Adaptive Intelligent Welding Manufacturing
YUMING ZHANG,QIYUE WANG,YUKANG LIU,
DOI: https://doi.org/10.29391/2021.100.006
IF: 2.2
2021-01-01
Welding Journal
Abstract:Optimal design of the welding procedure gives the desired welding results under nominal welding conditions. During manufacturing, where the actual welding manufacturing conditions often deviate from the nominal ones used in the design, applying the designed procedure will produce welding results that are different from the desired ones. Adaption is needed to make corrections and adjust some of the welding parameters from those specified in the design. This is adaptive welding. While human welders can be adaptive to make corrections and adjustments, their performance is limited by their physical constraints and skill level. To be adaptive, automated and robotic welding systems require abilities in sensing the welding process, extracting the needed information from signals from the sensors, predicting the responses of the welding process to the adjustments on welding parameters, and optimizing the adjustments. This results in the application of classical sensing, modeling of process dynamics, and control system design. In many cases, the needed information for the weld quality and process variables of our concern is not easy to extract from the sensor’s data. Studies are needed to propose the phenomena to sense and establish the scientific foundation to correlate them to the weld quality or process variables of our concern. Such studies can be labor intensive, and a more automated approach is needed. Analysis suggests that artificial intelligence and machine learning, especially deep learning, can help automate the learning such that the needed intelligence for robotic welding adaptation can be directly and automatically learned from experimental data after the physical phenomena being represented by the experimental data has been appropriately selected to make sure they are fundamentally correlated to that with which we are concerned. Some adaptation abilities may also be learned from skilled human welders. In addition, human-robot collaborative welding may incorporate adaptations from humans with the welding robots. This paper analyzes and identifies the challenges in adaptive robotic welding, reviews efforts devoted to solve these challenges, analyzes the principles and nature of the methods behind these efforts, and introduces modern approaches, including machine learning/deep learning, learning from humans, and human-robot collaboration, to solve these challenges.
metallurgy & metallurgical engineering