Multi-scale Siamese Network for Few-Shot Fault Diagnosis of Bogie Component

The diagnosis of the key components of train bogies is of great significance to the reliability and safety of train operation, and the intelligent diagnosis method shows great potential due to its excellent end-to-end capability. However, the traditional methods based on deep learning have the following limitations: 1. They require sufficient samples to drive model learning, and the acquisition cost of fault samples in actual engineering is high. 2. They did not fully take into account the deformation of features on a time scale under varying working conditions, resulting in the inability to improve the diagnostic accuracy. To overcome these limitations, a multi-scale siamese network is proposed in this paper. On the one hand, the siamese network realizes the effective learning of fault features by maximizing the similarity of similar samples and minimizing the similarity of different samples, and overfitting is effectively overcome. On the other hand, multi-scale convolution is introduced for siamese networks to extract time-scale complete features and achieve accurate diagnosis under variable operating conditions. The experimental results show that the proposed method has better diagnostic performance than the traditional deep learning network under the condition of changing working conditions and limited fault samples.