About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Surrogate Models of 3D Crystal Plasticity with Variable Microstructure and Loading Condition Using Recurrent Neural Networks |
| Author(s) |
Michael D. Atkinson, Mike D. White, Adam J. Plowman, Pratheek Shanthraj |
| On-Site Speaker (Planned) |
Michael D. Atkinson |
| Abstract Scope |
In fusion reactors, the plasma facing components are subject to extreme temperatures, loads and radiation damage during operation. The development of next-generation materials to withstand such harsh environments will require rapid assessment of material performance and how it is affected by the underlying microstructure. Recurrent neural networks (RNNs) have been demonstrated to produce accurate surrogates of physical models of materials mechanical behaviour under varying loading condition. These surrogates allow microscale material behaviour from complex crystal plasticity (CP) models to be embedded in component scale simulations and uncertainty quantification workflows to aid material qualification. Significant variation in the response of CP models is due to the explicit representation of microstructure, and we investigate the inclusion of microstructure in RNN surrogate models. The model produced is trained on a large dataset of 3D CP simulations and the accuracy of the model investigated and then demonstrated by embedding in larger simulations. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Modeling and Simulation, Machine Learning, Mechanical Properties |