Toward intelligent welding robots: virtualized welding based learning of human welder behaviors

Abstract

Combining human welders (with intelligence and sensing versatility) and automated welding robots (with precision and consistency) can lead to next-generation intelligent welding robots that can perform precision welding similar to or even outperform skilled welders. In this study, an innovative human-machine welding paradigm, virtualized welding, is proposed that can transfer human intelligence to welding robots. A welding robot is augmented with sensors to observe the welding process and performs actual welding. A human welder operates a virtual welding torch and freely adjusts its movement in 3D space based on the visual feedback from the sensors. The adjustments together with the reconstructed 3D weld pool surfaces are recorded, analyzed, and utilized to form models representing human welder intelligence. To demonstrate the concepts behind the virtualized welding system and modeling method proposed, human welder’s adjustments are “selectively” learned and transferred to the welding robot to perform automated gas tungsten arc welding (GTAW). Experimental results verified the effectiveness of the learned human response models. A foundation is thus established to rapidly extract human intelligence and transfer it into welding robots.