[Lookid]

Hello,

the norvegian hydropower centre realized experiments on an hydrofoil:
- https://www.ntnu.edu/nvks/f99-test-case3
- https://www.jstage.jst.go.jp/article.../_pdf/-char/en

At V=11m/s, the hydrofoil should vibrate with a frequency of 630Hz.
The vibration of the foil is due a von karman vortex appearing behind the trailing edge.

So far, I consider only the fluid zone, and try to have the proper frequency for the vortex. So I fix V = 11 m/s, let things run and check the frequency of the vortex. But I have a much lower value, about 480Hz instead of 630Hz.

Here are the known problems:
- The turbulence is not fully developed, making URANS difficult to use.
- URANS does not even predict the vortex.

After trying some 2D cases, with different models, I tried 3D with LES (first LES case I do).
I have a proper y+ on the hydrofoil, but my x+ and z+ are way too coarse for a proper LES (both about 1000 instead of [15-100]).

- Smagorinsky gives about 480Hz. I joined a picture to show some information.
- I read about Smagorinsky model problems and tried dynSmagorinsky and locOneEqEddy, but they crash after some time. My guess is that my mesh is too coarse for LES.
- Now also running kOmegaSSTSAS, starting from the Smagorinsky solution, but it seems to damp completely the vortex.

My BC is simply V = 11m/s.
I tried turbulentInlet, but it doubles the computation time and I failed to use directMapped in parallel.

So my questions are:
- What could be the cause of the low frequency?
- Any turbulence model recommended for this type of case? I use foam-extend 4.0.
Ideally I don't have to multiply by 10 my number of cells
Any comment appreciated.


EDIT: The quality of the joined picture is terrible sorry, here it is better on google drive https://drive.google.com/file/d/1SFE...ew?usp=sharing

Lilian

[misospider]

Not sure if attached image shows full domain. If it is, your outlet length is not long enough. This may be one critical reason of URANS convergence problem.

Check if vortices are mixing at outlet boundary, if yes, you need to extend domain for longer outlet distance.

[Gerry Kan]

Dear Lilian:

You might want to check the flow and pressure on the top and bottom boundaries. Ideally there should be almost no change there. If so, these boundaries are still interfering with the hydrofoil and you need to extend them further.

Do you expect the flow separate? If so, check if the URANS separation point is reasonablly represented. Another possibly are end effects. If you are using symmetry for the two side walls, you might be neglecting the additional energyat the edge.

Gerry.

[Lookid]

Sorry for the late answer! I got the mails in junk for some reasons...

@misospider : My outlet is placed way further, no worries about this

@GerryKan, the top and bottom boundaries are fixed dimensions, they match the experimental domain.

I got some improvement in frequency by :
- refining the number of cells around the edge (not refining y+, but simply having more cells around)
- changing the scheme from Gauss linear to Gauss filteredLinear 0.2 0. This second change helps to reduce some wavy-problem in the flow, due to the gauss linear scheme. It smoothed my results (by introducing a bit of upwind in the scheme) while keeping the vortex.

I use LES-Smagorinsky+vanDriest. RANS always damps the vortex.
I now have a vortex frequency of about 550Hz, "only" 13% less than the experimental value.

[Lookid]

Some more info, from literature, if it helps:

Vortex shedding frequency behind hydrofoil is always underestimated using RANS, by [15-25]%. It is because of the flow which is laminar > transitional > turbulent as we move from the leading edge to the trailing edge of the hydrofoil.

Since RANS assumes a fully turbulent flow, the boundary layer is way bigger than it should, and the vortex frequency is smaller.

It has been proved experimentally, by placing some rough patch at the leading edge in two papers:
1) https://www.researchgate.net/publica...ine_Stay_Vanes
2) https://www.researchgate.net/publica...cal_simulation

The frequency is reduced by ~20% compared to the smooth leading edge hydrofoil, and then RANS matches the experiment.

[1] obtained good results using the transitional kOmegaSST model.

All this is good but doesn't explain why using LES I still underestimate the frequency quite a lot ^^.
I would guess the fact of having a "curved" and not a "blunt" trailing edge doesn't help, but still.
Will continue and try again kOmegaSSTLM in OpenFOAM, will keep updating if better results are obtained.

Now a question: Physically, placing the rough patch at the leading triggers turbulence, and creates a bigger boundary layer at the trailing edge. Why then the frequency of the shedding is lower?
EDIT: smaller boundary layer > more energy

And of course, I'm open to any comments