Life Prediction of Electromagnetic Relay Based on GRU-1dCNN

Abstract

In recent years, the remaining life prediction of high-reliability and long-life electrical components such as electromagnetic relays has become the research focus and difficult research topic. According to the characteristics of electromagnetic relay performance parameters, a GRU-improved one-dimensional convolutional neural network electromagnetic relay remaining life prediction method is proposed. Firstly, collect the relay's performance degradation data through the electrical performance simulation experiment system of the contact material. In order to make the model more convenient to extract the data features, the Kalman filter algorithm is used to reduce noise and smooth the relay's performance parameters. Then, using the keras deep learning framework and convolutional neural network, the remaining life prediction model of the contact material is established, and the mean square error is used as the loss function to evaluate the performance of the model. The results show that the model's test set accuracy rate reaches 96%, which improves the prediction accuracy compared with the traditional one-dimensional convolutional neural network model.