1. Introduction
Mixed-matrix membranes (MMMs) have turned into an apparent morphology to prevail over the gas selectivity limitation by conventional polymeric membranes. In previous researches, several filler materials like silica [1] , zeolites [2] , conductive polymers [3] or carbon molecular sieves [4] [5] were studied to embed into polymer matrix. The notable attainment has been achieved to fabricate MMMs by using glassy polymers and zeolites as polymeric matrix and filler respectively [6] [7] . The selection of filler was based on its molecular sieve mechanism on the polymer matrix which relied on its stiffness and higher intrinsic selectivity. Nearly all MMMs experienced the deficiency of interaction between zeolite and glassy polymer; hence it was the outcome of investigating the effect of non-selective voids at the polymer-zeolite membrane interface [8] [9] . Several methods have been proposed and inspected to refine the interaction between polymer and zeolite [2] [10] . These approaches can be categorized into the adaptability of polymer during membrane fabrication and affinity between zeolite and polymer. Based on the research, the polyimides, polyesters, polysulfones and polyamides resulted in higher gas separation performance [11] [12] . The influences contributing to the result of membrane fabrication are the mixture concentration of polymer and solvent, type of solvent and effect of solvent evaporation on the permeability and selectivity of membranes [13] .
To date, many researches have examined to improve and optimize the membrane morphology [14] - [16] and gas separation attainment [17] through a statistical approach. In the research, the further investigation on the interaction of mixed matrix membrane variables is statistically designed with minimum experimental runs to locate ideal setting of mixture concentration factors for improving permeability of CO2 and N2 as well selectivity of CO2 over N2. The statistical approaches were ANOVA analysis, factorial design and fitted model equations to compute the complex interaction between the independent process factors [18] . Thus, the approachability of these membranes for the dissociation of industrially significant CO2 and N2 could be investigated and the transport mechanism through each membrane type could be explained thoroughly by mechanistic clarification on deviations from single gas permeability assessment of the membranes.
This research aimed to synthesize and investigate the flat sheet mixed-matrix membrane of polysulfone (PSF) and zeolite for CO2 and N2 gas separation by screening different concentration of mixture. The membranes were fabricated and characterized for their permeability and selectivity. In order to investigate permeability of CO2 and selectivity of CO2 over N2, this current study is aimed at illustrating the interaction between concentration of PSF, zeolite and NMP in diagnostic analysis using Design of Expert (DOE). Final regression models obtained from DOE are expected to be able to predict the optimum mixture concentration parameters in producing high permeability of CO2.
2. Experimental
2.1. Chemicals
The polymeric material used for the membrane preparation was analytical grade polysulfone, PSF, which was purchased from Sigma Aldrich. The polymer has a weight-averaged molecular weight of 35,000 and melt index of about 6.5 g/10 min. Analytical grade N-Methyl-2-pyrrolidone, NMP was used as solvent with purity of 99.5% were sourced from Sigma-Aldrich. It has the chemical formula of C5H9NO and boiling point is 202˚C per liter. The filler material was molecular sieve, 4Å with particle size of 8 - 12 mesh was purchased from Aldrich. They were dried at 80˚C for 24 hours before using in membrane preparation. Nitrogen and carbon dioxide with purity 99% were obtained from Air Products Malaysia SdnBhd for gas permeation.
2.2. Sample Preparation
The mixed matrix membrane was prepared by immersion precipitation method. The zeolite was prepared from 10 to 30 W/V% and dissolved in the NMP varied from 75 to 100 cm3. The mixture was then mixed homogenously in the shaker at 200 rpm and 25˚C for three hours. Polysulfone was weighed from 10 to 30 W/V% and added into mixture solution. In order to yield homogenous solution, mechanical stirrer was set at 300 rpm to agitate the mixture for another three hours at room temperature. The dope solution was settled in a storage vessel and degassed by using ultrasonic bath to withdraw any trace of bubbles generated by mechanical stirrer [19] . The glass plate was prepared for casting by using glass rod. The sample membranes were ready for the permeability and selectivity testing for CO2 and N2 separation.
2.3. Single Gas Permeation Test for CO2/N2 Separation
Permeability and selectivity of membrane were characterized by gas permeation equipment. The test module containing membrane with the diameter of approximately 6 cm each was prepared. The pure gas permeation study was measured based on the gas solubility, CO2 gas must be the last gas to be measured due to its encouragement of membrane plasticization [20] . Thus, the modules were connected to the gas permeation apparatus manufactured by Aba Manufacturing SdnBhd using purified nitrogen at two barg. The gas permeation rate from the permeate side was measured at room temperature (25˚C ± 5˚C) using soap-bubble meter manufactured by Dwyer. The procedures of conducting permeation test were repeated by using the carbon dioxide as the replacement of test gas at the pressure of 2 bar.
