A Loose Default Diagnosis Method for Oblique Bracing Wire in High-Speed Railway

摘要

Oblique Bracing Wire (OBW) is an important device of catenary support components (CSCs), which supports the lower steady arm and the registration tube to keep the overhead line high and the pull-out value within the specified range. OBW fault can cause unstable train operation or safety hazards to trains and passengers. With the development of deep learning, it has been tried to achieve the precise location and fault detection of the CSCs to ensure the stable operation of the train. In this article, three deep learning frameworks called Faster RCNN ResNet101, SSD (Single shot multi-box detector) and YOLOv2(You only look once) are used to achieve location for Bracing wire hook and Messenger wire base respectively. Through the comparison of the location effects of the three frameworks on CSCs, the Faster RCNN ResNet101 is chosen as the framework for location. Then curvature detection algorithm is used for loose default of OBW. The experiment results show that the proposed fault detection method has high diagnostic rate and universality.