Learning human welder movement in pipe GTAW: A virtualized welding approach

Robotic welding is often preferred for its outperformance over human welders who are subject to physical limitations to maintain the needed consistency. Unfortunately, industrial welding robots are basically articulated arms with a preprogrammed set of movements, lacking the intelligence skilled human welders possess. This paper aims to present a virtualized welding system that enables learning from human welder intelligence for transferring into a welding robot. In particular, a 6-DOF UR-5 industrial robot arm equipped with sensors observed the welding process and performed actual welding. A human welder operated a virtualized welding torch to adjust the welding speed based on the visual feedback from the sensors, and the motion of the virtualized torch was recorded and tracked by the robot arm. Nine such teleoperated welding experiments were conducted on pipe using gas tungsten arc welding (GTAW) under different welding currents to correlate the welding speed to the welding current. Robotic welding experiments, with the robot travel speed determined per the given welding current from the resultant correlation, verified that for top part of the pipe between 11 and 1 o'clock, adjusting the welding speed per the current used is adequate to generate acceptable welds. The obtained correlation between the welding speed and welding current could be used in human-machine cooperative control. It may also provide a constraint for automated welding process control. A foundation is thus established to utilize human intelligence and transfer it to welding robots.