Parallel Relation Network for Intelligent Fault Detection and Localization of Train Transmission Systems with Zero-fault Sample

In recent years, deep learning methods have empowered the fault detection of train transmission systems (TTS). However, the existing methods require a large number of fault samples, which are often difficult to obtain in TTS. In order to address the challenge, we propose a parallel relation network (PRN) for fault detection of TTS. The PRN method uses residual shrinkage network to extract the effective features from input samples and then form feature pairs to describe the relationship between the samples. Finally, Kolmogorov-Arnold networks (KAN) is used to assess the health states of the samples. In addition, we generate different auxiliary sample libraries to enhance the model's ability to extract features inherent to the health state. In experiments, we verified the feasibility of the PRN method, which achieved more than 98% accuracy and recall in different fault states of TTS.