To Örebro University

In modern engineering, weld joint excellence achieved through competent and on-line welding imperfections monitoring techniques. Online weld joint examination by non-destructive testing significantly demanded in vehicle, railways, aerospace, petrochemical, oil, gas, land transportation, shipbuilding, and marine industries. These industries demanded weld imperfection inspection as major part of their testing as physical inspection may confuse for appropriate justifications and lead to incorrect identifications. Hence, for error free examination automatic weld, this work proposed an autonomous technique for multi class weld imperfections namely crack, undercut, gas pores, porosity, slag, warm holes, lack of penetration and non defects are detected and classified in X- ray images by employing support vector machine and artificial neural network and confirm their high-performance accuracy. The proposed methods incorporated mainly four modules. In the first module weld image preprocess by median filtering to achieve protective edge of weld image by suppressing noises and by rectangular shaped histogram for dispersal grey levels to wider range with a brightness gradient achieved. Furthermore, second module propose canny edge operator to achieved complete and more continues edge of weld defect as compared to other edge operators. In the third module, ten texture feature extracted by grey level co-occurrence matrix. Moreover, cluster of signifiers identical to texture features measurements extract separated entity and stated as an input classifier. Finally for speedy advances of weld imperfection detection mechanization, support vector machine and artificial neural network detected and classified weld imperfections and confirmed their accuracy performance of 98.75 and 97.5% by confusion matrix. The autonomous technique for detection and classification of X-ray images reveals ideal calculation time without disturbing complete accuracy of features selection and presented comprehensive innovative technique with improved results.