Abstract:

Aimed at the current problem that the majority of seam tracking systems are very sensitive to splash, arc and other noise in the actual welding process, which result in loss of bonding accuracy, a line laser vision sensor which could detect characteristic images of weld seam in real time is designed, and an accurate measurement model based on its geometry model is built. By modeling the observation vectors using a linear representation model and using the affine transformation model to describe the motion of the weld in tracking process, a seam tracking algorithm based on the probability continuous model is proposed. To derive the objective function of seam tracking problem by taking advantage of the spatial consistency of laser stripe and background noise in the captured images and combining the first order Markov random field theory which reflects the relationship between the pixels in the neighborhood structure. Solve this function by iterative algorithm based on the least squares and max flow/min cut, and get the exact location of the weld ultimately. Built an experiment platform and apply the proposed algorithm to the welding seam automatic tracking experiment in the actual welding environment. Experimental results show that tracking precision is up to0.109 1 mm, the average processing time for each frame45 ms of this algorithm and the minimum distance between the laser stripe and weld pool is up to 24 mm, which prove that the proposed algorithm can overcome the strong noise and achieve real-time as well as accurate seam tracking.

Key words: max flow/min cut, probability continuous, seam tracking, spatial consistency, laser vision