A Multi-Model Collaborative Decision-Making Approach for Compound Fault Diagnosis

Compound faults inevitably occur in system-level equipment. Complex fault coupling and various combination types make diagnosis very challenging. This article proposes a multi-model collaborative decision-making approach for compound fault diagnosis, especially considering the generalization scenario for unseen compound fault combinations. The diagnosis results are determined by two indicators: decision quantity by category and model voting. In addition, batch sample diagnosis replaces the traditional one sample one-time diagnosis to further improve the accuracy of diagnosis and is beneficial for reducing false alarms. Experiments are conducted on a system-level subway bogie with compound faults. The results show that the proposed method achieves the highest accuracy of 90.95%, which is superior to advanced solutions.