Machine-learning based molecular dynamics simulation of crack propagation in bcc iron

Abstract

This study integrates machine-learning potentials (MLP) into crack propagation simulations, comparing the obtained atomic structural changes and stress distribution in four crack systems with the ones by embedded-atom-method (EAM) and modified-EAM (MEAM) potentials. The MLP predicts stress concentration at crack tips, with increased load leading to bond breakage and cleavage, implying brittle fracture. In contrast, EAM and MEAM suggest ductile failure, with observed dislocation ejection and gradual crack opening due to structural changes. Overall, MLP exhibits a distinct fracture tendency compared to conventional potentials.