Machine Learning Approach for Automatic Weld Defect Detection Using Radiographic Images

Radiographic images of weld beads are important to inspect subsurface weld defects for safety assurance. The diversified problems associated with the human analysis of radiographic images of the weldments paved the way to the use of machine learning approach for the detection of subsurface defects. Automatic weld defect inspection involves image pre-processing, defect detection, and classification of defects. The techniques of digitization of radiographs, image pre-processing, defect segmentation and classification are summarized. Advanced methods for radiographic weld defect image classification using machine learning and challenges associated with the current technology are discussed for future improvement.