A low-cost vision-based weld-line detection and measurement technique for robotic welding

Abstract

ABSTRACT To make robotic welding more flexible and intelligent, artificial intelligence-based systems are one of the most important developments. This paper introduces a computer vision-based algorithm for weld path detection, gap measurement, and weld length calculation. The proposed innovative approach employs various image processing techniques and mathematical operations, accurately determining weld attributes at seam points. Using the YOLO-based object detection algorithm, the model attains a remarkable average precision of 99.5% in identifying atypical weld regions. The study also introduces an efficient boundary line elimination method based on the Probabilistic Hough transform and mathematical logic. Methodology for classifying weld lines with or without significant gaps has been proposed, followed by adapting distinct set of algorithms for weld line identification and gap measurement. Rigorous testing on butt joints of diverse shapes (e.g., straight, zig-zag, and curve) and sizes verifies the robustness of the algorithm, with errors well within ±1 mm for length measurements. In testing conducted at three different points along the individual weld profile, the maximum error in estimating the weld gap was observed to be 0.11 mm. Weld seam information can be extracted effectively with the proposed algorithm, which proves its viability for industrial applications.