Hybrid multi-scale residual network for high-voltage circuit breakers fault diagnosis

In this letter, a novel Hybrid Multi-Scale Residual Network (HMSR-Net) is proposed to improve the accuracy of fault diagnosis in the spring operating mechanisms of High-Voltage Circuit Breakers (HVCBs). To facilitate enhanced fault classification across diverse operational conditions, five distinct fault states are simulated in addition to a baseline healthy condition, thereby capturing the nuanced and multi-scale features of vibration signals within varying working environments. The vibration signals generated during spring energy release are collected utilizing a test platform specifically for HVCBs. These one-dimensional time-series signals are transformed into two-dimensional time-frequency images by wavelet transform, thereby offering a precise and multi-scale representation of local signal characteristics. Subsequently, the generated time-frequency images are processed by the proposed HMSR-Net, extracting deep and discriminative features that are further classified with a Multi-Layer Perceptron (MLP) classifier. Experimental evaluations on field-measured fault vibration data reveal that HMSR-Net achieves a fault recognition accuracy of 98.2%, which notably outperforms other machine learning and deep learning approaches.