6FDA-DAM:DABA Co-Polyimide Mixed Matrix Membranes with GO and ZIF-8 Mixtures for Effective CO2/CH4 Separation

This work presents the gas separation evaluation of 6FDA-DAM:DABA (3:1) co-polyimide and its enhanced mixed matrix membranes (MMMs) with graphene oxide (GO) and ZIF-8 (particle size of <40 nm). The 6FDA-copolyimide was obtained through two-stage poly-condensation polymerization, while the ZIF-8 nanoparticles were synthesized using the dry and wet method. The MMMs were preliminarily prepared with 1–4 wt.% GO and 5–15 wt.% ZIF-8 filler loading independently. Based on the best performing GO MMM, the study proceeded with making MMMs based on the mixtures of GO and ZIF-8 with a fixed 1 wt.% GO content (related to the polymer matrix) and varied ZIF-8 loadings. All the materials were characterized thoroughly using TGA, FTIR, XRD, and FESEM. The gas separation was measured with 50:50 vol.% CO2:CH4 binary mixture at 2 bar feed pressure and 25 °C. The pristine 6FDA-copolyimide showed CO2 permeability (PCO2) of 147 Barrer and CO2/CH4 selectivity (αCO2/CH4) of 47.5. At the optimum GO loading (1 wt.%), the PCO2 and αCO2/CH4 were improved by 22% and 7%, respectively. A combination of GO (1 wt.%)/ZIF-8 fillers tremendously improves its PCO2; by 990% for GO/ZIF-8 (5 wt.%) and 1.124% for GO/ZIF-8 (10 wt.%). Regrettably, the MMMs lost their selectivity by 16–55% due to the non-selective filler-polymer interfacial voids. However, the hybrid MMM performances still resided close to the 2019 upper bound and showed good performance stability when tested at different feed pressure conditions.


Introduction
The energy demand of our severely industrialized world has led to an enormous amount of greenhouse gases emission. CO 2 has been determined to be one of the most concerning factors, and its atmospheric concentration is expected to increase to 450 ppm by 2035, which might cause an increase in the global temperature by 2 • C [1]. Each 188 million tons of industrial CO 2 emission contributes to the rise in atmospheric CO 2 concentration by 1 ppm and the level has reached ca. 414 ppm in December 2020 [2]. The overwhelming impacts from fuel combustion heat and electricity (>40% of global CO 2 emissions in 2014) emphasize the need for more effective CO 2 separation from natural gas resources and CO 2 capture in fossil fuel-based energy generation plants.

Co-Polyimide and Nanomaterials Syntheses
The co-polyimide was synthesized using a two-step poly-condensation method as presented earlier [31]. Firstly, a polyamic acid (PAA) was synthesized with an equimolar amount of diamines and dianhydride monomers in NMP under a N 2 atmosphere at room temperature. A combination of DAM and DABA diamines were used at the molar ratio of 3:1. The amount of reactants was calculated to obtain 6 g of 15 wt.% PAA solution. As the second step, thermal imidization was conducted to obtain the imidized 6FDA-DAM:DABA (3:1) polyimide ( Figure 1).
ZIF-8 was synthesized by adding a solution of Zn(NO 3 ) 2 ·6H 2 O (1.03 g in 70 mL methanol) into the 2-methylimidazole solution (2.07 g in 70 mL methanol), and the slurry solution was stirred for 1 h. The nanoparticles were collected through two methods, described previously [32], as follows: 1.
Dry method: The slurry solution was centrifuged at 20,000 rpm (20 min) and the supernatant methanol was removed and replaced with fresh methanol (30 mL), followed by an ultrasound sonication (Kraintek K-10LE, ultrasonic power 300 W at the frequency 38 kHz for 15 min) to re-disperse the nanoparticles in the fresh solvent. This procedure was repeated for 3 cycles and the final supernatant methanol was discarded. The obtained nanoparticles were dried at 90 • C overnight.

