A novel method of feature extraction for minor crack identification

Ultrasonic testing technique has been widely applied for monitoring the metal structure health. It is useful to detect and access the damage condition with the minor crack information being concealed in the ultrasonic signal. Although there has been a large amount of studies, the extraction of robust minor crack features is still a fundamental problem. In this paper, a novel crack identification algorithm is proposed by the wavelet packet transform (WPT) of received signal. With the calculation of sub-band signal energy, the most suitable decomposition level is decided. Then, the features are defined by the correlation coefficient between the damaged signal and undamaged signal. With principal component analysis (PCA), the feature extraction is achieved by reducing the overlapped and redundant ones. Finally, the extracted features are fed into support vector machines (SVM) classier and their outputs are employed to classify the damage type. The performance of the proposed method is confirmed with practical experiment. It indicated that compared with other methods, the proposed algorithm has a higher identification accuracy with more robust features.