About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Automated Microstructure Reconstruction, Mesh Generation, and Deep Learning Algorithms for Modeling Materials |
| Author(s) |
Soheil Soghrati, Mehran Shirazi, Kartik Kashyap, Md Shahneoug Shuvo, Zhe Yuan |
| On-Site Speaker (Planned) |
Soheil Soghrati |
| Abstract Scope |
We present a computational and AI-enabled framework for automated modeling and simulation of failure in materials with complex microstructures. The approach integrates microstructural characterization, virtual packing, optimization, finite element modeling, and the CISAMR algorithm to automatically generate conforming meshes that explicitly capture material interfaces. The framework is applied to diverse material systems, including structural adhesives, fiber-reinforced and woven composites, and additively manufactured alloys. We also introduce the Deep Learning-Driven Domain Decomposition (DLDł) method for predicting linear elastic responses across varying geometries and boundary conditions. Using an overlapping domain decomposition strategy and a novel Non-Uniform Fourier Neural Operator trained on millions of simulation samples, DLDł generalizes across geometries without retraining or transfer learning. Results show accurate predictions for complex porous microstructures while substantially reducing computational and operational costs. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
ICME, Composites, Computational Materials Science & Engineering |