Vibration signals from high-voltage circuit breakers (HVCB) typically contain complex background noise, and traditional fault diagnosis methods often neglect the temporal relationship between vibration signals and fault characteristics. To address these issues, an IMCEEMDAN-TSG fault diagnosis model based on vibration signals is proposed. First, Pearson correlation coefficient filtering is combined to improve the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (IMCEEMDAN) for adaptive multi-resolution analysis, which effectively separates the intrinsic mode function (IMF), thereby filtering out noise contained in the signal, suppressing mode aliasing, and preserving key signal features. Secondly, a TSG hybrid algorithm is constructed by combining the Temporal Convolutional Network (TCN) embedded with the Self-Attention Mechanism (SAM) and the Gated Recurrent Unit (GRU). This architecture facilitates the parallel feature extraction of multi-channel IMF and the capture of temporal relationships, thereby deeply modeling temporal dependencies and revealing the dynamic evolution of vibration signals. Experimental results demonstrate that the proposed model achieved a fault diagnosis accuracy of 100 % on the HVCB simulation datasets, surpassing the traditional Convolutional Neural Network (CNN) by 19.07 %. Furthermore, compared with conventional algorithms, significant improvements were observed across all classification metrics, providing an accurate and reliable solution for the mechanical fault diagnosis of HVCB.
Yu et al. (Thu,) studied this question.