Research On Weld Seam Detection And Tracking System Based On Laser Vision

With the improvement of industrial automation and the disadvantages of manual welding,teaching-type welding robots are gradually used in modern welding fields,such as the welding of complex components of automobiles and ships.In the actual welding environment,there are uncertain factors such as clamping errors and thermal deformation of thin-plate components,which will cause large errors between the actual trajectory of the welding seam and the teaching trajectory,and the welding trajectory cannot be corrected in real time,resulting in low welding efficiency and poor quality.Therefore,this research built an active vision robot weld seam detection and tracking system and proposed a weld seam tracking algorithm based on anchor-free classification regression siamese neural network to achieve accurate weld seam detection and tracking.Also complete the tracking path planning through an improved ant colony algorithm to improve welding quality and efficiency.(1)In order to obtain the conversion relationship of the weld seam between the two-dimensional pixel coordinates and the three-dimensional coordinates of the robot,this paper designs a three-line laser vision sensor based on optical triangulation and performs calibration.Then study the algorithms of camera calibration,laser plane calibration and the hand-eye calibration between the sensor and robot.Finally complete the precise calibration of each part with the Halcon vision software.The maximum calibration absolute error of 0.122mm meets the demand of welding seam tracking.(2)Aiming at the problem of welding seam detection and tracking under the interference of strong noise such as arc light and smoke,this paper proposes an anchor-free classification regression siamese-network tracking algorithm.First,before the welding starts,the initial weld feature points are detected by traditional image processing methods such as binarization and straight line detection,and establish the tracking target area.Then the siamese-subnetwork extracts the features and enters the anchor-free classification and regression subnetwork to perform similarity calculation and precise prediction of the target area of the weld seam.(3)For the purpose of improving the efficiency and safety of welding robot path planning in the process of welding seam tracking of complex thin plate parts,this paper uses the RRT~*algorithm to achieve collision-free local weld path planning,and then performs global path planning according to the improved ant colony algorithm.Finally the simulation experiment of seam tracking path planning is carried out through Matlab.The experimental results show that the path planned in this paper satisfies the requirements of directional constraints,no collision,and the shortest length.(4)In order to verify the feasibility of the tracking algorithm proposed in this paper,this paper designs and builds a six-axis welding robot laser vision weld seam tracking platform,and conducts visual tracking experiments on straight fillet joints welds and curved lap welds.The experimental results show that our algorithm can achieve accurate,robust and real-time tracking of weld seam feature points under intense noise such as arc light and smoke,with the average absolute tracking error of 0.36mm and 0.3mm,and the average frame rate is 90fps.Compared with traditional image processing and ECO algorithm,our algorithm performs better both in accuracy and speed,and effectively improves the welding efficiency and quality of complex thin plate components.