Discovery of ultra-high strength aluminum alloys with high damage tolerance via interpretable chain-based machine learning

The prolonged testing cycles and high costs associated with fatigue experiments make trial and error design and optimization improvement of complex alloys with high damage tolerance extremely difficult. This study proposes an interpretable chain-based machine learning (ICML) strategy for designing ultra-high strength aluminum alloys with high damage tolerance. First, a gradient boosting regression method was employed to establish a predictive model linking alloy composition (AC) and solution-aging processes (SAP) to tensile mechanical properties (TMP), including ultimate tensile strength (σb), yield strength (σy), and elongation (A). Symbolic regression was then utilized to identify key factors influencing fatigue strength and to derive an explicit quantitative relationship between TMP and fatigue strength (FS), expressed as FS=ασbA1/4. Subsequently, these models were integrated into a chain framework to construct a multi-scale interpretable predictive model AC + SAP → TMP → FS. Furthermore, combining SHAP analysis, multi-objective optimization, and thermodynamic calculations, the study achieves the integrated optimization design of AC and SAP. The newly developed ultra-high strength aluminum alloy with high damage tolerance Al-10.5Zn-2.34Mg-1.28Cu-0.11Zr-0.1Cr demonstrates outstanding mechanical properties, with a measured σb=794±2 MPa, σy=765±3 MPa, A=11.9±0.4%, FS=376±14 MPa. Compared to commercial AA7050 and AA7055 alloys, the novel alloy demonstrated over 20% enhancements in σb, σy, and FS, while maintaining equivalent or superior A. This work provides a novel approach for efficiently designing high damage tolerance alloys.