Machine learning descriptors for predicting the high temperature oxidation of refractory complex concentrated alloys

Abstract

Refractory Complex Concentrated Alloys (RCCAs) can exhibit exceptional high-temperature strength, making such alloys promising candidates for high-temperature structural applications. However, current RCCAs do not possess the high-temperature oxidation resistance required to survive in oxidizing environments for more than a few hours at or above 1000$^\circ$C, without relying primarily on an environmental barrier coating. Here, we present a machine-learning framework designed to predict the oxidation-induced specific mass changes of RCCAs exposed for 24 h at 1000$^\circ$C in air, in order to support the search for oxidation-resistant alloys over a wide range of compositions. A database was constructed of experimental specific mass change data, upon oxidation at 900-1000$^\circ$C for 24 h in air, for 77 compositions comprised of simple elements, binary alloys, and higher-order elemental systems. We then developed a Gaussian Process Regression (GPR) model with physics-informed descriptors based on oxidation products, capturing the fundamental chemistry of oxide formation and stability. Application of this GPR model to the database yielded a MAE (mean absolute error) test score of 5.78 mg/cm$^2$, which was a significant improvement in accuracy relative to models only utilizing traditional alloy-based descriptors. Our model was used to screen over 5,100 quaternary RCCAs, revealing compositions with significantly lower predicted specific mass changes compared to existing literature sources. Overall, this work establishes a versatile and efficient strategy to accelerate the discovery of next-generation RCCAs with enhanced resistance to extreme environments.