MONITORING OF WELD JOINT PENETRATION BASED ON WELD POOL GEOMETRICAL APPEARANCE

Weld joint penetration monitoring and control are fundamental issues i n automated welding. A skilled human operator can determine the weld p enetration from the geometrical appearance of the weld pool. To emulat e this using machine vision, a high-shutter-speed camera assisted with pulsed laser illumination is used to capture the clear image of the w eld pool. The pool boundary is extracted by the developed realtime ima ge processing algorithm. In order to emphasize the emulation of the hu man operator, general terms, i.e., size, shape and geometrical appeara nce, are used for the conceptual discussion, whereas more specific ter ms such as length, width, and rear angles are used in the detailed ana lysis. In particular, the size will be specified by the pool width and length, and the shape wi II be defined using the proposed rear angle of the weld pool. The geometrical appearance is described by a combina tion of the size and shape parameters. To investigate the relationship s, which could be complicated, between the weld penetration and differ ent parameters, neural networks are used because of their capability f or modeling complicated nonlinear functions. Extensive experiments hav e been developed to measure the weld penetration from the captured ima ge in 200 ms using the neural network and real-time image processing.