A Fault Detection Method of High Speed Train Pantograph Based on Deep Learning

The fault of pantograph will seriously reduce the reliability of power supply of train[1]. Faults such as spark discharge, abnormal pantograph dropping and pantograph defect can be displayed by image analysis. Based on the study of space breakdown mechanism of spark discharge between pantograph and contact wire, this paper explores the generation and development conditions of sparks in specific air insulation links, and proposes a fast and accurate method combined with deep learning to detect pantograph fault. This method uses the pantograph pictures obtained from different sources as training samples to train the deep learning convolution neural network with the improved yolov3 algorithm. After off-line training, generate the prediction model, which can extract the pantograph from the image captured by the high-speed vehicle camera[2]. Based on the extracted pantograph images, this paper designs an algorithm to eliminate the error of image shooting, analyses the color characteristics, and finally realizes the electric fault identification of pantograph.