High temperature alloys span multiple classes of materials including low alloy steels, stainless steels, nickel-chromium alloys, superalloys, aluminides, and, of more recent interest high entropy alloys, among others. Whereas many high temperature alloys deviate from the parabolic oxide growth law, the parabolic rate constant kp remains a useful indicator of the oxidation susceptibility for a given material. To design new classes of materials, and help with materials selection, it would be useful to directly predict the oxidation rate constants from materials features, such as composition and microstructure. With this goal in mind, parabolic rate constants have been collected from the literature for 75 alloys exposed to temperatures between 900 and 3000°F. Environments incorporated into the analysis include lab air, ambient and supercritical carbon dioxide, supercritical water, and steam. Predictive models for the oxidation rate constant were developed using machine learning and analyzed to provide insights into the leading factors producing corrosion resistance in these materials.
High-temperature service places severe constraints on materials selection due to a combination of factors including the formation of oxide films, spallation and volatilization, and deterioration in mechanical properties. Materials selection is principally informed by laboratory testing under simulated conditions of temperature, thermo-mechanical fatigue, and environment chemistry (such as the presence of steam, exhaust gas chemistry, or salts). Models for predicting the high temperature performance of materials a priori are an active area for development, and are currently focused on elements such as predicting oxide formation, microstructure evolution and reduced order models for creep.[1] Given that microstructure evolution in the near-surface region is a strong function of the reduction in near-surface concentration of the oxide-forming elements (principally Cr and Al), understanding the rate of alloy oxidation is a first step towards constructing life prediction models for high-temperature service. Herein, a machine learning approach is applied to understand the influence of materials composition variables on the oxidation rate constant for a variety of metal alloys for which data was collected from the literature. This work builds on a prior investigation [2] in which random forest regression was used to predict the oxidation rate constant. In this paper, feature selection, k-means clustering and artificial neural networks are used to identify key the composition variables that influence oxidation rate, organize the materials into "clusters" (alloy families), and develop quantitative models for predicting oxidation rate as a function of alloy composition and temperature. This approach provides a platform for further investigations that will begin to explore the role of gas chemistry as well as other potential factors such as materials microstructure and the role of minor alloying elements.
