A study of CO 2 /CO separation by sub-micron b -oriented MFI membranes

Separation of CO 2 and CO is of great importance for many industrial applications. Today, CO 2 is removed from CO mainly by adsorption or physical or chemical absorption systems that are energy-intensive and expensive. Membranes are listed among the most promising sustainable and energy-e ﬃ cient alternatives for CO 2 separation. Here, we study CO 2 /CO separation by novel sub-micron b -oriented MFI zeolite membranes in a temperature range of 258 – 303 K and at a feed pressure of 9 bar. Under all experimental conditions studied, the membranes were CO 2 -selective and displayed high CO 2 permeance ranging from 17 000 to 23 000 gpu. With decreasing temperature, the CO 2 /CO selectivity was increasing, reaching a maximum of 26 at 258 K. We also developed a mathematical model to describe the membrane process, and it indicated that the membrane separation performance was a result of selective adsorption of CO 2 on the polar zeolite. The heat of adsorption of CO 2 on the zeolite is more negative due to the high quadrupole moment and polarisability of the molecule as compared to CO. At the same time, di ﬀ usional coupling (correlation e ﬀ ects) at high adsorbed loadings was found to favour the overall CO 2 /CO selectivity of the membranes by reducing the di ﬀ usivity of the lighter CO molecule in the ca. 0.55 nm pores in the zeolite. The model also indicated that the separation performance was limited by the mass transfer resistance in the support and concentration polarisation on the feed side of the membrane.


Introduction
CO is used for manufacture of a wide range of valuable chemicals, such as oxo-alcohols, phosgene (intermediate for polyurethane), methanol, acetic acid or various liquid hydrocarbon fuels via the Fischer-Tropsch process. CO itself is oen produced by gasication (partial oxidation) of carboncontaining materials, e.g. coal, biomass, etc., or steam reforming of natural gas. In addition to CO, the produced gas mixture, referred to as synthesis gas, contains H 2 , CO 2 and minor impurities, e.g. N 2 , H 2 O, H 2 S, etc. To utilise for the synthesis of chemicals, CO usually has to be puried, mainly from CO 2 . For instance, the gas used for oxo-synthesis should contain #0.5% of CO 2 . 1 Separation of CO and CO 2 is, however, challenging since the properties of the two gases are rather similar. Most of the currently available CO 2 separation methods rely on adsorption or absorption, e.g. pressure swing adsorption or amine scrubbing. 2 Despite certain advantages, the sorptionbased separation systems are cumbersome, rather energyintensive and require large capital investments. 3 Moreover, the absorbents used today are oen corrosive and may release hazardous substances into the environment.
Membranes are an appealing alternative to many common separation technologies, including adsorption, absorption and cryogenic distillation. High efficiency, sustainability and low energy demand are dening characteristics of membrane-based separation methods. Moreover, membrane separation processes are simple and continuous, requiring a minimum of process equipment. For instance, we have recently demonstrated 4 that only one module housing 10 m 2 of membrane area could replace an entire amine scrubbing system for separation of 300 ton CO 2 per day from synthesis gas. It is worth pointing out that membranes have been listed among the most promising CO 2 separation and capture technologies. 3 The CO 2selective membranes currently available on the market are polymeric membranes. These membranes can be manufactured in large quantities at fairly low costs. At the same time, large membrane areas are a necessity for a given separation task because the membranes have low permeability, oen coupled with a fairly poor selectivity. For example, Polaris™ membranes have a CO 2 permeance of 100-300 gpu (1 gpu ¼ 3.35 Â 10 À10 mol s À1 m À2 Pa À1 ) and a CO 2 /CO selectivity of 10-20 at a temperature ranging from 268 to 298 K. 5 Even state-of-the-art CO 2 -selective polymeric membranes display a CO 2 permeance of at most 1000 gpu. 6 In addition, CO 2 has a detrimental effect on the membrane stability, especially at high concentrations (pressures), reducing the membrane performance and lifespan. 