Search

• The integration of laser positioning welding, trajectory fitting welding and real-time tracking welding was realized. • The identification algorithm with YOLOv5, ICUD and AFEA as the core could accurately identify the seam center. • The robot could achieve self-matching and self-switching of welding process depending on the needs. • The integrated welding scheme is essential for the robotic intelligent and flexible welding. The three welding modes of laser positioning, trajectory fitting and real-time tracking are key links for flexible robotic welding, because they directly affect the adaptability of welding robots for complex structural parts. Nevertheless, there are few studies on the integration of laser positioning, trajectory fitting and real-time tracking welding. Aiming at the drawback of that the robot cannot efficiently match the corresponding welding process between different working conditions such as short seam, spatial curve seam and long seam welding, a control scheme that can integrate the above three welding modes is proposed in this paper. First, the welding requirements of laser positioning, trajectory fitting and real-time tracking are transformed into a unified 3D coordinate recognition task for integrated welding, so the identification of welding points under three different processes can be obtained through an identical inference. The next, given the fast inference speed of YOLOv5, a joint ROI extraction algorithm with YOLOv5 as the core is applied. To compensate for YOLOv5′s inability to directly locate seam centers, iterative centerline unbiased detector (ICUD) and adaptive feature extraction algorithm (AFEA) are proposed without the need for complex image pre-processing and with very strong immunity to strong exposure and strong arc spatter. Finally, teaching trajectory correction model, B-spline curve fitting model, and 3D coordinate real-time tracking model are presented to enhance the adaptability of the welding robot and to flexibly match the appropriate welding mode in various working conditions. Experimental results indicate that the welding trajectory is basically consistent with the seam centerline when laser positioning and trajectory fitting welding. In real-time tracking welding, the real-time deviation is mainly controlled within the range of ± 0.2 mm, and the average error is 0.1160 mm. [ABSTRACT FROM AUTHOR]