2.
Wet method: At the third cycle of the dry method, the methanol was exchanged by 30 mL NMP and centrifuged at 20,000 rpm (20 min). The supernatant NMP was discarded and replaced by fresh NMP and the cycle was repeated 5 times, producing a solution of ZIF-8 nanoparticles in NMP. To determine the ZIF-8 powder concentration, 1 g of ZIF-8/NMP solution was spread onto a glass plate and kept dry in a vacuum oven at 90 • C for 24 h. The final dried weight was used to calculate the ZIF-8 concentration and determined at 0.074 g·mL −1 . The solution was tightly sealed and kept stirred at room temperature before the MMM preparation.
Please note that the 'dry' ZIF-8 nanoparticles were used for characterization only, whereas the 'wet' ZIF-8 was used to prepare MMMs.
In the preparation of GO, the graphite was oxidized via the improved Hummers method, also called the Tour method, according to the procedure described by Jankovský et al. [33]. The graphite (3 g) was mixed with 360 mL of H 2 SO 4 (98 wt.%) and 40 mL of H 3 PO 4 (85 wt.%). Subsequently, 18 g of KMnO 4 were added. The reaction mixture was stirred and then heated to 50 • C for 12 h. Afterward, the reaction mixture was quenched in ice (400 g) with 20 mL of H 2 O 2 (30 wt.%). The formed GO was separated by centrifugation, washed several times by deionized water until neutral pH, and finally lyophilized. The resulting powder was stored in a desiccator.
In the preparation of GO, the graphite was oxidized via the improved Hummers method, also called the Tour method, according to the procedure described by Jankovský et al. [33]. The graphite (3 g) was mixed with 360 mL of H2SO4 (98 wt.%) and 40 mL of H3PO4 (85 wt.%). Subsequently, 18 g of KMnO4 were added. The reaction mixture was stirred and then heated to 50 °C for 12 h. Afterward, the reaction mixture was quenched in ice (400 g) with 20 mL of H2O2 (30 wt.%). The formed GO was separated by centrifugation, washed several times by deionized water until neutral pH, and finally lyophilized. The resulting powder was stored in a desiccator.

Membrane Fabrication
In the preparation of neat membranes, 0.5 g of co-PI was dissolved in a pre-weighted solvent (THF), making a polymeric solution with a concentration of 7.0 wt.% (calculated from Equation (1)). The solution was magnetically stirred for 2 h, followed by 30 min sonication before pouring onto a casting glass and dried in a closed container of THF-saturated atmosphere for 24 h.
The dried membrane was detached by water and dried at 80 °C for 24 h before use. The procedure illustration is presented in Figure S1.

Membrane Fabrication
In the preparation of neat membranes, 0.5 g of co-PI was dissolved in a pre-weighted solvent (THF), making a polymeric solution with a concentration of 7.0 wt.% (calculated from Equation (1)). The solution was magnetically stirred for 2 h, followed by 30 min sonication before pouring onto a casting glass and dried in a closed container of THFsaturated atmosphere for 24 h.
The dried membrane was detached by water and dried at 80 • C for 24 h before use. The procedure illustration is presented in Figure S1.
All the membranes were casted and dried as above. The illustrations for MMM preparation are presented in Figures S2-S4. (2)

Characterization
The final polyimide was characterized by a Bruker Fourier-transform infrared spectroscopy (FTIR), equipped with an IFS 66v/s to confirm that the PAA's complete imidization was achieved. The spectral analysis was carried out from 400 to 4000 cm −1 wavenumbers at 4 cm −1 resolution. The same analysis was conducted for all membrane samples. ZIF-8 s crystallinity was determined by X-ray powder diffraction (XRD) analysis using an XRD-Diffractometer PANalytical X'Pert PRO (PANalytical Holland) using Cu-Kα radiation at a 40 kV voltage and 30 mA current.
All materials and membrane samples were imaged by a field emission scanning electron microscope (FESEM) JEOL-JSM-5600LV. The membrane samples were prepared through the freeze-fracturing method [24,35]. The membranes were immersed in liquid nitrogen for several minutes and fractured inside the liquid nitrogen for neatly fractured membrane cross-sections. Thermogravimetric analysis (TGA) was carried out using a Linseis STA 700LT where an 8-15 mg sample was placed into an alumina crucible and heated at 10 • C min −1 up to 700 • C under 20 mL min −1 N 2 flow. The decomposition temperature (T d ) was determined by the highest point of its first derivative of weight loss.

Gas Separation Measurement
Gas separation analyses were performed using a laboratory-scale permeation apparatus, with a Wicke-Kallenbach permeation cell. The set-up scheme is presented elsewhere [36]. The measurement was carried out at steady-state conditions using a CO 2 and CH 4 binary mixture as the feed and helium as a sweep gas. The permeate gas was analyzed using a FOCUS gas chromatograph equipped with a methanizer and a flame ionization detector (FID).
The base separation performance was performed using 50:50 vol.:vol. of CO 2 (>99.9%, SIAD) and CH 4 (>99.7%, Linde), at 25 • C. The gas permeabilities (reported in Barrer, 1 Barrer = 10 −10 cm 3 (STP)·cm·cm −2 ·s −1 ·cmHg −1 ) under mixed gas conditions were calculated using Equation (3), where y i and x i are the molar fraction of the corresponding gas in the permeate and the feed flow, respectively. F s is the calibrated sweep flow (cm 3 (STP)·s −1 ), l is the membrane thickness (cm), A is the membrane area (cm 2 ), and P f and P p are the feed and permeate side pressures (cmHg), respectively. These equations are derived from the cell mass balance, assuming the negligible cross membrane flow compared to the feed and sweep flow.
The separation factor of the membrane was calculated from Equation (4).

