Machine vision recognition of weld pool in gas tungsten arc welding

The weld pool and its surrounding area can provide a human welder with sufficient visual information to control welding quality. Seam tracking error and pool geometry can be recognized by a skilled human welder and then utilized to adjust the welding parameters. However, for machine vision, accurate real-time recognition of weld pool geometry is a difficult task due to the high intensity arc light, even though seam tracking errors can be detected. A novel vision system is, therefore, used to acquire quality images against the arc. A real-time recognition algorithm is proposed to analyse the image and recognize the pool geometry based on the pattern recognition technique. Despite surface impurity and other influences, the pool geometry can always be recognized with sufficient accuracy in 150 ms under different welding conditions. To explore the potential application of machine vision in weld penetration control, experiments are conducted to show the correlation between pool geometry and weld penetration state. Thus, pool recognition also provides a possible technique for front-face sensing of the weld penetration.