Few-Shot Mechanical Fault Diagnosis for a High-Voltage Circuit Breaker via a Transformer-Convolutional Neural Network and Metric Meta-Learning

High-voltage circuit breakers (HVCBs) are responsible for the vital tasks of control and protection in power grids. Strengthening research on the latent fault diagnosis of HVCBs is vital for improving their reliability in operation. However, current fault diagnosis models are all developed on sufficient samples, which is unrealistic for on-site HVCBs. In addition, these current models were developed on specific datasets and are difficult to generalize to other datasets, which restricts the development of HVCB fault diagnosis. To resolve this issue, a transformer and metric meta-learning (TML) model is proposed for few-shot on-site HVCB diagnosis. First, we propose a hybrid module of a transformer-convolutional neural network to extract fault features, which captures local and global features. Then, fault classification of HVCBs is achieved by using a prototype network (PN). In the PN, a prototype-rectified classification strategy is introduced to address the bias of intraclass prototypes. Moreover, near-neighbor boundary loss is introduced to correct for intraclass and interclass distributions of fault features, and the boundary of the class prototype is clarified. The experimental results reveal that the diagnostic accuracy of TML when applied to field HVCBs exceeds 95%, realizing high-precision and robust diagnosis of HVCB faults.