The SST model does not account for the effects of pressure gradient and streamline.

Moreover, excessive damping of the turbulence kinetic energy does not go well with it, because it can lead to an underestimation of the wall shear and overestimation of the separation bubble size. However, SST model need to be modified or supplemented with additional terms in order to capture the flow reattachment correctly.

I agree with what Chuck A. Arize said. The answer is no, the SST model cannot predict correctly the reattachment length and you will require a model which goes beyond the RANS assumption, such as LES. I am not sure if ANSYS offers an additional correction to take into account the reattachment.

Alternatively, you could use a hybrid LES/RANS approach and compare the results, please give a look to this article:

P. R. Spalart · S. Deck · M. L. Shur · K. D. Squires M. Kh. Strelets · A. Travin A new version of detached-eddy simulation, resistant to ambiguous grid densities

I don’t use RANS formulation, I can just address some general questions:

1) use at least 3-4 nodes within y+

I just found out that ANSYS Fluent offers some options to correct such missing in the original model. You might compare the results you get with and without this correction with experimental results taking into account what Filippo Maria Denaro said

Keep us posted

Chuck A Arize and Antonio Mezzacapo, thank you for your comments.

I agree that SST, similar to any other linear eddy-viscosity turbulence model, falls short of capturing some turbulent phenomena, such as enhancement/attenuation of turbulence due to streamlines curvature. I am also aware of correction models that try to compensate for such shortcomings, as well as more sophisticated models like GeKo or EARSM that may outperform SST in my study case. However, before trying them out, I would like to gain more physical insights. Are you aware of any research on the roots of SST failure in similar problems?

Hi Armin,

well I don't think I am able to answer to your question exhaustively but I will try to give my insights hoping it helps

As you pointed out, linear eddy-viscosity models employ a linear relation between Reynolds stresses and mean strains and they are known to fail in a number of flow situations although they are still widely used in industrial calculations

For instance, weak curvature can have a strong effect on turbulence energy level such as the case of a curved channel. The reason for this failure lies in the stress-strain relation:

uv= - mu_t ( dU/dy + dV/dx)

in a weakly curved flow with a small curvature as shown in the figure, you can notice that: dv/dx

Armin Hajighasem Kashani

Again, how do you can be sure of the experimental measurements you would compare? Have you already assessed that your numerical solution is grid independent (refining both dx and dy)?

Have you obtained any good solution in standard test-caselike the backward facing step flow?

Before discussing about any physical effect, one should be sure about the quality of the numerical solution!

Any turbulence model add something that is a mix of physical and numerical contribution. Do you want to see what the model add? Switch your case to laminar and check (on the same grid) your solution.

Be aware that in RANS you cannot see physical details about small structures and so on but only mean quantity. The more sophisticated model taking into account for curvature, rotation and so on cannot produce details that, by definition, cannot be present in the solution. An eddy viscosity model can be debated for several physical reasons. It was somehow abandoned also in LES, in favor of more realistic models.

Dear Filippo Maria Denaro,

I appreciate your comments.

I have not verified my model through other benchmark tests such as the backward-facing step. I did so using the experimental data from PIV and surface pressure measurements corresponding to the problem shown in the figure. And yes, I have done grid independence test.

As I mentioned earlier, for small values of draft t (see the figure for definition), the reattachment length Lr and the pressure distribution on the body are in good agreement with the experiments. For higher drafts, however, the numerical predictions start to deviate. Particularly, the fact that the reattachment length is overpredicted makes me believe that, due to some reasons that are unknown to me, the model is probably underestimating the turbulence level within the separated shear layer, something that could otherwise have assisted the separated shear layer to curve back towards the wall faster.

I totally agree that RANS is incapable of capturing the actual turbulence physics, but is designed to only emulate some flavours of it. Much as I do not expect any RANS model to do the miracle, I'm just looking forward to some corrections/more sophisticated models in the realm of RANS in the hope of a better representation of the missing physics. However, I need to gain some insights into the potential "missing physics" in my specific problem first. This will help me in pursuit of potential solutions.

Regards,

Armin

Dear Armin Hajighasem Kashani

why are you so confident in the experimental measurements you have? How about the details of the experiment? For example, you did a 2D simulation that would be correct if the real 3D problem has the spanwise direction of homogeneous condition. But what about the effect of confinement in the experiment when draft increases? Furthermore, how do you manage to match the real mean inflow condition of the experiment to prescribe in your simulation?

It is not so correct to think that any discrepancy between experiments and simulations is due to the turbulence model. You have to assess other things before that.

A failing in the model can be assessed only if you do a deep comparison with standard test-cases.

I do not state that your model is not a problem, but from your answers I would check also the details of the experiment to be sure you are matching the higher draft case.