A Novel Meta-Learning and Network Architecture Search Approach for Few-Shot High-Voltage Circuit Breaker Fault Diagnosis

In recent years, convolutional neural networks (CNNs) have achieved worth seeing results in mechanical fault diagnosis of high-voltage circuit breakers (HVCBs) due to their powerful classification capabilities. However, due to the discrepancy in the vibration signals of different voltage levels and types of HVCBs, it is difficult for the model developed on one dataset to be generalized and deployed to all scenarios, especially in the case of small samples in the field. To this end, this paper proposes a method for HVCB mechanical fault diagnosis based on meta-learning (ML) and neural architecture search (NAS). Firstly, NAS is adopted to automatically obtain the network model with the best existing modal performance. ML is then utilized to learn the design experience of the fault diagnosis model from the NAS process of existing modalities. When it is finally deployed in the field, the gradient update is performed on the basis of the learned design experience, that is, the HVCB fault diagnosis model can be quickly obtained under the condition of a small sample. The validity and feasibility of the proposed method are verified by laboratory and field data. The diagnostic accuracy of small samples on site reaches 94.15%, which provides a novel solution for HVCB fault diagnosis on-site.