Automatic classification of welding defects from ultrasonic signals using an SVM-based RBF neural network approach

Abstract

To improve the accuracy of classifying welding defects in ultrasonic non-destructive testing (NDT), a support vector machine-based radial basis function neural network (SVM-RBFNN) approach is presented in this paper. Based on the equivalence of a radial basis function neural network (RBFNN) and a support vector machine (SVM) in terms of structure, a modified RBF neural network model is established. An artificial bee colony (ABC) algorithm is employed to optimise the parameters of the SVM-RBFNN model. The optimised SVM-RBFNN model is applied to classify welding defects from ultrasonic signals. The experimental results demonstrate that the proposed approach is effective and feasible for the classification of welding defects in ultrasonic testing (UT). Moreover, the results show that the proposed approach is superior to RBFNNs and SVMs in terms of the classification accuracy, computation speed and generalisation achieved.