Search

Turbulent flows occur over a wide range of scales; this wide range becomes a challenge to resolve all the scales in turbulence for high Reynolds number industrial flows as required by direct numerical simulation (DNS) because the process is prohibitively computationally expensive. Large eddy simulation (LES) resolves the large energy-carrying scales in a turbulent flow, while the smallest scales, which can make up to 90% of the computational effort in DNS, are modelled. Consequently, LES is cheaper computationally than DNS, and a wall-resolved LES is used to investigate the development of heat transfer properties through a 90-deg pipe bend. The heat transfer along the outer wall of the bend is in good agreement with the mass transfer experimental data. However, the traditional correlation between heat and mass transfer breaks down along the inner wall due to the difference in the evolution of the flow and thermal fields. Despite the efficiency gains of LES against DNS, the computational cost remains very high for a fully wall-resolved LES. This work develops a wall function approach that allows for the deliberate reduction of the near-wall grid resolution for LES. A separate but smaller grid, which solves Reynolds-averaged Navier-Stokes equations (RANS) overlaps the near-wall LES domain to support the LES grid, where it is expected to be weak. Two variants of this approach have been tested. The more simple of the two variants, termed the numerical wall function for LES, is similar to the wall-modelled LES devised by Balaras et al. (1996) as the RANS grid computes a wall shear stress to correct the first cell at the wall of the under-resolved LES grid. The numerical wall function makes improvements over the wall-modelled LES by coupling consistent information at the interface between the LES grid and the top boundary of the secondary RANS grid. This procedure enables the computation of the full RANS equations, without simplification as done in traditional wall function approaches, and the specification of any advanced turbulence model in the RANS domain. The numerical wall function has been tested for a plane channel flow and a pipe bend flow. The second variant, which forms the major contribution of this project, has been the development of the subdomain wall function for LES to correct the under-resolved near-wall LES coarse grid beyond the first cell at the wall as is done in traditional wall function approaches. This method has a similar setup to the numerical wall function but uses ideas of the dual-mesh hybrid LES/RANS framework proposed by Xiao and Jenny (2012) to specify a weak source term in the LES momentum equation. The source term acts to readjust the partial mean LES fields in the near-wall near-wall region towards the equivalent fields in the RANS secondary grid. Predictions of the flow through numerous plane channels configurations, and flow through periodic hills and an asymmetric plane diffuser are in excellent agreement with reference data.

Despite the increasing computational power available to conduct turbulence-resolving Large Eddy Simulation (LES), it is still very expensive to simulate high Reynolds number flows as they appear in industrial problems. This thesis addresses the search for economic means to carry out LES by proposing a novel hybrid RANS/LES model, based on parallel LES and RANS simulations on two separate computational grids and a blended subgrid-scale model. The two-simulation framework enables the LES to be run on isotropic grids with low wall-parallel and wall-normal resolution, with the latter still being rare among present hybrid RANS/LES models. In contrast, the auxiliary RANS simulation is run on wall-refined grids with high-aspect ratio cells, where a full grid overlapping avoids complex boundary conditions. The seamless blending on the subgrid-scale level permits an unhindered movement of turbulent structures into and out of the RANS/LES subdomains. The Dual-Grid model is applied to a variety of wall-bounded flows with increasing flow complexity. Extensive testing on a heated plane channel flow for a range of Reynolds numbers overall shows a good prediction quality for velocity and turbulent stresses, although the grids are too coarse for conventional LES. Then, a periodic hill flow is simulated at different Reynolds numbers. Although the boundary layer separation is sensitive to the near-wall grid resolution and turbulence modelling, the model performance is competitive with other hybrid models on similar grid topologies. The Dual-Grid model is also demonstrated on a rib-roughened duct flow with a squared channel cross-section and high blockage ratio; a geometry usually found in turbine blade cooling applications. Again, the results are consistent with experimental data. Naturally, the computational cost is found to be higher than of the underlying LES on identical grids, yet, the prediction quality is increased. With possible improvements to the transfer of information across the simulations to reduce computational cost and to the heat transfer modelling, it is concluded that the proposed model has potential to become relevant to the industry.