The Research on Life Prediction Method of Safety Relay

Abstract

In order to improve the reliability and utilization of the safety relay, in this paper, failure mechanism and life prediction method of the railway signal relay were deeply studied. A variety of multivariate statistical analysis methods such as PCA and Mahalanobis distance are adopted to do data fusion and feature extraction of relay contact resistance, closing time, release time, super-path time, arc time and bounce time degradation parameters in the process of life test. Thus, parameters of the safety relay test are extracted and all of safety relays could be classified into different failure mechanisms. Then, the optimum identification parameters would be selected as prediction variables by Fisher discrimination criterion, and the failure threshold was defined. On this basis, we build the model of BP Neural Net Time Series to train the prediction variables and predict the change trend of variables. By comparing the prediction variables with failure threshold, the remaining life of relay is obtained. Finally, contrasting the real life of the failure relay with the life predicted by the regression models, we come out the precision of the prediction model. Before predicting data, we use the way of eliminating abnormal data, smoothing line and methods such as wavelet transform to eliminate noise. It will improve the accuracy of life prediction. This study can realize the accurate forecast of safety relay’s life in practical simulation.