Abstract: Body-centered cubic multi-principal element alloys (BCC MPEAs) face inherent strength-ductility trade-offs. Given their vast compositional space, identifying key factors governing strength and ductility, as well as developing novel strengthening-toughening models to accelerate the property-oriented design, remains an outstanding challenge. Here, we developed an interpretable machine learning (ML) framework for BCC MPEAs to identify the key factors governing yield strength (YS) and fracture elongation (FE). The results demonstrate that FE mainly arises from the synergistic effects of multiple factors (electronegativity difference Δχ^pauling, valence electron concentration VEC, and density ρ), while average shear modulus mismatch δG^ave is the dominant factor controlling YS. Using these screened features as inputs, we propose optimized YS and FE models that achieve high predictive accuracy (YS: R²=0.96, FE: R²=0.84) and outperform existing models. By transforming these ML insights into strengthening/toughening theories via feature-to-mechanical performance/element property mapping, we designed three novel Ti-Zr-based BCC MPEAs with exceptional properties: YS of 1.07-1.16 GPa, FE of 16.6 %-24.5 %, and specific yield strength of ∼170 MPa cm³ g^-1, surpassing most reported BCC MPEAs. This work not only provides a data-driven strategy to overcome the strength-ductility trade-off in BCC MPEAs but also establishes interpretable design principles for accelerating the discovery of advanced structural materials.

Key words: Multi-principal element alloys, Body-centered cubic, Machine learning, Mechanical properties