Research On Typical Weld Seam Feature Extraction And Real-time Seam Tracking Base On Active Vision

Robotic welding represents an inevitable trend of the development of future welding manufacturing.However,during the welding process,there may exist factors such as workpiece deformation under heat,the change of gap size and assembly errors which may cause deviation between welding seam position and robot teaching trajectory,thus affecting the final welding quality.It is very important to analyze and process the welding process images acquired by visual sensor and get the deviation information,then control the robot for the rectification movement.This will not only increase the ability to perceive external information and feedback control for the robot but also shows an important impact on the development of intelligentized manufacturing and the enhance of automation and intelligence in the process of welding.Nowadays,The visual sensor used in robot seam tracking is mainly divided into two types including passive vision and active vision.The active vision uses the laser as the auxiliary light source.By analyzing the deformation of the structural light at the weld location,the characteristic information of the weld can be obtained,and the laser stripe image can be easily collected without being influenced by the welding process.Compared with passive visual,it has simple procedure for feature extraction of images,more simple installation and wider application range.Additionally,it is more suitable for the effective promotion of the practical application of welding seam tracking in robot.This paper carried out the research of seam tracking based on active vision system.The line-structured laser vision sensor was designed and a 10% transmittance attenuator and a narrow-band filter with a central wavelength of 660 nm was used to filter ambient light and welding arc light which can ensure the obtain of clear welding images.Firstly,the weld image was smoothed by median filtering algorithm which can remove the noise and preserve the details of the laser stripe.Then morphological operation was used to eliminate the holes and small area spot in the laser stripe.Finally,the optimal threshold is obtained by Otsu adaptive method to separate the laser stripe from the surrounding background.According to four typical shape welds,the corresponding image feature extraction algorithm is adopted.The image processing speed can reach more than 10 images per second,which can meet the requirements of real-time seam tracking.In order to solve the problem that the welding arc is too strong or the welding splash which would affect the feature extraction of the welding seam and the deviation obtained,the image analysis method was used to keep the expected images and discard defect images to ensure the progress of the welding processing.In order to adjust the position of the robot smoothly and not affect the welding process,the trajected planning can be carried out by the method of piecewise interpolation.The robot's trajectory is interpolated with 3~5mm.When the sensor detects deviation information,the follow-up part of the trajectory will be recalculated and updated.Then the movement of the robot state was corrected to ensure that the torch is always in the weld center position.Finally,in order to verify the accuracy and stability of the whole seam tracking system,the corresponding tracking experiment is designed for the straight welds and curved welds with butt grooves and V-grooves.The results show that the welding seam tracking process is accurate and the error is within ±0.4mm.In the actual trajectory tracking experiments,some deviation was set between the actual welding trajectory path and the theory path to simulate the situation of deviation during the welding process.By observing the final shape of the weld,it can prove that the seam tracking system could meet the requirements for real-time weld seam tracking control.