Arc length–voltage behaviour in GMAW integrating a voltage model and deep learning

Abstract This paper investigates the relationship between arc length and voltage in aluminum GMAW using a combination of deep-learning-based high-speed video analysis and a physics-based voltage model. A deep learning segmentation model was applied to approximately 15,000 high-speed video frames per video synchronized with corresponding electrical signals to extract arc length metrics. The method enabled quantitative estimation of the cathode and anode fall voltages (19.77 ± 0.61 V) and the arc column electric field (0.41 ± 0.02 V mm^-1). The arc attachment definition employed to obtain the arc length values for the voltage model was the top of the molten consumable. Additional characteristic arc lengths were analyzed during droplet formation and detachment cycles, revealing a sharp voltage drop after detachment, concurrent with an increase in arc length and reduced metal vapour presence in the arc column. A critical molten consumable length equal to the wire diameter was identified as a threshold to establish a metal vapour column. Overall, the approach demonstrates value in the use of deep learning to validate numerical and analytical models of the arc region in GMAW.