3 Among inorganic CO 2 -selective membranes, zeolite membranes are particularly attractive. 7 These membranes have a well-dened pore system with pores ranging from 0.3 to 1.3 nm in size. 8 Hence, zeolite membranes are microporous membranes. Due to the microporous structure, the separation mechanism by zeolite membranes can rely on molecular sieving (size exclusion), differences in adsorption or differences in diffusion between the separated components. Being highly porous and thin, zeolite membranes can exhibit much higher uxes than polymeric membranes. 9 Currently, zeolite lms as thin as 100 nm can be prepared, displaying a high n-butane permeance up to ca. 4000 gpu. 10 As a consequence, much lower membrane areas would suffice for a given separation task. In addition, the chemical and thermal stability of zeolite membranes is vastly superior to that of polymeric membranes. 11 In spite of the great interest in separation of the main components of synthesis gas, i.e. CO, H 2 and CO 2 , the number of research reports on zeolite membranes for this separation is small. 7 In particular, CO 2 /CO membrane separation has scarcely been studied. Our research group 13 has previously evaluated a sub-micron randomly oriented highly siliceous MFI zeolite membrane for the separation of an equimolar mixture of CO 2 , H 2 and CO. The membrane was CO 2 -selective with a high CO 2 permeance up to ca. 24 000 gpu. At a temperature of 277 K and a feed pressure of 9 bar, a CO 2 /CO membrane selectivity of ca. 10 was reported. The highly siliceous MFI zeolite (also referred to as siliceous ZSM-5 with a unit cell represented by the general chemical formula [Si 96 O 192 ]-MFI) 14 has a threedimensional pore structure comprised of two types of intersecting pores, viz. tortuous and straight with an average pore diameter of ca. 0.55 nm. 12 The tortuous (sinusoidal) pores are running in the a-direction, whereas the straight pores are running in the b-direction, see Fig. 1. Molecules can also diffuse in the c-direction by alternating between the straight and sinusoidal pores. However, the diffusion in the c-direction is signicantly slower than in the a-and b-directions. 15,16 It has been shown 14,17 that b-oriented MFI lms may allow for somewhat faster transport of molecules, especially bulky molecules (e.g. xylenes) since the straight pores will be running along the lm thickness. The b-orientation has therefore been considered preferable.
Recently, we 18 have developed a sub-micron b-oriented MFI membrane employing a uoride route (neutral pH) for the synthesis. The developed membranes displayed high CO 2 /H 2 and CO 2 /CO selectivities up to ca. 110 and 20, respectively. Here, we extend our previous work on CO 2 /CO separation using submicron b-oriented MFI membranes by developing a model to elucidate the separation mechanism and describe the effect of the support and concentration polarisation on the membrane separation performance. In addition, the morphology of the membranes, especially the grain boundaries and grain size, was carefully studied by SEM. Permporometry data were used to estimate the width of the open grain boundaries and the grain size, and the ndings were compared with the SEM data.

Membrane synthesis
Sub-micron (ca. 0.5 mm) b-oriented MFI zeolite membranes were prepared as described in detail in our previous work. 18 The membranes were supported on a commercial graded a-alumina disc (Fraunhofer IKTS, Germany) with a diameter of 25 mm. A brief summary of the membrane preparation procedure is given below. In the rst step, the supports were seeded with plateshaped MFI crystals with a size of 500 Â 450 Â 200 nm. The seed crystals were synthesised from a mixture containing tetraethyl orthosilicate (TEOS), tetrapropylammonium hydroxide (TPAOH) and water with a respective molar ratio of 1 : 0.2 : 100. The mixture was kept at 403 K for 9 h. In the second step, zeolite lms were grown on the seeded supports by hydrothermal synthesis at 373 K for 48 h using a synthesis mixture with a molar composition of SiO 2 : 0.12TPAOH : 60H 2 O : 0.12HF. Aer the synthesis, the membranes were rinsed with a 0.2 M ammonia solution and dried at 323 K overnight. The following day, the membranes were calcined at 773 K for 6 h at a heating rate of 0.2 K min À1 and a cooling rate of 0.3 K min À1 .

