Machine-learning-aided density functional theory calculations of stacking fault energies in steel

A combined large-scale first principles approach with machine learning and materials informatics is proposed to quickly sweep the chemistry-composition space of advanced high strength steels (AHSS). AHSS are composed of iron and key alloying elements such as aluminum and manganese. A systematic exploration of the distribution of aluminum and manganese atoms in iron is used to investigate low stacking fault energies configurations using first principles calculations. To overcome the computational cost of exploring the composition space, this process is sped up using an automated machine learning tool: DeepHyper. Our results predict that it is energetically favorable for Al to stay away from a stacking fault, but Mn atoms do not affect the stacking fault energy and can stay in the vicinity of the fault. The distribution of Al and Mn atoms in systems containing stacking faults and the effects of their interactions on the equilibrium distribution are systematically analyzed.

Automated summary

This study proposes a combined large-scale first principles approach with machine learning and materials informatics to quickly explore the chemistry-composition space of advanced high strength steels (AHSS). The distribution of aluminum and manganese atoms in iron is systematically explored using first principles calculations to investigate low stacking fault energy configurations. The use of an automated machine learning tool, DeepHyper, speeds up the computational process. The study provides insights into the distribution of aluminum and manganese atoms in systems containing stacking faults and their effects on the equilibrium distribution.