Fault Diagnosis of Circuit Breakers Based on MCF-RPs and Deep Residual Knowledge Incremental under Distillation Learning

The core aspect of online monitoring the health status of low-voltage switching appliances is the fault diagnosis of circuit breakers. Contemporary intelligent fault diagnosis methods primarily focus on improving the accuracy of fault detection. However, the performance of incremental diagnosis in dynamic scenarios faces significant challenges due to the growing volume of data from industrial streaming and the continuous accumulation of new fault data for monitoring. Therefore, A multi-stage incremental learning fault diagnosis method is proposed. Firstly, the preprocessed three-channel one-dimensional vibration signal data is converted to the recurrence plot of multi-channel fusion (MCF-RPs) through a process that involves a combination of time-frequency decomposition and nonlinear dynamics analysis. This approach aims to reduce the impact of nonlinear disturbances on the diagnostic model while comprehensively capturing the characteristics of faults. Secondly, under the framework of distillation learning, the improved ResNeSt18 incremental learning model is constructed by combining the sample retention and transfer recall. This effectively extract fault characteristics, reduces the occurrence of catastrophic forgetting and knowledge drift. Finally, the results of incremental learning diagnosis experiment on multi-phase fault types demonstrate that the suggested approach significantly improves the scalability and intelligence of the diagnostic model. This development holds great promise for tackling the practical industrial issues.