For the common ultrasonic echo signals of porosity

unfused and defect-free aluminum alloy spot welded joints

noise reduction and reconstruction were first performed with empirical mode decomposition (EMD) analysis

and multi-scale eigenvalues in the time and frequency domains were extracted using statistical methods to analyze the eigenvalue variation patterns of different defects. The principal component analysis (PCA) and linear discriminant analysis (LDA) are used to optimize the feature values and obtain the principal element features of the defective echo signal

finally the principal element features are used as the input of the BP neural network to identify the defect signal. The experimental results show that the feature quantities constructed by the two dimensionality reduction methods have better classification results compared with those without dimensionality reduction

with PCA acting better and effectively improving the defect recognition accuracy of BP neural networks.