Abstract
The flaw classification has become one of important tasks in evaluating the seafloor petroleum transporting pipeline integrity. With the aim to classify different kinds of artificial flaws from ultrasonic signals, we develop a novel flaw classification system. In this system we employ wavelet packet decomposition (WPD) to extract the features of ultrasonic signals, employ a new optimization algorithm (chaotic genetic algorithm-CGA) to get rid of redundant and irrelevant features, and employ support vector machine (SVM) classifier to classify the flaws. We use a 5 MHz focal wideband transducer to test four kinds of artificial flaws and collect ultrasonic signals at a 40 MHz sampling rate by a high speed A/D card. Noise cancellation is implemented by adaptive filtering. For comparison purpose, we use different methods as feature selector. Then we compare their results. Through experiment, we can conclude that our system can improve the performance of the SVM classifier for flaws in seafloor pipeline significantly.