Research on Fault Detection Technology for Circuit Breaker Operating Mechanism Combinations Based on Deep Residual Networks

Due to the complex mechanical structure of the spring-operated mechanism, its failure mechanisms often exhibit a multi-faceted nature, involving various potential failure sources. Therefore, conducting a failure mechanism analysis for multi-source faults in such systems is essential. This study focuses on the design of composite faults in combination operating mechanisms and develops simulation scenarios with varying levels of fault severity. Given the challenges of traditional simulation methods in performing quantitative analysis of core jamming faults and the susceptibility of the core’s motion trajectory to external interference, this paper innovatively installs a spring-damping device at the extended core position. This ensures that, during the simulation of core jamming faults, the motion trajectory remains stable and unaffected by external factors, while also enabling precise control over the degree of jamming. As a result, the simulation more accurately reflects real fault conditions, thereby enhancing the accuracy and practicality of diagnostic model outcomes. This study employs the Morlet wavelet transform to convert the current and displacement signals in the time series into time-frequency spectrograms. These spectrograms are then processed using the ResNet50 deep residual neural network for feature extraction and fault classification. The results demonstrate that, when addressing the diagnostic problem of small-sample data for operating mechanism faults, ResNet50, with its residual structure design, exhibits significant advantages. The convolutional layer strategy, which first performs dimensionality reduction followed by dimensionality expansion, combined with the use of the ReLU activation function, contributes to superior performance. This approach achieves a fault recognition accuracy of up to 91.67%.