Interpretable Machine Learning Framework for Nb─Si Based Alloy Design with Enhanced Fracture Toughness

A continuing challenge in aerospace materials is the search for alloys that have desired functional properties to operate at higher temperatures with lower densities. This improves aero-engine efficiency and reduces CO₂ and other harmful emissions, aligning with aviation industry targets for emission reduction. Nb─Si alloys can operate at higher temperatures than current Ni-based superalloys and have lower densities. We develop a machine learning-driven design framework for Nb─Si based ultra-high temperature alloys using a three-step feature screening strategy to break the 20 MPa·m^1/2 fracture toughness barrier. Our model predicts fracture toughness (K_Q) with an error <7%, and we use SHAP(ley) and PDP analysis to interpret the model to guide alloy design. Five alloys with Si content ranging from 3 to 15 at.% were synthesized to validate model predictions. Sample #5 (Nb_38.5Ti_38.5Si₃Zr₁₈V₂) achieved a K_Q of 22.791 MPa·m^1/2, exceeding the typical range (below 20 MPa·m^1/2) for as-cast Nb─Si alloys. Microstructural analysis showed that the improved performance resulted from the transformation of brittle silicide phases to ductile Nbss phase and crack-bridging toughening. Strengthening mechanism analysis reveals that solid solution strengthening was dominant (68%-84%), with excellent strength-toughness balance.