Machine Learning Assisted Discovery of Efficient MOFs for One-Step C2H4 Purification from Ternary C2H2/C2H4/C2H6 Mixtures

Abstract

Purifying C₂H₄ from a mixture of C₂H₂/C₂H₄/C₂H₆ using a single adsorbent is crucial industrially. Yet, the challenge lies in their similar physicochemical properties, leading to low separation efficiency. Additionally, the lack of understanding regarding the structure–performance relationships hinders the development of high-performance metal–organic frameworks (MOFs). In this study, machine learning assisted high-throughput molecular simulation methods are employed to discover efficient MOFs for one-step C₂H₄ purification. The general material design strategies were proposed based on the analysis of 14,142 CoRE MOF simulation data. These include locking open metal sites, ensuring relative mass proportion of H atoms in the range of 2–4%, optimizing the largest cavity diameter to span 5–7 Å (ultramicropore), and fine-tuning φ within 0.5–0.6. Further using the computational insights obtained, 10 materials were identified with both C₂H₂/C₂H₄ and C₂H₆/C₂H₄ selectivities exceeding 3 from 137,953 hypothetical MOFs and 303,991 generated MOFs through additional molecular simulations. Our study not only provides screened and designed potential candidates for efficient one-step C₂H₄ purification from ternary C₂H₂/C₂H₄/C₂H₆ mixtures but also provides useful information for further performance improvement.