Nuclear thermal hydraulics analysis using devito’s code generation framework
File(s)
Author(s)
Nwegbu, Kene
Type
Thesis
Abstract
Simulating the fluid flow and heat transfer within nuclear power plants is an important part of nuclear safety and performance analysis. This could be to understand the performance of the reactor during normal operation or to study the reactors phenomena during severe accident scenarios. As nuclear power continues to be part of the global energy supply, new simulation codes are being developed with more accurate modelling capabilities. This has lead to an expansion in computational fluid dynamics (CFD) modelling for nuclear engineering. Despite this, thermal hydraulic codes are still the standard tool for performing simulations of a nuclear reactor core because they have been extensively validated.
A growing alternative to thermal hydraulic codes is the porous media approach within CFD. The reactor core is approximated as a porous media and uses thermal hydraulic correlations to capture the physics of the fluid flow. This flexibility allows porous media-computational fluid dynamics codes to be dual purposed as a thermal hydraulics code and a CFD code.
The aim of this thesis is to develop a porous-media computation fluid dynamics code that uses automatic code-generation to runs efficiently on GPU and CPU architectures. This has been developed with the following capabilities: a 3D numerical solver for the incompressible Navier-Stokes equations, subchannel models for subchannel scale thermal hydraulics analysis and an immersed boundary method for representing complex geometries. The efficient scaling on high performance computing is one of the key objectives of this project. This is why the solver has been written on Devito, a Python-based framework that supports parallel computing on CPU/GPU architectures. The partial differential equations can be written using a high-level syntax. From this syntax, the Devito compiler generates C code that can be optimized for different hardware architectures.
A growing alternative to thermal hydraulic codes is the porous media approach within CFD. The reactor core is approximated as a porous media and uses thermal hydraulic correlations to capture the physics of the fluid flow. This flexibility allows porous media-computational fluid dynamics codes to be dual purposed as a thermal hydraulics code and a CFD code.
The aim of this thesis is to develop a porous-media computation fluid dynamics code that uses automatic code-generation to runs efficiently on GPU and CPU architectures. This has been developed with the following capabilities: a 3D numerical solver for the incompressible Navier-Stokes equations, subchannel models for subchannel scale thermal hydraulics analysis and an immersed boundary method for representing complex geometries. The efficient scaling on high performance computing is one of the key objectives of this project. This is why the solver has been written on Devito, a Python-based framework that supports parallel computing on CPU/GPU architectures. The partial differential equations can be written using a high-level syntax. From this syntax, the Devito compiler generates C code that can be optimized for different hardware architectures.
Version
Open Access
Date Issued
2024-09-30
Date Awarded
01/12/2025
License URL
Advisor
Pain, Christopher
Gorman, Gerard
Smith, Paul
Sponsor
Engineering and Physical Sciences Research Council
Grant Number
2296238
Publisher Department
Department of Earth Science & Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)