Refractory high-entropy alloys (RHEAs) show great promise for extreme environments, but their development is hindered by the vast compositional space and the challenge of balancing multiple properties. This study presents an integrated machine learning (ML) framework for the efficient design of RHEAs for achieving both high hardness and excellent corrosion resistance. A comprehensive dataset was constructed, and multitask learning with tree-based ensemble algorithms was employed to develop predictive models for hardness, corrosion potential (Ecorr), and corrosion current density (Icorr). The models are trained for a narrowly defined Nb-Mo-Ta-W-V compositional space. Data from 36 publications were processed, with Ecorr cleaned systematically and Icorr purified electrochemically, yielding final datasets of 157(hardness), 93(Ecorr), and 187(Icorr) entries. Shapley additive explanations (SHAP) analysis revealed key descriptors, such as the d-electron concentration, average electronegativity, average melting point, and mixing entropy for hardness and the difference in the atomic size (δr) and electronegativity (Δχ) for corrosion resistance. The optimized models demonstrated high predictive accuracy (R2 was 0.91 for hardness and 0.83 for Ecorr and Icorr). Three novel RHEAs were designed and experimentally validated, the results revealed excellent agreement between the predicted and measured properties, with accuracies > 85 %. This work presents a robust ML-driven paradigm for multiobjective optimization of RHEAs.
扫码关注我们
求助内容:
应助结果提醒方式:
