I would like to know which are the turbulence models that can better cope with large regions of separated flow in unsteday turbomachinery flow fields. Example: simulation of the unsteady flow in compressors, from design operating conditions to near stall conditions. If you wish to reply, cancel the x in my e.mail address . Thanks

You'll have to use a first order or higher closure model, since algebraic models are completely incapable of describing flow with flows with separation or recirculation are incomplete and give no flow history. If computational resources are readily available as well as the model and the problem is 3D, I would recommend the use of either the,

First preference 2nd Order Closure Model. Second preference Algebraic Stress Closure Model.

If computational resources are available and model is 2D, these models should be sufficient to problem in hand, and should be loads of data readily available to compare against. First preference RNG first Order closure Model. Second preference standard first order closure model

Hi,

You are addressing two problems here: the ability of turbulence models to predict correctly separated flows in a strong adverse pressure gradient environment and furthermore, to give good time-accurate predictions.

I would guess that since you are carrying on time-accurate computations of turbomachinery flow fields, your CPU requirements will prevent you from using any "advanced" RSM models (6 + 1 additional equations to solve + numerical instabilities). You might use an EARSM model rather cheaper than RSM in terms of CPU time but still capable of capturing the strong anisotopy of the turbulence that exists in turbomachinery flows.

You might certainly use a two-equation linear eddy-viscosity model.

You should also keep in mind that the great majority of the turbulence models have closure coefficients which have been tuned after TIME-AVERAGED experimental data. Therefore, there is theoretically no warranty that the model will give you reliable results in an unsteady environment.


Anyway, to be more specific, you should have a look at two papers from F.R. Menter in the AIAA Journal: 1.) Vol.30, no 8, Performance of popular Turbulence models for Attached and Separated Adverse Pressure Gradient Flows

2.) Vol. 32, no 8, Two-Equation Eddy viscosity Turbulence Models for Engineering applications.

In the first one, Menter tested 4 models ranging from zero-equation to two equations models: the best was Johnson King's Model. Then, in the second paper, he modified his k-w model according to the main idea behind the J-K model: it is the so-called SST-k-w model that he validated even on a transonic bump test case and a NACA airfoil.

However, I would suspect that your main concern (especially in multistage environment with blade-row interference) is the prediction of transition and separation on the compressor blade suction side: unfortunately, as thousands of researcher would confirm, none of the most popular models are capable of predicting transition at the right location.

Good Luck

P.S: I am myself , from time to time, running unsteady computations of transonic fans : so any valuable information you may have on the subject of your question is more than welcome.

If you are using a commercial code, the answer is simple: select different turbulence model options and check out the results. If you are writing your code, the two-equation k-epsl model is the standard starting point. In this way, you will be able to get a lot of advice from other people. Since the turbulent separated flow is a very difficult field, I would suggest that you start with the steady-state flow first. Once you are satisfied with your results, you can then move on to the un-steady flow problems.

I am unsure of your precise interest. Do you wish to examine the flow over a stalled compressor stage, stalled aerofoil, or the flow over an aerofoil with a bit of a fat wake? They would benefit from different approaches. How much computing resource do you have available and what software is available?

To generalise greatly (so take it with a pinch of salt):

- if you wish to predict reasonably close to design conditions, a lot of time has gone into "tuning up" the simple turbulence models with empirical information.

- the k-e model is not well suited to this type of flow unless you do something about the turbulence structure in the impingement region, something about transition and, perhaps, something about the effects of curvature (and then it will still predict essentially solid body rotation in a recirculation).

- a Reynolds stress transport model will generally behave better in the impingement region, will treat the effects of flow curvature better but will still need a fix for transition and will not accurately predict the recirculating flow.

- the physics of the wake of the upstream stator is missing from these RANS predictions so there is little point pushing them too far.

- the future almost certainly lies with unsteady flow predictions but one needs access to a lot of computing resource.