DD3R zeolite membranes in separation and catalytic processes: Modelling and application

Around 2004 the annual energy consumption of the Dutch (petro-)chemical industry was estimated to be 460 PJ of which 200 PJ could be allocated to separation processes [1]. In 2009, 15% of the global energy consumption was required for separation and purification processes to produce commodities. Moreover, it is expected that in 2040 the global commodity demand is three times higher than in 2009 leading to an enormous energy demand increase in the coming decades related to separation processes [2]. These two examples clearly illustrate the need for the development of new innovative energy-efficient separation technologies. Membrane technology is considered a serious candidate to replace traditionally used thermally-driven separation processes, because of the large energy reduction that can be achieved [2-4]. An application particularly relevant for this thesis, is natural gas purification, which is by far the largest industrial gas separation application [5]. The focus of this thesis has been on a DD3R zeolite membrane. Zeolites are crystalline aluminosilicates with pores of sub-nanometer dimensions. Specific advantages of zeolite membranes are their high thermal and chemical stability and their molecular sieving ability. Two key aspects differentiate DD3R from other zeolites that make this zeolite particularly interesting to study: its small pore size and the possibility to synthesize it in an all-silica form. The 8-ring window has approximate dimensions of 0.36 ? 0.44 nm which makes this material very interesting for separation of light gases which is not possible with larger-pore zeolites, like zeolite MFI. Other 8-ring zeolites are available, however very few have been synthesized successfully in all-silica form. Since it appears to be extremely challenging to make high quality membranes for gas separation from aluminum containing zeolites [6], the all-silica nature of DD3R is a clear advantage. Anticipating the special properties of the DD3R topology in gas separation, the goal of the project has been to study the application of DD3R zeolite membranes in separation and catalysis. Special attention is paid to the understanding and modeling of the mass transport across such a membrane. The thesis objective has been approached by performance testing of a disc- and tubular-shaped DD3R membrane supplied by NGK-insulators in several gas separations and in one reactive separation: the dehydrogenation of isobutane in a DD3R zeolite membrane reactor. The permeation properties of a series of light gases and mixtures thereof have been analyzed as a function of temperature and pressure and compared to mass transport mechanisms available in literature. Because currently available mass transport models did not lead to satisfactory results a new approach to describe diffusion in zeolites has been proposed (the Relevant Site Model). Diffusivity data calculated from MD simulations have been used to develop and verify this new model. Permeation and separation characteristics (Chapter 2 & 3) Adsorption experiments of a series of light gases on DD3R crystals revealed the following order in amount adsorbed at 303 K and 120 kPa: CO2 = N2O >> Kr ? CH4 > CO > N2 ? O2 ? Ar > H2. In case of Ne no significant adsorption could be detected. These isotherms could be modelled well by a single or dual-site Langmuir isotherm. The permeation and separation characteristics of light gases through DD3R membranes can be explained by taking into account: (1) steric effects introduced by the window opening of DD3R leading to molecular sieving and activated transport (Figure 1), (2) competitive adsorption effects, as observed for mixtures involving strongly adsorbing gases, and (3) momentum exchange between diffusing molecules in the zeolite. Momentum transfer appears only relevant in mixtures below 373 K that involve strong adsorbing component (CO2 or N2O). Competitive adsorption is also found only in case of mixtures that involve strong adsorbing components. Suppression of the flux of the weakly adsorbing component can be very strong and becomes manifest at low temperatures ( 500 @ 303 K) and high CO2 fluxes, confirming the results of Tomita et al. [8]. The selectivity decreases with increasing total feed pressure, but up to 1500 kPa total feed pressure the selectivity remains above 100 up to 373 K. The N2/CH4 separation factor is quite good, but the low N2 flux is clearly limiting successful application. Since the separation is predominantly due to molecular sieving, the O2/N2-separation factor is relatively low (? 