Reflections on RANS Modelling

The article begins with a reminder that the need for RANS modelling in 2009 is not any lower than had been predicted in 1997, when Detached-Eddy Simulation was created against a wave of predictions that Large-Eddy Simulation (LES) would soon make RANS redundant. The CFD community, in large part, has embraced DES and other hybrid RANS-LES approaches for flows other than thin shear layers, while noting their high cost and the uncertainty in arriving at the right numerical resolution and estimating the accuracy of the results. However these approaches leave RANS in charge of the boundary layers. In addition, wall modelling below an LES has the nature of RANS modelling. In short, RANS is in great demand. Key points are then drawn from a special RANS issue of International Journal of CFD, with the author as guest editor. The articles confirmed the static state of modelling for practical applications since the mid-1990's, and the near-failure of efforts aimed at making RANS modelling more rigorous, using math or DNS data. They also show that in the applied community pure RANS will be tried for many flows, even those where there is little hope for success, just because of computing cost. This is especially the case for a turbulent flow region that is a small part of a larger system; say the antenna on a car. Therefore, even knowing the strong misgivings which apply to Unsteady RANS (URANS) on conceptual grounds and based on poor results with bluff-body flows, there is reason to continue exploring URANS in 2D and 3D, and there would be great value in RANS models which perform better with it. The meaning of Steady RANS solutions in flows which many observers consider "too unsteady" is discussed, using examples such as the circular cylinder and airfoils at very high angles of attack. Attention then shifts to the treatment of laminar regions in RANS CFD, with emphasis on proper and reproducible use of the models, without hidden influences of inflow values, numerics, or domain sizes. The promise of transition prediction by local equations is examined. Finally, the issue of acceptance of new RANS models is addressed in terms of how well the scientific publication system and the code industry are prepared to recognize and implant the next good idea in RANS modelling, especially if it comes from a less-established source without a long record and strong resources, and/or if it conflicts with "common wisdom."