A New Machine Learned Interatomic Potential for Simulating the Effect of Alloying Content on the Alpha-Beta Transition Range of Ti-Al-V

Within the last decade, significant advances have been made in the application of machine learning principles to the generation and improvement of new classes of interatomic atomic potentials for use in classical molecular dynamics simulations. In the current work, the so-called machine learned interatomic potential generation process implemented in the MedeA environment was applied to the creation of a spectral analysis neighbor potential for an exemplar manufacturing titanium chemistry, Ti-Al-V. Requisite training set characteristics, validation against first-principles and experiment, and implementation within high-throughput workflows will be discussed. Leveraging the new interatomic potential for Ti-Al-V, a robust validation effort was successful in showing the ability to predict the relative shift in the crucial thermodynamic alpha to beta phase transition as a function of alloying element concentration.