About this Abstract |
| Meeting |
2025 AWS Professional Program
|
| Symposium
|
2025 AWS Professional Program
|
| Presentation Title |
Advancements in Deposition Rate Characterization in Gas Metal Arc Welding Using Deep Learning |
| Author(s) |
J. Eduardo Alvarez Rocha, Lucas Taípe, Patricio F. Mendez |
| On-Site Speaker (Planned) |
J. Eduardo Alvarez Rocha |
| Abstract Scope |
Weld performance is directly influenced by current and voltage parameters. Understanding the impact of these parameters on metal transfer is essential for optimizing power source design and welding output prediction. Recent advancements in computer vision enable quantification of welding features from high-speed videography, yielding results that manual measurements cannot achieve at scale. This study employs a modified U-Net deep learning model to quantify arc region characteristics such as droplets, molten consumable (attached droplet), as well as metal vapour and shielding gas regions. A large-scale dataset of up to 15,000 frames from each high-speed video of welds using aluminum ER4043 consumable spanning globular and spray transfer modes was processed through the U-Net model. The output images from the U-Net model were synchronized with corresponding current and voltage measurements and processed to determine the effects on droplet area, detachment frequency, and molten consumable length. Also, a notable increase in voltage was measured when metal vapours appear, which occurred when the molten consumable reached a threshold length. The use of this computational tool provides valuable insights for optimizing waveform design and to minimize metal vapor formation. |
| Proceedings Inclusion? |
Undecided |