Intelligent Seam Tracking of an Ultranarrow Gap During K-TIG Welding: A Hybrid CNN and Adaptive ROI Operation Algorithm

Abstract

Deep-penetration keyhole tungsten inert gas (K-TIG) welding uses a large current for welding. The extraction of the deviation between the position of the welding torch and the centerline of the weld is the prerequisite for the realization of seam tracking. However, K-TIG welded workpieces are often assembled without a groove and the gap between each weld is super narrow (0.2–1 mm), making it difficult to accurately identify the seam. In addition, traditional methods of identifying seams based on a fixed region of interest (ROI) are likely to lead to seam tracking failures when the type of weld changes. A narrow gap seam tracking algorithm suitable for different weld shapes is developed to address these factors and increase the robustness. To minimize the interference caused by strong arc lights, the welding images are captured by an high dynamic range (HDR) camera. A you only look once (YOLOv5)-based object detection algorithm is used to determine the position of the torch by the center of the keyhole entrance. Then, an image processing algorithm based on the adaptive ROI operation is used to adaptively extract the weld centerline and obtain the weld deviations. The experimental results proved that the welding deviation detected by the algorithm fluctuated within ±0.122 mm and that the average error was within ±0.043 mm. Moreover, the real-time performance of the algorithm can reach up to 45 frames per second (FPS), which is sufficient to meet the real-time and accuracy requirements of the K-TIG seam tracking system.