Materials Characterizations
3.1.1. 6FDA-Copolyimide and Nanoparticle Characterizations FTIR spectra of the synthesized 6FDA-DAM:DABA (3:1) PAA and its imidized co-PI are presented in Figure 2a. A complete polyimide formation was indicated by the disappearance of amide functional group, -CONH at~1510 cm −1 into imide, -NH-at 1638 cm −1 [24], and the disappearance of PPA's -COOH and -NHCO broad convoluted stretching band between 2500 and 3500 cm −1 [37]. Other defining co-PI peaks are the symmetric and asymmetric C = O stretching at~1724 cm −1 and~1788 cm −1 , as well as the -CN-stretching peak at~1358 cm −1 . The co-PI was then used to fabricate neat flat sheet membranes and MMMs with GO (1-4 wt.%), ZIF-8 (5-15 wt.%), and their mixtures. species from the near surfaces [40,41]. In the case of GO, the initial weight loss around 90-120 °C is attributed to the dehydration of the nanosheets, followed by the main decomposition profile that peaks at ~240 °C and showing ca. 23.6% mass loss, attributed to the carboxylic decomposition, releasing CO2 gas [16]. The removal of hydroxyl and carboxylic groups will leave space vacancies and topological defects throughout the nanosheets planes, with smaller interlayer spacing and referred to as reduced GO (rGO).    (222), respectively. The primary peak at (110) indicates ZIF-8 face orientation and its high intensity is attributed to the stable rhombic dodecahedron shape in ZIF-8 formation, which resembles the final stage of ZIF-8 structure growth [11,34]. The FESEM images of ZIF-8 show very small nanoparticles in size range of ca. 37.1 ± 8.4 nm (see Figure 3c,d). On the other hand, GO nanosheets (FESEM images in Figure 3a,b) show a sharp 2θ peak at 11.8 • (Figure 2b), attributed to the (001) GO direction. These values can be used to determine the nanoparticle pore opening or GO interlayer or polymer packing distances using Bragg's law (2d hkl sin θ = nλ; d hkl = distance) [26]. As for the 6FDA-DAM:DABA (3:1) co-PI, the pristine membrane shows a broad prominent characteristic peak of an amorphous polymer with a d-spacing of 5.7 Å, smaller than pristine 6FDA-DAM membranes (d-spacing = 6.8-7.0 Å [38,39]), which will directly associate with the polymers' smaller free volumes.
Referring TGA profile in Figure 2c, the decomposition temperature (T d ) of neat co-PI is at ca. 557 • C, determined by the peak of the final weight loss (~41.5%) that designates the decomposition of a PI backbone. The earlier weight loss (~4.0%), initiated at~351-490 • C, is related to the decarboxylation of DABA moieties in the 6FDA-DABA homo-polyimide, releasing CO 2 and CO [23]. The earliest weight loss (<100 • C), which is almost negligible, is related to the evaporation of surface and near-surface trapped moisture and volatile solvent (THF, b.p. 66 • C). The following that is a near-plateau weight loss (~5.3%) up to~350 • C associated with the deep-matrix residual removal [26]. As for ZIF-8, the analysis showed that the nanoparticles are stable up to ca. 420 • C, similar to the reported thermal stability values for nano-sized ZIF-8 in N 2 [40]. The earlier weight loss (~1.6%) is attributed to the removing trapped moisture, guest molecules, and potentially the unreacted reactant species from the near surfaces [40,41]. In the case of GO, the initial weight loss around 90-120 • C is attributed to the dehydration of the nanosheets, followed by the main decomposition profile that peaks at~240 • C and showing ca. 23.6% mass loss, attributed to the carboxylic decomposition, releasing CO 2 gas [16]. The removal of hydroxyl and carboxylic groups will leave space vacancies and topological defects throughout the nanosheets planes, with smaller interlayer spacing and referred to as reduced GO (rGO).

Membrane Characterizations
TGA was conducted to determine the decomposition profiles and effect of filler incorporation on Td of membranes. FTIR spectroscopy was used to determine a possible chemical interaction between the filler and polymer, while FESEM imaged the membranes' cross-section microstructures (thickness of 50-65 μm).
In all the GO MMMs, the Td (555-557 °C, see Table 1) is similar to that of the neat membrane (557 °C); indicating GO addition brings no significant effect on the overall MMM thermal stability. Their FESEM images showed continuous and undisruptive phases as can be seen in Figure S5. Contrarily, in ZIF-8-containing MMMs, the Td (553-556 °C) slightly decreases, probably due to its higher degree of polymer chain disruption upon ZIF-8 addition and the insufficient or poor interface interaction, which can be mainly observed in the FESEM images of high loading MMMs (see Figure S6, defects are highlighted in red circles). Evidently, the addition of 'wet ZIF-8′ in the NMP solution may have contributed to the more considerable amount of deep-matrix trapped solvents (9.5-20.0% in ZIF-8-containing MMMs, see Table 1), which was removed between 100-350 °C. The value is much smaller in neat and GO MMMs. In the FESEM image of GO/ZIF-8 (15 wt.%) (Figure 4c), it can be clearly seen that the MMM possesses a good filler distribution. The round cavities surround the nanoparticles suggest that the NMP may have encapsulated the ZIF-8 nanoparticles during the preparation stage and when cured, the NMP was removed (not entirely, as evidenced in TGA), leaving space voids without filler-polymer interaction. This defective microstructure will influence the gas separation performances and be discussed accordingly in the next section. decomposition temperatures (Td) of the 6FDA-DAM:DABA (3:1) and its respective MMMs. Their decompoand weight loss first derivative curves are presented in Figure S7.

