Accelerating Finite Element Analysis of Dynamic Impact Response of TiN/Ti Multilayer Coatings Based on Small Sample Machine Learning

Engine high-pressure turbine blades operating in extreme environments, such as deserts, are subjected to long-term high-velocity impacts from sand particles carried by hot combustion gases, significantly reducing their service life. Owing to its high hardness and toughness, the TiN/Ti multilayer coating has emerged as a preferred surface coating material for such blades. However, its erosion resistance is highly dependent on structural parameters, and traditional experimental trial-and-error methods and finite element simulations are often time-consuming and labor-intensive. To address this challenge, this study proposes a TiN/Ti multilayer coating design framework that integrates small-sample machine learning (ML) with finite element analysis. Multiple regression algorithms were evaluated, and Gaussian process regression (GPR) was selected for its superior performance, enabling high-accuracy prediction of the maximum intralayer stress and the maximum plastic strain in the substrate under dynamic impact conditions (with R² values of 0.88 and 0.85, respectively). The modelʼs fitting capability was further enhanced through residual and uncertainty analyses. Moreover, shapley additive explanations (SHAP) analysis was employed to elucidate the contribution of each feature to the target variables. Finally, eight new coating structures under varying impact conditions were designed and simulated to validate the predictive accuracy of the ML model. This framework offers a data-efficient and computationally economical solution for impact-resistant coating design in high-dimensional parameter spaces.