A numerical study into the effects of turbulent flows around full-scale buildings
This thesis describes the development and application of a numerical predictive procedure for turbulent flows around full-scale buildings. Two different turbulence models were considered: a complete Reynolds stress model with two alternative proposals for the pressure-strain correlations and a k-E model used in conjunction with both linear and nonlinear stress-strain relationships. The governing differential equations were discretized using finite-volume techniques and a co-located variable-storage arrangement. A multigrid method was introduced and was found to reduce the computational time by nearly a factor of ten. Both the Reynolds-stress transport model and the non-linear k-€ model were implemented in a form suitable for use with body-fitted coordinates on a co-located grid. When using the Reynolds-stress models, a number of techniques were utilized to stabilize the solution process and attain rapid convergence. The atmospheric boundary layer at inlet to the computation domain, traditionally specified from empirical correlations, was simulated here using a full Reynolds-stress model in conjunction with a marching integration procedure. Due account was taken of the terrain roughness which matched that for the full-scale tests. The outcome of those simulations consisted of profiles of mean velocity and turbulence quantities that were self sustaining and in close accord with the few full-scale measurements available. The turbulence models' performance was assessed first through detailed comparisons with various benchmark flows including the backward-facing step in both straight and divergent channel, the two-dimensional rib, the circular cylinder and the three-dimensional cube. Detailed model verification was then carried out by comparisons with full-scale measurements on various structures including a single-span low-rise building, a semicylindrical greenhouse and a multi-span glasshouse. It was found that the Reynolds-stress models consistently produced more accurate simulations than the k-E models. Moreover, it was demonstrated that a recently proposed model for the pressure-strain correlations yields very satisfactory results without the use of wall-reflection terms. Parametric studies were performed to determine the sensitivity of the average pressure loading to various design parameters such as the height and width of the building, the eaves geometry (sharp or curved) and the height and location of a solid windbreak placed upstream of the building. Finally, a method is proposed for representing the unsteady nature of the flow around full-scale buildings. The unsteady pressure loading is recovered and so is the peak loading which far exceeds the steady-state value. The method utilizes classical turbulence modelling techniques and is shown to yield results that are in qualitative accord with the field measurements.