Membrane Characterizations
TGA was conducted to determine the decomposition profiles and effect of filler incorporation on T d of membranes. FTIR spectroscopy was used to determine a possible chemical interaction between the filler and polymer, while FESEM imaged the membranes' cross-section microstructures (thickness of 50-65 µm).
In all the GO MMMs, the T d (555-557 • C, see Table 1) is similar to that of the neat membrane (557 • C); indicating GO addition brings no significant effect on the overall MMM thermal stability. Their FESEM images showed continuous and undisruptive phases as can be seen in Figure S5. Contrarily, in ZIF-8-containing MMMs, the T d (553-556 • C) slightly decreases, probably due to its higher degree of polymer chain disruption upon ZIF-8 addition and the insufficient or poor interface interaction, which can be mainly observed in the FESEM images of high loading MMMs (see Figure S6, defects are highlighted in red circles). Evidently, the addition of 'wet ZIF-8 in the NMP solution may have contributed to the more considerable amount of deep-matrix trapped solvents (9.5-20.0% in ZIF-8containing MMMs, see Table 1), which was removed between 100-350 • C. The value is much smaller in neat and GO MMMs. In the FESEM image of GO/ZIF-8 (15 wt.%) (Figure 4c), it can be clearly seen that the MMM possesses a good filler distribution. The round cavities surround the nanoparticles suggest that the NMP may have encapsulated the ZIF-8 nanoparticles during the preparation stage and when cured, the NMP was removed (not entirely, as evidenced in TGA), leaving space voids without filler-polymer interaction. This defective microstructure will influence the gas separation performances and be discussed accordingly in the next section.
These findings are further supported by FTIR analysis, where all membranes (GO, ZIF-8, GO/ZIF-8 MMMs) show no significant shift in the specific co-PI functional group signals; imide -NH-at~1638 cm −1 , symmetric and asymmetric C = O stretching at~1724 cm −1 and~1788 cm −1 , and -CN-stretching peak at~1358 cm −1 , indicating there is no strong filler-polymer interaction. The spectra can be referred to in Figure S8.  These findings are further supported by FTIR analysis, where all membranes (GO, ZIF-8, GO/ZIF-8 MMMs) show no significant shift in the specific co-PI functional group signals; imide -NH-at ~1638 cm −1 , symmetric and asymmetric C = O stretching at ~1724 cm −1 and ~1788 cm −1 , and -CN-stretching peak at ~1358 cm −1 , indicating there is no strong filler-polymer interaction. The spectra can be referred to in Figure S8.

