A comprehensive review of welding defect recognition from X-ray images

The evaluation of radiographic indications in welds plays a critical role in the quality assurance of the manufacturing process for metal products. The traditional visual approach for the evaluation of defects is inefficient and inconsistent. Various techniques for automated defect recognition of indications in weld radiographs have been proposed in the last three decades. In recent years, notable progresses have been made with the development of deep learning-based techniques. However, to date, the literature still lacks a comprehensive review of automated defect recognition in radiographic images. Therefore, this paper reviews the automated defect recognition in X-ray weld inspection, including traditional and deep-learning-based techniques. The review of traditional techniques is outlined from the perspective of image pre-processing, feature extraction, and defect analysis and evaluation. Deep-learning-based methods are reviewed from the perspective of datasets and networks structures, discussing the techniques employed to solve the small datasets problem, segmentation and classification of defects in welds. Finally, potential advancements in automated weld inspection techniques are drawn.