A Robust Pantograph–Catenary Interaction Condition Monitoring Method Based on Deep Convolutional Network

Abstract

A pantograph–catenary system (PCS) is an important part of a railway power supply system, which is an interface of the power supply system and the electric locomotive. The quality of a traction power supply depends on the stability of the contact between a pantograph and a catenary. Therefore, it is necessary to monitor the contact state by detecting the contact point (CPT) between the pantograph and the catenary. Recently, automatic CPT detection methods based on video monitoring have been introduced to improve the railway operation safety. However, the existing methods were still not stable enough in complex backgrounds. To improve the stability of CPT detection, we proposed a method combining a deep convolutional network with handcrafted features to detect the CPT. The proposed method consists of two stages. First, a deep pantograph network (DPN) was adopted to segment the pantograph strip. The DPN was mainly composed of a deep pantograph detection network (DPDN) and a deep pantograph segmentation network (DPSN). Then, the edge detection and the Hough transform were used to detect the contact line above the pantograph. Concretely, the CPT was obtained by finding the intersection of the contact line and the upper surface of the pantograph strip. The experimental results demonstrated the robustness and the accuracy of the proposed method.