A Novel Data Generation and Quantitative Characterization Method of Motor Static Eccentricity With Adversarial Network
Wei Sun, Haowen Wang, Ronghai Qu
Abstract
Eccentricity can cause motor damage. It is necessary to obtain the degree of motor eccentricity for health management or eccentricity suppression. Due to the complex electromagnetic mechanism, it is hard to obtain the degree of motor eccentricity in traditional way. The neural network is suitable for obtaining the degree of eccentricity, but the eccentricity data are too few to train network. In this article, a data generation and quantitative characterization method for motor eccentricity based on improved generative adversarial network is proposed. The mathematical model is improved for pretraining, and the network and objective function are improved for data generation and eccentricity characterization. The results show the validity of the proposed method.