| dc.description.abstract | Abstract. Arc-erosion faults in fretting electrical contacts may pose a serious threat to the reliability of electrical connectors. However, accurately evaluating arc-erosion severity online, particularly identifying weak arc erosion, remains challenging. To address this problem, continuous wavelet transform (CWT) scalograms were employed to capture the nonstationary time–frequency characteristics of vibration signals. The CWT scalograms were further integrated with a visual geometry group (VGG) convolutional neural network to develop an intelligent arc-erosion severity evaluation method, termed CWT-VGG. Experimental results indicate that CWT scalograms offer markedly stronger discriminative power for arc-erosion severities than time-domain vibration signals. Notably, compared with five other representative methods, the proposed CWT-VGG method yields the highest average evaluation accuracy (97.03%) and the most stable performance across repeated trials. This study holds significant value for advancing early condition-based maintenance of electrical connectors. | |
