High-throughput computational screening of metal-organic framework membranes for upgrading of natural gas

Highlights

• A single-step separation of both CO₂ and N₂ from a three-component CO₂/N₂/CH₄ mixture.

• Integration of computational techniques including geometry analysis, Monte Carlo and molecular dynamics simulations.

• Structure-performance relationships between the geometrical descriptors of MOFs and membrane performance.

• Multivariate statistical tools including principal component analysis, multiple linear regression and decision tree modeling.

Abstract

We report a computational study to screen 4764 computation-ready experimental metal-organic frameworks (CoRE-MOF) for the membrane separation of a ternary gas mixture (CO₂/N₂/CH₄) at 298 K and 10 bar. Combining Monte Carlo and molecular dynamics simulations, the adsorption, diffusion and permeation of the gas mixture are predicted. The structure-performance relationships are established between the geometrical descriptors of MOFs (pore liming diameter, density, void fraction and volumetric surface area) and the evaluation criteria of membrane performance (permeability and permselectivity). Furthermore, principal component analysis is used to assess the interrelationships among the descriptors, then multiple linear regression is applied to quantitatively determine the respective effects of descriptors on performance. In addition, decision tree modeling is adopted to define a clear effective path for screening. Finally, seven best MOF membranes are identified for the single-step separation of both CO₂ and N₂ from CH₄. The microscopic insights and structure-performance relationships from this computational study can facilitate the development of new MOF membranes for the upgrading of natural gas.

Introduction

CO₂ emissions from the combustion of fossil fuels have been considered to cause adverse environmental effect [1]. With less carbon footprint, natural gas is environmentally more benign energy carrier [2]. Nevertheless, impurities (e.g. CO₂ and N₂) in natural gas remarkably decrease the heat value and thus have to be removed. A handful of techniques exist for the purification of natural gas, such as solvent scrubbing, cryogenic distillation, porous solid adsorption, membrane separation and their combinations [3]. Among these, adsorption and membrane-based approaches are energetically and economically feasible because of easy operation, low cost and high efficiency. A variety of porous materials such as carbons, zeolites, polymer, and metal oxides have been tested as adsorbents or membranes [4]. However, they are either insufficiently selective or difficult to be regenerated [5]. Therefore, there is continuous quest for new materials with improved performance.

In the past two decades, metal organic frameworks (MOFs) have emerged as promising alternative [6]. They can be synthesized from a very wide range of inorganic and organic building blocks, and the diversity and multiplicity of MOF structures are far more extensive than any other porous materials. Consequently, MOFs have been considered versatile materials for gas storage, separation, catalysis, etc [7]. Among tens of thousands of experimentally synthesized MOFs, a larger number of them have been investigated for the separation of natural gas and other gas mixtures [8]. For instance, ZIF-8 membrane was fabricated on Al₂O₃ support with high permeability and selectivity for CO₂/CH₄ [9]. In a zeolite-like MOF, high adsorption selectivity was predicted by molecular simulation for CO₂/CH₄, CO₂/N₂ and CO₂/H₂ mixtures [10]. The efficacy of MOFs for CO₂ capture was revealed to strongly depend on process and the performance could not be simply determined without considering a specific process [11]. On the other hand, computational studies have been reported to screen suitable candidates from a large collection of MOFs for gas separation. Sholl and coworkers used pore size analysis and simulations to screen 504 MOFs for H₂/CH₄ separation [12] and 1163 MOFs for CO₂/N₂ separation [13]. Snurr and coworkers simulated the adsorption of pure CO₂, N₂ and CH₄ in 137953 MOFs and proposed relationships between structural descriptors, as well as chemical functionality, with adsorbent evaluation criteria for CO₂/CH₄ and CO₂/N₂ separation [14]. From hundreds of thousands of zeolites and zeolitic MOFs, Smit and coworkers identified many potential adsorbents for CO₂/N₂ separation [15]. We screened 4764 computation-ready experimental (CoRE) MOFs for CO₂/CH₄ and CO₂/N₂ separation, and proposed relationships between metal type and adsorbent evaluation criteria [16].

The above studies were focused on the separation of binary mixtures. In natural gas, however, there exist other impurities like N₂ in addition to CO₂. Currently, CO₂ and N₂ are separated from natural gas via two sequential energy intensive steps (amine scrubbing and cryogenic distillation) [17]. As an alternative, we recently reported a computational study to screen 137953 hypothetical MOFs (HMOFs) for the single-step membrane separation of a CO₂/N₂/CH₄ mixture [18]. Because of the larger number of MOFs involved, the screening consisted of four steps: (1) MOFs with a pore liming diameter (PLD) of 3 ~ 4 Å were chosen; (2) simulations of adsorption and diffusion were conducted for pure gases at infinite dilution; (3) MOFs were narrowed down based on permselectivity/permeability; (4) simulations of a three-component CO₂/N₂/CH₄ mixture were finally performed at 298 K and 10 bar to identify the best MOFs.

In the current study, we aim to computationally screen CoRE-MOFs for the membrane separation of a CO₂/N₂/CH₄ mixture. Nevertheless, it is different from our recent study [17] in several aspects. First, the CoRE-MOFs considered here have been experimentally synthesized, unlike the HMOFs; second, the simulations of adsorption and diffusion were conducted directly for a three-component CO₂/N₂/CH₄ mixture at 298 K and 10 bar, not at infinite dilution; third, principal component analysis (PCA) and multiple linear regression (MLR) are integrated to evaluate the interrelationships among MOF descriptors and the respective effects of descriptors on separation performance; fourth, a decision tree (DT) model is proposed to define a simple path to identify the best MOFs. It is worthwhile to note that PCA, MLR and DT modeling are not commonly used in the studies of MOFs, however, they are instrumental to provide insightful and quantitative information. Following this introduction, the molecular models of MOFs and gases are described in Section 2, followed by simulation methods. In Section 3, the adsorption, diffusion and permeation of a CO₂/N₂/CH₄ mixture, as well as the corresponding selectivities, are presented. The relationships between geometrical descriptors and separation performance are estimated; then PCA and MLR are conducted, along with DT modeling. Finally, the best MOFs are identified. The concluding remarks are summarized in Section 4.