Abstract: In the weld tested ultrasonic signal, noise is the leading factors that interfering defect detection. In order to suppress ultrasonic noise and recognize defect signal effectively, an improved non-linear wavelet shrinkage (NWS) method is proposed on the basis of the cross-correlation of sub-waves. The proposed method inherits the merit of multi-resolution of wavelet, and takes cross-correlation between defects waves in the two adjacent signals into account at the same time. By employing the conventional and improved NWS methods, both computer simulated data and weldment tested ultrasonic signal are de-noising processed. The effects of mother wavelets and threshold estimation methods on the noise suppression are researched. Compared with conventional NWS, the improved method is robust to mother wavelets and threshold estimation methods, and shows wide adaptability to noise of complicated frequency. The proposed method has the technical advantage of both noise suppression and defect wave reconstruction, and can effectively identify and quantitatively detect of defects in ultrasonic tested signal.

Key words: Defect detection, Noise suppression, Ultrasonic signal, Wavelet