Predicting the microhardness of alumina-based ceramics using machine learning methods

Abstract

To mitigate the substantial labor, time, and material costs associated with laboratory testing, this study proposes predicting the microhardness of Al2O3-based ceramics using machine learning methods. A database was compiled from a comprehensive analysis of literature to predict the properties of alumina ceramics. The input variables include chemical composition, density, sintering temperature, and dwell time for alumina ceramics doped with ZrO2, ZrO2−Y2O3, CeO2, MgO, CaO, and SrO. The microhardness of Al2O3-based ceramics was predicted using Support Vector Regression (SVR), Random Forest, Gradient Boosting, and Ridge Regression models. To determine the predictive performance of the models, the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Maximum Error (Max Error), and the coefficient of determination (R2) were calculated, quantifying the deviation of the predicted microhardness values from the actual ones. The microhardness prediction model based on Support Vector Regression (SVR) is characterized by high predictive efficiency, as evidenced by a high coefficient of determination (R2).