Mechanism-informed prediction of γ-TiAl alloy mechanical behavior via molecular dynamics and explainable machine learning

ZW Zhang and QH Song and J Qin and YP Li and KY Li and ZQ Liu, MATERIALS TODAY COMMUNICATIONS, 48, 113630 (2025).

DOI: 10.1016/j.mtcomm.2025.113630

gamma-TiAl alloys are widely used in aerospace applications due to their high-temperature mechanical performance. However, their complex deformation mechanisms has somewhat limited their further development. In this study, we present a mechanism-informed modeling framework that integrates molecular dynamics (MD) simulations with interpretable machine learning (ML) algorithms to characterize the mechanical behavior of gamma-TiAl alloys across a range of temperatures, strain rates, and grain sizes. Atomistic stress-strain datasets generated by MD are used to train six supervised ML models. Among these, the Random Forest algorithm exhibits superior predictive performance, achieving an R2 value of 0.999 and a mean absolute error (MAE) of less than 0.02. To enhance interpretability, SHAP-based feature attribution analysis is employed to quantify the relative importance and interaction effects of input variables. The results indicate that strain is the dominant factor influencing stress response, while temperature and strain rate exhibit complex nonlinear coupling. The model successfully captures key physical mechanisms including the inverse Hall-Petch softening effect, the strengthening effect induced by high strain rates, and the degradation of strength caused by thermal softening. These results highlight the capability of MD-ML integration to deliver physically grounded predictions and reveal deformation mechanisms. The approach offers a generalizable framework for predictive modeling of complex alloy systems under demanding thermomechanical conditions.

Return to Publications page