Multi-class Weld Defect Detection and Classification by Support Vector Machine and Artificial Neural Network

TLDR: In this paper, a group of four descriptors matching to texture measurements extracted segmented entity and specified input to classifiers, classifier trained to classify entity from one of the defects classes and support vector machine and artificial neural network classifiers confirmed accuracy performance of 92 and 87% by confusion matrix.

Abstract: Online weld examination by non-destructive testing significantly demanded specially for aerospace, petrochemical, shipbuilding and nuclear power industries. Mostly, X-ray testing accepted by accuracy and consistency in weld bead examinations and approving part quality. In radiography, the texture feature extraction by grey level co-occurrence matrix plays key role for surface texture examination. This works projected technique for detection and cataloguing of imperfections in weld joint. This technique identify detects and differentiates weld images that look like to improper signs or deficiencies such as crack, slag, incomplete fusion, incomplete penetration, porosity, gas cavity and undercut. A group of four descriptors matching to texture measurements extracted segmented entity and specified input to classifiers. Then, classifier trained to classify entity from one of the defects classes. At last, support vector machine and artificial neural network classifiers confirmed accuracy performance of 92 and 87% by confusion matrix.