Exploration of the oxidation and ablation resistance of ultra-high-temperature ceramic coatings using machine learning

Abstract

Carbon fiber-reinforced carbon matrix composites (C/C) will be easily oxidized in high temperatures, which will have a great negative effect on their performance. Preparing ultra-high-temperature ceramic (UHTC) coatings is a well-established method to improve the oxidation and ablation resistance of C/C. However, it is time-consuming and costly to obtain these coatings through the traditional experimental method. Motivated by the outstanding performance of machine learning (ML) algorithms in many fields, this study adopts ML algorithms based on historical experimental datasets to build a model. This model will predict the oxidation and ablation resistance, represented by mass ablation rate. For this purpose, variables that affect the mass ablation rate and are easily accessible were used as input features. That includes the chemical composition and essential physics/chemistry properties of coatings and experimental parameters. Seven different ML algorithms were used to establish the model; namely, ridge regression (Ridge), lasso regression (Lasso), kernel ridge regression (KRR), support vector regression (SVR), random forest regression (RFR), AdaBoost regression (ABR), and bagging regression (Bagging). The results show that RFR has the optimal generalization performance with a mean absolute error (MAE) of 0.55, mean-squared error (MSE) of 0.71 and coefficient of determination (R2) of 0.87 on the testing set. SHapley Additive exPlanations (SHAP) analysis of the RFR model explained how these input features affect the mass ablation rate and further provided the critical features for performance prediction. The model established in this study can predict coating performance accurately and accelerate the development of UHTC-coated C/C composites from a data-driven perspective.