2). Interestingly, CO2 and N2O behave (almost) identical in DD3R and cannot be separated (selectivity ~ 1). The ideal H2/CO and CO2/CO selectivities range from 3 to 12 and 10 to 2 between 303 and 673 K, respectively. These mixture selectivities were always below 5 and much lower than the ideal selectivities because of non-differential operation of the tubular membrane. The H2/isobutane mixture selectivity at 101 kPa total feed pressure is ~ 400 in an equimolar binary mixture over a broad temperature range (303 – 773 K). The high separation factor is due to exclusion of isobutane from the DD3R pores. Isobutane dehydrogenation (Chapter 4) Alkane dehydrogenation reactions are industrially very relevant, but they are also class of reactions where the conversion can be (severely) equilibrium-limited at practical high temperature conditions necessary to perform the reaction. Low conversions lead to a large flow of alkane/alkene mixtures that needs to be separated and recycled. Particularly the separation of alkanes/alkenes is very energy intensive [9]. An approach to increase the single-pass conversion is by using a membrane reactor (MR). By in situ removal of the product H2 an apparent equilibrium shift can be accomplished. Because of the excellent H2/isobutane selectivity and a reasonable H2 permeance (~ 4.5?10-8 mol m-2 s-1 Pa @ 773 K) the dehydrogenation of isobutane has been studied in a DD3R zeolite membrane reactor (MR) at 712 and 762 K. Experiments in a conventional packed bed reactor (PBR) served as benchmark, Cr2O3 on Al2O3 is used as catalyst. At low residence times isobutene yields above the equilibrium yield based on feed conditions could be obtained. At 762 K and 0.13 kgfeedkgcat-1h-1, the isobutene yield in the membrane reactor (MR) is 0.41, where the equilibrium yield is ~ 0.28. The increased yield is attributed to removal of H2 from the reaction zone by the membrane. The removal of H2 mildly promotes coke formation, suppresses hydrogenolysis reactions and slightly reduces the catalyst activity. During several months of high temperature operation the membrane quality did not change notably. The membrane permeation parameters and reaction rate constants have been estimated independently from membrane permeation and PBR experiments, respectively. From these parameters the behaviour of the MR can be simulated well. Two important dimensionless parameters determine the MR performance primarily, the Damköhler (Da) and membrane Péclet number (Pe?). For a significant improvement of the MR performance as compared to a PBR Da ? 10 and Pe? ? 0.1. DaPe? should be ? 1 to optimally utilize both catalyst and membrane activity. In the current MR design both the hydrogen removal capacity and catalyst activity stand in the way of successful application. Using a more active catalyst and a more favourable area to volume ratio could greatly improve the MR performance. Operation at a higher feed pressure could be a possible solution. Since membranes with higher fluxes are already available, it is the limited catalyst activity and stability under relative low temperature and H2 lean conditions that is an important limiting factor regarding application of MRs in dehydrogenation reactions. Mass transport modelling: The Relevant Site Model (Chapter 5) Together with excellent separation results a remarkable strong loading dependency of diffusion of light gases in DD3R is found. It is well known that the diffusivity in zeolites is dependent on the loading, but simulation and experimental data point out that especially for small-pore 8-ring zeolites very strong loading dependencies of the diffusivity are quite common [10-12]. This strong loading dependency complicates macroscopic modelling of permeation behaviour of such zeolite membranes, which is required for module design. The best model currently available, based on the so-called Reed Ehrlich approach, turned out only modestly successful to model CH4/CO2 and CH4/N2 mixture permeation across an all-silica DD3R membrane [13]. Therefore, a new model has been introduced in this thesis to capture the loading dependency of the diffusivity in zeolites in the formulation of a macroscopic transport model. The model is formulated around the idea of segregated adsorption in cage-like zeolites, i.e. that molecules are located either in the cage or in the window. Furthermore, it is assumed that only the molecules located at the window site are able to make a successful jump to another cage. This so-called Relevant Site Model (RSM) is based on the Maxwell-Stefan framework for mass transport but includes one extra parameter that describes the adsorption properties of the ‘relevant site’. Key feature of the RSM as applied to mixtures is that competitive adsorption effects and ‘speeding up and slowing down’ (momentum exchange) effects between guest molecules are related to the RS loading instead of the overall loading, which