Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Driven Prediction and Uncertainty Quantification of PWR Crud-Induced Power Shift Using Convolutional Neural Networks

Aidan Furlong, Farah Alsafadi, Scott Palmtag, Andrew Godfrey, Xu Wu

TL;DR

The paper tackles the challenge of predicting Crud-Induced Power Shift (CIPS) in PWRs by introducing a reactor-specific, data-driven surrogate based on a 3D convolutional neural network. It fuses core-model inputs (3D pin powers, boron concentration, and core exposure) with measured CIPS indices from CNS-1 cycles to predict, on an assembly basis, both the occurrence and timing of CIPS, while quantifying uncertainty via Monte Carlo dropout. The model achieves about $92.3\%$ accuracy and strong discrimination metrics, with rapid inference (~$17$ ms per prediction batch) and improved calibration when uncertainty quantification is included. Practically, the approach offers a fast first-order tool for core-design prototyping, capable of guiding loading-pattern decisions, though its generalization to other reactors requires additional data and validation.

Abstract

The development of Crud-Induced Power Shift (CIPS) is an operational challenge in Pressurized Water Reactors that is due to the development of crud on the fuel rod cladding. The available predictive tools developed previously, usually based on fundamental physics, are computationally expensive and have shown differing degrees of accuracy. This work proposes a completely top-down approach to predict CIPS instances on an assembly level with reactor-specific calibration built-in. Built using artificial neural networks, this work uses a three-dimensional convolutional approach to leverage the image-like layout of the input data. As a classifier, the convolutional neural network model predicts whether a given assembly will experience CIPS as well as the time of occurrence during a given cycle. This surrogate model is both trained and tested using a combination of calculated core model parameters and measured plant data from Unit 1 of the Catawba Nuclear Station. After the evaluation of its performance using various metrics, Monte Carlo dropout is employed for extensive uncertainty quantification of the model predictions. The results indicate that this methodology could be a viable approach in predicting CIPS with an assembly-level resolution across both clean and afflicted cycles, while using limited computational resources.

Data-Driven Prediction and Uncertainty Quantification of PWR Crud-Induced Power Shift Using Convolutional Neural Networks

TL;DR

The paper tackles the challenge of predicting Crud-Induced Power Shift (CIPS) in PWRs by introducing a reactor-specific, data-driven surrogate based on a 3D convolutional neural network. It fuses core-model inputs (3D pin powers, boron concentration, and core exposure) with measured CIPS indices from CNS-1 cycles to predict, on an assembly basis, both the occurrence and timing of CIPS, while quantifying uncertainty via Monte Carlo dropout. The model achieves about accuracy and strong discrimination metrics, with rapid inference (~ ms per prediction batch) and improved calibration when uncertainty quantification is included. Practically, the approach offers a fast first-order tool for core-design prototyping, capable of guiding loading-pattern decisions, though its generalization to other reactors requires additional data and validation.

Abstract

The development of Crud-Induced Power Shift (CIPS) is an operational challenge in Pressurized Water Reactors that is due to the development of crud on the fuel rod cladding. The available predictive tools developed previously, usually based on fundamental physics, are computationally expensive and have shown differing degrees of accuracy. This work proposes a completely top-down approach to predict CIPS instances on an assembly level with reactor-specific calibration built-in. Built using artificial neural networks, this work uses a three-dimensional convolutional approach to leverage the image-like layout of the input data. As a classifier, the convolutional neural network model predicts whether a given assembly will experience CIPS as well as the time of occurrence during a given cycle. This surrogate model is both trained and tested using a combination of calculated core model parameters and measured plant data from Unit 1 of the Catawba Nuclear Station. After the evaluation of its performance using various metrics, Monte Carlo dropout is employed for extensive uncertainty quantification of the model predictions. The results indicate that this methodology could be a viable approach in predicting CIPS with an assembly-level resolution across both clean and afflicted cycles, while using limited computational resources.
Paper Structure (22 sections, 15 equations, 11 figures, 7 tables)

This paper contains 22 sections, 15 equations, 11 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Training Data and ML Model
  3. Training Data
  4. ML Model for CIPS Prediction
  5. Methods
  6. General Approach
  7. Network Architecture
  8. Performance Metrics
  9. Uncertainty Quantification
  10. Overview of MCD for UQ of DNN
  11. Uncertainty Metrics
  12. Results
  13. Results for Base Model
  14. Base Metrics
  15. Exposure-Dependent Efficacy
  16. ...and 7 more sections

Figures (11)

  • Figure 1: CIPS values for the 49 fuel assemblies over Cycle 8.
  • Figure 2: General workflow.
  • Figure 3: Visualization of the ANN architecture. Values below each component indicate the dimensions of each layer, such as the number of neurons in fully-connected stacks (e.g., boron and exposure). In the 3D convolutional layers, the three spatial dimensions are first indicated, followed by the number of filters. The dimensions of the flatten and concatenation layers are interpreted as the length of their output vectors. The values in this figure describe how a single set of inputs are processed, and do not indicate the volume of input data.
  • Figure 4: Visualization of active dropout layers in a simplified neural network.
  • Figure 5: Training and validation loss curves for fold $1$. The other folds showed similar behavior.
  • ...and 6 more figures