Model Classification of Guided Wave Signal based on the Visibility Graph and SVM

Abstract

Model signals of guided wave contains considerable defect information. Model classification is a critical process for online monitoring. Visibility graph (VG) is proposed to transform the model signal into network graph. The topology characteristics of network graph are taken as the new features, and are put into support vector machine (SVM). Three categories dataset consisting of A0 model signal, S0 model signal and the noise signal are considered in this paper. The optimal VG-SVM model is constructed when 13 degrees of cumulative degree distribution are selected. The classification accuracy of optimal VG-SVM model is 95.9%, which is higher than LDA-SVM, PCA-SVM and SVM with the same defect dataset. The experimental results demonstrate that the visibility graph could extract more model information for SVM classifier.