Federated Transfer Learning for Fault Diagnosis of High-speed Train Bogie With Data Security and Training Optimization

In the field of high-speed train fault diagnosis, great achievements have been made in recent years. Although it has achieved considerable accuracy, most of them require a large amount of labeled training data, which is difficult to collect and obtain in actual industrial environment. To solve this problem, this paper proposes a fault diagnosis framework based on federated transfer learning. Under the premise of protecting data privacy and security, to create a model with excellent generalization capabilities which can diagnose the client without the need for data labels, two clients work together to train the model. Additionally, an enhanced multi-scale convolution fusion technique is employed to extract and fuse fault features at various sizes, so as to further improve diagnostic accuracy in local model training. By verifying the fault data of the bogie, a diagnostic accuracy of 92.51% has been achieved for railway bogie with unlabeled data. This provides an effective solution for safety zero-shot fault diagnosis of high-speed train bogie, especially in the current situation of "data island" in the railway industry.