Membrane characterisation
The morphology of the membranes was characterised by Scanning Electron Microscopy (SEM) using a Magellan 400 SEM (the FEI Company, The Netherlands). Cross-sections of the membranes were obtained by fracturing the membranes with a pair of cutting pliers. No conductive coating was applied to the samples.
The quality of the membranes in terms of defects was evaluated by n-hexane/helium permporometry 19 as described in detail in our earlier work 20 and in brief here. The membranes were mounted in a stainless steel cell using graphite gaskets (Eriks, The Netherlands) for sealing. In order to remove any adsorbed compounds, the membranes were heated to 573 K at a heating rate of 1 K min À1 and kept at this temperature for 6 h in a ow of pure helium (99.999%, AGA). The permporometry experiment was performed at a temperature of 323 K, a total feed pressure of 2 bar and a total permeate pressure of 1 bar. The relative pressure of n-hexane (99%, Alfa Aesar) in the feed was increased in a step-wise manner from 0 to ca. 0.4. At each relative pressure, the system was allowed to equilibrate. In order to remove n-hexane from the permeate stream, a condenser kept at 233 K followed by a column packed with activated carbon were employed. A soap bubble ow meter was used to measure the n-hexane-free permeate volumetric ow rate. For each n-hexane relative pressure, a defect width was calculated by either the Horvàth-Kawazoe equation (micropore-range defects) or the Kelvin equation complemented by the Harkins-Jura equation accounting for the thickness of the adsorbed layer (mesopore-range defects). For each defect interval, the average defect width was then calculated. Based on the average defect width, the average helium diffusivity in each defect interval was estimated using the gas-translational model. Knowing the diffusivity, the helium molar ux was further calculated from Fick's law. The area of defects was estimated as the ratio between helium molar ow and ux through the defects in that particular interval. In the nal step, the relative area of defects was calculated by dividing the area of defects with the total membrane area. More details about the evaluation procedure of the permporometry data can be found in our previous work. 20 In order to identify the type of defects detected by permporometry, a geometrical model of the membrane microstructure was used. It is worth noting that this model is only a rough approximation of the microstructure. In the model, the grains were assumed to have a cuboid habit with a square cross-section with a width of x, see Fig. 2. It should be noted that the cuboid habit with a square cross-section was only assumed to simplify the calculation of the grain surface area from the permporometry data. The grain width was assumed to be constant throughout the lm thickness. The micropore defects with a weighted arithmetic average width of d were assumed to be evenly distributed between the grains, as shown in Fig. 2. Therefore, the relative area of micropore defects a can be calculated as Since d is much smaller than x, eqn (1) can be simplied to The relative area of micropore defects a can be estimated from permporometry data obtained experimentally. Knowing the relative area, the grain size x, and hence the grain surface area, can then easily be calculated using eqn (2).

CO 2 /CO separation experiments
Mixed-gas separation experiments were carried out using a 50/ 50 (v/v) mixture of CO 2 (99.999%, AGA) and CO (99.97%) at a total feed pressure of 9 bar and a total permeate pressure of 1 bar. The membrane was kept in the cell used for the permporometry test. The experiments were carried out at a temperature ranging from 258 to 303 K. Prior to the experiments, the membrane was dried at 573 K for 6 h using a stream of pure helium (99.999%, AGA) to remove any adsorbed species. The volumetric ow rate of the permeate was measured with a drumtype gasmeter (TG Series, Ritter Apparatebau GmbH), and the composition of the permeate was analysed with a mass spectrometer (GAM 400, InProcess Instruments).