Gas Permeability and CO2/CH4 Selectivity
The thin dense 6FDA-DAM:DABA (3:1) membranes show a good CO2 permeability (PCO2) of 147.4 ± 6.1 Barrer and CO2/CH4 selectivity (αCO2/CH4) of 47.5 ± 4.0. The selectivity value is higher than the 6FDA-DAM:DABA (3:1) membrane reported earlier by our group [31]; nonetheless, the performances are comparable to several other studies (the values are summarized in Table 2). The synthesis procedure does not guarantee the regular distribution of DAM and DABA in polymeric chains, which leads to their random sequence distribution, hence explains the separation performance discrepancy. Despite the fact that in global the ratio of 6FDA-DAM and 6FDA-DABA sequences is 3:1, we can also expect in polymeric chain the presence of segments where the sequences of 6FDA-DAM are more cumulated and vice versa. In an extreme case, it can result in the presence of some fraction of 6FDA-DAM and 6FDA-DABA homo-polyimides in the 6FDA-DAM:DABA co-PI. From the data presented in Table 2, we may speculate that higher gas permeabilities are obtained in the 6FDA co-PI membranes with a higher fraction of 6FDA-DAM homo-polyimide or presence of longer segments with 6FDA-DAM sequences. It is known that the bulkier DAM moieties in polymers contributes to their higher free volumes, FFV (6FDA-DAM of 18-24% [39,42,43]) > 6FDA-DABA of 18.3% [44]) and bigger inter-chain distances, d-spacing (6FDA-DAM of 6.8 Å > 6FDA-DAM:DABA of 5.6 Å > 6FDA-DABA of 5.1 Å [38]), affected by its inefficient chain packing. This would influence the ability of small molecules to diffuse through the glassy polymer matrix. As for the lower permeability 3.2. Gas Transport Properties 3.2.1. Gas Permeability and CO 2 /CH 4 Selectivity The thin dense 6FDA-DAM:DABA (3:1) membranes show a good CO 2 permeability (P CO2 ) of 147.4 ± 6.1 Barrer and CO 2 /CH 4 selectivity (α CO2/CH4 ) of 47.5 ± 4.0. The selectivity value is higher than the 6FDA-DAM:DABA (3:1) membrane reported earlier by our group [31]; nonetheless, the performances are comparable to several other studies (the values are summarized in Table 2). The synthesis procedure does not guarantee the regular distribution of DAM and DABA in polymeric chains, which leads to their random sequence distribution, hence explains the separation performance discrepancy. Despite the fact that in global the ratio of 6FDA-DAM and 6FDA-DABA sequences is 3:1, we can also expect in polymeric chain the presence of segments where the sequences of 6FDA-DAM are more cumulated and vice versa. In an extreme case, it can result in the presence of some fraction of 6FDA-DAM and 6FDA-DABA homo-polyimides in the 6FDA-DAM:DABA co-PI. From the data presented in Table 2, we may speculate that higher gas permeabilities are obtained in the 6FDA co-PI membranes with a higher fraction of 6FDA-DAM homo-polyimide or presence of longer segments with 6FDA-DAM sequences. It is known that the bulkier DAM moieties in polymers contributes to their higher free volumes, FFV (6FDA-DAM of 18-24% [39,42,43]) > 6FDA-DABA of 18.3% [44]) and bigger inter-chain distances, dspacing (6FDA-DAM of 6.8 Å > 6FDA-DAM:DABA of 5.6 Å > 6FDA-DABA of 5.1 Å [38]), affected by its inefficient chain packing. This would influence the ability of small molecules to diffuse through the glassy polymer matrix. As for the lower permeability shown by 6FDA-DAM:DABA (2:1) compared to (3:2) in Table 2, it is merely caused by the higher feed pressure, which is related to the gradual saturation of permeating gases inside the polymer permanent voids, affecting the overall gas mobility through the membrane matrix rather than the competitive sorption [28].
It is known that a dry GO film is impermeable for all gas molecules and only water vapor permeation is allowed [45]. However, in the presence of defects in their sheetlike morphology, GO nanosheets act like a sieve that permits relatively smaller sized CO 2 to permeate but restricts comparatively larger-sized CH 4 molecules to pass through the pores. As expected, CO 2 permeability in GO MMMs only increases at low loadings (1 wt.% GO MMM, P CO2 = 179.4 ± 3.6 Barrer, α CO2/CH4 = 50.8 ± 3.6) (see Figure 5a). A similar observation was reported in the highly permeable Pebax ® 1657 [46] MMMs with GO. At the higher loadings (2 and 4 wt.% GO), the membranes lost their gas permeabilities. Higher permeability would be expected at higher GO loadings, as observed in many studies [22,47,48], but we observed otherwise. This discrepancy can be explained by the two transport pathways, as presented by Ibrahim and Lin [15] and illustrated in Figure 6a. The inter-sheet pathway A comprises randomly distributed nanoscale wrinkles and intergalleries between stacked GO sheets, which will increase the permeating gas's tortuosity pathway and thus decrease the gas permeability. Whereas the inner-sheet pathway B constitutes of GO sheet structural defects, assumed to be aligned like straight channels, and produces a much smaller tortuosity factor for pathway B and faster gas permeation rate. As for ZIF-8 addition, the 6FDA co-PI MMMs showed a tremendous and continuous increase of CO2 and CH4 permeabilities, as shown in Figure 5a,b. The suggested gas transport model of porous filler addition; for fillers such as ZIF-8 [34,53], UiO-66 [26], HKUST-1 [54] is presented in Figure 6b, wherein a superlative case, the porous fillers will selectively enhance the permeation of smaller gas molecules and thus increase the overall MMM selectivity. Regarding the ideal case of MMM without any interface defect morphologies [9], the addition of porous selective filler should increase the gas permeability and its selectivity. Still, our observation differs from these findings (see Figure 5c). Thus, it can be suggested that there is an occurrence of non-selective interface voids, which causes the permeability increase and selectivity decrease. This phenomenon was suggested to be caused by the poor filler-polymer interaction and the polymer matrix's inability to fully surround the large filler particles (especially agglomerates), which both lead to the non-selective permeation pathways [55]. In the highest degree of voidage, it may lead to a leaking phenomenon where the MMMs completely lost their selectivity [9]. In a very recent study, Knebel et al. [56] showed the leaking phenomenon in MMM can be resolved by using a ZIF-based porous liquid [57]. The author also ensured good filler distribution even at high particle loadings (20-50 wt.