An Ultrasonic Defect Identification Method Based on Wavelet Packet and PCA

In ultrasonic testing, qualitative analysis of defects is the key content of nondestructive testing and evaluation. In this paper, a detection method of defect classification is proposed, in which the recognition of the defect type is realized by extracting characteristic quantities of different defect wave signals. Firstly, the air-coupled ultrasonic detection system is used to collect defect-free signals and three different types of defect-wave signals, and the time-domain dimensionless parameters and wavelet packet energy coefficients of the signals are extracted to form multi-dimensional feature vectors. Then principal component analysis (PCA) is used to reduce the dimensionality of multi-dimensional feature vectors to obtain the feature fusion quantity. Finally, BP neural network system is input to classify the defect types, and compare with the test results without PCA processing. The experimental results show that the PCA treatment has higher accuracy and shorter test time.