CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Turbulence models (http://www.cfd-online.com/Forums/main/1537-turbulence-models.html)

Reza Imani November 15, 1999 15:54

Turbulence models
 
In some turbulence models we contact to "Y" that introduce distance from rigid body surface (like asASM). If our problem contain more than one rigid body in domain how we can do ?

Jonas Larsson November 15, 1999 16:07

Re: Turbulence models
 
Good question, unfortunately there is no good general answer. This is the reason why many general purpose 3D codes avoid using models that depend on Y+. There is no correct way to do this - these models were not tuned for cases with two walls close by. What most people do is that they use the Y distance to the closest wall. Finding this in an unstructured 3D code takes some extra coding though. I'd recommend that you use a model that is dependent on Re_t or similar instead.

Sung-Eun Kim November 15, 1999 17:45

Re: Turbulence models
 
Although still questionable, distance to the nearest wall would work reasonably well especially if your wall distance dependency is such that it is significant only in the immediate neighborhood of walls, which is usually the case in many models attempting to take wall proximity effects into account.


Jonas Larsson November 16, 1999 10:24

Re: Turbulence models
 
I agree, but let me add some topics to discuss about.

A more correct solution I guess would be to use the closest y+ distance (not y), this is very inefficient though since y+ is dependent on the flow (wall shear stress) and thus can vary during the simulation and would have to be recomputed after each iteration. However, even this solution is probably less correct than using a model based on Re_t. With a Re_t based model you can at least hope to predict some influence from *both* walls. Since Re_t is dependent on transported quantities it will be influenced by all walls. If the influence will be correct is another question though.

Anyone done any simulations of "corner flows" etc. where this could be an important issue?

Fred Uckfield November 16, 1999 11:23

Re: Turbulence models
 
Corner effects are handled sweetly by Spalding's LVEL model.

Fred.

John C. Chien November 16, 1999 11:27

Re: Turbulence models
 
(1). This can be as big as the turbulence modeling itself, because the turbulence modeling deals mainly with the distribution of length scale (or scaling factor). (2). This can also be a limited issue related the sublayer behavior of a wall in a turbulent flow. In this case the Y+ effect is limited to the close to wall region. I mean, if you look from the low Reynolds number modeling point of view. The wall distance (Y+) dependence has been established long time ago. (3). In the applications, even if the mesh is not normal to the wall, sometimes Y+ is taken from the mesh directly rather than the actually normal distance to the wall. This is because the sublayer is relatively thin near the wall and the effect further away from the wall disappears quickly. This is normally associated with the paralle cut H-type grid, the distance is taken along the mesh line instead of the computed normal distance. (it takes a little more effort to compute the normal distance). (4). Around the corner of two walls, such as the blade and hub junction in a turbine passage, the minimum distance is normally used with the Baldwin-Lomax type model. This is not really a problem, because the minimum distance can be uniquely determined. (5). The use of a non-distance related parameter can also be developed, but it is difficult to know that whether such parameter will produce an unique indicator like the distance. The numerical error alone will make the equivalent distance "warped or stretched". In this case, the uncertainty is as bad as that related to the use of distance directly. This is especially true because of the high gradients of flow and turbulence variables near the wall. (6). So, the use of the physical distance or Y+, has a positive effect to restrain or control the run-away behavior in the near wall calculations. The use of the two-layer model takes advantage of this positive behavior near the wall. (7). There are other problems like a thin boundary layer interaction with a thick boundary layer case. It becomes a general turbulence modeling issue. (8). I think, away from the sublayer region, it is a good idea to use non-distance related models. Near the sublayer, or for specific types of boundary layer, distance-related model or terms can be used. (9). For the interaction of turbulence between two bodies, maybe the best answer can be derived from the racing car drivers. For safety reasons, try to keep at a safe distance from each other.

Sung-Eun Kim November 16, 1999 11:37

Re: Turbulence models
 
It would be of great value and convenience to have universal "pointwise" turbulence models which do not depend upon wall distance. Apparently, local turbulent Reynolds number, Re_t is, more attractive and, implementationwise, far less ambiguous than Re_y or y+. Yet, only if such Re_t based turbulence models consistently give better predictions.

Most of the existing "phenomenological" turbulence models are all semi-empirical nature and still ought to heavily rely upon experimental data (and DNS more recently) for near-wall flows. And the majority of near-wall data are analysed using some sort of wall-distance based parameter, whose ultimate example is the "law-of-the-wall". Imaginably, distance from wall is the most physically obvious and compelling parameter experimentalists would choose to play with. And this explain the wealth of near-wall data analysed and compiled on the basis of wall-distance based coordinates.

With this in mind and considering that wall-distance is not terribly expensive to compute (it has to be computed only once, unless the mesh deforms), I wouldn't frown at all at having to use wall-distance based turbulence models if they give reasonable results. Yes, there are simulations where use of wall distance becomes questionable such as at junctions. But Re_t based models are not free from problems and ambiguities. After all they are tuned for simple canonical equilibrium flows in the hope that they will work for complex flows. But as we all know well, in situations where transports of turbulence is significant, equilibrium based turbulence models disappoint us. A quinteseential example of this is the anomalous wall heat transfer predictions by many Re_t based turbulence models behind backward-facing step that predict erroneous heat transfer coefficient (or Nusselt number) at and near reattchment points.

What I'm getting at is that I wouldn't mind whether the models are wall distance based or local turbulent Renolds number based. What I'm rather interested in is the fidelity of those turbulence models for wide range of flows and heat transfer.


All times are GMT -4. The time now is 19:37.