Fusing machine algorithm with welder intelligence for adaptive welding robots

Current industrial welding robots are mostly articulated arms with a pre-programmed set of movement. These robots lack the adaptation and intelligence skilled human welders possess. With the recent developments in measuring 3D weld pool surface in GTAW, the critical process feedback available to human welders also becomes available to welding robots. To effectively use this critical feedback to equip welding robots, welders' responses to this critical feedback have been modeled. It is arguable that human may have better robustness against process certainties but machine algorithms may be designed to be quicker in response. Switching among them appropriately may improve the performance but should base on a scientific criterion. Further, fusing them together may even better perform than simply switching. In this paper, a criterion is established to automatically rate the performance from each of the three algorithms being fused - one machine algorithm and two human response models/algorithms. A fuzzy system has been proposed and established to fuse the decisions from these algorithms per their performance ratings. Simulation confirms the superiority of the fusion method to any of these three algorithms. Automated welding experiments conducted by welding robot verifies that the fusion based welding speed controller was robust under different welding currents and input disturbances. A foundation is thus established to developing next generation intelligent welding robots by fusing machine and human intelligence. (C) 2017 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.