RANS modelling into a second century

The purpose of this special issue is a broad-based look at the performance of RANS models, and at the more pressing needs for improvement. As Guest Editor (GE), this author felt a need to reach out of his home industry to learn some of the good and bad experiences and the requests of a large body of users, best accessed through the considered opinions of expert providers of CFD software and support. Suppliers of RANS models were also solicited, to be sure emerging types of modelling would be covered. All generously donated their precious time, first as writers and then as reviewers, and I am personally very grateful for this. This spirit suggest that we have a meaningful turbulence-modelling community with ideas being shared quite freely; this is sensible, considering the extreme intellectual difficulty of the task, and its importance to society. As could be expected, the word ‘perfection’ was not used, and ‘excellence’ only a few times (typically, the GE requested extra evidence for that!). RS tactfully report that their field ‘is teeming with opportunities for CFD improvement’ (articles will be cited simply with the authors’ initials, and f for a figure). There was a general expectation that stagnation over the last decade or more would be reported, and it is implicit in the prevalence of the SA and SST models, from 1992, even outside the fields influenced by NASA and Boeing. It would be easy to argue that models with few equations ‘should’ not work well at all. Fresh ideas have been published during the decade, but have not spread into active codes. Relative to 1992 expectations, the payoff of deep work aimed at more rigorous modelling, and of input from Direct Numerical Simulations and Coherent-Structure thinking, must be described as disappointing. Moreover, in a sense the assessment of even the best-known models is not complete yet, and each model’s ‘primary flaw’ is not a matter of consensus (in contrast, at least in Aerospace, late separation can be viewed as the ‘primary flaw’ of the Baldwin-Lomax and the k e models). In fact, it is 2009, and 1992 models are being tested against 1979 experiments. At the simplest level of searching for flaws, we now have cases in which the models are blamed for causing too much dissipation or mixing (in adverse pressure gradients), and cases with blame for too little dissipation or mixing (after reattachment). Too much mixing causes late separation, and vice-versa, but reattachment is also on the users’ minds. Methods with LES content fare better than RANS in reattaching flows (Mf8), but are much costlier and were outside the focus of this special issue. A consequence of this weakness after reattachment is that shear-stress limiters like the one in SST are double-edged: they help one flow, and can hurt the other. Separation with SA may be a little late on smooth surfaces especially if shockinduced, and early in corners, especially on windtunnel walls (HT, and recent Drag-Prediction Workshops). Therefore, a modification which always nudges separation in the same direction would not please all users. Such a consensus on ‘primary flaws’ would be most helpful for the model suppliers to improve their products, and for new entrants to assemble the set of test cases they use to establish their new model. More than one committee has been formed to establish this minimum set of cases, but no list has become de rigueur. One reason for this is that a truly sufficient list is discouraging, and scientists are eager to publish rapidly to gain recognition, funding and helpful feedback. Entry into the field is certainly difficult, but most of the established modellers would rather see a wonderful new model displace theirs and drive progress, than preserve their oligopoly. Traditionally, the prediction of incipient separation was the principal challenge, and this is still the case in the primary application of Navier–Stokes CFD in Aerospace, particularly as it sets the position of shock waves, the drag, and the maximum lift. Heat transfer may rate higher in other fields. Two relatively new themes which seem to run through this issue and other recent work are: inaccurate reattachment, and poor treatment of massive separation by RANS, whether steady or unsteady (URANS). These are discussed below. Of course, transition prediction is desired by the more careful CFD users, and remains a great weakness of CFD (except in a few finely-tuned 2D codes). However, it is mentioned only in passing (Mf5), partly based on the opinion that it will remain technically separate from RANS turbulence modelling and therefore outside the scope of the special issue. It also happens to be kept partly secret for strategic advantage International Journal of Computational Fluid Dynamics Vol. 23, No. 4, April–May 2009, 291–293