Defect Classification With SVM and Wideband Excitation in Multilayer Aluminum Plates

Abstract

This paper presents a nonconventional excitation method with white noise to detect surface and subsurface cracks in aluminum plates, and the performance of the method was compared with the multifrequency excitation method. In a second stage, the best excitation method was combined with the machine learning algorithm support vector machine (SVM) to classify the location and depth of subsurface cracks in multilayer aluminum structures. The experimental measurements were performed on two stacked aluminum plates with their thickness equal to 4 and 3 mm. Several experimental tests were performed for the classification of the subsurface crack location, as well as the depth classification of the cracks. The selected features to train and test the SVM algorithm for classification are reported in this paper. The results obtained from the SVM approach include a classification obtained by training 72% of the experimental measurement data with linear, quadratic, polynomial, and Gaussian radial basis kernels and by testing the remaining 28% of the collecting measurement data.