can be very different. In addition to the RSM the concept of free space relevant for diffusion has been introduced. Because the diffusivity often approaches very small values when the loading in the zeolite approaches its saturation loading a ‘confinement’ factor is introduced to indicate the available free space. Now it is argued that due to, for example, side pockets or positional rearrangements not all free space is relevant for diffusion. A method has been put forward to account for these effects in modelling work. Application to zeolite DDR (Chapter 5B & C) Firstly, the RSM has been applied to a set of single component diffusivity data of CO2 and N2 in DDR computed using molecular dynamic (MD) simulations. The RSM describes the Maxwell Stefan diffusivity data very well up to saturation. The observed strong diffusivity loading dependency is explained by the relative low window site occupancy that is typically much lower than the total occupancy at lower loadings. The RSM is successfully applied to non-isothermal diffusivity data of CO2 and N2 in DDR. Relating intermolecular correlation effects (momentum exchange) to the RS occupancy instead of the total occupancy leads to a quantitative prediction of the observed correlation effects and, consequently, the self diffusivity. Analysis of the N2 data suggests positional rearrangements in the DDR cages in a certain loading range. These effects have been incorporated in the model successfully using the concept of free space relevant for diffusion. Then, the RSM has been subjected to an extensive set of diffusivity data of N2/CO2 and Ne/Ar mixtures in zeolite DDR, directly computed using molecular dynamics. A large part of the considered data has been taken from literature [14]. It has been shown that the RSM provides excellent mixture diffusivity predictions from single component diffusivity data. The results are comparable to the ‘Reed-Ehrlich’ approach as put forward by Krishna and co-workers. A clear improvement by the RSM is found in the case of the N2 diffusivity in N2/CO2 mixtures (Figure 6), attributed to the specific window blocking effect by CO2 which is inherently incorporated in the RSM by relating adsorption to the relevant (=window) site. Extension to other zeolites (Chapter 5D) After the successful application of the RSM to describe the loading dependency of diffusion in zeolite DDR it has also been successfully applied to a variety of light gases (CH4, CO2, Ar and Ne) and binary mixtures thereof in other zeolite topologies, DDR, CHA, MFI and FAU, utilizing the extensive diffusivity dataset published by Krishna and van Baten for this variety of zeolite-guest systems (e.g. [14])[Chapter 5D]. From the RS approach a measure for the level of adsorption segregation is derived: the ratio of the RS and total occupancy. The predicted level of adsorption segregation correlates well with the level of confinement of a molecule at the RS: the ratio of molecule diameter to zeolite pore diameter. Moreover, the predicted degree of adsorption segregation of the studied light gases in DDR is in good agreement with molecular simulations results, indicating the physical meaningfulness of the estimated RS adsorption parameters. The binary mixture diffusivity modelling points out that in case of the small-pore zeolites (DDR and CHA) the data is described best with equal RS saturation loadings for both components. For the large pore zeolite FAU the ratio of the RS saturation loadings equals that of the bulk saturation loadings. The geometry of the RS strongly influences the RS saturation loading: in case of the small-pore zeolites the RS (= window site) is restricted to only one molecule but when the RS becomes larger more then one molecule can be found at the RS. Application to DD3R membrane permeation data (Chapter 5E) Having demonstrated the usefulness of the RSM using simulated diffusivities, the model has been applied to membrane permeation data. Single component (CO2, CH4 and N2) and equimolar binary mixture (CO2/CH4, N2/CH4 and CO2/Air) permeation data across a disc-shaped all-silica DDR zeolite membrane have been the subject of a thorough modelling study over a challenging broad temperature (220-373 K) and feed pressure (101-1500 kPa) range. Also here a comparison with the Reed Ehrlich approach is made. Both the RSM as the RE approach yield an excellent model fit of the single component permeation data. However, for both models the N2 and CH4 single component permeation data did not allow an accurate estimation of the model fit parameters. Both models can lead to a good prediction of comparable quality of the mixture permeation data based on the single component model fit parameters. The RE approach is very sensitive towards the model input parameters and the