Machine-Learning-Assisted High-Throughput Screening of High-Performance MOFs for Multicomponent Gas Separation

To develop a novel method for rapid and accurate prediction, achieving efficient screening of optimal MOFs for the competitive adsorption of different concentrations of multicomponent gases, this study initially identified 1,956 metal–organic frameworks (MOFs) structures from a database of 14,142 core MOFs through high-throughput screening. The single-component gas adsorption capacity of these MOFs adsorbents was calculated using grand canonical Monte Carlo (GCMC) simulations, along with the competitive adsorption capacity of the multicomponent mixture. Subsequently, single-component (CO2, CH4, N2, H2) and mixed-gas competitive adsorption capacities (CO2/CH4/N2/H2 = 25/5/5/65 vol %) were rapidly predicted using both the “Pore Volume Method” and Machine Learning (ML) modeling. Finally, among the 50 most promising MOF structures for gas separation, time-cost correlations were calculated, based on the experimental testing and computational simulations of each structure. Cu-BTC and Mg-MOF-74 were selected for experimental validation to assess the accuracy of the Pore Volume Method and the machine learning model.