Abstract: A robotic arc welding system has an integrated vision sensor to image and analyze the weld scene in real time. The most effective weld puddle geometry discriminants for use in an adaptive feedback control, to assure full penetration welds, are weld puddle area and maximum width. The adaptive control system determines a puddle geometry error and a correction to nominal welding current to change the heat input to the weld pool, regulating a combination of puddle area and width, or only puddle area under some welding conditions, to control weld quality during the welding process. Arc voltage is modulated to reflect changes in welding current and maintain constant arc length.

Abstract: A computationally efficient technique to determine weld puddle area and maximum width in real time from noisy measurements of the puddle trailing edge. Using the torch electrode as the origin, image intensities are sampled in radial directions, and potential pool edge points are extracted. The boundary data is prefiltered to remove extraneous points, and a least-squares algorithm is used to estimate the area and width of the elliptical puddle. The values are sent to a vision-based arc welding robot control system which regulates these pool parameters to assure full penetration welds. Angular orientation of the weld puddle is estimated in an analogous manner.