| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Materials Informatics to Accelerate Nuclear Materials Investigation
|
| Presentation Title |
Machine Learning for Predicting Reactor Pressure Vessel Embrittlement |
| Author(s) |
Dane Morgan, Ryan Jacobs, G. Robert Odette, Takuya Yamamoto |
| On-Site Speaker (Planned) |
Dane Morgan |
| Abstract Scope |
In this talk we discuss our recent work on predicting Reactor Pressure Vessel (RPV) embrittlement using machine learning methods. We have integrated yield stress and ductile to brittle transition temperature shift data from across multiple databases (including PLOTTER-22 surveillance, PLOTTER-15 MTR, IVAR, BR2, ATR2, and RADAMO) to obtain over 4500 data points and develop a model applicable to a very wide range of alloys and radiation conditions. We have utilized a flexible ensemble neural network approach combined with calibrated uncertainties to obtain accurate predictions and error estimates. When compared to hand-tuned and physics-based approaches we demonstrate that the model is numerically as good or better on average. However, the model has orders of magnitude more fitting parameters and some potentially unphysical behavior that occurs in regions of poor training data sampling. We finish with a discussion of pros and cons of machine learning vs. more hand-tuned and physics-based approaches. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, |