About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Working Towards a Buildable and Transferable Deep Learning Model Simulating Full-Field Micromechanical Evolution of Polycrystalline Materials |
| Author(s) |
Ashley Lenau, Reeju Pokharel, Alexander Scheinker, Stephen Niezgoda |
| On-Site Speaker (Planned) |
Ashley Lenau |
| Abstract Scope |
High energy diffraction microscopy (HEDM) is a non-destructive characterization technique that studies a material’s evolution during a mechanical load. As valuable as HEDM is, the extensive planning, data collection and time needed for a successful experiment can make it an expensive endeavor. Numerically based crystal plasticity simulations may allow for better planning but are too slow to be used in real-time with an experiment. A deep learning model would allow for in real-time feedback that could focus data collection and increase the design space for experimental planning. However, deep learning is currently limited by the small datasets available. This work proposes a U-Net model to predict the full-field micromechanical evolution of a 3D Cu polycrystal and the transferability of this network is demonstrated on three different materials. The possibility of using the Cu-trained network as a building block to incorporate additional materials into the model’s capability is explored. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Mechanical Properties, Machine Learning, Computational Materials Science & Engineering |