Frontiers | A cable insulation defect classification method based on CNN-transformer

Abstract

Cable insulation defect detection ensures electrical safety, prevents accidents, extends equipment life and guarantees stable system operation. For the traditional cable insulation defect detection and identification of difficult problems, this paper proposes the use of ultrasonic cable insulation defect detection and combined with the Convolutional Neural Network (CNN)-transformer model of cable insulation defect classification method. Firstly, the ultrasonic probe is used to obtain different cable insulation defect signals, and then the CNN-transformer model is used to classify different cable insulation defects. The CNN is used to initially extract the characteristics of the cable insulation defects from the input signals, and then the multi-attention mechanism in the time series Transformer is used to extract the transient local and periodic global characteristics of the cable insulation defect signals. The deeper transient local features and periodic global features of the cable insulation defect signal are extracted by the multi-attention mechanism in the time series Transformer; finally, the recognition results are outputted by the fully connected layer and softmax classifier. The results show that ultrasonic reflection and transmission phenomena occur at the defects, and different defects can be accurately reflected by the defect echo time and amplitude, and the accuracy of cable insulation defect recognition using the CNN-transformer model reaches 100%, with good generalization ability.