Ultrasonic Flaw Signal Classification using Wavelet Transform and Support Vector Machine

This paper presents a ultrasonic flaw signal classification system by using wavelet transform and support vector machine (SVM). A digital flaw detector is first used to acquire the signals of defective carbon fiber reinforced polymer (CFRP) specimen with void, delamination and debonding. After that, the time domain based ultrasonic signals can be processed by discrete wavelet transform (DWT), and informative features are extracted from DWT coefficients representation of signals. Finally, feature vectors selected by PCA method are taken as input to train the SVM classifier. Furthermore, the selection of SVM parameters and kernel function has been examined in details. Experimental results validate that the model coupled with wavelet transform and SVM is a promising tool to deal with classification for ultrasonic flaw signals.