1D Convolutional Neural Networks For Fault Diagnosis of High-speed Train Bogie

Abstract

With the development of high-speed train (HST), fault diagnosis of bogie has become a research hotspot in the field of train stability. In this paper, a pattern recognition method is presented, which uses one-dimensional convolutional neural network to extract the deep features of HST fault signal. The proposed CNN model consists of 8 layers besides the input layer and output layer, including three convolutional layers, three downsampling layers, and two fully connected layers. This model can automatically complete the feature extraction and selection of the original data, thus achieving the classification of the faults of the bogie under 7 working conditions, i.e., normal, air spring fault, lateral damper malfunction, anti-yaw damper failure, and three mixed fault types generated by the combined influence of each two different single fault types. Experimental results show that the classification accuracy achieves 96.4%, which verifies the validity of the proposed method.