Integrating Machine Learning with High-Throughput Screening Technology to Drive Molecular Sieve Design for Air Separation Oxygen Production

High-altitude oxygen generation via pressure swing adsorption is hindered by inefficient adsorbent development and limitations of conventional materials, including escalating lithium costs and high-pressure performance bottlenecks. To address this, a comprehensive feature database was established using molecular fingerprinting and simulation-derived physical characteristics, encompassing the structures of two distinct material types (CoRE-MOF 2019 and the International Zeolite Association database). A highly accurate binary classification machine learning model, developed from these data, elucidated structural features critical for N2/O2 separation, focusing on the A-type and X-type zeolite structures currently primarily used for air separation, revealing synergistic interactions between alkaline earth metal fragments and the FAU zeolite framework. This insight facilitated the development of a low-cost, high-performance Sr-LSX oxygen adsorbent. Its exceptional performance was experimentally confirmed, exhibiting a N2 adsorption capacity of 26.54 cm3 g–1 and an IAST selectivity of 7.40 (298 K, 1 bar). This data-driven strategy offers a universally applicable paradigm for addressing N2/O2 separation challenges.