2.4. Screening Study of Membrane for CO2/N2 Separation
The permeability and selectivity of membrane were screening by using Design of Experiment (DOE) software, version 7.1. Three variables were investigated in the study, coded as concentration of zeolite (X1), concentration of PSF (X2) and concentration of NMP (X3) in the software. The result of gas content of nitrogen and carbon dioxide diffused through membrane housed in a Single Gas Permeation Test was analyzed as response which coded as Y in the software. All experiments were set in random order to reduce the influence of unexplainable variability in the observed responses due to insignificant aspects. Mathematical models containing the notable terms were developed for each response parameter using the multiple linear regression analysis (MLRA) and analysis of variance (ANOVA).
3. Results and Discussion
3.1. Gas Performance Analysis
The permeability of CO2, permeability of N2 and selectivity of CO2/N2 were performed based on the single gas permeation measurement. In this study, single gas permeation measurements for N2 and CO2 gasses were performed for all fabricated membranes. At least three samples from eight different casting solutions for each of the membrane formulation were used with vary concentration of PSF, zeolite and NMP. The complete eight numbers of experiment design matrix and the output responses were shown in Table 1.
The proportion differentiation between the permeability data for the whole design matrix was below 5% based on gas permeation performed on three samples. However, it was in the range of 5.1% - 8.4% for the selectivity of CO2/N2. Thus, the membrane preparation and testing methods were confirmed to be reproducible in this research. The single gas permeation was tested at the pressure of 2 bar. From the Table 1, the permeability of CO2 through the membrane was in higher range from 9.873 to 11.641 GPU. The solubility coefficient of CO2 was greater at a lower operating pressure [20] . The solubility coefficient was a significant element in gas carrier mechanism through a polymeric dense layer because the increasing of gas solubility promoted a greater permeation rate. Thus, low pressure was preferable in single gas permeation test [9] [20] . From Table 1, gas component permeation obtained was lower than the values reported in the previous reseach for PSF membrane [12] . The
Table 1. Experimental design matrix and response results.
X1, concentration of PSf; X2, concentration of Zeolite; X3, concentration of NMP; Y1, permeability of CO2; Y2, permeability of N2; Y3, selectivity of CO2/N2.
difference may be due to the lengthy evaporation time in the research, thus it fabricated thicker dense top layer structure which influenced the permeability of CO2.
3.2. Fourier Transform Infrared Spectroscopy (FTIR) Analysis
FTIR spectra were used to investigate the interactions between polymers and zeolite particles. In this respect following the samples with higher permeability of CO2 with 11.641 GPU was determined to reveal the significant morphology of membrane polymerization based on the fabrication of polymer concentration. Figure 1 characterized that the FTIR spectra of PSF membrane with higher permeability of CO2. The peak at 3596.17 cm−1 corresponds to O-H bound in zeolite. This peak expressed the existence of an interaction between polymer and zeolite. In high zeolite loadings, the peak intensity increased and the bounds between two phases become durable. However, this interaction was not strong enough to eliminate the void spaces between the polymer matrix and zeolite [22] . The spectra showed two intense bands associated with aromatic ?CH vibration at approximately 3859.91 cm−1 and 3748.7 cm−1.
The FTIR bands at 1013.36 cm−1 and 1295.62 cm−1 on the PSF spectrum corresponded to ether diaryl C-O-C stretch and anhydride C-O stretch groups respectively. Peak at 1000 - 1100 cm−1 was related to O = S = O bound of polysulfone. This bound revealed the existence of polysulfone [22] . The band located at 1322 cm−1 was associated with sulfone S = O vibration of the membrane. The band observed at 1680.16 cm−1 was assigned to the aromatic carbon double bond vibration which indicates the bonds were not disrupted and still remained in the polymerisation structure. The disappearance of aromatic C-H bond and CH3 alkene bend with 3000 cm−1and 1375 cm−1 respectively in the spectra. It reflects the completion of polymeriasation reaction. Table 2 summarized the characteristics adsorption of membrane sample with selectivity of CO2/N2 of 1.222.
Figure 1. FTIR spectra of polysulfone based membrane with higher permeability of CO2 and higher selectivity of CO2/N2.
Table 2. Functional group of membrane sample with selectivity of CO2/N2 of 1.222 in comparison of characteristics adsorption [29] .
