| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI/Data Informatics: Computational Model Development, Verification, Validation, and Uncertainty Quantification
|
| Presentation Title |
Big Microstructure Datasets for Materials Informatics: Using Statistically Conditioned Generative Models to Curate Big Datasets |
| Author(s) |
Andreas E. Robertson, Adam Generale, Surya Kalidindi |
| On-Site Speaker (Planned) |
Andreas E. Robertson |
| Abstract Scope |
Materials Informatics rely on leveraging expressive and diverse microstructure datasets. However, collecting statistically diverse heterogeneous microstructures is not straightforward. This is especially true if one is interested in ‘higher order’ statistical diversity – practically: diversity in the types of microstructural features and their spatial arrangements. We propose a framework for curating statistically diverse microstructure datasets utilizing statistically conditioned generative models. Specifically, we combine our recently proposed Local-Global Decomposition generative models, novel methods for identifying salient inputs to the models (i.e., 2-point statistics and neighborhood distributions), and efficient Design of Experiments techniques. The crux of the proposed framework is a novel suite of algorithms for generating salient 2-point statistics – without needing prior examples – to condition the LGD generative models. Beyond providing a foundational framework, we curate a baseline dataset – as well as both thermal and elastic mechanical homogenized properties – as a resource for benchmarking materials informatics algorithms. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, ICME, Computational Materials Science & Engineering |