Computational fluid dynamics (CFD) and affordable computing power have
advanced considerably in recent years, bringing full 3D simulation of complex
high Reynolds number flows within reach of industry. However, providing
accurate and trustworthy results in diverse flows with constraints on computational
resources is still a considerable challenge. Owing to the complexity
of commonly-encountered turbulent flows, robust turbulence models are required
which do not have to be manually tuned to specific flow conditions to
ensure their accuracy.
In this regard, a highly effective approach is unstructured mesh adaptivity
which automatically refines or coarsens the mesh locally in order to achieve
a desired accuracy with minimum computational effort. However, the use
of such adaptive meshes with turbulence models raises questions about the
origins and interactions of various errors. This thesis describes the development, verification and validation of robust turbulence models suited to high
Reynolds number single-phase turbulent flow using unstructured adaptive
meshes.
The main focus of this thesis is a new tensorial dynamic large eddy simulation
(LES) model. The novel combination of the dynamic LES method
with a tensorial definition of filter width is ideal for capturing the anisotropy
and inhomogeneity of turbulence. This model is designed for use with unstructured
mesh adaptivity, which enables the simulation of turbulent flow
with high efficiency in terms of mesh resolution. Furthermore, the model is
robust since both the resolution and the sub-filter-scale (SFS) stresses adapt
to local flow conditions so that little a priori knowledge of the flow is required. Verification tests of the filtering method and validation of the new
LES model in the 3D backward-facing step are presented.
To provide context for the research, the contribution made by CFD simulations
to the analysis of nuclear reactor safety and performance is discussed.
The practicalities of performing simulations on high performance computing
(HPC) facilities are also discussed. Background theory necessary to understand
the research is presented, including a mathematical description of
turbulent flow and the classes of CFD methods used to approximate it. A review of turbulence models, discretisation methods, boundary conditions
and adaptive meshing methods is included.
The construction and testing of a Reynolds-averaged Navier-Stokes
(RANS) k - ε turbulence model and a scale-adaptive very large eddy simulation
(VLES) model in the open-source CFD code Fluidity are also described.
The development of a law-of-the-wall boundary condition for turbulent flow
in variational (weak) form is also presented. Verification tests are performed
to establish that the k - ε model has been coded correctly. Validation of
the RANS model and the wall function using fixed and adaptive meshes is
carried out in the 2D backward-facing step.
Finally, results of simulations of a vortex diode device using various turbulence
models are presented and compared to results from the commercial
CFD code CFX and experimental results. This study was carried out during
the industrial component of the Engineering Doctorate, which was intended
to further the development and understanding of CFD at Rolls-Royce Nuclear.
The device presents a challenging test case for CFD but some useful
conclusions are reached about how to model it. The thesis concludes with a
summary of findings and proposals for further research.