Human-Machine Cooperative Control of GTAW Process

In manual gas tungsten arc welding (GTAW) process, a human welder can perceive the weld penetration from his observation of the weld pool surface. He makes decision on how welding parameters be changed to achieve the desired penetration based on experience learned from previous practice/operation. The proposed control system aims at using a machine vision system to assist a human welder to better assure the weld quality. To this end, a vision system is set up to measure the weld pool surface. Specifically, a dot-matrix laser pattern, attached on welder’s torch, is projected on the specular weld pool surface. The reflection is intercepted/imaged by an imaging plane fixed on welder’s helmet and captured by a camera installed on the helmet. The laser dots in the acquired images contain geometrical information about the weld pool surface. Using 3D wireless accelerometers, the location and orientation of the imaging plane and the laser reflection dots are determined in the world coordinate system. A reconstruction scheme is developed to determine the weld pool surface based on the reflection law. Real-time experiments show that the image processing and reconstruction process can be completed within 30ms. A data analysis shows that the welding parameters (e.g., welding current, voltage, and travel speed) are nonlinearly correlated with the weld pool which is specified by the characteristic parameters proposed. On-line dynamic model is developed to predict the weld pool surface. To model welder’s reaction such that the system can predict what actions the welder may take and overwrite welder’s decision, an unskilled welder operates the process to conduct experiments. The dynamic model of his reaction (adjustment on the welding parameters) to the weld pool surface is identified using data from the experiments.