Online defect recognition of narrow overlap weld based on two-stage recognition model combining continuous wavelet transform and convolutional neural network

Abstract

To improve the defect identification of eddy current detection signals for narrow lap welds, we propose a method, which combines continuous wavelet transform (CWT) with convolutional neural network (CNN). Firstly, a two-dimensional time-frequency diagram is generated from the one-dimensional eddy current signal through CWT, and then the time-frequency diagram of the eddy current signal is obtained as the input of CNN. In order to meet the real-time requirement of narrow overlap weld defect recognition in practical inspection, a two-stage defect recognition model is further proposed. The first stage is to detect the anomaly of narrow lap weld based on CWT and CNN, and the second stage is to classify and identify defects by combining CWT and CNN as well. Through case study, the accuracy of our method is 96.94%, which is nearly 10% higher than the traditional method. Furthermore, the actual average detection time is only 2.4 s, making the proposed approach capable of being deployed for enterprises’ online operation.