Accurate Ti-Al-Nb ternary interatomic potential development using deep neural networks for TiAl PST single crystals.

Abstract

A deep learning-driven Ti-Al-Nb ternary interatomic potential is developed continuously through DP-GEN framework, combining first-principles accuracy with molecular dynamics scalability. The neural network potential demonstrates exceptional transferability in predicting critical properties of Nb-dopedγ-TiAl andα₂-Ti₃Al phases. Nb influence on shear deformation inα₂-Ti₃Al is investigated. Meanwhile, Nb-dopedα₂/γinterface tensile perpendicular to the interface and shear simulations along 1/2[11¯0] and 1/2[112¯] are performed in order to simulate the local configurations in Ti-Al PST single crystals. This model provides a computational framework for interfacial engineering in lamellar TiAl alloys.