Abstract
High-entropy alloys (HEAs) exhibit comprehensive properties such as high strength, outstanding corrosion resistance, and excellent thermal stability due to their multi-component chemical composition, offering broad application prospects in industrial fields. However, for hexa-component HEAs, the combination space based on element concentrations encompasses approximately 9.65×10⁷ possible compositions. Traditional trial-and-error approaches are extremely inefficient in exploring this vast chemical space, severely limiting rational design and performance optimization for such alloys. To address this, this study proposes a data-driven dual-task prediction framework for both ’phase’ and ’hardness’. Focusing on the Al-Co-Cr-Cu-Fe-Ni system, which offers both excellent mechanical potential and lower cost, the framework enables rapid prediction of phase and hardness for a large-scale dataset of HEAs with unknown properties. The framework establishes separate machine learning (ML) models for phase classification and hardness regression, achieving 93% accuracy and a correlation coefficient of 0.949. Interpretability analysis further reveals the dominant roles of key descriptors such as valence electron concentration and atomic size difference, validating the model's physical reliability. Finally, the trained model enabled high-throughput prediction for over 48,000 potential alloys, yielding high-hardness HEAs (e.g., BCC: Al0.350Co0.124Cr0.310Fe0.086Ni0.068Cu0.061, FCC: Al0.102Co0.221Cr0.285Fe0.283Ni0.057Cu0.052, BCC+FCC: Al0.348Co0.065Cr0.236Fe0.088Ni0.056Cu0.207). A comprehensive ”composition-structure-property” correlation map was constructed, identifying key characteristics influencing phase composition and hardness in HEAs. This study establishes an efficient, scalable composition screening workflow capable of rapidly identifying HEAs with ideal phase structures and high hardness potential. It not only provides an efficient, precise computational tool for designing HEAs in the Al-Co-Cr-Cu-Fe-Ni system, but also constructs a collaborative prediction framework and methodology with strong transferability. This approach offers a paradigm reference for performance-driven reverse design in other multi-component alloy systems.