Developing interatomic potentials for mechanical properties of multi-component alloys using machine learning technique

Lobzenko, I.   ; Shiihara, Yoshinori*; Mori, Hideki*; Matsunaka, Daisuke*; Tsuru, Tomohito   

Refractory multi-component alloys (MCA) form an important class of materials with high potential for use in severe conditions. One of the main problems hindering the application of these alloys is the low ductility inherited from the body-centred cubic (BCC) crystal structure. Dislocation motion is the factor significantly influencing the ductility of the material, so a comprehensive understanding of the dislocation dynamics in refractory MCAs should be achieved to pave the way for designing refractory alloys with increased ductility. To achieve high accuracy in classical molecular dynamics simulations of dislocation motion, we apply the technique of machine learning (ML) for interatomic potential development. It is known that alloys with hexagonal closed-packed (HCP) elements such as Zr exhibit higher ductility, which is why two medium-entropy alloys, MoNbTa and ZrNbTa, were chosen to study the influence of elements' constitution on dislocations dynamics. The inter-atomic potentials for MCAs built using ML need a specific dataset. In the process of the potential development, we identify which structures contribute to a better quality of materials' mechanical properties prediction by the potentials. Results of the simulations have shown qualitative and quantitative differences between the two alloys under study. One example of that difference can be seen in the shapes of the screw dislocation core. In contrast to MoNbTa, ZrNbTa demonstrates a non-compact core with an extension on a (110) plane.