Machine learning-assisted high-throughput screening of MOFs for efficient adsorption and separation of CF4/N2

Abstract

There is an imperative need for top-performing materials with extraordinary adsorption selectivity and working capacity, in order to achieve productive adsorption of CF4 in a CF4/N2 mixture. In this work, the High-Throughput Grand Canonical Monte Carlo (HT-GCMC) simulation method and the Machine Learning (ML) method were employed to predict and screen the adsorption performance of 10 143 computation-ready experimental metal-organic frameworks (CoRE-MOFs) for separating CF4/N2 mixed gas. Through computational simulation and ML prediction, 15 and 73 highly promising adsorbents were selected out of the 690 randomly sampled MOFs and the CoRE-MOFs database. The selection process was based on criteria that balanced favorable CF4 selectivity, working capacity, and regenerability: selectivity >60, working capacity >70 mg g−1 (0.8 mmol g−1) and regenerability >70%. The maximum observed capacity of the 15 top evaluated metal-organic frameworks (MOFs) was: 52.85 mg g−1 (0.6 mmol g−1) at 1 bar; and 204.90 mg g−1 (2.3 mmol g−1) at 10 bar. The maximum working capacity was 152.06 mg g−1 (1.7 mmol g−1) and the highest selectivity reached was 118.12 (YEGCUJ) and 101.80 (VEHLIE) at 1 bar and 10 bar, respectively. Notably, the most promising MOFs exhibited elevated Zn content relative to the overall MOF population and also possessed a significant nitrogen content. This result should serve as a compelling motivation to further investigate the utilisation of MOFs with a high Zn content (e. g. zeolitic imidazolate frameworks), for enhanced adsorption applications.