estimated mixture loading, which both can be very hard to determine accurately in practice. The RSM does not suffer from both these issues, which is an evident advantage with respect to application of this model. Reconciliation with dynamically corrected Transition State Theory (Chapter 5F) The RSM closely resembles the well-known concept of dynamically corrected Transition State Theory (dcTST) which is often used in molecular simulations to study the dynamics of rare events. Therefore we investigated this connection in detail. It turns out that the ratio of the RS and total occupancy and a factor containing the exchange effects and free space available for diffusion in the RSM are directly related to those in dcTST, i.e. the probability that a molecule is on top of the free energy barrier and the transmission coefficient . Therefore, the RSM provides a direct link between properties at the molecular scale and the macroscopic Maxwell-Stefan diffusivity. Concluding remarks & future outlook What is the added value of this thesis in a broader perspective? To start with, a large part of this thesis is devoted to the introduction of a new model to describe mass transport in zeolites: the relevant site model (RSM). This model adequately captures mass transport phenomena in small-pore cage-like zeolites like DD3R. Moreover, this model appears useful to describe mass transport in other types of zeolites as well. The RSM offers handles to incorporate effects on the micro-scale into an engineering model (Maxwell Stefan approach to mass transfer). This is a definite step forward regarding mass transport modelling for design. A powerful tool in the development of this model has been Molecular Dynamic (MD) simulations. Where experimental membrane permeation data can be influenced by several phenomena, like crystal grain boundaries or surface barriers, in MD simulations the intrinsic diffusion phenomena can be studied separately. Although always a reality check needs to be made with respect to experimental data, these methods have developed in the last decades in such a way that impressive results can be obtained, particularly for all-silica zeolites. Zeolites, and other crystalline materials like metal organic frameworks, are excellently suited for this kind of approach due to their ordered structure. It has been demonstrated that the studied DD3R membranes are of excellent quality for gas separation applications. Moreover, stable operation is demonstrated, also at high temperatures. (Reactive) Separations where small molecules, like CO2, H2 or H2O [15], can be removed based on molecular sieving can be an attractive application. However, in case of H2, the flux seems currently still too low to compete with other available membranes (e.g. Pd). Clearly natural gas purification and biogas treatment seems the target application with DD3R membranes. Note that this should involve CO2/CH4 separations. Although the N2/CH4 separation factor is high, the current N2 flux can be considered too low for successful use for this bulk application. Also CO2 removal of flue gas can be interesting given the reasonable CO2/air separation factor, but the CO¬2 flux decrease with increasing temperature also reveals clear limitations for applications at elevated temperatures. Although this small-pore zeolite membrane retains its molecular sieving properties at high temperatures, the fluxes at these conditions are relatively low. The high CO2 flux at lower temperatures is due to the strong adsorption in the zeolite. At high temperatures these adsorption effects disappear and the size of the molecule appears the decisive factor in the obtained magnitude of the flux. It seems that the small molecules H2 and He provide an upper limit of the flux at high temperatures. If then the H2 flux is too low for applications, it can be doubtful if the current generation of this type of small-pore zeolite membrane will be suitable for any high temperature application. It is then the classical flux-selectivity trade-off [16] that makes small-pore zeolite membranes unsuitable for these type of applications, at least when applied at the macro-level. Note that the very small H2O molecule with its relative high boiling point could be a positive exception. But, finally it is good to emphasize that the excellent separation performance in CO2/CH4 separations can be considered a breakthrough in the development of zeolite membranes for gas separation applications. This application could be a driver for a more widespread application of zeolite membranes. This development inclines us to be optimistic that, after the first application of a zeolite membrane in alcohol dehydration, also the first gas separation applications appear to be within reach.DelftChemTechApplied Science