| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Friction Stir Welding and Processing XIII
|
| Presentation Title |
Machine Learning Implementations for Predicting Weld Strengths on Aluminum 7075 |
| 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 of high-strength aluminum alloys has been studied for several decades. However, weld development and qualification remains costly with the industry relying on extensive post-weld destructive evaluation. A machine control system that can predict the strength of a weld using process feedback data could potentially lower these costs. This study explores the use of data from previously welded and destructively tested aluminum 7075 butt-welds to predict joint efficiency during the welding process. Transformer and gMLP architectures were used to learn patterns in sequences of time-series data and then use them to predict strength at a particular point in the weld. A third post-weld ML technique was used to predict weld strength based on Phased-Array Ultrasonic Testing scan images. Results from the three methods are compared and discussed. Strong performance was seen in predicting yield strengths and higher variance in some of the Ultimate Tensile Strength predictions. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Aluminum, Machine Learning, Modeling and Simulation |