Methods of Characterizing Gas-Metal Arc Welding Acoustics for Process Automation
Recent developments in material joining, specifically arc-welding, have increased in scope and extended into the aerospace, nuclear, and underwater industries where complex geometry and hazardous environments necessitate fully automated systems. Even traditional applications of arc welding such as off-highway and automotive manufacturing have increased their demand in quality, accuracy, and volume to stay competitive. These requirements often exceed both skill and endurance capacities of human welders. As a result, improvements in process parameter feedback and sensing are necessary to successfully achieve a closed-loop control of such processes. <br ><br /> One such feedback parameter in gas-metal arc welding (GMAW) is acoustic emissions. Although there have been relatively few studies performed in this area, it is agreed amongst professional welders that the sound from an arc is critical to their ability to control the process. Investigations that have been performed however, have been met with mixed success due to extraneous background noises or inadequate evaluation of the signal spectral content. However, if it were possible to identify the salient or characterizing aspects of the signal, these drawbacks may be overcome. <br ><br /> The goal of this thesis is to develop methods which characterize the arc-acoustic signal such that a relationship can be drawn between welding parameters and acoustic spectral characteristics. Three methods were attempted including: Taguchi experiments to reveal trends between weld process parameters and the acoustic signal; psycho-acoustic experiments that investigate expert welder reliance on arc-sounds, and implementation of an artificial neural network (ANN) for mapping arc-acoustic spectral characteristics to process parameters. <br ><br /> Together, these investigations revealed strong correlation between welding voltage and arc-acoustics. The psycho-acoustic experiments confirm the suspicion of welder reliance on arc-acoustics as well as potential spectral candidates necessary to spray-transfer control during GMA welding. ANN performance shows promise in the approach and confirmation of the ANN?s ability to learn. Further experimentation and data gathering to enrich the learning data-base will be necessary to apply artificial intelligence such as artificial neural networks to such a stochastic and non-linear relationship between arc-sound and GMA parameters.
Cite this version of the work
Joseph Tam (2005). Methods of Characterizing Gas-Metal Arc Welding Acoustics for Process Automation. UWSpace. http://hdl.handle.net/10012/859