Assessment of Machine Learning-Based Critical Heat Flux Models in the CTF Subchannel Code for Square Rod Bundle Prediction
Machine learning models in CTF subchannel code significantly improved critical heat flux prediction for square rod bundles.
Key Findings
Methodology
This study employs pure ML and hybrid residual correction models, evaluated in local and semilocal formulations. Using the EPRI rod bundle CHF database, models are deployed in CTF subchannel code to assess performance in square rod geometries.
Key Results
- The local hybrid LUT model showed the best performance, while the semilocal pure ML model was also competitive. Compared to Bowring, W-3, and 2006 Groeneveld LUT, tube-trained models significantly improved rod bundle predictions.
- ML models outperformed traditional methods across most geometries and conditions.
- Significant improvements in rod bundle CHF prediction are possible even when models are trained exclusively on tube data.
Significance
This study provides the first large-scale assessment of ML models in square rod bundles within a production-level subchannel analysis environment, supporting their broader application in reactor thermal hydraulic analysis. By improving CHF prediction accuracy, it enhances fuel performance and reactor safety.
Technical Contribution
The study demonstrates the transferability of ML models to complex geometries, particularly from tubes to rod bundles. Hybrid residual correction models combine the strengths of traditional methods and ML, offering higher prediction accuracy and model interpretability.
Novelty
This is the first large-scale assessment of ML models applied to square rod bundles in a production-level subchannel analysis, showcasing the potential of ML methods in complex geometries.
Limitations
- Model performance was suboptimal under certain geometries and conditions, indicating a need for further optimization.
- Sensitivity to turbulent mixing assumptions may affect prediction accuracy.
Future Work
Future research could explore model optimization under different geometries and conditions, and assess the potential of other ML methods in rod bundles.
AI Executive Summary
In nuclear thermal hydraulics, accurately predicting critical heat flux (CHF) is crucial for fuel performance and reactor safety. Traditional empirical correlations and lookup tables often show inconsistent predictive accuracy across different geometries and conditions. Recently, machine learning (ML) methods have shown potential in significantly improving CHF prediction accuracy, especially in tube data applications. However, their applicability in reactor-relevant rod bundle geometries remains largely unexplored.
This study evaluates ML models deployed in the CTF subchannel code using the EPRI rod bundle CHF database. Results indicate that even when trained exclusively on tube data, the models perform well in square rod bundle applications, outperforming traditional methods. The local hybrid LUT model showed the best performance, while the semilocal pure ML model was also competitive.
These findings support the broader application of ML models in production-level subchannel analysis environments, demonstrating their potential in reactor thermal hydraulic analysis. Future research could further optimize models and explore other ML methods in different geometries and conditions.
Deep Analysis
Background
Critical heat flux (CHF) is a key parameter in nuclear thermal hydraulics, directly impacting fuel performance and reactor safety. Traditionally, CHF prediction relies on experimental measurements and empirical correlations, but these methods often show inconsistent accuracy across different geometries and conditions. Recently, machine learning (ML) methods have shown potential in improving CHF prediction accuracy.
Core Problem
Most ML models are developed and evaluated using tube data, and their applicability to reactor-relevant rod bundle geometries remains largely unexplored. Rod bundles experience additional physical phenomena, such as turbulent mixing and non-uniform power distribution, making the performance of tube-trained models in bundle environments uncertain.
Innovation
This study provides the first large-scale assessment of ML models applied to square rod bundles in a production-level subchannel analysis environment. It employs hybrid residual correction models, combining the strengths of traditional methods and ML to improve prediction accuracy and model interpretability.
Methodology
- �� Use the EPRI rod bundle CHF database for model evaluation.
- �� Employ pure ML and hybrid residual correction models, evaluated in local and semilocal formulations.
- �� Assess model performance across different geometries and operating conditions.
Experiments
Experiments use the EPRI rod bundle CHF database, covering various geometries and conditions. Models are deployed in the CTF subchannel code to assess performance in square rod bundles. Comparisons are made with traditional methods like Bowring, W-3, and 2006 Groeneveld LUT.
