| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI/Data Informatics: Computational Model Development, Verification, Validation, and Uncertainty Quantification
|
| Presentation Title |
A Data-driven Active Learning Paradigm to Model Dislocation Mobility From Atomistics |
| Author(s) |
Yifeng Tian, Soumendu Bagchi, Liam Myhill, Giaocomo Po, Danny Perez, Enrique Martinez-Saez, Yen Ting Lin, Nithin Mathew |
| On-Site Speaker (Planned) |
Nithin Mathew |
| Abstract Scope |
Plastic deformation in crystalline materials is primarily accommodated by the motion of dislocations. Relationship between dislocation velocity and the driving force, known as mobility law, is an essential component in describing dislocation motion in mesoscale models such as discrete dislocation dynamics. Conventionally, mobility laws are fitted to atomic scale simulations of a select few configurations under varying conditions of temperature and stress. Such modeling efforts can quickly become cumbersome and suffer from loss of accuracy especially in materials governed by complex nonlinear effects, e.g., BCC metals and alloys. In this talk, we present a novel, data-driven paradigm for deriving dislocation mobility laws, from automated high-throughput large-scale molecular dynamics simulations, using Graph Neural Networks (GNN) with a physics-informed architecture. Accelerated by an active learning framework, we demonstrate how our approach can deliver a mobility law which is more accurate compared to existing phenomenological models in BCC metals. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, ICME, Machine Learning |