Unsteady turbulence instability
 

Hi to all,

I have been simulating a case in 3D unsteady, with standard RANS model (k-ω) and a hybrid RANS/LES model.

The case is shown in both RANS_result.png and in LES_result.png. The geometry simulated is axis-symmetric, with the axis of symmetry indicated by the dashed line. Dimensions are in mm and max. velocity is ~70m/s (fluid water at 20C). Both simulations are 3D unsteady and a slice is shown on a symmetry plane. Unfortunately there are no experimental flow data yet; though we may obtain them in the following months.

The RANS results show a single recirculation zone (R1) and with a smooth velocity jet formed on the wall at the right side. On the other hand, the RANS/LES hybrid detects several vortices in two main zones (shown as R1, R2) and also show an instability downstream at Y~25mm; the jet detaches and strikes the opposite side (left side). The same happens further downstream with the jet changing side once more. The position of these instabilities changes over time - they move, fade out and re-generate.

Now, I understand that RANS generally oversmooths the solution, due to the inherent time-averaging, so such instabilities are hidden. However, has anyone experienced a similar phenomenon, i.e. a fluid jet/stream moving parallel to the wall and then suddenly detaching and hitting the other wall, especially in a tight space (note here that the space between the walls is 2.5mm, so the flow is rather constrained)?

If this effect is true and not a computational artifact, does it have a specific name (like the Koanda effect)? I was unable to find something like that in literature.

I am going to check with a finer mesh and see how it will behave and maybe with pure LES (I haven't done that already because my mesh is not that fine especially upstream.). The existing mesh consists only of hexahedra and has an inflation layer with y+~1.

Any comments/ideas are welcome
Thank you in advance.

LES and URANS simulations give you different solutions as one would theoretically expect from solving equations for different velocities.
That is what you would see even for a very simple case as the backward facing step flow.
However, you for LES you should not show a slice because the flow can be not simmetric also for symmetric geometry.
A way to see a comparison between the two simulations is to run the LES case in order to cover a proper time-period and compute the time-averaging. I believe this would produce a single separation zone

Thank you for your response Filipo; it is obvious that you have significant experience with LES since your replies are always well placed.

So, to the point:
It is true that I have shown instantaneous fields and I will perform averaging of the consecutive time steps. However do you think that after averaging I will get a similar flow field as with RANS? I mean RANS models have their problems (e.g. k-e is problematic at recirculation / detachment, k-ω is sensitive at free stream conditions, etc.). I am hoping that I will be able to get a better physics representation with LES.

The reason why I am using LES-related models is because I believe that there is a form of vortex shedding after the turn, following the flow. This shedding is not captured with RANS or URANS, since as depicted from the pictures, only a single steady vortex is found downstream. I hope that LES models will be able to capture it (provided that I am able to run the appropriate mesh for long enough).

Again thank you.

The key to understand the feature of the two solutions is that you do not have to capture secondary vortices in a RANS solution because, by definition of Reynolds average, they will be modelled. That does not mean that your RANS solution is wrong or not able to capture them but simply that the RANS formulation models globally their effects on the main flow.
If you perform a long time averaging of the LES solution, you should get a solution quite similar to the RANS one. Theoretically, there is the filter effects that still distinguishes the two solution but generally is quite disregardable.
Furthermore, for LES formulation I suggest to use the dynamic SGS procedure.

Your comment is correct; RANS models are supposed to represent the underlying behavior (time averaged) of vortices by increasing the viscosity.

Well, maybe I should have told from the beginning, as a second step I am planning to test cavitation on this device. I believe that the largest vortices shed play a fundamental role in:
- carrying bubble clumps with the flow
- creating new bubbles since pressure at the vortex core is lower - even lower than the vapor pressure.

I am afraid that no matter what I do with RANS/URANS I will not be able to get this effect; I will have bubbles following the flow, but I won't have the additional nucleation sites at the vortex cores, or the bubble clumping.

Of course it might be possible to model the effect of the low pressure at the vortex core, by linking it to some relation including the turbulent kinetic energy or the turbulent viscosity. Unfortunately, I have tried that and it didn't work well...

Thank you for your time.