High-Throughput Screening of Metal–Organic Frameworks for Ethane–Ethylene Separation Using the Machine Learning Technique

A hybrid approach combining machine learning algorithms with molecular simulation is utilized to screen hypothetical metal–organic framework (h-MOF) database for the best material to separate ethane (C2H6) and ethylene (C2H4). In particular, we rationalized the relation between structural and chemical properties of h-MOF with the C2H6/C2H4 selectivity. 8% h-MOFs were chosen randomly from the h-MOF dataset as a training set. The simulations were conducted at 298 K and 1 bar using a multicomponent grand-canonical Monte Carlo method to obtain the C2H6/C2H4 selectivity. Based on the training set, the random forest (RF) model was developed to predict the selectivity of the rest of the h-MOFs. Among all the chemical and structural properties, void fraction plays a significant role in predicting the equilibrium C2H6/C2H4 selectivity. The trained machine learning model can reasonably predict the C2H6/C2H4 selectivity of the remaining h-MOF materials with an RF score of 0.89. Four h-MOFs have shown the best performance, which was compared with the previously discovered materials. The top four h-MOFs were further simulated at different pressures to obtain the adsorption isotherms. Further, the energy contribution of secondary building units and the local density profiles were analyzed to understand the enhanced interaction between h-MOF atoms and C2H6.