Signal Analysis Method for Automatic Flaw Classification on Pipeline Girth Weld Inspection by Ultrasonic Phased Array System

Presently, flaw classification of ultrasonic phased array systems is completely manual. It is dependent on the operator's testing experiences and errors are very easily introduced. In this article, "energy-status" method based on wavelet packet transform, having time-frequency signal analysis character, is applied to achieve ultrasonic echo signals' feature. Then, a feature library is built. Distribution regularity of the energies in the decomposed frequency bands is researched. In virtue of the following neural network and ISODATA dynamic cluster pattern recognition algorithms, automatic defect recognition is realized. Experiment is implemented on a flaw testing pipeline girth weld block with four common defects in it. Data collected by the ultrasonic phased array system prove the efficiency of the method.