Abstract: Capturing high-quality welding images and extracting deviation information are two important steps of welding deviation detection.In this paper,on the basis of imaging characteristics analysis in the process CO₂ arc welding,real-time welding images were acquired clearly and steadily by using a wide dynamic range camera.Then,according to the characteristics of welding images,an improved Canny algorithm was developed to detect groove edges,and both Hough transform and prior knowledge were used to connect these edges.Moreover,in order to fit out the edge of weld pool,a differential evolution ellipse detection algorithm on the basis of condition restric- tion was proposed,with which welding deviation can be successfully calculated.Experimental results show that the proposed algorithm possesses strong robustness and high precision,so that it meets the requirements of seam track- ing and welding automation well.

Key words: vision sensing, CO₂ arc welding, welding deviation detection, DE algorithm, improved Canny algo- rithm