The ux of component i, J i (mol s À1 m À2 ), was estimated from the measured molar ow rate of the corresponding component through the membrane, F i (mol s À1 ) as where A is the membrane area (m 2 ). The permeance of component i, P i (mol s À1 m À2 Pa À1 ), was calculated from the ux of the corresponding component through the membrane as where DP i (Pa) is the partial pressure difference of component i across the membrane. It should be noted that for gas permeation and separation, permeance is normally reported in gpu (gas permeation units), where 1 gpu is equal to 3.35 Â 10 À10 mol s À1 m À2 Pa À1 . The CO 2 /CO membrane selectivity, a CO 2 /CO , was estimated as

Modelling
For composite membranes, as in this work, high ux may cause both concentration polarisation on the feed side and a pressure drop over the support. These effects may result in a decrease in the performance of the membrane, in terms of both ux and selectivity. [21][22][23] To assess to what extent these effects may have inuenced the separation performance of the membrane in the present work, a mathematical model was used. The model has been described in detail elsewhere 21,23,24 and will only be described briey below.
The effect of concentration polarisation may be evaluated by determining the concentration polarisation index (CPI) as where n b is the mole fraction in the gas bulk, k c is the mass transfer coefficient, J v is the volumetric ux and n p is the mole fraction in the permeate. The mass transfer coefficient may be obtained from mass transfer correlations relating the Sherwood (Sh) number to the Reynolds (Re) and Schmidt (Sc) numbers. Perdana et al. 25 have reported a correlation for mass transfer in a Wicke-Kallenbach cell, which is used in the present work.
where w is the compartment height, i.e. the thickness of the gas lm on the feed side of the membrane and D C is the diameter of the membrane inside the gaskets, for more details see Perdana et al. 25 The mass transfer coefficient was nally retrieved from the Sherwood number: where d h is the hydraulic diameter of the cell and D is the gas phase diffusivity of CO 2 . The pressure drop across each layer of the support was determined using procedures described in detail elsewhere. 21,[26][27][28] The transport through the 3 mm thick base layer with 3 mm pores was assumed to occur via viscous ow, whereas transport through the 30 mm thick top layer was assumed to occur via a combination of viscous ow and Knudsen diffusion. Effective permeabilities and Knudsen transport coefficients for each of the support layer have been reported in our previous work. 21 Gas phase diffusivities and viscosities were adjusted to the desired temperature using the D i f T 3/2 relationship and Sutherland's equation, 29 respectively. The viscosities of the mixtures were calculated as molar fraction weighted averages. The ideal adsorbed solution theory (IAST) 30 was used to estimate adsorbed loadings on the feed side of the membrane during the separation experiment. Saturation loadings, Langmuir affinity coefficients and adsorption enthalpies were taken from the literature. [31][32][33][34] Results and discussion Membrane characterisation Fig. 3 shows cross-sectional and top-view SEM images of an assynthesised membrane. It is evident that the MFI crystals comprising the lm were mainly b-oriented. The MFI lm was even and continuous, and the crystal length in the b-direction, i.e. the lm thickness, was ca. 0.5 mm. The crystals composing the dense part of the lm appear to be well-intergrown with a smooth top surface. However, some of the crystals were sticking out of the dense part of the lm, displaying a typical rounded boat shape. 35 The in-plane dimensions of the crystals composing the dense part of the lm were found to vary considerably. However, the maximum lengths along the a-axis and c-axis were ca. 0.5 mm and 2 mm, respectively. Defects in the form of cracks and pinholes, which are normally larger than 5 nm, were not detected by SEM. The absence of such defects indicates high quality of the membrane. In Fig. 3a, part of the top layer of the support can also be observed. The grain size of the support is about 150 nm and the pore size is about 100 nm. No growth of the zeolite into the support, oen referred to as invasion, 36 could be detected, which is desired as the invasion would reduce the permeability of the support. The absence of the invasion can most likely be attributed to the good seeding, resulting in a dense seed layer protecting the support. The boriented MFI phase comprising as-synthesised membranes was corroborated by X-ray diffraction (XRD), and the data were reported in our previous work. 18 Table 1 shows the helium permeance through the synthesised membrane measured at each relative pressure of nhexane in the permporometry experiment. The table also shows the relative areas of defects estimated for each defect interval. In the permporometry experiment, the initial measuring point is recorded at a relative pressure of n-hexane of 0. The helium permeance at this point denotes the combined permeance through zeolite pores and defects. For this membrane, the initial helium permeance was high amounting to ca. 20 000 gpu, indicating that the zeolite pores are open and permeable. With increasing relative pressure of n-hexane, rst MFI pores and then defects were blocked by n-hexane causing the helium permeance to decrease. The total relative area of defects in the membrane was very small, scarcely above 0.1% of the total membrane area, indicating a very high quality of the membrane. The few detected defects were essentially only micropore defects, accounting for ca. 99.6% of all defects. Virtually no mesopore defects were detected by permporometry, which is consistent with the SEM observations.
