Abstract
To understand the potential of intelligent confirmatory tools, the U.S. Nuclear Regulatory Committee (NRC) initiated a future-focused research project to assess the regulatory viability of machine learning (ML) and artificial intelligence (AI)-driven Digital Twins (DTs) for nuclear power applications. Advanced accident tolerant fuel (ATF) is one of the priority focus areas of the U.S. Department of Energy (DOE). A DT framework can offer game-changing yet practical and informed solutions to the complex problem of qualifying advanced ATFs. Considering the regulatory standpoint of the modeling and simulation (M&S) aspect of DT, uncertainty quantification and sensitivity analysis are paramount to the DT framework's success in terms of multi-criteria and risk-informed decision-making. This chapter introduces the ML-based uncertainty quantification and sensitivity analysis methods while exhibiting actual applications to the finite element-based nuclear fuel performance code BISON.
Introduction
As the nuclear industry moves toward Digital Twins (DTs)—virtual mirrors of physical assets—regulators like the U.S. Nuclear Regulatory Commission (NRC) require rigorous methods to assess the reliability of these models. Nuclear environments are notoriously difficult to monitor in real-time, and data for new materials like Accident Tolerant Fuels (ATF) is often scarce.
The challenge is twofold: first, material properties such as density or thermal conductivity exist on a distribution rather than as fixed values, necessitating an understanding of how these input uncertainties propagate through complex simulations. Second, traditional "brute force" methods like Monte Carlo simulations require thousands of runs, which is computationally prohibitive for high-fidelity codes like BISON (a finite element-based nuclear fuel performance tool) in real-time Digital Twin applications.
Methodology and Model Setup
To address these challenges, this research employs a non-intrusive Polynomial Chaos Expansion (PCE) method. PCE serves as a fast-running mathematical surrogate or "meta-model" that decomposes the random response of the system into a series of orthogonal polynomials. This "black box" approach allows researchers to train the PCE model using a limited number of BISON outputs (100 samples in this study) without modifying the underlying source code.
The study generated synthetic data using the BISON code, varying four key parameters: fuel thermal conductivity, fuel density, cladding thermal conductivity, and cladding density. The model utilized a simplified 3D mesh of a fuel pin consisting of $UO_2$ fuel and cladding, as shown in Figure 1. Following the creation of the PCE surrogate, Sobol sensitivity analysis was performed to quantify the contribution of each input variable to the total output variance.
Key Results and Sensitivity Analysis
The sensitivity analysis yielded specific rankings for the impact of various parameters on fuel performance metrics. A primary finding was that fuel density is the dominant contributor to uncertainty for both cladding surface temperature and fuel centerline temperature.
Key insights from the analysis include:
- Dominance of Fuel Density: As illustrated by the input distribution in Figure 2, fuel density significantly influences temperature uncertainties.
- Fission Gas Drivers: For fission gas production, both fuel density and cladding thermal conductivity played major roles.
- Temporal Uncertainty: The study revealed that uncertainty is dynamic. While temperature uncertainty stabilizes quickly after reactor startup, uncertainty in fission gas production begins to climb rapidly only after approximately $10^6$ seconds of operation.
- Efficiency: The PCE method proved significantly more computationally efficient than traditional Monte Carlo methods, making it a viable engine for real-time Digital Twin applications.
Conclusion and Future Work
This framework demonstrates a robust path for integrating Uncertainty Quantification (UQ) and Sensitivity Analysis (SA) into nuclear Digital Twins. While highly efficient, the authors note that PCE faces the "curse of dimensionality" when handling more than 20 unknown parameters. Future work will focus on full integration—linking real-time sensor data from operating reactors back into this UQ/SA framework to create a truly synchronized, living model of nuclear fuel performance.
Summary updated