%) by using ZIF-67-IDip, a type III porous liquid in 6FDA-DAM [56,57]. As for ZIF-8 addition, the 6FDA co-PI MMMs showed a tremendous and continuous increase of CO2 and CH4 permeabilities, as shown in Figure 5a,b. The suggested gas transport model of porous filler addition; for fillers such as ZIF-8 [34,53], UiO-66 [26], HKUST-1 [54] is presented in Figure 6b, wherein a superlative case, the porous fillers will selectively enhance the permeation of smaller gas molecules and thus increase the overall MMM selectivity. Regarding the ideal case of MMM without any interface defect morphologies [9], the addition of porous selective filler should increase the gas permeability and its selectivity. Still, our observation differs from these findings (see Figure 5c). Thus, it can be suggested that there is an occurrence of non-selective interface voids, which causes the permeability increase and selectivity decrease. This phenomenon was suggested to be caused by the poor filler-polymer interaction and the polymer matrix's inability to fully surround the large filler particles (especially agglomerates), which both lead to the non-selective permeation pathways [55]. In the highest degree of voidage, it may lead to a leaking phenomenon where the MMMs completely lost their selectivity [9]. In a very recent study, Knebel et al. [56] showed the leaking phenomenon in MMM can be resolved by using a ZIF-based porous liquid [57]. The author also ensured good filler distribution even at high particle loadings (20-50 wt.%) by using ZIF-67-IDip, a type III porous liquid in 6FDA-DAM [56,57].  [15] and (b) the diffusion pathways through porous filler phases in the polymer matrix, where a discontinuous (left) or continuous (right) particle distribution is formed. The permeability in the order of Pb > Pa [5]. All images are reproduced with permissions.
In both GO-and ZIF-8-addition, we believe that the gas separation observation well agrees with their SEM images (see Figures S5 and S6), where the GO MMMs show a firm, continuous and non-disruptive phase whilst the ZIF-8 MMMs show gaps and defective interfaces around the fillers, as previously discussed.  [15] and (b) the diffusion pathways through porous filler phases in the polymer matrix, where a discontinuous (left) or continuous (right) particle distribution is formed. The permeability in the order of P b > P a [5]. All images are reproduced with permissions.
Based on these transport models, it can be concluded that in the 1 wt.% GO MMM the pathway B arrangement is primarily generated. However, at the higher GO loadings, the GO nanosheets tend to agglomerate due to GO sheets crosslinking, leading to gas diffusion channels' blockage. The aggregated GO sheets possess a complex tortuosity [46], that reduces the CO 2 and CH 4 permeabilities (a similar finding is shown in Figure 5a,b). In a study using graphene sheets and PIM-1 [49], the authors simulated the MMM system and their final visualization concluded that sheets, when not agglomerated, were arranged in parallel to the polymer fragments. The arrangement constrained the polymer chain mobility, blocked and/or occupied the polymer free volumes, causing the reduced gas permeability. The phenomenon also might have occurred in our GO/6FDA-copolyimide MMMs. Meanwhile, for the other studies [22,47,48], it can be concluded that their continuous increase in gas permeability with GO loadings is caused by the non-selective voids present at GO-polymer interfaces, which were evidenced in their loss of gas pair selectivity (CO 2 /CH 4 and CO 2 /N 2 ). Overall, the reduced gas permeabilities cause the CO 2 /CH 4 selectivity to decrease at these higher loadings (Figure 5c). * CO 2 permeance in gas permeation unit (GPU) and CO 2 /N 2 selectivity.
As for ZIF-8 addition, the 6FDA co-PI MMMs showed a tremendous and continuous increase of CO 2 and CH 4 permeabilities, as shown in Figure 5a,b. The suggested gas transport model of porous filler addition; for fillers such as ZIF-8 [34,53], UiO-66 [26], HKUST-1 [54] is presented in Figure 6b, wherein a superlative case, the porous fillers will selectively enhance the permeation of smaller gas molecules and thus increase the overall MMM selectivity. Regarding the ideal case of MMM without any interface defect morphologies [9], the addition of porous selective filler should increase the gas permeability and its selectivity. Still, our observation differs from these findings (see Figure 5c). Thus, it can be suggested that there is an occurrence of non-selective interface voids, which causes the permeability increase and selectivity decrease. This phenomenon was suggested to be caused by the poor filler-polymer interaction and the polymer matrix's inability to fully surround the large filler particles (especially agglomerates), which both lead to the non-selective permeation pathways [55]. In the highest degree of voidage, it may lead to a leaking phenomenon where the MMMs completely lost their selectivity [9]. In a very recent study, Knebel et al. [56] showed the leaking phenomenon in MMM can be resolved by using a ZIF-based porous liquid [57]. The author also ensured good filler distribution even at high particle loadings (20-50 wt.%) by using ZIF-67-IDip, a type III porous liquid in 6FDA-DAM [56,57].
In both GO-and ZIF-8-addition, we believe that the gas separation observation well agrees with their SEM images (see Figures S5 and S6), where the GO MMMs show a firm, continuous and non-disruptive phase whilst the ZIF-8 MMMs show gaps and defective interfaces around the fillers, as previously discussed.
