Research on Weld Defect Identification Technology Based on EMD and BP Neural Network

Abstract

Aiming at the research problem of weld defect type recognition based on ultrasonic signals, an automatic recognition method was proposed based on the combination of empirical mode decomposition (EMD) and BP neural network. Firstly, EMD was used to decompose the ultrasonic A-scan signals of different weld defects, and some intrinsic modal functions (IMF) of the defect signals were obtained. Then the correlation between the IMF and the original signal is carried out, and dimensionality reduction is carried out based on the eigenvalues of the parameters of the IMF. The final weld defect using BP neural network as a classifier, the intrinsic mode function of time domain and frequency domain features as input parameters to the BP neural network for training decisions, and aim to achieve the defect types automatic recognition. The experimental results show that the method can accurately classify weld internal defect information, comprehensive recognition accuracy rate reached 94%, It has good engineering application value.