| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Artificial Intelligence Applications in Integrated Computational Materials Engineering
|
| Presentation Title |
Machine Learning Facilitated Integration of Characterization Data and Simulations to Generate Residual Stress Distributions |
| Author(s) |
Kranthi Balusu, Shadab Anwar Shaikh, lei li, Ayoub Soulami |
| On-Site Speaker (Planned) |
Kranthi Balusu |
| Abstract Scope |
Residual stresses are “hidden” features of load-bearing components in many material processing methods, including friction stir processes. Like microstructure, they are known to play an important role in determining properties but are often not integrated into process development. Efficient determination of residual stresses is essential for rectifying this oversight.
This work first presents results from two approaches: experimental characterization and process simulations. Experimental characterization is demonstrated to be labor-intensive and costly, while simulation predictions from two different simulation models, like all other models, are shown to be inherently uncertain. These results justify a third approach, a novel machine learning (ML) based approach that combines both previous approaches.
We propose an ML-based method using U-Net, which leverages simulation results as training data and selected characterization data as prompt to generate full-field stress distributions. Comparison with full-field experimental data showcases this approach's potential to provide accurate and efficient stress predictions. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Process Technology, Characterization, Machine Learning |