In our previous work, we showed that the micropore defects detected by permporometry in our randomly oriented high-ux MFI membranes were open grain boundaries. In order to identify whether the detected micropore defects in the b-oriented MFI membranes prepared in the present work were as well open grain boundaries, we used a rough geometrical model of the membrane microstructure described in the Experimental. The average width of micropore defects d was estimated from the permporometry data as a weighted arithmetic mean to be 0.79 nm. For this particular membrane, the relative area of micropore defects a was ca. 0.13% yielding a grain size x of ca. 1.2 mm (see eqn (2)), and consequently a square grain with a surface area of ca. 1.4 mm 2 . In spite of the great simplicity of the model, the grain size estimated from the permporometry data was in excellent agreement with the grain sizes observed by SEM, i.e. the maximum lengths along the a-and c-axis of ca. 0.5 and 2 mm corresponding to a square grain with a surface area of 1 mm 2 . The latter indicates that the micropore defects detected by permporometry in the prepared membrane should mainly be open grain boundaries. Fig. 4 shows mixed-gas permeances of CO 2 and CO as a function of temperature. High mixed-gas permeances up to 23 000 gpu were observed, and the permeance of CO 2 was higher than that of CO in the entire temperature range. With decreasing temperature, the permeances of both CO 2 and CO were decreasing. Bakker et al. 37 studied temperature dependence of pure-gas permeation of various gases through MFI membranes. It was found that the diffusion of CO 2 and CO in the membranes was activated surface diffusion, for which the diffusivity is increasing with increasing temperature, following an Arrhenius-type expression. The same type of temperature dependence was also found by Kärger et al., 38 who studied the intracrystalline self-diffusion of CO 2 and CO in MFI zeolite by NMR. Thus, the decrease in permeance with decreasing temperature observed in the present work should emanate from the decrease in the diffusivities. It should be noted, however, that even at the lowest studied temperature of 258 K the permeance of CO 2 was as high as ca. 17 000 gpu. Fig. 5 shows CO 2 /CO mixed-gas membrane selectivity as a function of temperature. At the highest studied temperature  of 303 K, the CO 2 /CO selectivity was ca. 6, and it was increasing with decreasing temperature, reaching a maximum of 26 at the lowest studied temperature of 258 K. Since the size of both CO 2 (0.33 nm) and CO (0.38 nm) 37 is smaller than the MFI pore size (0.55 nm), the separation mechanism should rely on the differences in either adsorption or diffusion for the two molecules, and not the molecular sieving. To further elucidate the ndings, a modelling work was carried out, and the results will be discussed below. Table 2 reports the CO 2 and CO uxes, and the concentration of CO 2 and CO in the permeate stream. The observed CO 2 ux was very high, i.e. 320-440 kg m À2 h À1 , in the entire temperature range. The high CO 2 ux is most likely a result of the very low thickness of the zeolite lm, strong CO 2 adsorption and high CO 2 diffusivity in the zeolite pores, and a relatively high CO 2 partial pressure difference of 3.5 bar across the membrane.

