A novel approach based on support vector machine to forecasting the quality of friction welding

The traditional quality evaluation methods for friction welding joints suffer from problems of complicated testing process, difficult evaluating criteria, low accurate ratio and off-line implementation. In this study, a new approach of computation intelligence using support vector machine (SVM) arithmetic to predict the quality of the welding bond is presented. The features from technique parameters are directly extracted and a radial basis function (RBF) is selected as kernel function to construct a SVM classifier. The utilization quality or the most important property in service is acted on as a mere criterion to precisely evaluate the performance of FRW bond, which decides the classification rules for SVMs. The new technique performs better than conventional evaluation methods with advantages of high efficiency, lower cost and easy implementation online. It is also proved that the SVM classifier is superior to RBF neural networks in prediction precision and generalization. The approach provides a novel technique for nondestructive properties evaluation of friction welding joints.