Results
The local hybrid LUT model showed the best performance, while the semilocal pure ML model was also competitive. Tube-trained models significantly improved rod bundle predictions, outperforming traditional methods.
Applications
The study supports the broader application of ML models in reactor thermal hydraulic analysis, particularly in CHF prediction under complex geometries and conditions.
Limitations & Outlook
Model performance was suboptimal under certain geometries and conditions, indicating a need for further optimization. Sensitivity to turbulent mixing assumptions may affect prediction accuracy.
Plain Language Accessible to non-experts
Imagine a kitchen where traditional methods are like using old recipes that make food but don't always taste great. Machine learning is like a smart chef assistant that adjusts recipes based on ingredients and cooking conditions to make tastier dishes. In a nuclear reactor, critical heat flux (CHF) is like the kitchen's temperature control; too high and the food burns. Machine learning models help predict and control this temperature more accurately, ensuring the food (fuel) is safe and tasty (performing well).
ELI14 Explained like you're 14
Imagine you're playing a game where the goal is to keep your character's health from dropping. Traditional methods are like using old guides that help you win but not always with high scores. Machine learning is like a smart AI assistant that adjusts strategies based on game progress, making it easier to score high. In a nuclear reactor, critical heat flux (CHF) is like the character's health; too high and the game fails. Machine learning models help predict and control this value more accurately, ensuring the game runs smoothly.
Glossary
Critical Heat Flux (CHF)
The point in a nuclear reactor where heat transfer rapidly deteriorates, leading to potential safety issues.
Used to evaluate the prediction accuracy of machine learning models.
Hybrid Residual Correction Model
Combines traditional models and machine learning to improve prediction accuracy by learning residuals.
Used to enhance CHF prediction accuracy.
EPRI Rod Bundle CHF Database
A large experimental dataset used to evaluate CHF model performance.
Serves as the benchmark dataset for model evaluation.
CTF Subchannel Code
A computational tool for nuclear reactor thermal hydraulic analysis.
Used to deploy and evaluate machine learning models.
Local and Semilocal Formulations
Different input feature combinations in models affecting prediction capabilities.
Used to assess model performance across various geometries and conditions.
Open Questions Unanswered questions from this research
- 1 How to optimize machine learning model performance across different geometries and conditions?
- 2 What is the specific impact of turbulent mixing assumptions on model prediction accuracy?
- 3 How to further enhance model transferability?
Applications
Immediate Applications
Nuclear Reactor Safety Analysis
Improving CHF prediction accuracy enhances reactor safety and operational efficiency.
Fuel Performance Optimization
More accurate CHF predictions aid in optimizing fuel design and improving fuel utilization.
Long-term Vision
Intelligent Reactor Control
Using machine learning models to achieve smarter reactor control systems, enhancing overall safety and efficiency.
Abstract
The prediction of critical heat flux (CHF), a key safety-related quantity in nuclear thermal hydraulics, remains an important challenge due to its direct relationship with fuel performance and reactor safety. Recent studies have demonstrated that relative to traditional empirical correlations and lookup tables (LUTs), machine learning (ML) methods can substantially improve CHF prediction accuracy. Most ML-based CHF models, however, have been developed and evaluated using tube databases, leaving their applicability to reactor-relevant rod bundle geometries largely unexplored. This study evaluates ML-based CHF models deployed within the CTF subchannel code using the Electric Power Research Institute (EPRI) rod bundle CHF database. Both pure and hybrid residual correction models are considered in local and semilocal formulations. The tube-trained ML CHF models generally transferred favorably to rod bundle applications and outperformed traditional CHF methods across most geometries and operating conditions. The local hybrid LUT model produced the strongest overall performance, and the semilocal pure ML model remained highly competitive. Comparison against the Bowring correlation, W-3 correlation, and 2006 Groeneveld LUT demonstrated that substantial improvements in rod bundle CHF prediction are possible even when models are trained exclusively on tube data. These findings provide one of the first large-scale assessments of ML-based CHF models in square rod bundles within a production-level subchannel analysis environment and support their broader application in reactor thermal hydraulic analysis.