| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Bridging Scale Gaps in Multiscale Materials Modeling in the Age of Artificial Intelligence
|
| Presentation Title |
Coarse-Graining Atomistic Simulation Data with Physics-Guided Gaussian Process Regression |
| Author(s) |
Ryan Sills, Yating Fang, Qian Qian Zhao, Ahmed Aziz Ezzat |
| On-Site Speaker (Planned) |
Ryan Sills |
| Abstract Scope |
A key challenging when scale bridging in materials science is obtaining coarse-grained constitutive laws suitable for higher scale models (e.g., finite elements) from lower scale simulations (e.g., molecular dynamics), especially when the underlying physical processes are stochastic in nature. Such coarse-graining is usually accomplished by applying some ad hoc averaging procedure using an averaging window whose width is manually tuned. This leads to constitutive laws that are sensitive to the arbitrary window size and also not easily represented via a closed-form expression. In this talk, we demonstrate how Gaussian process regression, a non-parametric machine learning technique, provides a rigorous methodology for coarse-graining with no arbitrary constants that yields closed-form expressions. Furthermore, the method allows for physics information to be incorporated into the training process via the so-called deterministic mean function. We demonstrate the technique on atomistic data of fracture in metals. |
| Proceedings Inclusion? |
Planned: |