2-D defect profile reconstruction from ultrasonic guided waves signals adopting fuzzy wavelet packet and LS-SVM

The nondestructive testing is an important part of quality analysis and risk evaluation. Ultrasonic guided wave testing is extensively used in detecting and characterizing defects in natural gas and oil transmission pipelines. Defect profile reconstruction means establishing the profile parameters and constructing defect profile. In this paper, practical experiments and numerical simulations on propagation of L (0,2) guided wave testing are conducted to build the sample database for profile reconstruction. The echo signals, containing the defects information, are filtered the noise by a fuzzy wavelet packet denoising method. Considering the exiting limitations in defect profile reconstruction methods for ultrasonic guided wave inspection, the method based on least square support vector machine (LS-SVM) is established. Then, the denoised echo signals are adopted as the input data of LS-SVM and the parameters of defects are used as the output data. The reconstruction of 2-D profiles at axial width and radial depth of defects is achieved. Finally, the reconstruction ability between the LS-SVM method and the RBF Neural Network is compared. The comparison indicates that the reconstruction method based on LS-SVM processes high precision, robustness against noise, and good generalization ability. The proposed methods are effective approaches to defect detection and profile reconstruction in the ultrasonic guided waves inspection.