Numerical simulations have been here carried out for turbulent flows in geometries relevant to electronic systems. These include plane and ribbed channels and a central processor unit (CPU). Turbulent flows are random, three-dimensional and time-dependent. Their physics covers a wide range of time and space scales. When separation and reattachment occur, together with streamline curvature, modelling of these complex flows is further complicated. It is well known that, when simulating unsteady flows, the traditional, steady, linear Reynolds-averaged Navier-Stokes (RANS) models often do not give satisfactory predictions. By contrast, unsteady, non-linear RANS models may perform better. Hence the application of these models is considered here. The non-linear models studied involve explicit algebraic stress and cubic models. The Reynolds Stress Model (RSM) has been also evaluated. Modelling strategies more advanced than RANS, i.e. Large Eddy Simulation (LES) and zonal LES (ZLES), have also been tested. Validation results from URANS, LES and ZLES indicate that the level of agreement of predictions with benchmark data is generally consistent with that gained by the work of others. For the CPU case, flow field and heat transfer predictions from URANS, LES ; and ZLES are compared with measurements. Overall, for the flow field, ZLES and LES are more accurate than URANS. Zonal low Reynolds number URANS models (using a hear wall k-l model) perform better than high Reynolds number models. However, for heat transfer prediction, none of the low Reynolds number models investigated performed well.
展开▼
