dear cfd friends,

does anybody now about improved turbulence models - perhaps already implemented and tested in commercial solvers - for stronly separated but incompressible flow ?

You are talking about strongly separated flows. This means that the assumption of isotropic turbulence doesnt fit. Whenever you want to get closer to reality you have to use models wich take into acount the anisotropic effect. This means nonlinear or reynolds stress models. whenever you use a linear k,epsilon - tau - omega model you neglet the physics of anisotropie which ich very strong in seperated flow regions. For instance you don't get the secondary flow strucutres in a simple tube with a linear model. This nonlinear models are implemented in some commercial codes. For my knowledge Star-Cd has one implemented. But on the other hand for the engineering approach I am not shure if there is a great benefit in using this modells presumed you can end in converged solution.

Regards Debatin

There are two types of flow separation : (1). Flow separation from a well-defined sharp corner, such as the flow over a backward-facing-step, and (2). Flow separation over a smooth surface, such as the separated diffuser flow and the flow over airfoil at high angle of attack. In the first case, a distinct shear layer is generated from the corner and re-attached at a distance downstream to form a closed re-circulating bubble. In this case, the shear layer development, the size of the re-circulating region, the behavior of the re-attachment region and the re-development region, and the surface pressure and the skin friction distributions are important parameters for turbulence modelling. In the second case, the point of separation is unknown and the real flow normally is unstable. For the internal separated flows, most of the time the separation is non-symmetric even though the geometry is symmetric. In this case, the prediction of the boundary layer behavior through adverse pressure gradient, the separation point location and the subsequent development of the separation bubble or open-ended bubble are key parameters for turbulence modelling. Naturally, one would like to use a low Reynolds number turbulence model to predict the separation point in this case. Even in the first case, it is also a good idea to use a low Reynolds number model to eliminate the uncertainty of using wall function approach. ( whether a low Reynolds number result is more accurate than the wall function result is itself a separate issue.) The level of turbulence modeling which falls into the above requirements is at least two-equation model with a low Reynolds number treatment ( a model which can cover the whole flow field including the wall sub-layer). The heavy demand of fine mesh near the wall for the low Re model can be a tough task for some commercial codes because of the mesh and the memory requirements. Otherwise, the low Re model technology has been available ( for use and modification) since early 70's.

Look at http://sgp.me.umist.ac.uk/~mcjtsda/turb.htm