The fault diagnosis of catenary system based on the deep learning method in the railway industry

Abstract

The catenary system plays a vital role in the railway industry, which is associated with the security and efficiency of the train operation. The fault diagnosis and anomaly detection of the catenary system is of significance. The current carrying ring and dropper are important parts of catenary and attract attention in the inspection process. Based on the image processing technique and deep learning method, the fault diagnosis method of the catenary system is presented. The fault diagnosis of catenary system consists of three parts, top current carrying ring, dropper and bottom current carrying ring detection. The feature pyramid network is applied for the various scales units of catenary system in image from inspection vehicle. Based on the modified CenterNet, the current carrying ring is detected. The results of the located rings are chosen through specific selection. Then the selected top and bottom rings are matched further through the location relationship. Based on the matched rings, the dropper is located and then classified by the classification network. According to the experiments on the plenty of catenary image datasets, it shows that the method have efficient and satisfied performance on the fault diagnosis of the catenary system.