3.3. ANOVA Analysis
The models expressed by the variables of amount of polysulfone (Y1), amount of zeolite (Y2) and concentration of NMP (Y3) as a function of permeability of CO2 gas (A), permeability of N2 gas (B) and selectivity of CO2 over N2. Based on the statistical point of view, there were two analysis involved to assess the model like test of significance of factors (SOF) and R-squared (R2) test. Table 3 exhibited the result of the test of SOF and interactions for theY1, Y2, and Y3 after discarding the significant factors with Pro > F was less than 0.05. In other words, it reflected that the statistical proof with the valid result (P < 0.5). The elements of lack of fit and pure error were ignored in the fitted models. The models were analysed to have great (>0.70) determination coefficient of R2 indicating that the variation in the response could be justified by the mathematical model. Table 3 revealed the original R2, adjusted R2 and predicted R2 for the models after eliminating the insignificant terms. For each of the models, the predicted R2 for permeability of N2 was in an acceptable concurrence with the adjusted R2 because they were within 0.2 with one another. The mathematical models were also investigated for its conviction by correlated with the experimental data and the predicted data provided by the models. The data was observed in Table 3.
The data proved that the model provided an accurate description of the experimental data, indicating the connection between the variables and output data. The results demonstrated that there were tendencies in the linear regression fit, and the model explains the experimental range studied adequately. The fitted regression equation showed good fit of the models and was successful in capturing the correlation between the three concentration variables. From all these validity tests, the model was found to be adequate for predicting the optimized production of polysulfone based mixed matrix membrane within the range of the preparation mixture concentration studied.
3.4. Empirical Model Analysis
This empirical model was well-fitted to the experimental results, as the great value of R2 of 0.9 gave a convinced justification to the reliability of regression models for membrane’s permeability of carbon dioxide, as shown in Table 3. The model equations were analysed by F-test ANOVA which indicated that these regressions were statistically significant at 95% confidence level. The P-value for the permeability of CO2, permeability of N2 and selectivity of CO2/N2 model were less than 0.5 expressing that LOF for the mathematical models were significant. This problem has been previously reported [23] . This condition arised when the model was well fitted to the data but the measurement method was very accurate and complex. The models generated for the permeability of CO2 and N2 and selectivity of CO2/N2 by discarding the insignificant effects which was obtained after performing eight experiments, were listed in Table 4.
It reflected the single concentration variable has low effect, however the mixture concentration interaction was capable to bring the better selectivity of CO2 over N2 with F value of 0.012 if compared with concentration of PSF and zeolite. At F value of zeolite of 0.006, it was considered as minor effects on the selectivity of CO2/N2. At high zeolite loadings of 30%, the size of interfacial voids was usually greater than gas molecular
Table 3. ANOVA and R-squared (R2) statistics for the fitted models.
Y1, permeability of CO2; Y2, permeability of N2; Y3, selectivity of CO2/N2; SS, sum of squares; DF, degree of freedom and MS, mean square. *Significant at “Prob > F” less than 0.05.
Table 4. ANOVA for the regression model equation and coefficients of permeability of CO2, permeability of N2 and selectivity of CO2/ N2.
A, concentration of zeolite; B, concentration of PSf; C, concentration of NMP.
diameter thus gas molecules were able to penetrate into these voids with lower diffusion resistance instead of selective pores of zeolite. This leaded the molecular sieve role of zeolite pores less adequate [24] . As the result, the rate of permeability growth for larger molecules, N2was lower than smaller molecules, CO2 and this in turn caused to faster selectivity CO2/N2 increasing in Table 4.
The function of interaction between concentration of PSF and NMP (BC) had considerable effect on the selectivity of CO2/N2, NMP presented a high degree of polarity and hydrogen bonding which was able to overcome the solvent volatility [25] . Hence, the outermost surface of the membrane fabricated with high polymer concentration as a more volatile solvent was discharged during the evaporation process. The more solvent evaporated, the thicker the concentrated polymer region which then caused to a thicker selective skin and a minimization in permeation rate [26] . Thus, affinity between PSF and NMP potentially leaded the permeability of CO2 and N2 as well as selectivity of CO2/N2. Association of NMP and PSF can increase with increasing polymer concentration. Increasing in solvent-polymer interaction can result decreasing of the coagulation value [25] . Thus, those effects encouraged delay demixing.
The concentration of zeolite and PSF had the significant effect on the selectivity of CO2 over N2 due to the F value with 0.023, however the interaction had higher permeability of CO2 with the F value of 0.69. The cause of low selectivity was adverse association between the zeolite and the polymer influencing to a great free volume within the membrane [27] . During casting, the PSF was detached from the Zeolite 4’s surface causing micro-cavities throughout the membrane. Although it enhanced the permeability of CO2, but it limited the sieving mechanism which resulted low selectivity of CO2 over N2. For mixed matrix membranes to perform effectively, it is required to functionalize molecular sieves to have great interaction between the polymeric phase and sieves.
