This thesis presents numerical computations of two- and three-dimensional incompressible turbulent flows around prismatic obstacles. These flows are not only relevant to engineering applications related to wind engineering and cooling of electronic components, but also exhibit most of the complex and challenging features associated with multiple flow separation and reattachment. The thesis addresses the two central issues in the computational modeling of such flows:​ numerical accuracy and turbulence modeling. The computational method is based on a finite-volume solution of the time-averaged Navier-Stokes equations. In order to ensure numerical stability without introducing unacceptable levels of numerical diffusion, a QUICK-type higher-order upwind-biased differencing scheme was formulated and implemented. This scheme is bounded, stable, and retains transverse curvature (TC) correction terms for both cross-stream directions. To achieve an improved representation of turbulence transport processes and the eddy viscosity, a multiple-time-scale (MS) model which is based on a variable partitioning method of the turbulent energy spectrum is implemented in the computer code.

Two- and three-dimensional laminar obstacle flows are used to validate the computer code and the different elements of the numerical method. The numerical predictions using the QUICK-TC scheme are found to be in excellent agreement with available experimental data and show that i) three-dimensional effects can be important in the recirculation regions, and ii) the use of higher-order differencing is essential for such flows.

A comprehensive assessment of the multiple-time-scale approach versus the classical single-scale k — e turbulence model is performed for various two- and threedimensional configurations at high Reynolds numbers. These include the backwardfacing step, the surface-mounted rib, the square cylinder, and three-dimensional prismatic obstacles. In the case of the benchmark backward-facing step flow, the QUTCKTC/MS computations result in a reattachment length within the certainty bound of the measurements, and excellent overall agreement with measured velocity profiles and shear stress distributions. For flow over surface-mounted obstacles, the present computations resolve successfully the complex features observed in previous flow visualization studies.

Detailed comparisons with available experimental data show that the MS model is superior to the k — e in predicting mean flow quantities in all regions, and in particular, in regions with high non-equilibrium turbulence. Fluctuating velocity components remain, however, underpredicted. Overall, it is demonstrated that the present computational model is capable of predicting complex three-dimensional flows involving separation and reattachment with reasonable computing requirements. Extension of the computational method to take into account large-scale unsteadiness should bring further improvements in the prediction of the turbulent stress components.