Defect Inspection of an Arc Welded Bead Using a Support Vector Machine and a Neural Network

Abstract

This paper presents a method for a defect inspection of an arc welded bead using a support vector machine (SVM) and a neural network (NN). In our approach, these classifiers are trained to classify arc welded beads as a non-defect class or a defect class. We use intensity, frequency and edge features extracted from arc welded bead images. After extracting these features, principal component analysis (PCA) is used to reduce the number of dimensions of the extracted features. PCA is a statistical tool, which is useful to extract dominant features from a set of multivariate data. Experiments show that both the SVM and the NN have over 91% detection rate for the non-defect and over 82% detection rate for the defect.