Tribological properties and machine learning prediction of FeCoCrNiAlN high entropy coatings

Abstract

High-entropy nitride ceramic coatings exhibit excellent wear resistance and are regarded as promising coatings in tools. Predicting the coefficients of friction (COF) and related microstrain under friction condition is critical for their application. In this study, nine different FeCoCrNiAlN high-entropy coatings were prepared with varying substrate bias voltage and nitrogen gas pressure using PVD CAIP (Physical Vapor Deposition-cathode arc-ion plating). The bias voltage and nitrogen pressure significantly influenced the wear resistance of the coatings. Low nitrogen gas pressure results in insufficient ceramic phase content, leading to lower hardness and the generation of more grooves. However, coatings with high hardness exhibit excellent wear resistance. Groove wear and debris wear are closely related to the surface particles and mechanical properties of the coating. Alterations in friction load and velocity were also conducted to investigate the tribological performance of various coatings. Results suggest that the COF increased with the friction load and velocity. Furthermore, 53 sets of COF data for machine learning (ML) predictions were obtained. Seven ML regression models, including Multilayer Perceptron (MLP), were employed to predict the COF using input features, including coating deposition process, microstructure features such as elemental composition, and grain size, as well as mechanical features such as nanohardness, H/E (Hardness/Young's modulus), Lc1 (Critical load). Friction conditions include friction load and velocity. Simultaneously, the significance of multiple features such as the coating preparation processes, microstructural, mechanical properties, and friction conditions in affecting the COF of the coatings, was analyzed by the feature importance model. The results indicate that the XGBoost regression algorithm exhibited the highest R2 value and the lowest RMSE, MAE, and MSE values, which were 0.8629, 0.0659, 0.0567, and 0.0047, respectively. This model demonstrated a favorable capability in predicting the COF of FeCoCrNiAlN high-entropy coatings. Furthermore, feature importance analysis revealed that the load, Lc1, velocity and Cr have the most significant impact on the COF of the coatings.