Active learning boosted computational discovery of covalent–organic frameworks for ultrahigh CH4 storage. Issue 11 (29th August 2022)

• Record Type: Journal Article
• Authors: Tang, Hongjian; Jiang, Jianwen
• Is Part Of: AIChE journal. Volume 68:Issue 11(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-08-29
• Subjects: active learning -- covalent–organic frameworks -- methane storage -- molecular simulation -- parallel acquisition; Chemical engineering -- Periodicals; Génie chimique -- Périodiques; 660.28
• DOI: 10.1002/aic.17856
• Languages: English
• ISSNs: 0001-1541

Abstract

Abstract: As an environmental‐benign fuel, methane (CH4 ) has received considerable interest for developing high‐capacity energy storage systems. Herein, we aim to rapidly discover covalent–organic frameworks (COFs) for ultrahigh CH4 storage among 530, 000+ COFs, including one experimental (Curated) and two hypothetical (Berkeley and Genomic) databases. First, the feature space of all the three COF databases is projected by t‐Distributed Stochastic Neighbor Embedding (t‐SNE) technique, which reveals a potential but unexplored regime in Genomic COFs. Subsequently, an active learning (AL) approach is developed by integrating parallel acquisition with molecular simulation to efficiently explore Genomic COFs. The parallel AL model demonstrates remarkable screening efficiency and shortlists top COFs by evaluating only 50 out of 445, 845 Genomic COFs. A record‐breaking Genomic COF is identified with CH4 deliverable capacity of 222.2 v/v, surpassing the current world record (208.0 v/v from experiment and 217.9 v/v from simulation). Our AL approach is significantly faster than brute‐force simulation and conventional machine learning, it would accelerate the discovery of advanced porous materials for broad applications.