Abstract:

To address the problems that single diagnostic signals of air-type AC contactors are easily disturbed and difficult to extract features from,a fault diagnosis method based on sound-vibration fusion and an improved temporal convolutional network (TCN) is proposed.The sound and vibration signals are denoised using parameter-optimized feature mode decomposition (FMD) and an improved wavelet threshold algorithm,and a fault database is established.According to the closing mechanical characteristics,the signals are divided into three stages for time-domain and local energy feature extraction,followed by t-SNE-based dimensionality reduction and visualization.Based on TCN,bidirectional gated recurrent unit (BiGRU) and attention mechanisms are introduced to construct an improved TCN composite diagnostic model for enhancing long-term dependency modeling,bidirectional dynamic feature extraction and representation of key fault segments.Experimental results show that the diagnostic accuracies for three contactors reach 97.33%,94.67%,and 96.00%,respectively,outperforming traditional models and demonstrating the effectiveness and engineering applicability of the proposed method.

Key words: air-type AC contactor, fault diagnosis, closing mechanical characteristics, feature extraction, temporal convolutional network (TCN) composite model