To further understand the potential of the 6FDA co-PI MMM, we explored a third possibility by making MMM with 1 wt.% of GO and varied ZIF-8 contents (5-15 wt.%). The obtained results of gas separation measurements are presented in Figure 7.
With the addition of 1 wt.% of GO and 5 wt.% of ZIF-8, the CO 2 permeability was increased to 1607.2 ± 17.6 Barrer and the CH 4 permeability to 40.8 Barrer, translated into 990% and 1216% increment respectively. The simultaneous high increase of CO 2 and CH 4 permeabilities is related to the interface void formation (as discussed above), indicates that the contribution of 1 wt.% GO in the binary filler MMM is marginal to improve its CO 2 /CH 4 selectivity (reduced by 17% to a value of 39.4 ± 1.3), compared to the improved selectivity in 1 wt.% GO MMM. Further increase of ZIF-8 loading to 10 wt.% shows a similar effect where both gas permeabilities increase with a further reduction of CO 2 /CH 4 selectivity. At the highest ZIF-8 loading (15 wt.%), the partial filler blockage and interface rigidification (other types of defective MMM morphology) can be concluded to be presence [9,26] where the gas permeability is reduced due to the blocked or stunted permeation pathways, possibly caused by the large agglomerates and polymer rigidification. The reduction instantaneously improves the overall gas selectivity by 33% (α CO2/CH4 = 28.2 ± 0.8), compared to the previous loading MMM, as presented in Figure 7b. All the gas permeation numerical data are summarized in Table S1. To further understand the potential of the 6FDA co-PI MMM, we explored a third possibility by making MMM with 1 wt.% of GO and varied ZIF-8 contents (5-15 wt.%). The obtained results of gas separation measurements are presented in Figure 7. With the addition of 1 wt.% of GO and 5 wt.% of ZIF-8, the CO2 permeability was increased to 1607.2 ± 17.6 Barrer and the CH4 permeability to 40.8 Barrer, translated into 990% and 1216% increment respectively. The simultaneous high increase of CO2 and CH4 permeabilities is related to the interface void formation (as discussed above), indicates that the contribution of 1 wt.% GO in the binary filler MMM is marginal to improve its CO2/CH4 selectivity (reduced by 17% to a value of 39.4 ± 1.3), compared to the improved selectivity in 1 wt.% GO MMM. Further increase of ZIF-8 loading to 10 wt.% shows a similar effect where both gas permeabilities increase with a further reduction of CO2/CH4 selectivity. At the highest ZIF-8 loading (15 wt.%), the partial filler blockage and interface rigidification (other types of defective MMM morphology) can be concluded to be presence [9,26] where the gas permeability is reduced due to the blocked or stunted permeation pathways, possibly caused by the large agglomerates and polymer rigidification. The reduction instantaneously improves the overall gas selectivity by 33% (αCO2/CH4 = 28.2 ± 0.8), compared to the previous loading MMM, as presented in Figure 7b. All the gas permeation numerical data are summarized in Table S1. Figure 8 shows the performances of the best 6FDA-DAM:DABA (3:1) membranes obtained in this study, with 1 wt.% GO, 10 wt.% ZIF-8 and mixture of 1 wt.% GO and 5 wt.% ZIF-8, against the CO2/CH4 upper bounds [4, 58,59]. As seen, the addition of the GO/ZIF-8 mixture pushes the 6FDA-DAM:DABA (3:1) co-PI to perform greater than the 2008 upper bound [58] and closing to the current one [59]. Additionally, for comparison, the MMM also performed better than the co-PI with other fillers such as small pore size zeolite, SSZ-16 [31] and 3D disordered mesoporous silica (DMS) [60]. Furthermore, it also performs better than other 6FDA-based polyimides with ZIF-8 [34,61,62].  Figure 8 shows the performances of the best 6FDA-DAM:DABA (3:1) membranes obtained in this study, with 1 wt.% GO, 10 wt.% ZIF-8 and mixture of 1 wt.% GO and 5 wt.% ZIF-8, against the CO 2 /CH 4 upper bounds [4, 58,59]. As seen, the addition of the GO/ZIF-8 mixture pushes the 6FDA-DAM:DABA (3:1) co-PI to perform greater than the 2008 upper bound [58] and closing to the current one [59]. Additionally, for comparison, the MMM also performed better than the co-PI with other fillers such as small pore size zeolite, SSZ-16 [31] and 3D disordered mesoporous silica (DMS) [60]. Furthermore, it also performs better than other 6FDA-based polyimides with ZIF-8 [34,61,62].   Figure 9 shows the gas separation performances of the neat 6FDA-DAM:DABA (3:1) and its GO/ZIF-8 MMMs when tested with 50 vol.% of CO 2 in binary CO 2 :CH 4 feed mixture at 2-8 bar transmembrane pressure, at 25 • C. In all samples, a continuous decrease in CO 2 permeability is observed when the transmembrane pressure increases. The reduction is related to the well-known dual sorption model of the membrane and has been presented countlessly over the years [23,28,63,64]. When the feed pressure is increased, the polymer's CO 2 solubility rises, leading to polymer free volume saturation. The CO 2saturated polymer leads to a reduced diffusion due to reduced gas transport driving force in the membrane matrix and consequently affects the permeability. Interestingly, when comparing CO 2 permeability reductions between 2 and 8 bar of the neat membrane (-38%) to the MMMs (−34-36%), a similar reduction percentage is presented (Figure 9a). This could be explained by the fact that the membranes possess a high degree of interface defects (evidenced by its gas permeation behavior and FESEM images), and the defects greatly influence the overall permeation, where the gases more likely to only permeate through the voids rather than selectively through the fillers. This prevents a precise observation of the positive effects of ZIF-8 addition in the MMM. Several studies had shown that several MMMs with porous fillers (e.g., SSZ-16 in 6FDA-DAM:DABA [31], UiO-66 in 6FDA-DAM [26]), when not defective, demonstrated lesser CO 2 permeability reduction at high pressure, due to the presence of 'strong' filler-polymer interface interaction.  Table S2.

