Optimization of Power Frequency Withstand Voltage Characteristics of Thermal Electrochemical Oxide Ceramic Film Based on Machine Learning

The thermal electrochemical oxidation ceramic film layer on the surface of aluminum alloy has high thermal conductivity and insulation properties. The purpose of this paper is to determine and optimize the dielectric withstand voltage characteristics of the ceramic layer through machine learning. Three methods, least squares support vector machine (LSSVM), BP neural network, and linear regression are used to establish the model between power frequency breakdown voltage and key process parameters: production line speed, current density, and electrolyte temperature. The average relative error and the maximum relative error between the predicted value and the true value of the prediction data set are used to evaluate the accuracy of the three models. The average prediction relative errors of the three models are 4.72%, 5.46%, 13.92%, and the maximum prediction relative errors were 9.69%, 11.49%, and 32.21%, respectively. The prediction results show that the LSSVM model has the highest accuracy. Based on the above research, the particle swarm optimization (PSO) algorithm is used to optimize the LSSVM model to find the maximum breakdown voltage of 551V, thereby increasing the power frequency breakdown voltage.