An Accurate Detection and Location of Weld Surface Defect Based on Laser Vision

Abstract

In order to effectively improve the efficiency of automatic detection and subsequent processing of welding defects in the construction field, this paper proposes a method for detecting and locating weld surface defects based on machine vision and laser vision. YOLOv5 is used for the initial detection and identification of weld hole defects to obtain the approximate location of the defect. Subsequently, the detailed features of the defect sites are extracted by scanning the approximate range of defect locations with a line laser 3D sensor based on the identification of weld defect holes. Finally, the defect location and depth are accurately located based on the extracted features. Experimental results show that the proposed method is capable of identifying weld surface hole defects with an accuracy rate of over 94%. Furthermore, the combination of the system with the line laser 3D sensor detection can significantly improve the accuracy compared to pure 2D visual inspection, while the manual measurement is neither convenient nor accurate. This indicates that the proposed system can be used for rapid and accurate feature information extraction of weld hole defects, making subsequent remedial welding in actual engineering more automatic and efficient.