Machine-learned accelerated discovery of oxidation-resistant NiCoCrAl high-entropy alloys

TL;DR

This work addresses long-standing challenges in designing oxidation-resistant NiCoCrAl high-entropy alloy bond coats by predicting the parabolic oxidation constant $k_p$ across a wide compositional space. A data-driven framework integrates 743 experimental measurements, physically meaningful descriptors, and Wagner-based kinetics to train multiple models, with XGBoost achieving the highest accuracy ($R^2\approx0.91$, RMSE $\approx1.87$). Interpretability analyses reveal that temperature, Al, and Cr are key drivers of protective oxide formation, while reactive elements Hf and Y offer composition-dependent benefits, notably enabling robust protection in NiCo-lean regimes. The model identifies optimal compositions, notably Ni17Co23Cr30Al30 for high-temperature bond coats, and combines ML with Thermo-Calc phase stability to accelerate the design of oxidation-resistant HEA coatings for high-temperature service.

Abstract

The development of oxidation-resistant high-entropy alloy (HEA) bond coats is restricted by the limited understanding of how multi-principal element interactions govern scale formation across temperatures. This study uncovers new oxidation trends in NiCoCrAl HEAs using a data-driven analysis of high-fidelity experimental oxidation data. The results reveal a clear temperature-dependent transition between alumina- and chromia-dominated protection, identifying the compositional regimes where alloys rich in Al dominate at $\ge1150$ °C, mixed Al-Cr chemistries are optimal at intermediate temperatures, and, unexpectedly, Cr-rich low-Al alloys perform best at 850 °C-challenging the assumption that high Al is universally required. The effects of Hf and Y are shown to be strongly composition-dependent with Hf producing the largest global reduction in oxidation rate, while Y becomes effective primarily in NiCo-lean alloys. Y-Hf co-doping offers consistent improvement but exhibits site-saturation behavior. These insights identify new high-performing HEA bond-coat families, including $\mathrm{Ni_{17}Co_{23}Cr_{30}Al_{30}}$ as a substitute for conventional mutlilayer thermal barrier coatings.

Figures (9)

• Figure 1: (a) Compositional counts of each element in the final dataset; (b) Gaussian distribution of $\ln k_p$ across the dataset; (c) Pearson coefficient matrix calculated for descriptors. Note: alloy components are excluded; (d) Model architecture development.
• Figure 2: (a) Parity plots of the ML predicted $\ln k_p$ versus the experimental $\ln k_p$ using five different ML models; (b) SHAP summary plot of the contribution of each feature to the model; (c) SHAP dependence plot of the top 20 features that correlate with the target variable.
• Figure 3: Predicted compositional dependence of oxidation resistance in the NiCoCrAl HEA system. The ternary projection shows the atomic fractions of Co, Ni, Cr and Al. Data points are colored by the predicted $\ln(k_p)$, where oxidation resistance increases with decreasing $\ln(k_p)$ values.
• Figure 4: Predicted oxidation behavior of NiCoCrAl alloys grouped into four compositional clusters: CrAl-free, CrAl-containing (CrAl), Cr-free, and Al-free. Alloys containing both Cr and Al exhibit the lowest $\ln(k_p)$ values, highlighting their synergistic role in forming stable, protective oxide scales, whereas CrAl-free alloys show the highest oxidation rates.
• Figure 5: Predicted isothermal oxidation behaviour of NiCoCrAl alloys at various temperatures after 250 h exposure, showing the variation of $\ln(k_p)$ with composition. Lower values indicate improved oxidation resistance, with Cr–Al-rich compositions maintaining superior performance across the temperature range.