Data-driven property-oriented composition design and feature analysis of lightweight high-entropy alloys

• Published in: Journal of alloys and compounds Vol. 1037; p. 182197
• Main Authors: Song, Tianpeng, Cui, Puchang, Xia, Tao, Liu, Yong, Zhu, Jingchuan
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 10.08.2025
• Subjects: Lightweight high-entropy alloys; Machine learning; Mechanical properties; SHAP; Mechanical properties; Lightweight high-entropy alloys; SHAP; Machine learning
• ISSN: 0925-8388
• DOI: 10.1016/j.jallcom.2025.182197

Lightweight high-entropy alloys (LHEAs), characterized by low density, high strength, and excellent comprehensive mechanical properties, show great promise for demanding industrial applications in automotive and aerospace engineering, where material lightweighting and high strength are critical. However, the vast and complex compositional space of LHEAs remains largely unexplored, leaving numerous compositions with potentially exceptional properties untapped. In this study, a property-oriented design strategy for rapid multi-component alloy optimization is proposed by integrating a machine learning (ML) model with a particle swarm optimization (PSO) algorithm. A prediction model was constructed for Al-Ti-V-Zr-Nb-Cr-Ni system LHEAs. Thirteen feature descriptors were introduced and screened in two stages via Pearson correlation analysis and principal component analysis (PCA). Using four machine learning regression algorithms, the relationship between input features and room-temperature mechanical properties was modeled by integrating material composition and descriptors. The PSO algorithm was used for multi-objective reverse optimization design of the trained model, and several low-density (5–5.63 g/cm3) LHEAs with room-temperature specific yield strengths of 221–271 MPa·cm3/g were designed. Among them, Al20Ti30V18Nb25Cr7 exhibited over 50 % compressive deformation at a peak stress of 2534 MPa without fracture. SHapley Additive exPlanations (SHAP) analysis revealed the influence mechanisms of key features on alloy properties. Results demonstrate that this property-oriented alloy design approach effectively guides the development of LHEAs to meet lightweight and high-performance requirements. •A property-oriented design strategy for rapid material optimization of multi-principal element alloys.•Two-step dimensionality reduction method for filtering feature descriptors.•A low-density lightweight high-entropy alloy with high specific yield strength and excellent compression plasticity.•Effect of elemental content and key characteristic descriptors on alloy properties.