New flaw feature extraction method for ultrasonic phased array nondestructive testing

To realize automatic flaw classification of an ultrasonic phased array system, the key is to extract flaw feature that can mostly represent flaw´s nature from ultrasonic signals. As ultrasonic echoes are nonstationary signals, wavelet packet transform (WPT) is a common feature extraction method. But it is fit for off-line signal analysis. In this article, lifted wavelet transform (LWT) is studied to improve the computation speed, not at the cost of decreasing the computation precision. Separability measure based on Euclidean distance is introduced as the evaluation criterion to assess flaw feature extraction performance. An ultrasonic phased array system is implemented to collect ultrasonic echoes of four common types of flaws machined in a pipeline girth weld block. Both extraction algorithms are tested on the same echo signals. The experiment results indicate the classification speed of LWT-based feature extraction algorithm is almost two times of the WPT-based feature extraction algorithm, and the separability measure of the former is similar to the latter, which means classification precision is not reduced. The proposed algorithm can find applications in other similar fields of signal processing.