No abstract

In terms of modern industrial welding, the teaching programming mode currently is still the major method applied with the welding robots, whereas as a matter of fact the operation of this method is comparatively time‐consuming, easily affected by human uncertainty and thus, requires experienced workers. In an effort to develop the autonomous and adaptive abilities of the welding robots, enhancing their flexibility as well as accomplishing more efficient industrial welding, this paper proposes an intelligent robot welding system based on laser vision and machine learning, applying a self‐designed laser scanning sensor. The crucial procedures of system calibration are illustrated step by step, involving camera calibration, laser plane calibration, hand‐eye calibration and etc. AI tracing method is implemented via computer vision and machine learning to substitute for teaching programming method to actively lead the laser scanning sensor of the welding robots to trace a welding seam for gathering the spatial coordinates of seam points. The method is tested on the complex intersection welding joint, demonstrating its reliability and superiority. Commonly, during the industrial welding process, some environmental factors, e.g. overheating, can cause some deformation of the welding seam. In order to ensure that welding torch is controlled to still follow and perform welding on the correct trajectory, smart welding seam tracking is proposed to tackle this welding issue. In addition, seam pattern mismatching detector based on Learnable Template is proposed to be used with Kernelized Correlation Filter (KCF) for accurately tracking the positive seam pattern as well as detecting the signal of mismatching to make KCF stop tracking the negative seam pattern. The corresponding experimental results have demonstrated the effectiveness, dependability and efficiency of the intelligent robot welding system.