Papers by Keyword: V-SVM

Abstract: Aiming at the difficult question of qualitative analysis for welding flaws in ultrasonic testing, a novel method of flaw classification utilizing the combination of wavelet packet transform (WPT) with the layer multi-class classifier based on -SVM is presented in this paper. First, WPT is applied to extract features for ultrasonic echo-signals of welding flaws, and flaws are automatically classified by using the layer multi-class classifier based on -SVM. Then, its classification results are compared with that of several widely used multi-class SVM classification methods. To validate the method above, a series sample of butt girth welds of seamless steel tube with four types of welding flaws is made, and some experiments are performed. The results show that the proposed method is superior to other multi-class SVM classification methods in the aspects of classification accuracy, training and testing time.

Abstract: Because of steel cord conveyor belt with high load operating and complex conditions of coal mine, it is prone to cause conveyor belt horizontal rupture. It will bring tremendous hazards for coal mine production. Twelve time domain features of joints signals, broken wires signals and abrasion signals for steel cord conveyor belt were extracted with weak magnetic detection system. The algorithm of combining rough set based on information entropy with multi-classbased on binary tree was proposed to classify the three categories signals. The experiment results show that rough set reduction algorithm based on information entropy can effectively achieve feature reduction and classification speed of multi-classclassification algorithm based on binary tree can be improved by rough set feature reduction without changing classification accuracy.