Performance at Various CO 2 Partial Pressure
(a) (b) (c)

Conclusions
In this study, we presented a high-performance 6FDA-copolyimide for CO2/CH4 separation. This polymer's gas transport properties were investigated and further enhanced with the incorporation of GO nanosheets, ZIF-8 nanoparticles, and a mixture of the two fillers. GO addition presented both CO2 permeability and CO2/CH4 selectivity improvement at low loading, whereas the ZIF-8 incorporation resulted in tremendous permeability improvement at the expense of its selectivity. The findings indicate that the MMMs consist of interfacial defects that act as non-selective gas permeation pathways. With the use of GO/ZIF-8 filler mixture, similar shortcomings were encountered. Nevertheless, the MMM performed close to the 2019 CO2/CH4 performance upper bound and showed good performance stability when tested at different feed pressure. Based on these outcomes, it further emphasizes the importance of non-defective MMM morphologies for small molecule separation and strengthening our knowledge in the MMM field.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1. Figure S1. The illustration of preparation procedure of co-PI membranes; Figure S2. The illustration of preparation procedure of co-PI with GO MMMs; Figure S3. The illustration of preparation procedure of ZIF-8/co-PI MMMs; Figure S4. The illustration of preparation procedure of ZIF-8/GO/co-PI MMMs; Figure S5.  For the pressure dependence of the CH 4 permeability (Figure 9b) over the measured pressure range, a similar case applies with regards to the continuous CH 4 permeability reduction (as observed in the neat membrane). However, at 4-8 bar, all the MMMs showed increased permeability and it is more prominent in 5 wt.% and 15 wt.% ZIF-8 MMMs by 59% and 70%, respectively, whereas it is only 3% increase for 10 wt.% ZIF-8 MMM. As already known, the dissolved CO 2 promotes chain mobility (which can be explained by dynamic swelling of glassy polymer matrices upon CO 2 exposure [65]). It increases the polymer free volume pathways, causing the lower permeating CH 4 to permeate faster [39,63]. This also suggests that the effect of chain mobility induced by absorbed CO 2 is more prominent in the readily disrupted polymer chains in MMMs and the absence of good filler-polymer interactions. The CH 4 permeability increase directly influences the higher CO 2 /CH 4 selectivity reductions in the said membranes (−59% in 5 wt.% ZIF-8 MMM and -70% in 15 wt.% MMM), compared to the neat (−12%) and 10 wt.% MMM (−28%) (Figure 9c). We can safely conclude that MMMs with 5 wt.% and 15 wt.% are more defective with substantial interface voids than the 10 wt.% MMM. The permeation numerical data are presented in Table S2.

Conclusions
In this study, we presented a high-performance 6FDA-copolyimide for CO 2 /CH 4 separation. This polymer's gas transport properties were investigated and further enhanced with the incorporation of GO nanosheets, ZIF-8 nanoparticles, and a mixture of the two fillers. GO addition presented both CO 2 permeability and CO 2 /CH 4 selectivity improvement at low loading, whereas the ZIF-8 incorporation resulted in tremendous permeability improvement at the expense of its selectivity. The findings indicate that the MMMs consist of interfacial defects that act as non-selective gas permeation pathways. With the use of GO/ZIF-8 filler mixture, similar shortcomings were encountered. Nevertheless, the MMM performed close to the 2019 CO 2 /CH 4 performance upper bound and showed good performance stability when tested at different feed pressure. Based on these outcomes, it further emphasizes the importance of non-defective MMM morphologies for small molecule separation and strengthening our knowledge in the MMM field.