APPLICATION OF COMPUTER VISION TECHNOLOGIES FOR AUTOMATED DETECTION OF WELD DEFECTS IN SHIPBUILDING

Abstract

The article examines modern computer vision technologies used for automated detection of weld defects, with particular emphasis on shipbuilding applications The analytical basis comprises eleven peer-reviewed publications issued in 2021–2026 and three documents of the International Association of Classification Societies and Lloyd’s Register governing weld quality and non- destructive examination Conventional image processing, convolutional neural networks, segmentation methods, the YOLO family, multi-sensor systems, and unsupervised learning are compared It is shown that the transition from image- level classification to defect localization and segmentation increases the practical value of automated inspection; however, the reliability of the result depends on the representativeness of training data, imaging quality, and the consistency of algorithmic output with regulatory acceptance criteria For shipbuilding, the most promising direction is the integration of computer vision with radiographic and visual- optical inspection, robotic inspection platforms, and a vessel digital twin