Machine learning-assisted design of Ti–V–Nb–Mo refractory high-entropy alloys with higher ductility and specific yield strength

The performance of refractory high-entropy alloys (RHEAs) is closely related to the content of their constituent elements, which makes compositional exploration through traditional trial-and-error methods a challenging and time-consuming endeavour, with the goal of developing an alloy that exhibits both high ductility and high specific yield strength. A dataset of the alloys' performance parameters was established by applying first-principles and molecular dynamics calculations. The combination of the aforementioned dataset with the solid solution strengthening (SSH) model and the D (γ_s/γ_usf) parameter enabled the construction of a highly accurate strength-ductility prediction model for the alloys through the use of an XGBoost algorithm. The model was employed to predict the compositions of two novel RHEAs and their mechanical properties were verified by experiments. The predicted results are in general agreement with the trends of the experimental data. The Ti₃₅V₃₅Nb₁₀Mo₂₀ alloy exhibiting excellent comprehensive performance, achieving a specific yield strength of 149.55 kPa m³/kg, which is 10.97% higher than that of traditional equiatomic alloy, and a compressive strain exceeding 50%. In conclusion, this work presents an effective alloy design strategy, offering a new approach for the future design of high-performance RHEAs.