Conference Agenda

Simulating nuclear transients with Computational Fluid Dynamics (CFD) poses significant computational challenges due to complex physics and disparate temporal scales. These lead to high computational costs, necessitating advanced techniques for efficiency. This work explores two strategies: multi-rate timestepping with overset grids and reduced-order modeling (ROM).

First, we present an overlapping domain capability within NekRS, a GPU-accelerated spectral element CFD solver. This feature enables independent solution of spatial regions, improving efficiency for large-scale transients in complex geometries. We demonstrate its scalability through the TALL-3D experiment, a benchmark for thermal-hydraulic behavior in liquid metal reactors. Multi-rate timestepping significantly accelerates simulations, addressing a key CFD bottleneck.

Second, we develop ROMs as a computationally efficient alternative for high-fidelity transient simulations. ROMs support digital twin technologies, enabling rapid simulations with high accuracy. Using the NekROM framework, we implement Proper Orthogonal Decomposition (POD)-based ROMs to tackle challenges in nuclear modeling, including (1) thermal striping, which induces fluctuating thermal stresses affecting structural integrity, and (2) molten salt reactor (MSR) modeling, an emerging reactor technology. Both cases involve long time scales, benefiting from ROM acceleration. Our results show POD-based ROMs achieve significant computational speedup while maintaining essential accuracy.

These advancements in multi-rate time-stepping and ROMs represent a major step forward in transient simulation. By improving computational feasibility, they enable more efficient and accurate simulations of complex nuclear systems, enhancing reactor design, safety analysis, and operational decision-making.

In the realm of computational fluid dynamics (CFD), the high computational cost has always been a significant challenge, particularly in fields like nuclear safety analysis where complex flow problems are common. However, scientific machine learning (SciML) has demonstrated its effectiveness in solving real-world problems, including those in CFD. In recent studies, the issue of residual divergence in long-term simulations when using a single training approach has been identified. Our goal is to develop a flexible framework to optimize the state-of-the-art hybrid ML algorithm; residual based physics informed transfer learning (RePIT) developed by J. Jeon et al. This is a promising technique which has accelerated the simulation and ensures long-term stability using neural networks. However, its performance was only demonstrated on one network architecture known as finite volume method network (FVMN), also manual intervention was involved while switching between ML and CFD computation. Our projection is that we can further reduce the computational time by automating the switching process and also adding several network benchmarks where we could be able to implement state-of-the-art neural network architectures and choose the best performing one. In this work, we verified (1) the integrity of the framework by using the FVMN and compared the results with the original case study and (2) that various ML models could be loaded in the RePIT framework. In particular, DeepONet-RePIT shows the best acceleration performance. We believe that this hybrid approach is the most practical SciML utilization for robust CFD acceleration.

Typical prismatic core High Temperature Gas Reactor (HTGR) designs feature many individual coolant channels that are long in the axial direction and circular in cross section. When predicting fuel temperatures during initial concept design iterations, it is appropriate to model the steady, single-phase flow and heat transfer within each of these coolant channels using simple correlations. The challenge is then providing appropriate thermal boundary conditions along the walls of every individual channel in the core (e.g. accounting for power shapes). For complex core layouts, conjugate 3D CFD models of the entire assembly could be used to address this at the expense of significant computation time.

In this work we demonstrate a hybrid approach that uses a fluid-as-a-solid approximation to enable 3D simulations of entire core assemblies at reduced computational cost. This eliminates the need to solve the Navier-Stokes equations in the fluid, while leveraging existing functionality available within the CFD code STAR‑CCM+ to minimise implementation time and enable rapid design studies.

The approximation involves the fluid in each channel being modelled as having a suitable anisotropic effective thermal conductivity. This study considers gas flows within geometries relevant to prismatic HTGR cores. Predicted temperatures are compared between baseline CFD simulations and simulations employing the fluid-as-a-solid approximation. The fluid-as-a-solid approximation predicts similar temperature profiles to the baseline simulations. A reduction in computation time of around two orders of magnitude was achieved. This approach is expected to be valuable to preliminary concept designers, wherein rapid turnarounds are desirable across a range of designs.

In advanced reactor designs, dual-cooled annular fuel assembly has attracted significant attention due to its unique thermohydraulic characteristics. However, its complex structure poses challenges for the traditional analysis methods. In this paper, a Physics-Informed Coarse-Mesh method is proposed and applied to the analysis of annular fuel assembly. Given that annular fuel consists of internal and external channels, coarse meshes are employed to capture the primary geometric features, thereby limiting computational costs. Widely validated empirical correlations are used to correct wall friction and heat transfer, ensuring simulation accuracy. By developing a conjugate heat transfer method and a one-platform multi-scale coupling strategy with the fine-mesh CFD method, the issues of flow and heat distribution within annular fuel assembly are resolved. Based on the design parameters of the OPR-1000 reactor, a comparison of the thermohydraulic characteristics between cylindrical and annular fuel assemblies is conducted. The results show that annular fuel assembly exhibits lower central fuel temperature and higher pressure drop. The flow rate between different internal channels remains consistent. Furthermore, comparisons with existing programs demonstrate that this method can accurately simulate the flow and heat transfer characteristics of annular fuel assemblies, providing robust support for the design and optimization of advanced nuclear reactors.

The Advanced Test Reactor (ATR) at Idaho National Laboratory (INL) is a research reactor capable of delivering large-volume, high-flux thermal neutron irradiation in a realistic environment. Its design enables comprehensive studies on the effects of intense radiation on reactor materials and fuels. To ensure these experiments are conducted under specific conditions while maintaining safety standards, rigorous programmatic and safety analyses are required. These analyses typically consider coupled physics such as thermal-hydraulics, neutronics, structural analysis, and fuel performance. In this work, the Multiphysics Object Oriented Simulation Environment (MOOSE) framework and MOOSE-based applications are used for developing coupled multiphysics simulations for the ATR core. Regarding thermal-hydraulics analyses, traditional high-fidelity computational fluid dynamics (CFD) are often computationally expensive and the available codes do not have the physics models required for the simulation of the other aspects necessary for reactor analysis. This study leverages the coarse-mesh CFD capabilities in Pronghorn for conducting coupled thermal and neutronics analyses of the ATR core using a simplified porous-medium approach. This method homogenizes solid and fluid regions, enabling streamlined geometry and accelerated simulation times. This paper aims to: i) develop a 3D model of the ATR using publicly available data; ii) create a corresponding coarse-mesh CFD model; iii) verify simulation results against benchmark calculations; and iv) evaluate the use of the porous-media methodology for simulating the ATR. The results indicate that the coarse-mesh CFD capabilities provide accurate predictions for the temperature difference and pressure drop at the core, and fuel temperature distribution of the ATR, with improved run-time.