Vision-based neuro-fuzzy control of weld penetration in gas tungsten arc welding of thin sheets

This paper develops a vision-based neuro-fuzzy system to control the weld joint penetration in Gas Tungsten Arc Welding (GTAW) of thin sheets. To this end, a camera equipped with a specially designed composite light-filter is used to observe the weld pool from the topside of the workpiece so that comparatively distinct images of the weld pool are obtained. As the Backside weld Width (BW) reflects the degree of the weld joint penetration, a model describing the relationship between the weld pool surface geometrical parameters (which can be extracted from the weld pool images) and the backside weld width is constructed. A neuro-fuzzy controller and a learning algorithm are developed to address dynamic and non-linear characteristics of the welding process. The controller can learn fuzzy rules and adjust the fuzzy rules with the variation of welding conditions automatically. Simulation and control tests demonstrated the effectiveness of the developed control system.