About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Additive Manufacturing Modeling, Simulation and Artificial Intelligence
|
| Presentation Title |
SciML for Modeling Fatigue Indicator Parameters Near Voids in AM IN718 |
| Author(s) |
Jonas Merrell, Josh Urbonas, Krzysztof Stopka, Michael Sangid, Jacob Hochhalter |
| On-Site Speaker (Planned) |
Jonas Merrell |
| Abstract Scope |
Fatigue indicator parameters (FIPs) can be used to predict the number of cycles to initiate microstructurally-small fatigue cracks. FIPs can be computed using crystal plasticity finite element models (CPFEM), which is computationally demanding. Machine learning (ML) can be used to develop more efficient FIP surrogate models that are dependent on crystallographic texture, grain interconnectivity, grain geometry, void geometry, and loading. To accelerate ML model training, we use data acquired from low-fidelity elastic simulations and high-fidelity CPFEM simulations. Crystallographic information is characterized by EBSD, whereas void geometries, locations, and grain shapes are characterized using X-ray CT. We combine symbolic regression (SR) with graph neural networks (GNNs) to develop a hybrid approach that balances accuracy and interpretability of the FIP surrogate model. Specifically, GNNs are used to model the high-order coupled features, such as the grain interconnectivity and shape, while SR models low-order features, such as texture, void geometry, and loading. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Additive Manufacturing, Machine Learning, Other |