CO 2 /CO separation experiments
In general, performance of supported zeolite membranes is dened by the properties of the active zeolite lm and the support. If the thickness of the lm is high (e.g. a few microns), the effect of the support on the membrane performance can be neglected since essentially all mass transfer resistance will occur over the zeolite lm. In other words, both the overall selectivity and ux of such a membrane will be dened by the properties of the lm, which is preferable. However, the high lm thickness results in very low ux, 7 making the membrane impractical. In contrast, if the lm thickness is low (i.e. below 1 mm), high uxes, sufficient for a practical application, can be achieved, but the membrane performance may be inuenced by the support, as its contribution to the overall mass transfer becomes signicant. Roughly speaking, the lower the lm thickness, the stronger the support may inuence the membrane performance. The effect of the support can also be enhanced if the separated components are small, highly permeable molecules, such as CO 2 and CO, as their transport through the lm will be even faster. Since the support is normally non-selective or sometimes adversely selective, its contribution to the mass transfer will reduce the separation performance of the membrane. Hence, the effect of the support on the performance of high-ux membranes should be taken into account. In our previous works, [21][22][23] we showed that the pressure drop over the support adversely affected the performance of our high-ux membranes for liquid separation of alcohols and water, and gas separation of helium and nitrogen at cryogenic temperature. It is therefore important to study the effect of the support on CO 2 /CO separation reported in the present work. In addition, high-ux membranes may suffer from concentration polarisation on the feed side, reducing the driving force over the lm, and therefore selectivity and ux. 23 In the present work, we have studied the effects of these two phenomena for three different temperatures, viz. 303, 278 and 258 K. The selected points represent high, medium and low ux, respectively. Table 3 shows the results of the evaluation. The relative pressure drop over the support, DP, (i.e. pressure drop over the support/pressure drop over the whole composite membrane) was estimated to be 33%, 27% and 21% at 303, 278 and 258 K, respectively. These values are consistent with our previous ndings for high-ux MFI zeolite membranes, [21][22][23] indicating that the support adversely affected the performance of the membrane. Most of the resistance was found to be in the thin top layer of the support having 100 nm pores. Therefore, in order to improve the membrane separation performance, the negative effect of the support should be eliminated or at least minimised. One feasible way of minimising this effect would be to reduce the thickness of the top layer from 30 to, e.g. 5-10 mm. 24 The concentration polarisation index was determined to be 0.83, 0.87 and 0.90 at 303, 278 and 258 K, indicating that concentration polarisation reduced the performance of the   membrane, especially at the higher temperatures, where the ux was higher. In order to minimise the concentration polarisation effect and improve the membrane performance, higher feed ow rates would be required. Additionally, turbulence generating internals in the membrane cell could be used to improve the mass transfer on the feed side. It is also worth noting that if the lm thickness were even lower, the negative effect of the support and concentration polarisation on the separation performance would be even higher, and it would thus be even more important to minimise these effects. The permselectivities for the zeolite lm aer correcting for concentration polarisation and pressure drop over the support, a perm,lm , were ca. 15, 24 and 41 at 303, 278 and 258 K, respectively. These values differ from the overall selectivities, ca. 6, 14 and 26, measured at the same temperatures (see Fig. 5). It is therefore evident that the membrane selectivity was signicantly reduced by concentration polarisation and pressure drop over the support.
