基于深度学习的高铁接触网旋转双耳开口销钉缺失故障检测

Fault Detection of Missing Split Pins in Swivel with Clevis in High-speed Railway Catenary Based on Deep Learning

针对高速铁路接触网支撑装置中旋转双耳开口销钉缺失故障检测的问题,提出一种基于深度卷积神经网络和集成学习的故障检测方法。通过Faster R-CNN网络对旋转双耳整体进行精确定位;在整体定位结果的基础上,进一步完成对开口销钉的精确定位,最大程度上降低背景对故障检测的干扰;通过多个深度卷积神经网络提取开口销钉图像的多种特征,最终由多个线性SVM构成的集成分类器实现开口销钉缺失故障检测。实验结果表明:本方法能在复杂的接触网支撑装置图像中实现旋转双耳开口销钉的精确定位,并且在销钉的缺失故障检测中表现出较高的可靠性。

A fault detection method based on deep convolution neural network and ensemble learning was proposed to detect the fault of missing split pins of swivel with clevis,which is an important part of high-speed railway catenary support system.First,the swivel with clevis was precisely located by the Faster R-CNN network.Then,based on the overall locating results,the precise location of the split pins was further accomplished,to minimize the interference of the background to fault detection.Finally,a variety of features of split pins image were extracted through a number of deep neural networks,and the split pins missing fault detection was achieved by the ensemble classifier,composed of a number of linear SVM classifiers.The experimental results show that the proposed method can realize the precise location of the swivel with clevis and the split pins in the complex catenary support system image,and can achieve the split pins missing fault detection with high reliability.