Multiple investigations have been performed on the main influences that may affect Navier-Stokes solutions such as turbulence modelling, grid effects and wind tunnel correction parameters. It has been shown that the quality of the results obtained indeed may strongly depend on these influences. There was some significant scatter in the results presented. Concerning the mesh influences and the performance of turbulence models, some basic tendencies became visible.
The mandatory grid that has been used in most of the tests seemed to affect the solutions to a certain degree in general. The main disadvantage of this grid may be seen in the absence of adequately condensed streamwise step sizes within the shock region, preventing by this a more satisfactory resolution of the lambda structure at the shock foot. Hence, better agreement was achieved between computations and the experiments just by using different grids of higher density.
The turbulence models employed had the most significant impact on the solutions. There were drawbacks in all models and their performances sometimes turned out to be different in different flow situations. Modifications of the most common Baldwin-Lomax model, such as the Granville extension or the Goldberg backflow model, have been shown to perform to a significantly improved level of accuracy in the separated flow regime. The 'half equation' nonequilibrium models of the Johnson-King class turned out to be very promising, although the well known deficiencies that cause an unsatisfactory skin friction representation still prevent a 'break through' of this model type. The Johnson-Coakley variant, originally designed to overcome this drawback, had been tested also, but, for the test cases investigated here, this model did not succeed.
Due to the non-condensed mandatory grid that most of the computations have been performed within, the more complex turbulence models such as transport and stress models did not succeed to a superior degree over the algebraic models, which was caused by the impossibility to resolve for the shock lambda structure.
The results computed depended on the way the wind tunnel corrections were introduced. This, however, is not really a numerical effect acting on the solutions, and so there is little that can be done as long as the applications aim for free flight conditions rather than to account for the wind tunnel environment (which, however, is possible but is connected with considerable increase of computational effort).