Recognition of Internal Weld Defects by Defect Model

Abstract

In order to improve the automation of defect inspection by careful examination by the naked eye, we consider that defects should be understood as special featured regions, and we present a new approach for detecting such defects. The approach uses "a defect model" for recognizing defects. The defect model is carefully constructed based on visual features of defects, including their sizes, shapes, and intensity or shade, and also the inspector's knowledge and experience. This approach is applied to detect internal weld defects which appear in X-ray radiographic films. On the basis of visual features and size distribution of observed defects in 107 films and the global feature of the weld zone, an efficient defect model is constructed geometrically and arithmetically. The results show success in detecting very small and fuzzy defects.