An Automatic Fracture Defect Detection Approach for Current-carrying Rings of Catenary Droppers Using A Multi-task Neural Network

Abstract

Catenary droppers play an essential role in the catenary system. They both connect messenger wires and contact wires to stabilize contact wires and transmit the working current and short-circuit current. And the current carrying rings of catenary droppers can avoid the direct connect of droppers and dropper clips and reduce electrical faults and arcs when the locomotive draws current through the pantograph. At present, researchers mainly focus on the loosen defect of droppers, but still do not pay attention to the fracture defect of the current-carrying ring of droppers. Therefore, this paper proposes an automatic fracture defect detection approach for the current-carrying ring of droppers using a multi-task neural network. Compared to traditional solutions based on the connected domains, the method based on neural networks is more automated and robust. First, this network consisted of three function head networks of a classification head network, a regression head network, and a mask head network is performed to get a classification score and a segmentation score, respectively. And then, by counting the scores of normal and faulty components, a fault criterion for evaluating the current-carrying ring is proposed. Experiment results show that the proposed method is highly accurate and automatic for the state detection of catenary droppers.