| About this Abstract |
| Meeting |
TMS Specialty Congress 2024
|
| Symposium
|
Symposium on Digital & Robotic Forming 2024
|
| Presentation Title |
Application of Scientific Machine Learning for Robotic Forming |
| Author(s) |
Yeping Hu, Bo Lei, Victor Castillo |
| On-Site Speaker (Planned) |
Yeping Hu |
| Abstract Scope |
We will discuss the development of fast-running reduced-order models (ROMs) based on thermomechanical simulations and graph neural networks (GNNs). We use Serac, a high-order nonlinear thermomechanical simulation code, to investigate robot/metal interaction under a variety of process conditions. These simulations are used to train GNNs, specifically MeshGraph networks, to replicate the results in a fraction of the time. Although the simulations and GNN training require large high-performance computing platforms, the ROMs can run on the edge to inform real-time path planning. We have used MeshGraphs to successfully solve complex fluid dynamics problems. Elastoplastic systems present additional complexities that will be discussed. |
| Proceedings Inclusion? |
Definite: Other |