An intelligent nondestructive detection method based on wavelet processing and principal component analysis

To improve the performance of the acoustic nondestructive detection, an intelligent method was put forward. By using the wavelet transform (WT) with the optimal basis, the original acoustic resonance spectroscopy (ARS) signal was projected to the wavelet subspace at first, and then the signal was represented by a matrix of wavelet coefficients. To reduce the amount of calculation, the principal component analysis (PCA) was performed: The feature vector was obtained by Karhunen-Loeve transformation (K-L transformation), serving as the input of the neural network. Finally, a radial basis function (RBF) neural network was developed as a classifier using the recursive localized least square method. Simulation and experimental results showed that the proposed method is accurate and have good generalization ability.