Design And Implementation Of 4C Fault Detection Software For High-speed Railway Catenary Based On Deep Learning

With the continuous increase of high-speed railway mileage in China,the application demand for railway operation and maintenance automation has increased dramatically.The catenary is an important device for electrified railways and an important part of ensuring the safe and reliable operation of high-speed railways.Once the catenary system fails,it will cause a major traffic accident and bring huge threats and losses to people's lives and property.At this stage,the railway department's detection of catenary failure mainly relies on the technical personnel to judge whether there is a fault in the catenary images collected by the inspection vehicle in the manner of human eyes.Such detection methods are limited by the uncontrollable factors such as the level of personal experience of the technicians,the concentration of attention,ect.and there are problems such as long failure screening time and high risk of missed detection.This thesis takes images collected by the high-speed railway catenary's suspension state detection and monitoring system(4C system)installed on the catenary inspection vehicle as research object.The cutting-edge deep learning and computer vision object detection and recognition technology is applied to the fault detection task of high-speed railway catenary's suspension and support device parts.This thesis has studied and completed the following two tasks: First,the SSD network and the ResNet network are cascaded to realize the automatic fault detection of the high-speed railway catenary's flat wrist arm insulator.After the high-speed railway Beijing-Guangzhou line's Yueyang East to Changsha South section actual data testing,the insulator fault detection model is used to achieve 93.9% average precision and 94.2% fault classification accuracy,On the server with NVIDIA TITAN Xp Graphics Processing Unit,the average time for processing an image of the insulator fault detection model obtained in this thesis is only 33.8ms.Second,after investigating the power supply section units of several railway departments on the BeijingGuangzhou line,through in-depth demand acquisition and analysis,a set of high-speed railway catenary 4C fault detection software was designed and implemented,which can eliminate the 4C invalid data of high-speed rail catenary and automatic detect insulator and suspension string's failure,and reserved programming interface for extended integration of more component fault detection algorithms through strategic design mode.The first version of the software has been tested in the Guangzhou Power Supply Section of Guangzhou Railway Group,which laid a solid foundation for the iterative update and application of the software.