Predicting thermophysical properties and fracture toughness of Zirconia-based thermal barrier coating by interpretable machine learning method

Traditional thermal barrier coatings (TBCs) consisting of 6–8 wt% yttria-stabilized Zirconia remain significant challenges particularly at elevated temperatures, due to phase transitions that result in excessive volume expansion and crack formation. There is a pressing need to develop alternative TBC materials which exhibit better thermophysical and mechanical properties. Thermal conductivity (TC), thermal expansion coefficient (TEC), and fracture toughness (K_ic) are the three key properties that determine the long-term stability and performance of TBC materials. However, conventional methods such as trial-and-error experiments, density functional theory, and empirical models, suffer from low efficiency and high cost. To address these challenges, this study develops a machine learning (ML) framework to predict these three critical properties. A comprehensive dataset with 135 descriptors for K_ic and 136 descriptors for TC and TEC is curated, and a four-step feature selection process including Pearson correlation, Random Forest, Recursive Feature Elimination, and Best Subset Selection, is employed to identify the most influential features. The optimal feature subsets for TC, TEC, and K_ic are successfully reduced to 7, 5, and 5 features, respectively. Eight ML algorithms are evaluated, and model performance is assessed using determination coefficient (R²) and root mean square error (RMSE). Our results show robust predictive performance for TC and TEC, with R² values exceeding 0.95 for both, while K_ic predictions also achieve good accuracy. SHapley Additive exPlanations (SHAP) analysis is further used to interpret the contribution of individual features, revealing the significant impact of factors like ionic radius variation and temperature on material properties. This study demonstrates the effectiveness of ML in predicting TBC properties.