Nuclear Core Loading Pattern Prediction via Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining nuclear core loading patterns in pressurized water reactors are time-consuming and resource-intensive, failing to adequately account for mechanical deformations of fuel assemblies during operation and maintenance, which can lead to safety risks and operational disruptions.
Innovation Solution
A method using an automatic learning algorithm, specifically a neural network trained on data from previous loading patterns, to predict fuel assembly bowing and evaluate potential core loading patterns based on predetermined criteria, allowing for the selection of an optimal pattern that minimizes deformations and enhances safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If finite elements mechanical models and coupled fluid-structure CFD models are used to calculate coolant flow and mechanical deformation, then the accuracy of deformation prediction is improved, but the calculation time and resource consumption increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing deformation data from finite elements and CFD models for various loading patterns in a training dataset before actual core loading decisions are made. This pre-computation allows the neural network to quickly predict deformations during operation without performing time-consuming calculations in real-time, thus resolving the contradiction between accuracy and calculation time.
Solution Approach 2:
The patent creates a simplified copy of the complex physical system by training a neural network model on data from detailed finite elements and CFD simulations. The neural network learns to replicate the deformation prediction functionality of these complex models but executes much faster, effectively copying the accurate prediction capability while eliminating the computational burden.
2Reliability
If detailed CFD and finite elements calculations are performed for each potential core loading pattern, then the reliability of safety assessment is improved, but the productivity of the determination process deteriorates
Solution Approach 1:
The patent performs preliminary calculations by pre-generating a comprehensive training dataset containing deformation results from finite elements and CFD models for multiple loading patterns before the actual determination process. This allows the neural network to be trained once and then rapidly evaluate multiple potential core loading patterns without repeating the expensive calculations, thus maintaining reliability while improving productivity.
Solution Approach 2:
The patent replaces the mechanical calculation system (finite elements and CFD models) with a computational intelligence system (neural network). The neural network substitutes the physics-based mechanical calculations with pattern recognition based on previously learned data, achieving similar reliability in safety assessment but with dramatically improved computational efficiency and productivity.
3Productivity
If engineering experience input is used in a hybrid process instead of complete calculation, then the productivity is improved, but the reliability and completeness of the calculation process deteriorates
Solution Approach 1:
The patent creates a neural network model that copies the decision-making capabilities embedded in engineering experience but formalizes them into a systematic calculation process. The neural network learns from training data that incorporates engineering judgment while producing consistent, complete, and reproducible results, thus maintaining productivity while improving reliability and calculation completeness.
Solution Approach 2:
The patent transforms the qualitative engineering experience into quantitative parameters that can be processed systematically by the neural network. By converting engineering judgment into learnable patterns from training data, the system maintains the insights of experienced engineers while applying them in a complete, systematic, and reliable calculation framework rather than as ad-hoc estimates.
Data Source
AI summary
A method of determination of a nuclear core loading pattern defining the disposition of fuel assemblies. The method includes defining at least one potential core loading pattern and calculating predictive bowing of the fuel assemblies at the end of the operation cycle for each potential core loading pattern. The calculation is carried out by an automatic learning algorithm trained on a training data set that includes a plurality of other core loading patterns. The set also includes, for each of the other core loading patterns, measurements of bowing of fuel assemblies at the end of operation cycle. The method also includes evaluating the at least one potential core loading pattern based on the predictive bowing calculations and at least one predetermined criteria. The method further includes selecting one of the potential core loading patterns based at least in part on the evaluating.