Abstract: The liquid metal embrittlement (LME) of advanced high-strength steels caused by zinc (Zn) has become a critical issue hindering their widespread application in the automotive industry. In this study, atomic-scale simulations are carried out to elucidate the underlying cause of this phenomenon, namely grain boundary embrittlement due to Zn segregation at iron (Fe) grain boundaries. A machine learning moment tensor interatomic potential for the Fe-Zn binary system is developed, based on which the thermodynamics of grain boundary segregation is evaluated. The yielded segregation energy spectrum of Zn in BCC Fe reveals the quantitative relationship between the average segregation concentration of Zn at Fe grain boundaries and the macroscopic Zn content, temperature, and fraction of grain boundary atoms. It suggests a strong thermodynamic driving force for Zn segregation at the Fe grain boundaries, which correlates directly with the grain boundary energy: high-energy grain boundaries can accommodate a large amount of Zn atoms, while low-energy grain boundaries exhibit a certain degree of repulsion to Zn. Kinetically, Zn enters the grain boundaries more easily through diffusion than by penetration. Nonetheless, the grain boundary embrittlement caused by Zn penetration is more severe than that by Zn diffusion. The embrittlement effect generally increases linearly with the increase in Zn concentration at the grain boundary. Altogether, it suggests that the LME in steels induced by grain boundary segregation of Zn emerges as a combined consequence of Zn melt penetration and solid-state diffusion of Zn atoms.

Key words: Liquid metal embrittlement, Machine learning interatomic potential, Grain boundary segregation, Grain boundary penetration, Grain boundary diffusion