Optimizing the performance of the ZnO-based varistors using the neural networks technique

Abstract

This study aimed to propose a scientific method depending on neural networks technique to determine the optimal proportions of some impurities (Bi2O3, Nb2O5, MnO2, Co3O4, Cr2O3, NiO, La2O3, Ce2O3) used in manufacturing the ZnO-based varistors. The technique enables obtaining varistors with high structural homogeneity, high performance, and surge energy absorption capability. A nonlinear neural model between the nonlinear coefficient and the proportion of each impurity was accurately created. Subsequently, the neural models have been trained to approximate the correlation factor to the optimum value (R = 1), which proves the accuracy of the models. Then, using the optimization toolbox in MATLAB, the proportion of each impurity producing the highest nonlinear coefficient was determined. Afterwards, a nonlinear neural model representing the relationship between the most influential impurities combined (MnO2, NiO, Ce2O3, La2O3) and the nonlinear coefficient was built. Through training the neural model, the correlation factor was R = 0.99516, and the mean square error was MSE = 0.1078, proving the model's accuracy. Depending on the optimal solutions based on the Lagrange-Newton algorithm in MATLAB, the optimal proportions of impurities combined corresponding to the highest value of the nonlinear coefficient were determined. The accuracy of the proposed method was practically confirmed by preparing a varistor according to the optimal proportions and performing the measurements. The relative error between the theoretical value (α = 99.7522) and the practical value (α = 99.41) of the nonlinear coefficient was RE = 0.0035. That confirms that the proposed method has high reliability and accuracy to be adopted.