Real-Time Robot Trajectory Correction through Computer Vision: a Welding Application in the Automotive Industry

To ensure the optimal performance of welding robots in industrial applications, the elements to be welded must be accurately positioned. Additionally, elements that do not meet the quality standards can negatively influence the welding process. In this sense, failures may result in incorrect welding, leading to various structural issues within the assembly, where manual rework may be necessary to correct the errors. This work consists of developing an embedded image processing system to collect the appropriate coordinates where welds should be made on the structure of a car backrest, to adapt the trajectory of a welding robot, improving the quality and reducing the need for reworking the parts. To accomplish this application, an image acquisition system was developed. The image was processed so that it could be sent to the robot controller, which modifies the coordinates of certain movement points to correct the welding trajectory by applying an offset along the required points. To assess the effectiveness of the system, 160 weld points were tested, and 159 were found to be within the required margin of error, with only one point exhibiting a variation of 0.07 mm greater than the average in one direction. Another test performed was considering three wrong positioning of the pieces, the intention was to verify if the system could absorb these errors, thus proving the project's operation.