About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Fast Fatigue Lifetime Prediction for as Manufactured Single Crystal CMSX4-PLUS Specimens Versus Experimental Results |
| Author(s) |
David Ryckelynck, Henry Proudhon, Axel Aublet, Clément Remacha |
| On-Site Speaker (Planned) |
David Ryckelynck |
| Abstract Scope |
The fatigue lifetime of structural components is sensitive to manufacturing defects at various scales. In this study, we present a fast data-driven method for predicting fatigue lifetime using 3D geometrical twins of single-crystal CMSX4-PLUS specimens undergoing mechanical cyclic loading at high temperatures. The fast numerical predictions of fatigue lifetime are compared to the results of experiments. The results show that geometrical defects observed by 3D X-ray computed tomography can capture crack initiation using a reduced-order model setup in crystal plasticity. Due to the wide range of possible defective geometries, we present a methodology that combines physics-based modeling, model order reduction and machine learning. The overall computational complexity is limited to high-fidelity simulations of a few finite-element digital twins in the dictionary of reference models. The methodology aims to retrieve a convenient reduced-order model from the dictionary of models depending on the specimen's true geometry. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Machine Learning, Characterization |