VISION-AIDED ROBOTIC WELDING - AN APPROACH AND A FLEXIBLE IMPLEMENTATION

Conventional, nonadaptive, robot welding systems can only be used when the workpiece are highly repeatable and well fixtured, An approach for vision-aided adaptive robotic welding and an implementation of a flexible, sensory-controlled robotic welding system are presented in this paper. Structured lighting, in the form of a steerable cone of laser light, and machine vision are used for sensing of the weld joint location and determining the detailed three-dimensional weld joint surface geometry ahead of the welding torch. This sensory feedback is used for the off-line detection of large fixturing errors before welding starts (part finding); the real-time correction of robot paths to compensate for thermal distortion and loose part tolerances during single-and multipass welding conditions to correct for weld joint shape variations (adaptive welding). A general approach for modeling and calibration of the novel cone-of-light structured lighting sensor is presented. Robust vision-processing schemes for the detection and recognition of laser stripe features in noisy images are developed and implemented using a pipelined processing architecture. Three-dimensional (3D) offsets from the taught path are computed in real time along the actual path, and a 3D surface model is developed and used to determine user-defined joint cross-sectional features and dimensions during welding. Approaches are proposed and implemented for incorporating the visually determined offsets in robot path planning and control and for controling the welding process parameters on the basis of the monitored cross-sectional dimensions.