Advancing micromechanical property characterization in ceramic multilayer coatings via hierarchical machine learning

This study focuses on a numerical data-driven machine learning (ML) approach applied to predict critical parameters, including hardness, Von Mises stress, and equivalent plastic strain in various ceramic multilayer coatings on a Ti alloy substrate, such as Ti/TiN, Ti/TiVN, Ti/TiZrN, Cr/CrN, Cr/CrAlN, and Ta/Ti-Zr-Ta, through the nanoindentation process. The regression analysis demonstrated the model’s effectiveness in predicting these parameters, with heightened accuracy in hardness and stress compared to plastic strain. The remarkable efficiency of the proposed hierarchical ML model derives from its ability to unravel complex interdependencies within the dataset, revealing subtle relationships that traditional models often overlook. The outcomes also revealed a direct correlation between increases in output targets, such as hardness and average Von Mises stress, and the amplification of weight factors associated with processing parameters. Conversely, heightened values of equivalent plastic strain demonstrated a proportional increase in weight factors associated with material properties. This observation underscores the individual contributions of processing parameters and material characteristics in modeling the mechanical behavior of multilayer coatings. Moreover, the ML model significantly enhanced the predictive performance for multilayer coatings by providing a detailed relevance score for the material properties of the layers. These properties included factors such as elastic modulus, hardness, Poisson ratio, and yield strength.