Mining the relationship between microstructural characteristics and dynamic compression properties of dual-phase titanium alloys via data-driven random forest and finite element simulation

To optimize the dynamic compression properties of titanium alloys, it is necessary to reveal the internal relationship between microstructural characteristics and dynamic mechanical properties. In this work, a dynamic compression numerical simulation approach, based on realistic microstructures and parametric modeling was proposed and validated experimentally. Following this, 4075 sets of dynamic compression simulation results for dual-phase TC6 titanium alloys were calculated by high-throughput simulation. Subsequently, a regression model and a four-classification model, aiming to predict the dynamic strength (σ_D) and dynamic plasticity (ε_f) of titanium alloys, were established by the data-driven random forest algorithm. The regression model attained a goodness-of-fit metric of 0.99, while the four-classification model achieved an F1-score of 0.88. Further, combined with the Shapley additive explanations (SHAP), it was found that the width of secondary α phase (Sw) and the volume fraction of primary α phase (Pf) were the most critical microstructural characteristics. Specifically, Pf was negatively correlated with σ_D and ε_f, whereas Sw was negatively correlated with σ_D but positively correlated with ε_f. Meanwhile, intrinsic mechanisms behind the above laws were revealed through local stress and adiabatic shear sensitivity analyses of typical microstructure models. Finally, the range of microstructural characteristics of excellent dynamic mechanical properties (a Sw of 1 μm and a Pf ranging from 0.1 to 0.2.) was determined by further analysis of datasets without dynamic plastic fracture. These findings can provide a significant reference for subsequent experimental efforts to optimize the dynamic mechanical properties of titanium alloys.