AI-Powered Experimental Discovery of Metal-Organic Frameworks for n/i-Butane Separation.

Abstract

There are significant challenges in developing efficient adsorbents as alternatives to the energy-intensive distillation processes for n/i-butane separation. Metal-organic frameworks (MOFs) hold great potential in addressing this issue. However, the vast diversity of MOFs makes the discovery of high-performance materials akin to searching for a needle in a haystack. Here, the high-throughput screening based on artificial intelligence (AI) is employed to accelerate the identification of MOFs for n/i-butane separation. An integrated descriptor system, accessible via both experiments and simulations, is proposed and broadly validated, demonstrating better performance over those widely-used descriptors. In addition, an optimization strategy for training dataset is proposed based on similarity, allowing for the efficient model training with only 10% samples from the entire database and thus significantly reducing the costs. Leveraging the integrated descriptors and optimization strategy, MOFs with exceptional n/i-butane separation performance are successfully identified through neural network model. As a proof of concept, SIFSIX-3-Zn is synthesized for validation because it has the largest n-butane capacity among top 20 MOFs. The SIFSIX-3-Zn demonstrates outstanding n/i-butane separation performance with nearly zero uptake of i-butane. This work introduces a novel research paradigm integrating AI, simulation and experiment, and presents an efficient process with broad applicability for material discovery.