According to your experience, what is the best turbulence model to be used for turbomachinery applications with turbulent, separated flow together with shocks? Both physical and computational topics are to be considered. Thanks

(1). Good question. (2). But I don't have the answer.

(1). I have the answer. (2). I'm not going to tell you.

Well, some people are using zero equations models (Baldiwn-Lomax and Cebeci-Smith) to calculate these flows. They are supposed to be robust and reasonably easy to code. But then, some other people claim that these turbulence models cannot predict separated flow...Commercial codes use mainly two equations models (k-eps or k-w) and calculate turbomachinery flows without any problem (they say). In middle it is possible to implement 1 equation models (Spalat-Almaras)...Any clues on which one it is most suitable in turbomachinery flows??

(1). Try a low Reynolds number two-equation model, whether it is k-epsilon, k-omega, or a two-layer model. (2). These are basic requirement. Beyond that point, you will have to do some fine adjustment and modeling. (3). Below this requirement, you can come up with your own execuse and use the Baldwin-Lomax,Cebeci-Smith, or ...models. (4). Since your requirement is turbulent, separated flow with shocks, item-1 is the minimum requirement. (5). Check NASA/Ames works around mid-80's in this area. Turbomachinery flow is always 3-D (and transient?, well forget about it for now), so the issue is very complex. (even with shocks ignored)

(1). I kind of agree with you, because I had developed low Re two-equation model for separated flow more than ten years ago. It's validated for only subsonic flows without shocks. It's been sealed and in the closet for ten years. And I am not ready to release it yet. I need to do more validation on it if I can find the time.(unlikely for now)

Hi, i feel low reynold number K-W is very good for the problem u mentioned. K-epsilon is not that much relaible. People are also using Baldwin-Lomax and other algebric models. but the main problem with this models is they will not capture exact size of the bubble behind the shock. usually they will give double the size of the bubble. also separtion will atart littile bit earlier with these models. by rajani

So k-omega model is good for the physics and can capture shock location and bubbles accurately. What happened with the numerical aspects? Computational time and convergence rate are important. Is it still superior k-omega to Cebeci-Smith regarding these issues?

it is all depends on your requirement. if you want more accuracy then you should go for K-omega model. if you want to capture the general physical phenomena then you should go for badlwin-Lomax or Cebeci-Smith models. computational time is less in the case of these algebaric models. also convergence rate. if you want better results from K-omega first you run your program with algebric models and then use this data as intial values for the K-omega case. u will get better convergence rate and also can save computational time. infact if you want more accuracy you should go for LES (deconvolution) procedure. which is very relaible and proved one for boundary layer separation. but computation time is more. so you have to deside what you want. by rajani

(1). A turbulence model is just a model. It is not a real thing. So, it is hard to say that it is more physically valid. Just like a big bang theory. (2). A turbulence model is a consequence of "turbulence modeling" which is part of the CFD activities. This is very important. (3). So, in CFD, you are dealing with "turbulence modeling" and you will be changing and modifying the model and coefficients to meet your solution requirement, which is case-dependent. (4). So, in the "turbulence modeling', it is perfectly all right to use Baldwin-Lomax model, as long as you can make it fit your solution requirement. It is not going to give you vey good result, if you just apply the original model to an arbitrary case.

For me it is very important computational time and the possibility to get fast results so that it is feasible to compare a large number of them. Si it seems that I should go for the algebraic models even if there is room for improvement in the physics. Thanks

(1). The complexity of the model used should be consistent with the user's background in the first place. (2). In case of doubt, take a simpler model. Actually, a constant eddy viscosity model (same as the laminar flow, with user specified viscosity level or simulated turbulent Reynolds number) can also be used to simplify the "analysis". (3). But if you are trying to predict the turbulence flow variables, then it is a different story. So, your goal of analysis will determine the type of models used.