Research on Federated Learning Privacy Protection Model for Valve Fault Intelligent Diagnosis

Control valves are the core equipment in the braking system and fluid lines of rail transportation, and control valve troubleshooting can effectively enhance the service reliability of rail transportation equipment, improve the work safety, and reduce the maintenance cost. Current control valve fault diagnosis methods usually require sufficient high-quality condition monitoring data for model training, and the engineering application of diagnostic models often faces failure problems due to the difficulty of obtaining labeled sample data, “data silos” caused by data privacy protection, and the demand for cloud-side-end collaborative operation and maintenance. To this end, we propose a different specification control valve fault diagnosis method, GSI-FL, for multi-user siloed private data federated model migration, which performs a grayscale graph transformation on multi-user traffic data, uses convolutional neural networks to extract features and construct user local models, improves parameterization strategies to reduce the number of local model parameters, and collaborates with multiple users to construct a shared model for migration learning and sends it to the users, who use the local data to fine-tune the shared model to improve its performance. Users use the local data to fine-tune the shared model to realize fault diagnosis of control valves with different specifications. The test verifies the accuracy and effectiveness of the proposed method.