Human Welder 3-D Hand Movement Learning in Virtualized GTAW: Theory and Experiments

Abstract

Combining human welder (with intelligence and sensing versatility) and automated welding systems (with precision and consistency) can lead to next generation intelligent welding systems. This paper aims to present a data-driven approach to model human welder hand movement in 3-D, and use the learned model to control automated Gas Tungsten Arc Welding (GTAW) process. To this end, an innovative virtualized welding platform is utilized to conduct teleoperated training experiments: the welding current is randomly changed to generate fluctuating weld pool surface and a human welder tries to adjust the torch movements in 3-D (including welding speed, arc length, and torch orientations) based on the observation on the real-time weld pool image feedback. These torch movements together with the 3-D weld pool characteristic parameters are recorded. The weld pool and human hand movement data are off-line rated by the welder and a welder rating system is trained, using an Adaptive Neuro-Fuzzy Inference System (ANFIS), to automate the rating. Data from the training experiments are then automatically rated such that top rated data pairs are selected to model and extract “good response” minimizing the effect from “bad operation” made during the training. ANFIS model is then utilized to correlate the 3-D weld pool characteristic parameters and welder’s torch movements. To demonstrate the effectiveness of the proposed model as an effective intelligent controller, automated control experiments are conducted. Experimental results verified that the controller is effective under different welding currents and is robust against welding speed and measurement disturbances. A foundation is thus established to learn human welder intelligence, and transfer such knowledge to realize intelligent welding robot.