Skilled human welder intelligence modeling and control: Part II-analysis and control applications

Transferring a skilled human welder's experiences and skills is an essential step in developing the next-generation intelligent welding machines. In the first part of this study, a skilled human welder's responses to a 3D weld pool surface was modeled. In this second part of the paper, the proposed skilled human welder model is first compared with the novice welder model. The model is then implemented as an intelligent controller to feedback control the gas tungsten arc welding process to maintain a consistent, complete joint penetration. After the initial open-loop control period, the welding current is adjusted by the skilled welder model based on the real-time measured weld pool surface characteristic parameters as well as the welder's previous response. The resultant current waveform, front-side weld pool characteristic parameters, and back-side bead width are recorded/measured and analyzed. It is found that the skilled human welder model can adjust the current appropriately to control the welding process to a desired penetration level despite different initial currents. The controller is also robust against various welding process disturbances, including welding current, arc length, and welding speed disturbance. Compared to the novice welder, the skilled human welder model has a faster convergence time. In addition, no noticeable overshoot is observed. A foundation is thus established for exploring the mechanism and transferring of a skilled human welder's intelligence into a robotic welding system.