| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Frontiers of Materials Award Symposium: Physics-Informed Machine Learning for Modeling and Design of Materials and Manufacturing Processes
|
| Presentation Title |
Physics-Aware Recurrent Convolutional Neural Networks for Modeling Hotspot Formation and Growth in Energetic Materials
|
| Author(s) |
Stephen Baek |
| On-Site Speaker (Planned) |
Stephen Baek |
| Abstract Scope |
In this presentation, I will present my team’s recent work on physics-aware recurrent convolutional neural networks (PARC). PARC is a deep neural network that embodies a neural architecture mirroring how traditional numerical simulation packages solve physics governing differential equations. We found PARC capable of predicting the complex, extreme thermomechanics of energetic materials such as explosives and propellants, with high resolution of details and accuracy comparable to those of traditional direct numerical solvers. Yet, PARC presented multiple orders of magnitude acceleration in computing speed from hours on high-performance clusters for traditional solvers to a few seconds on a laptop for PARC. This presentation will delve into the innovations and potential that PARC presents to the energetic materials community, as well as its scientific contribution in the context of other previous physics-informed machine learning models. It will also discuss the current limitations and research opportunities in the mathematical formulation and computational implementation of PARC for EM research.
|
| Proceedings Inclusion? |
Planned: None Selected |