking (kGy)
xylene permeability (×10-12 m2/s‚bar)
DSKS (×10-10 m2/s)
1 2 3 4 5 6 7 8 9 10 11 12 13 14
2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 1.0 10.0 1.5 2.0 2.0 2.0
80 80 80 80 80 80 80 80 80 80 thermal 50 100 200
2.839 2.817 2.811 2.706 2.756 2.956 2.113 1.740 1.928 1.820 0.829 2.832 1.990 1.576
6.252 6.204 6.190 5.959 6.069 6.510 4.653 3.832 4.246 4.008 1.826 6.237 4.382 3.471
degree of swelling. Our previous work,9 Paul et al.22 and Vankelecom et al.25 are three examples that highlight that the solvent flux is governed, in part, by the degree of membrane swelling, with higher fluxes obtained with higher degrees of solvent-induced swelling. Because the extent to which the membrane is cross-linked determines the degree of swelling (more cross-linking corresponds to less swelling and vice versa4), the results shown in Table 3 are in qualitative agreement with those of previously published work. At this point, it is worth noting that the degree of cross-linking of the remaining membrane samples cannot be quantified in terms of an irradiation dose. However, the relationship between permeability and the degree of swelling infers that the degree of cross-linking can be qualitatively characterized by the solvent flux itself. At relatively low pressures (
