| About this Abstract |
| Meeting |
2021 TMS Annual Meeting & Exhibition
|
| Symposium
|
Phase Transformations and Microstructural Evolution
|
| Presentation Title |
Image-driven Discriminative and Generative Machine Learning Algorithms for Establishing Microstructure-processing Relationships |
| Author(s) |
Wufei Ma, Elizabeth Jane Kautz, Arun Baskaran, Aritra Chowdhury, Vineet Joshi, Bulent Yener, Daniel Lewis |
| On-Site Speaker (Planned) |
Elizabeth Jane Kautz |
| Abstract Scope |
Methods of microstructure representation for the purpose of predicting processing conditions from image data were investigated via discriminative and generative machine learning methods. A U-10Mo (wt%) metallic nuclear fuel was studied for building predictive models to link microstructure to processing conditions through image processing, object recognition and characterization. This study includes testing different microstructure representations and evaluating model performance based on the F1 score. A F1 score of 95.1% was achieved, indicating that our microstructure representation describes image data well, and the traditional approach of utilizing area fractions of different phases is insufficient for distinguishing between multiple classes using a small, imbalanced original data set of 272 images. To explore applicability of generative methods for supplementing limited data sets, generative adversarial networks were trained to produce artificial microstructure images. Challenges and best practices associated with applying machine learning to limited image data sets will be discussed. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Characterization, Machine Learning, Nuclear Materials |