Parametric Fault Diagnosis Using Transfer Cerebellar Model Neural Network for Power Electronic Converter

The intelligent fault diagnosis methods have become increasingly popular in the field of power electronic converters fault diagnosis. However, in industrial applications, only a small amount of labeled data is generally available, which results in a decrease in diagnosis accuracy. In this article, a novel method called transfer cerebellar model neural network (TCMNN) is proposed for a dual-buck bidirectional dc–ac converter parametric fault diagnosis with few training samples. First, the feature transfer strategy is used to reduce the distribution difference between the target domain samples and the auxiliary samples from different working conditions. Second, a fault classifier based on improved cerebellar model neural network (CMNN) is designed to achieve more effective and accurate classification of fault information. Third, pretrain a fault classifier using the auxiliary samples after feature transfer, and transfer its parameters to the target model. Finally, the target model is fine-tuned with few training samples after parameters transfer in the target domain to obtain a fault classifier adapted to the unlabeled testing samples. The simulation and experimental results suggest that the proposed TCMNN still has better fault diagnosis performance even with few training samples, which is superior to other intelligent diagnosis methods. Most importance is the developed fault diagnosis system with few training data is more useful for general practical systems.