| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Friction Stir Welding and Processing XIII
|
| Presentation Title |
Visual Temperature Estimation and Flash Detection of Aluminum 7075 Welds Using Neural Networks |
| Author(s) |
Daniel Langan, Michael Hall, Sasha Schrandt, Jason Grafft, Ryan Schuette, Ryan Jason Tedjasukmana |
| On-Site Speaker (Planned) |
Daniel Langan |
| Abstract Scope |
Friction Stir Welding is an effective solid-state joining method but the barrier to entry remains high due to the technical skills required and the cost to reach production. Use of machine learning techniques has the potential to reduce the barrier to entry through enhanced operator feedback and closed loop control. Temperature control has been shown to be an effective method to maintain a desired grain structure and properties of metal alloys. The purpose of this effort was to create a non-contact temperature estimation method and surface quality classification by evaluating weld surface images. Using process images of the weld surface, thermocouple data, and expert labels, two ResNet-Style Convolutional Neural networks were trained to predict temperature of an AL-7075 weld using images alone as well as evaluate surface finish and flash. Both techniques show promise with both their reliability of results and viability in use in future real-time closed loop control. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Aluminum, Machine Learning, Modeling and Simulation |