Effect of computational grid on performance of two-equation models of turbulence for film cooling applications

The effect of the computational grid on the performance of two-equation models of turbulence in predicting film cooling effectiveness has been investigated. The four turbulence models are: the standard, RNG, and realizable k-s, and the standard k-co models in the CFD code FLUENT. The geometry and flow-field were taken from an experimental study on discrete-hole film-cooling on a flat plate and are simulated in FLUENT. Some of the parameters are: Pitch-to-diameter ratio of 3.04, injection tube length-to-diameter ratio of 4.6. Two blowing ratios of M = 0.5 and 1.5 were considered. In earlier studies presented in Turbo Expo 2006 and 2007, a hexahedral grid for the main flow and tetrahedral grid for the short injection tubes and plenum were used as suggested by the FLUENT users' manual. The results of predictions both under low and high freestream turbulence and for high and low blowing ratios were not that good. It was decided that some of these differences arise from the use of "hybrid mesh" with a non-conformal interface boundary between them. Also tetrahedral mesh in the injection tube could not resolve the boundary layer there leading to incorrect injection profiles at the hole exit. In order to correct this deficiency, in the current study a single hexahedral grid is used both in the injection tube, plenum and the main flow regions eliminating the need for merging two different meshes and also leading to correct boundary layer resolution in the tube resulting in more realistic jet exit profiles. The difference between the model calculations and data for effectiveness are reduced to within 40 % in the near field (~x/D < 8) of the jets and 10% in the far field (~x/D >8). This shows the importance of using correct mesh which fits the problem physics and expected flow pattern for the accurate solutions and indicated the necessity of using CFD codes such as FLUENT with care and experience. There were still differences between the results for four models. It looks like k-s and k-co models give the best predictions and the trend of the data for the centerline effectiveness in the near field, whereas all models give good (within 5-10%) predictions of the centerline effectiveness in the far field.