Application of Artificial Neural Network to Flaw Classification in Ultrasonic Testing
Aiming at the difficult question of flaw qualitative analysis during industrial ultrasonic testing, a method of flaw classification based on the combination of wavelet packet transform (WPT) with artificial neural network (ANN) is proposed in this paper. Firstly, WPT is applied to feature extraction of ultrasonic flaw echo signals, and then BP neural network (BPNN), RBF neural network (RBFNN) and probabilistic neural network (PNN) are respectively used to perform flaw classification by means of the features. To validate the method above, some experiments of feature extraction and flaw classification are performed utilizing a series sample of butt girth welds of seamless steel tube with four types of welding flaws, such as crack, stomata, incomplete penetration and slag inclusion. The results show that the accuracy of flaw classification by three kinds of neural networks respectively reached to 91.25%, 92.50% and 93.75%, and the better classification effect is obtained.
Liangchi Zhang, Chunliang Zhang and Zichen Chen
Y. Chen and H. W. Ma, "Application of Artificial Neural Network to Flaw Classification in Ultrasonic Testing", Advanced Materials Research, Vols. 328-330, pp. 1876-1880, 2011