To better understand the experimental ndings, we estimated the adsorption selectivity on the feed side of the membrane using the IAST, aer removing the effect of the support and concentration polarisation, see Table 3. The CO 2 /CO adsorption a ads selectivity was increasing with decreasing temperature. This is expected because the heat of adsorption of CO 2 (À24.1 kJ mol À1 ) is more negative than that of CO (À17.9 kJ mol À1 ) due to the high quadrupole moment and polarisability of the CO 2 molecule as compared to CO. 32 This also shows that the adsorption of CO 2 is stronger than the adsorption of CO. Moreover, the estimated adsorption selectivities were very similar to the permselectivities. The latter fact indicates that the membrane selectivity to CO 2 should mainly be ascribed to the stronger adsorption of CO 2 in the zeolite lm. It should also be noted that the adsorption selectivity can probably be increased by increasing the polarity of the zeolite, 39 i.e. by decreasing the Si/Al ratio, resulting in even higher membrane selectivity to CO 2 . Further, the permselectivity for zeolite membrane may be expressed as the product of the adsorption and diffusional selectivities: The diffusional selectivities may therefore be retrieved from the experimental data shown in Table 3. The diffusional selectivity only varies marginally, from 0.82 at the highest temperature to 0.89 at the lowest temperature. Only a few reports exists on transport of CO through MFI membranes, however Bakker et al. 37 have determined pure-gas diffusivities and activation energies of diffusion of CO 2 and CO in silicalite-1 from pure-gas membrane permeation data. For CO 2 and CO, they reported pure-gas diffusivities of 0.7 Â 10 À8 and 0.9 Â 10 À8 m 2 s À1 and activation energies of diffusion of 9.6 and 7.1 kJ mol À1 , respectively. Noticeably, the ratio of the pure-gas diffusivities (D CO 2 /D CO ) yields 0.78, which is strikingly close to the diffusional selectivity of ca. 0.82 that we obtain in the present work at the higher temperatures. As the activation energy of diffusion for CO 2 is larger than that for CO, the diffusivity of CO 2 should decrease faster than that of CO with decreasing temperature. Consequently, the ratio of pure-gas diffusivities should also decrease with decreasing temperature, e.g. by 19% for a temperature decrease from 299 to 248 K. However, our diffusion selectivity was slightly increasing with decreasing temperature, from ca. 0.82 to 0.89. We ascribe this to increased diffusional coupling (correlation effects) at reduced temperatures. 39 As the adsorbed loading of CO 2 increases, it becomes more effective in slowing down the diffusion of CO (the faster diffusing species), thereby leading to a slight increase in the diffusional selectivity with decreasing temperature. It is also worth noting that since the kinetic diameter of CO (0.38 nm) is larger than that of CO 2 (0.33 nm), the overall CO 2 /CO selectivity can probably be increased further by using a smaller-pore zeolite, e.g. CHA zeolite, due to the additional molecular sieving effect favouring the permeation of CO 2 . It would thus be interesting to develop high-ux CHA zeolite membranes for this separation.
To conclude, the adsorption selectivity was shown to be the dominating factor for the permselectivity observed in this work. In addition, at the high adsorbed loadings obtained at the lower temperatures, diffusional correlation effects also favour the overall CO 2 /CO selectivity, by keeping the diffusional selectivity up (closer to unity).

Conclusions
Sub-micron b-oriented high-ux MFI zeolite membranes were prepared and carefully characterised by permporometry and SEM. By comparing the permporometry and SEM data, the micropore defects detected in the membranes by permporometry were found to be open grain boundaries. The membranes were then evaluated for CO 2 /CO separation at a temperature ranging from 258 to 303 K, a feed pressure of 9 bar and a permeate pressure of 1 bar. The membranes were CO 2 -selective in the entire temperature range. The highest CO 2 /CO selectivity of 26 was observed at the lowest studied temperature of 258 K. The selectivity of the membrane to CO 2 was attributed to the stronger CO 2 adsorption, which was supported by a mathematical model based on IAST. The adsorption selectivity, and hence the overall CO 2 selectivity can probably be increased further by increasing the polarity of the zeolite. It was also found that diffusional coupling (correlation effects) at high adsorbed loadings resulted in a slight increase of the diffusional selectivity, keeping it closer to unity. In addition, the model indicated that the separation performance of the membrane was reduced by the pressure drop across the support and concentration polarisation on the feed side. Hence, in order to improve the separation performance, these effects should be minimised by, for instance, reducing the thickness of the support and increasing the feed ow rates. Without these effects, a CO 2 /CO selectivity of 41 would be achieved at the lowest studied temperature of 258 K.