Multiscale Catalyst Design for Steam Methane Reforming Assisted by Deep Learning

Computational design of high-quality catalysts targeting specific operation conditions is a challenging task due to the mechanistic, structural, and environmental complexities across multiple length and time scales. A multiscale method of a catalyst design linking ab initio calculations, microkinetics, and multiphysics modeling was proposed to address this challenge. The chemistry-based analytical model derived from a microkinetic model assisted by first-principles-based deep neural networks efficiently bridged zero Kelvin ab initio microscopic descriptors and multiphysics modeling. We applied the multiscale method to the design of carbon-resistant steam methane reforming catalysts, successfully identifying a few cost-efficient bimetallic alloys for CH4 internal reforming solid oxide fuel cells. The multiphysics modeling demonstrates that catalysts of relatively low activity such as NiZn are actually beneficial for fuel efficiency, highlighting the importance of the multiphysics model for a multiscale computational catalyst design.