| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Friction Stir Welding and Processing XIII
|
| Presentation Title |
Machine Learning Based In-process Defect Detection in Refill Friction Stir Spot Welding |
| Author(s) |
Jordan Andersen, Ruth Belnap, Damon Gale, Taylor Smith, Yuri Hovanski |
| On-Site Speaker (Planned) |
Jordan Andersen |
| Abstract Scope |
This paper investigates using machine learning for in-process defect detection and classification in refill friction stir spot welding (RFSSW). RFSSW is increasingly favored over resistance spot welding due to its lower energy consumption, reduced machine maintenance, and enhanced weld consistency, particularly for aluminum joints. However, RFSSW exhibits distinct defect patterns such as inadequate filling or mixing inside of the joint. Current non-destructive testing methods are largely untested, time-consuming, and costly. Our study leverages feedback data including forces, tool RPMs, and position to predict defect presence and classify defect types. We evaluate various feature extraction and data cleaning methods alongside different machine learning models. We also compare binary classification to multi-class classification of defect types. Three datasets are utilized: one with welds on AA6061-T4, another with AA2029-T7, and a third involving welding dissimilar alloys (AA2024 to AA7075) and thicknesses. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Joining, Aluminum |