Contrasting CFD for electronic systems modeling with that for aerospace

With aerospace industries wishing to take maximum advantage of computational modeling CFD (computational fluid dynamics) has found increasing use at off-design conditions. These are characterized by significant areas of separated flow and the need for the resolution of small scale complex geometrical features. This new aerospace CFD direction has increased the common links between electronics CFD needs. This link is explored. As part of this, it is demonstrated how for typical non-separated aerospace flows with traditional turbulence models (RANS) CFD yields high solution accuracy (integrated stresses to within 1%). More separated aerospace flows, and the modern eddy resolving approaches now used to solve these are then considered. Despite the relatively simple level of geometrical complexity (in relation to electronics) and aerodynamic nature of geometries a significant reduction in solution accuracy is shown. However, even so the predictive accuracy could be better than 10 %. The same model types are then applied, by the same research group, with high accuracy well validated CFD codes (again with aerospace Rolls Royce - origins) to a non-isothermal ribbed channel flow, cube flow and complex geometry CPU system flow. With increasing geometrical complexity solution accuracy is shown to rapidly deteriorate. For the most complex geometry, turbulence models are shown to give a vast range of heat transfer levels deviating from measurements by around /spl plusmn/100 %. The important question of solution uniqueness is considered (for both aerospace and electronic geometry flows) and the matter of how different solvers can give qualitatively different flow solutions even for very simple multiple boundary isothermal flows noted.