Large eddy simulations of an asymmetric diffuser characterized by complex 3-D flow separation for which RANS models provide qualitatively wrong predictions have been performed.. An incompressible, turbulent, fully-developed flow in a rectangular duct (aspect ratio 1:3.33) expands into this diffuser. Two such diffusers were constructed by deflecting a pair of adjacent walls for the experiments in Cherry et al. [1, 2] (2006, 2008) and Buice, C. U. and Eaton, J. K. [3]. Most of our simulations consider Diffuser 1 with wall deflection angles 11.3° and 2.56°. In the experiments, flow begins to separate at the corner formed by the two deflected walls and then spreads so that flow is separated from the wall at the larger deflection angle. In simulations with RANS models, flow separates from the wall with the smaller deflection. It has been possible to obtain solutions with LES where flow separates correctly, off the wall at the larger deflection angle, as in the experiment. The LES finds a qualitatively correct separation, with characteristics in close quantitative agreement (within 5%) with the experimental values for Diffuser 1. The effects of variations in grid aspect ratio, grid refinement, inlet length, number of flow passes, and secondary flow structure upstream of the diffuser on solutions were determined. An LES was carried out for Diffuser 2 (deflection angles of 9° and 4° respectively), applying all lessons learnt in Diffuser 1 studies. It was found that the results for Diffuser 2 are not as quantitatively close to the experimental results as in case of the Diffuser 1, but the discrepancies appear to have a similar origin in some finer aspect of diffuser inflow conditions.

This content is only available via PDF.
You do not currently have access to this content.