Understanding dislocation velocity in TaW using explainable machine learning

Abstract

The present work calculated the velocity of edge dislocations in the Ta–W system using molecular dynamics (MD) simulations and through machine learning (ML), identified the key parameters influencing the velocity. To achieve this, MD simulations were conducted at various values of the extrinsic parameters—temperatures and applied stresses ($${\tau }_{\text{app}}$$ τ app ), and the intrinsic variables—slip systems and alloying contents of tungsten in tantalum. Configurations containing edge dislocations on {110}/{112}/{123} planes were employed, and dislocation velocities were subsequently estimated. The MD results were processed using ML models, specifically extreme gradient boosting and SHapley Additive exPlanations (SHAP). SHAP analysis identified $${\tau }_{\text{app}}$$ τ app as the most influencing parameter affecting velocity, followed by slip plane, temperature, and W addition. SHAP estimated the base velocity value ($${v}_{\text{b}}$$ v b ) to be 1376 m·s-1. $${v}_{\text{b}}$$ v b was calculated by training SHAP on a parameter-less model. $${v}_{\text{b}}$$ v b could be increased by applying $${\tau }_{\text{app}}$$ τ app of at least 1 GPa, through slipping on the {112} and {123} planes, at temperatures of 0 and 300 K, and in configurations with 0 wt.% and 5 wt.% W. The importance of $${v}_{\text{b}}$$ v b on deformation was established.