High temperature Nb-Si alloys using data science: optimization of fracture toughness and high-temperature strength

Abstract High temperature Nb-Si based alloys face a critical challenge: achieving adequate room-temperature fracture toughness ( > 18 MPa·m 1/2 ) for processing while maintaining high-temperature strength, properties that typically compete with each other. Here, we overcome this inherent trade-off through machine learning-guided alloy design, employing a three-step feature screening strategy to identify 6 key descriptors from 200 initial features. SHAP analysis reveals how melting enthalpy and atomic radius mismatch control property outcomes, enabling targeted multi-objective optimization via NSGA-II algorithm. The optimized Nb-12.26Si-21.35Ti-1.98Al-1.96Cr-0.51Hf-4.34Zr-4.35 V alloy achieves an as-cast fracture toughness of 18.92 MPa·m 1/2 while maintaining 322 MPa strength at 1250 °C, surpassing all reported as-cast Nb-Si alloys. Microstructural analysis shows that the superior properties originate from the dispersed distribution of nanoscale γ′-Nb 5 Si 3 phase and crack deflection at phase boundaries with 67.6% lattice mismatch. Our results demonstrate that combining machine learning techniques with mechanistic understanding can accelerate the discovery of high temperature materials.