Turbulent models and prediction of turbulence
 

Hi,

My question is regarding the prediction of turbulence (intensity and transition) in computation, which is a new area to me.

I know that there are many turbulent models like Spalart-Allmaras, k-epsilon, k-omega etc. All these are available in commercial packages including fluent with many input parameters. One important user input there is turbulence intensity, which I believe is a measure of free-stream turbulence level.

If that is the case, are these models sufficient to capture the additional turbulent intensity created in a domain ? For example, if a geometry contains wall roughness or flow reattachment (which are known examples of turbulent stimulation), will the models correctly account for these induced turbulence ?

Sorry if it was silly question, but I want to make sure that, giving a certain intensity level at the beginning does not restrict the computations from capturing the induced turbulence.

Thanks in advance and waiting for a reply.

In the models you listed the turbulence intensity is going to be used to set the value of a variable (usually k) at the boundary condition where you're setting the intensity. So if the boundary in question is the inflow, then you can liken it to the freestream intensity. I am not familiar enough with commercial packages to say whether they have the option to apply clipping to k anywhere, but i imagine that if they do it will be separate from specifying intensity.

Wall roughness will not be natively dealt with in any of the models you listed; you would have to apply additional modeling assumptions to the boundary conditions on the wall in question. I am not familiar enough with commercial solvers to say what options you have there. Wall roughness is not an area I've spent much time on, but I imagine the approach would be very different depending on whether you are using wall functions or not.

Reattachment on the other hand can happen without additional modeling. The accuracy of the reattachment will probably depend a lot on the model in question and the mesh.

Thank you very much Robo, for your reply.

Kindly correct me if I am wrong: Setting up zero intensity should mean laminar free stream. If a turbulent model is able to generate a non zero intensity based on the physics of the problem, it means that transition is happening somewhere in the domain.

Does this mean that I can use an available turbulent model to predict transition if it is processed well ? I thought we could not, especially when we read about complicated CFD codes to predict transition.

Hello,
to predict transition you should not think about a good model but about a good formulation of the equations.
Of course DNS can be used as no modelling is introduced, but also LES can be used to somehow predict transition. That especially true if high order scheme and modern SGS model are used.

Very different is the framework of RANS/URANS formulations where, by definition of their statistical averaging, transition is a part of the modelling you provide..

Thank you very much Denaro

generally no, you can't just use an off the shelf model to predict transition, especially with RANS models. Typically you will require other terms or model equations be coupled.

RANS models work by adding additional equations to the system; the turbulence intensity is just a means to set boundary conditions on those additional equations. Setting zero intensity in the freestream will reduce the turbulence viscosity to zero, however as soon as any non-zero value is introduced, it is transported through those equations and will yield a non-zero turbulent viscosity. This process isn't guaranteed to be the same as the physical transition process, due to the assumptions used in formulating the different RANS models. There might be situations where you can get away with a zero freestream, but that will depend on what specifically you are trying to simulate. If simulating transition is your goal, you definitely need something more comprehensive.