Machine learning aided design of single-atom alloy catalysts for methane cracking

• Published in: Nature communications Vol. 15; no. 1; pp. 6036 - 9
• Main Authors: Sun, Jikai, Tu, Rui, Xu, Yuchun, Yang, Hongyan, Yu, Tie, Zhai, Dong, Ci, Xiuqin, Deng, Weiqiao
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 18-07-2024; Nature Publishing Group; Nature Portfolio
• Subjects: 119/118; 140/133; 140/146; 147/135; 147/143; 147/3; 639/638/77/885; 639/638/77/887; 639/925/357/537; Ball milling; Carbon; Catalysis; Catalysts; Catalytic converters; Catalytic cracking; Deposition; Energy of dissociation; Free energy; Heat of formation; Humanities and Social Sciences; Hydrogen; Hydrogen production; Learning algorithms; Machine learning; Methane; multidisciplinary; Noble metals; Rhenium; Science; Science (multidisciplinary); Single atom catalysts; Transition metal alloys; Transition metals
• ISSN: 2041-1723; 2041-1723
• DOI: 10.1038/s41467-024-50417-7

The process of CH
4
cracking into H
2
and carbon has gained wide attention for hydrogen production. However, traditional catalysis methods suffer rapid deactivation due to severe carbon deposition. In this study, we discover that effective CH
4
cracking can be achieved at 450 °C over a Re/Ni single-atom alloy via ball milling. To explore single-atom alloy catalysis, we construct a library of 10,950 transition metal single-atom alloy surfaces and screen candidates based on C–H dissociation energy barriers predicted by a machine learning model. Experimental validation identifies Ir/Ni and Re/Ni as top performers. Notably, the non-noble metal Re/Ni achieves a hydrogen yield of 10.7 gH
2
gcat
–1
h
–1
with 99.9% selectivity and 7.75% CH
4
conversion at 450 °C, 1 atm. Here, we show the mechanical energy boosts CH
4
conversion clearly and sustained CH
4
cracking over 240 h is achieved, significantly surpassing other approaches in the literature.
The process of CH
4
cracking into H
2
and carbon has garnered significant attention for hydrogen production, but traditional catalytic methods are hampered by severe carbon deposition. Here, a machine-learning model has been developed to expedite the screening of CH
4
cracking catalysts from 10,950 types of single-atom alloy surfaces.