Laser vision method for online assessment and measurement of weld gaps

Purpose One of the prerequisites for enabling welding robots to achieve autonomous welding is the ability to accurately recognize the real-time state of the welding seam. This paper aims to focus on an arc welding robot used to weld the bead of the hydraulic bracket’s box structure and explores a method for online assessment and measurement of the weld bead gap using laser vision. Design/methodology/approach First, a binocular laser vision measurement system was developed to meet the requirements for measuring weld gaps at various positions within the box structure. Second, the images of weld gaps from these structures were collected to create data sets for both a classification network and a semantic segmentation network, evaluating and comparing each model. Finally, the MobileNet V2 network was selected to assess the validity of the welding gap, and the optimized DeepLab V3+ network was used to design the welding gap measurement algorithm. Findings The results indicate that the accuracy of MobileNet V2 is 95.43% and the response time is about 1.72 s. The limit error of the designed measurement algorithm is about 0.081 mm and the response time is about 1.77 s, which meets the actual measurement needs. Originality/value The method can not only realize the accuracy and measurement of the welding gap of the hydraulic support box structure, but also provide a new idea for the online measurement of welding conditions and lay the foundation for the realization of robot autonomous welding.