ML23290A100
| ML23290A100 | |
| Person / Time | |
|---|---|
| Issue date: | 10/20/2023 |
| From: | Matthew Homiack, Raj Iyengar, Matrachisia J, Pillar R, Savara A, Starr M, Verzi S, Villarreal T Office of Nuclear Regulatory Research, Oak Ridge, Sandia |
| To: | Office of Nuclear Regulatory Research |
| References | |
| Download: ML23290A100 (12) | |
Text
John Matrachisia1 Tristan Villarreal1 Engineering Applications of AI/ML for Mechanical Systems and Component Performance Aditya Savara1 Raj Iyengar1 Matthew Homiack1 Stephen Verzi2 Rishi Pillai2 Michael Starr2 1 The views expressed in this paper are those of the authors and do not reflect the views of the U.S. Nuclear Regulatory Commission. This material is declared a work of the U.S.
Government and is not subject to copyright protection in the United States. Approved for public release; distribution is unlimited.
2 This report was prepared as an account of work sponsored by an agency of the U.S. Government. Neither the U.S. Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for any third partys use, or the results of such use, of any information, apparatus, product, or process disclosed in this report, or represents that its use by such thirdparty would not infringe privately owned rights. The views expressed in this paper are not necessarily those of the U.S. Nuclear Regulatory Commission.
U.S. Nuclear Regulatory Commission Sandia National Laboratories Oak Ridge National Laboratories
System Performance Monitoring The NRC is exploring AI/ML methods to support safety research efforts for mechanical systems and component performance.
1 Probabilistic Fracture Mechanics 2
Materials Compatibility 3
System Performance Monitoring Using severe accident simulator data, an ML model was trained to detect degraded system performance.
Recirculation Pump A
A long short-term memory neural network within the structure of an autoencoder can train on time-series data.
=
Output Layer
Input Layer
Hidden Layer Encoder Decoder Latent Space Simulator Data
=
The trained ML model detects the degraded system performance via change in the mean squared error beyond the set tolerance.
Degraded System Performance Detected
Using probabilistic fracture mechanics code input and output data, ML models were used to conduct sensitivity analyses.
Inputs Outputs Relate Using Machine Learning
The random forest regression model can rank the most influential inputs with respect to both uni-and multi-variate output sets.
Most Important Input Random Feature
The random forest regression model could also be used as a surrogate model to support time-series prediction.
Linear Regression Multilayer Perceptron Random Forest Regression Gradient Boosting Regression
A surrogate ML model was trained on physics-based thermodynamic calculation data for iron alloys being considered for molten salt reactors.
Corrosion of C276 iron alloy in molten salt
The trained piecewise Gaussian process (p-GP) model was 95%
percent accurate in predicting the element thermodynamic activities.
0.55 0.6 0.65 0.7 0.75 13.95 14.45 14.95 15.45 15.95 16.45 16.95 Cr Activity Concentration of Cr (Atomic % )
Predicted Actual
The p-GP model was much faster than the physics-based model and more accurate than ordinary least squares regression.
p-GP Model CALPHAD Physics-Based Model Ordinary Least Squares Regression Predicted Cr Activity Actual Cr Activity Predicted Cr Activity Actual Cr Activity
These studies show how industry may use AI/ML for nuclear safety applications and how the NRC staff need to be ready for similar uses.