3.5. Verification on Statistical Models and Diagnostic Statistic
Responses surface methodology has the gain of observing the interaction effect among independent variables. The model was able to easily investigate the influences of binary combination of combining two independent factors. From Figure 2, it revealed the variable interaction for all responses. The non-parallel curvatures indicated that there was an interaction between the element of zeolite (A) and PSF (B) in Figure 2. Based on the model, the maximum operating condition of the membrane stretching was predicted around the region of higher concentration of PSF and zeolite. The range of screening PSF and zeolite concentration implied from 22.30 to 27.45 W/V% for the better selectivity of CO2/N2.
From Figure 3(a), it displayed the normal probability of the residuals, to prove whether the standard deviations between the actual and the predicted response values do obey the normal distribution [28] . The generalimpact from the graph demonstrated that the underlying errors were distributed normally as the residuals fall on the side of straight line, and thus, there was no severe evidence of non-normality of the experimental results. The plot of the residuals versus predicted response for the response was presented from Figure 3(b). All points of experimental runs were scattered randomly within the constant range of residuals across the graph within the horizontal lines at point of ±1.50. This indicated that the model proposed was adequate and the constant variance assumption was confirmed.
Reliability and adequacy of empirical model from respective responses was confirmed when the actual values obtained from experimental results were compared to the estimated values from regression models from Figure 3(c). Points above the diagonal line were those which were over-estimated and vice versa. The graph shown from Figure 3(c), it generally implied that all experiment design points were distributed along the diagonal line. Responses from the experimental results were well-fitted in acceptable variance range when collated to the predicted values from respective empirical models. This proved that error occurred was due to uncontrollable experimental error could be disregarded. Thus, regression models obtained from DOE can be further used as a predictor for the screening of membrane concentration variables in order to produce a high selectivity of CO2/N2.
3.6. Model Equations Based on Screening Effect
In general, all concentration factors had first-order effect. The linear model was performed on membrane permeability of CO2 and N2 as well as selectivity of CO2/N2. Three of the mixture concentration factors had significant model terms, which were A: concentration of zeolite, B: concentration of PSF and C: concentration of NMP. The model equations were assessed by F-test ANOVA analysis which expressed that these regressions were statistically significant at 95% confidence level. Table 5 showed the ANOVA for the regression model
(a) (b) (c)
Figure 2. Interaction via concentration of PSf and zeolite for (a) permeability of CO2; (b) permeability of N2 and (c) selectivity of CO2/N2.
(a) (b) (c)
Figure 3. (a) Normal probability plot of residual; (b) Plot of residual versus predicted response and (c) Predicted vs. actual values plot.
Table 5. The model equations for permeability of CO2, permeability of N2 and selectivity of CO2/N2.
Y1, permeability of CO2; Y2, permeability of N2; Y3, selectivity of CO2/N2; A, concentration of zeolite; B, concentration of PSf; C, concentration of NMP.
equations and coefficients for permeability of carbon dioxide.
4. Conclusion
The polysulfone-zeolite based mixed matrix membranes were fabricated in order to investigate the effect of polymer concentration on the permeability of CO2 and N2 as well as selectivity of CO2/N2. The polymer mixtures were miscible and formed completely homogenous matrix. In accordance to our expectation, FTIR results revealed that the intensity at 3596.17 cm−1 described the existence of an interaction between polymer and zeolite. Maximum permeability of CO2 and selectivity of CO2/N2 of the membrane were observed at 11.641 GPU and 1.397 respectively. The DOE results revealed that these matrix membranes were affected by the interaction of PSF and NMP to offer higher permeability and acceptance selectivity using ANOVA analysis. The range of screening PSF and zeolite concentration was from 22.30 to 27.45 W/V% for the better selectivity of CO2/N2. The model regression equations were developed in Table 5 for screening the permeability of CO2 and N2 as well as selectivity of CO2/N2 based on the effect of polymer concentration.
Acknowledgements
Thank you for UMP Grant (RDU 130310) for support this research.
Symbols
P: permeability GPU
A: effective membrane area cm2
T: temperature ˚C
Tg: glass transition temperature ˚C
M: molecular weight of the gas
xi: mole fraction of gas component i in the feed side
yi: mole fraction of gas component i in the permeate side
Δp: trans membrane pressure difference Bar
ρ: density of the gas
Greek Letters
δ: membrane thickness μm
α: selectivity
Subscripts
DOE: Design of experiment
DPE: Diphenyl ether
FTIR: Fourier transform infrared
GC: Gas chromatograph
MMM: Mixed matrix membrane
NMP: N-methyl-2-pyrrolidone
PSf: Polysulfone
NOTES
*Corresponding author.