About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
Advanced Characterization Techniques for Quantifying and Modeling Deformation
|
Presentation Title |
Predicting Dislocation Distribution from XRD Measurements through Machine Learning and 3D Discrete Dislocation Dynamics Simulations |
Author(s) |
Dylan Madisetti, Christopher Stiles, Jaafar El-Awady |
On-Site Speaker (Planned) |
Dylan Madisetti |
Abstract Scope |
Utilizing our database of X-Ray diffraction patterns produced from dislocation evolution in large-scale 3D Discrete Dislocation Dynamics (DDD) simulations, we have developed a machine learning model for predicting per slip system dislocation density from the spread, intensity, and character of imaged individual single crystal FCC diffraction spots. Our model, underpinned by Physics Informed Neural Networks, enables an in-situ relation of diffraction patterns to dislocation presence. The model receives image data, spectra, diffraction information, and material character as input to enable future experimental application, and the ability to be used on historical data. This model is shown to be easily cross-trainable, as we demonstrate transfer learning from a dataset of high strain-rate plasticity in copper to 316L stainless steel cooling after additively manufactured. More-over, we utilize explainable machine learning techniques (integrated gradients) to help inform a 2D analog to the modified Williamson-Hall equation. |
Proceedings Inclusion? |
Planned: |
Keywords |
Computational Materials Science & Engineering, Characterization, Additive Manufacturing |