open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

I'm using k-epsilon model in my simulation (steady case). The subroutine KEPCL3 has been modified, in which the parameter FICTIFUET has been set to 200.0. That's because we are trying to use a very thin layer in the bottom.

But the problem raised due to this change. The vertical profiles of k and epsilon are no longer correct. It gave strange results. I've also tried to initialize k and epsilon by using the values obtained from previous runs but it still can't give good profiles.

Does this mean I can't use very thin layer in the bottom? Attached is the plot of the two profiles obtained from two cases.



Thanks in advance!

Best regards,
Qilong

Hello,

The classical value of FICTIFUET is 2., it is so because U* (UETCAR is U* squared) has been computed at mid first layer of elements. At this mid layer the finite difference derivatives along the vertical are second order, so this is presumably a good choice for finding UETCAR.

For epsilon we are facing the problem that it is infinite at Z=0 so we say that our bottom is in fact slightly above the real bottom, and we choose the position where UETCAR has been computed. Other points of view may give other values than 2, e.g. there are some arguments in favour of 2.71828**2. To sum up it seems that we have here an arbitrary parameter which is an obscure side of the k-epsilon model. Another way is to have a boundary condition which is not Dirichlet, in the form d(epsilon)/dn=..., this is another variant of the k-epsilon model.

In your case, if you are modifying the mesh to have thin layers near the bottom, you should not change FICTIFUET, as the point for computing epsilon at the bottom will naturally get closer to the bottom. Moreover, with very thin layers (let's say less that 1 mm) two things may happen :

1) the procedures checking elements crushed by tidal flats may start to work (this is for sure for elements less than 0.1 mm high).
2) the second plane may no longer be in the logarithmic profile, in which case the computation of U* may be wrong.

I hope this helps,

Jean-Michel Hervouet

Thank you very much for the explanations.

Now I modified the mesh definition instead of changing the parameter FICTIFUET. I use MESH TRANSFORMATION=2 (zstar) and then the vertical positions of each plane become

Plane 1: 4.00E-03
Plane 2: 9.00E-03
Plane 3: 1.40E-02
Plane 4: 2.40E-02
Plane 5: 0.20396
Plane 6: 1.0038
Plane 7: 2.0036
Plane 8: 6.0028
Plane 9: 12.0016
Plane 10: 20

Then I got strange profiles again.





Does this mean the bottom layer is too thin to get correct profiles?

Best Regards,
Qilong

Hello,

Hello,

You are getting into terra incognita but I asked colleagues and it turns out that we had a good logarithmic profile in k-epsilon model with a first layer of 2 mm, provided that the grain size corresponding to the friction is less that this first layer. This was in version 6.2, but another colleague reports difficulties in version 6.1. You should look at the velocity profile, it could be that it is not so bad despite the values of epsilon near the bottom.

Regards,

JMH

Hello,

I have the same experience, that a to thin bottom layer (as Jean-Michel mentions, depending on the grain size, ks-value) in 6.1 gives you some problems and crashes.
But I experience also that for correct computations the bottom-most layer is not necessary to be as thin.

cheers
Annalena

Thank you Jean-Michel and Annalena.

It seems that version 6.1 cannot solve the thin lay at bottom correctly. But this is what we would like to do because we are trying to include low-Reynolds effects near the bottom and low-Reynolds effects always happen in such thin bottom layer. If we don't use thin bottom layer, we may not capture those effects.

I tested several cases with different thickness of this bottom layer, from 2.5mm to 100mm. As you can see from the plot, the k profile becomes more smooth when the bottom layer become thicker. For epsilon, the thicker bottom layer I have, the smaller values near the bottom I got.





I also checked the velocity profiles of 2.5mm cases and both of them look OK.



Do you suggest I can only improve the profiles in version 6.2? Is there anything currently I can do to improve my result?

Best regards,
Qilong

Hello,

I found Annalena's Fortran subroutine princi_20_levels2.f in the old thread. In her code, a 20 vertical planes have been initialized with very thin layer at bottom (3mm). I used this 20 vertical layers in my model. The result gets better but there is still one point (bottom node) not consistent with the theoretical profile. It seems that the distributions of each vertical plane affect the results.



Best regards,
Qilong

Hello,

For the 20-layer case, it doesn't work for our modified low-Reynolds k-epsilon model. It's very similar to the situation I posted at the beginning of this thread 3 days ago.

Best regards,
Qilong

Hello,

I’ve tested the fine layers at bottom again. But this time I tried two different cases , one is without initialization of k and epsilon, and the other one is initialized by using the values obtained from previous runs. In the initialized case, the profiles seem ok except the bottom node. I've checked the steering files and the FORTRAN files I used in both cases. The only difference is the initialization of k and epsilon.












This also makes me surprised. I have encountered two situations generally. One is that the turbulence kinetic energy evolves through the whole water column since the beginning (situation 1). The other situation is that the turbulence only develops in the two or three layers down to the bottom and the k and epsilon above these layers stay at the minimum value no matter how much time it takes (situation 2).

In some cases, I didn’t initialize the k and epsilon and they evolve through the whole water column like I did in the comparison between the different bottom layer thickness from 5mm to 50mm. But in this case, I have the second situation if I don’t initialize k and epsilon.

Best regards,
Qilong