Plasma electrolytic oxidation (PEO) coatings on AZ31 magnesium alloy: An interpretable machine-learning study of nano-SiC and process parameters

This study presents a machine learning-based model used in predicting coating thickness and corrosion current in Plasma Electrolytic Oxidation (PEO) coating on AZ31 magnesium alloy. The goal is to use the previously published experimental data to develop accurate and interpretable models that can be used to aid in parameter selection and process optimization. The key input variables are electrolyte family (aluminate-silicate or phosphate), current density, processing time, and the presence of nano-SiC. Several models of machine learning, such as Multiple Linear Regression (MLR), Support Vector Regression (SVR), Artificial Neural Networks (ANN), Random Forest Regression (RFR) and XGBoost, were trained and tested, and the hyperparameters were optimized to maximize the predictive accuracy. It was demonstrated that ANN and SVR models were the most accurate to predict coating thickness (R 2 = 0.95, MSE = 0.0041) and SVR and RFR models were the most effective to predict corrosion current (R 2 = 0.94). Visualization methods such as kernel density estimation (KDE) plots and process heatmaps were employed, in order to investigate correlations between inputs and outputs. Trends were seen that were expected due to known physical behaviors: reduced corrosion current with SiC addition and increased coating thickness with greater current densities. Even though the study does not imply new experimental validation, the proposed framework helps to provide a reliable and transferable method of conducting experimental designs and enhancing the optimization of the PEO processes with the help of machine learning.