Dynamic analysis and identification of gas tungsten arc welding process for weld penetration control

In this study, gas tungsten arc welding is analyzed and modeled as a 2-input (welding current and arc length) 2-output (weld depression and width) multivariable process. Experiments under a number of typical welding conditions are performed to excite and identify the process characteristics and variations. It is observed that the model parameters vary in a large range with the experimental conditions. A real-time model frame with only a few parameters to be identified on-line is proposed. Based on the obtained models, the process characteristics in terms of inertia, delay, nonminimum phase, and coupling are given. These characteristics suggest an adaptive predictive decoupling control algorithm. By designing and implementing the suggested control algorithm with the real-time model, excellent results have been achieved for both simulation and practical control. This shows that the dynamic analysis and identification provide sufficient process information for design of the control system.