| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Dynamic Behavior of Materials X
|
| Presentation Title |
Limited Neural Networks for the Prediction of Shockwave Initiation of Energetic Materials |
| Author(s) |
Brenden W. Hamilton, Timothy C Germann |
| On-Site Speaker (Planned) |
Brenden W. Hamilton |
| Abstract Scope |
The shockwave compaction of energetic materials results in heterogenous localization of energy that initiates chemical reactions. Numerous mechanisms, dependent on a variety of microstructural features, lead to these localizations. While atomistic simulations are a powerful tool, their expense is too great for realistic microstructures.
Machine learning techniques offer a ubiquitous method to develop models for physical processes, and by limiting/altering the input microstructural information while assessing relative changes in error, the importance of different descriptors and mechanisms can be inferred. Here we start by using initial density fields as network inputs. By altering the spatial resolution, predictions of final temperature fields are made. The inclusion of additional descriptors is used in a ‘leave one feature out’ scheme to assess what hotspot mechanisms they are related to. Additional ML dimensionality reduction techniques are applied to help predict other local properties such as shear stress and intra-molecular strains. LA-UR-23-22355 |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Modeling and Simulation, Machine Learning, Other |