| About this Abstract |
| Meeting |
Materials in Nuclear Energy Systems (MiNES) 2021
|
| Symposium
|
Materials in Nuclear Energy Systems (MiNES) 2021
|
| Presentation Title |
Developing Neural Network Model for Automated Analysis of Radiation-induced Grain Growth in UO2 |
| Author(s) |
Xinyuan Xu, Zefeng Yu, Arthur Motta, Xing Wang |
| On-Site Speaker (Planned) |
Xinyuan Xu |
| Abstract Scope |
In the context of a research project designed to investigate the effect of in-situ Kr ion irradiation on grain growth in UO2, a large number of microscopy images have been generated at a range of temperature and doses. To aid in the processing of this large dataset and to reduce human bias, we developed a U-Net model to automatically recognize grains in dark field transmission electron microscopy (TEM) images and measure the grain sizes. U-Net is a convolutional neural network with a unique architecture that makes it efficient in image segmentation and particle analysis. The use of data augmentation through rotation, zooming, and shearing methods improved the model accuracy from 95% to 97%. The U-Net model successfully reproduced the grain growth kinetics from human experts with a much shorter processing time. The model can be further improved to analyze in-situ TEM videos and grain growth of other nanocrystalline materials. |
| Proceedings Inclusion? |
Undecided |