Fillet weld recognition based on monocular structured light and RANSAC algorithm

Objective For the problems of manual, low efficiency and poor anti-interference of traditional fillet weld detection methods, an automatic recognition method based on monocular structured light and RANSAC algorithm is proposed, aiming to improve accuracy and efficiency of welding trajectory planning. Methods Fillet weld is modeled as the intersection of two planes. RANSAC is used to fit the plane, and intersection line is calculated. Heron’s formula is used to select adjacent point set of intersection line, and weld location is optimized with quadratic RANSAC fitting. Through hand-eye calibration, camera coordinate system is converted to base coordinate system of robot arm, and automatic planning of welding trajectory is realized. Results The experimental results show that average recognition error of V-shaped and T-shaped fillet welds is 0.909 mm with the proposed method, and length error between the planned trajectory and the actual trajectory is less than 1 mm. Moreover, welding task can be completed without manual teaching, which significantly improves welding efficiency and consistency. Conclusion The proposed method solves the problem of accurate location of fillet welds in complex environment by combining 3D reconstruction of monocular structured light with robustness algorithm, which has practical application value.