Optimizing high-temperature oxidation resistance in Ti-50Al-W alloys: A first-principles and machine learning approach

Designing oxidation-resistant γ-TiAl alloys remains a longstanding challenge due to their poor oxidation resistance at high temperatures. Here, we propose an integrated strategy that combines machine learning, first-principles calculations, and experimental validation to accelerate dopant selection and performance optimization. Fifteen candidate elements were initially evaluated using a random forest classifier and SHAP analysis based on oxidation-relevant descriptors. Tungsten (W) emerged as the most effective dopant. Density functional theory (DFT) calculations show that W preferentially substitutes Ti atoms, stabilizes the lattice, weakens Ti–Al bonding, and increases the oxygen diffusion barrier. A Gaussian process regression (GPR) model trained on CI-NEB data accurately predicts diffusion trends (R2 = 0.896), enabling data-efficient screening. Experimental results confirm that a 3 at.% W addition significantly improves oxidation resistance, reducing mass gain by ∼ 97 % and forming a dense Al2O3-rich oxide scale after exposure at 900 °C. This integrated approach provides a rapid, interpretable, and scalable pathway for the design of oxidation-resistant TiAl-based alloys.