CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   unrealistic velocity and vortex filed, how to remove this effect? (http://www.cfd-online.com/Forums/main/102764-unrealistic-velocity-vortex-filed-how-remove-effect.html)

lakeat June 1, 2012 11:53

unrealistic velocity and vortex filed, how to remove this effect?
 
Has anyone seen something like this? I got very unpleasant vortex traces in front of the building, when I was doing a 3D exterior LES incompressible flow simulation. It seems this is related to the mesh, but I guess there would be a way to remove this effect through convection term discretization scheme...
Could anyone who is familiar with LES look at this,

https://plus.google.com/photos/10298...PuUroPlqp7N-QE

In this link, one is the mesh, another is the vortex (extracted by Q method.)

Any ideas? Thanks

FMDenaro June 1, 2012 12:27

Quote:

Originally Posted by lakeat (Post 364319)
Has anyone seen something like this? I got very unpleasant vortex traces in front of the building, when I was doing a 3D exterior LES incompressible flow simulation. It seems this is related to the mesh, but I guess there would be a way to remove this effect through convection term discretization scheme...
Could anyone who is familiar with LES look at this,

https://plus.google.com/photos/10298...PuUroPlqp7N-QE

In this link, one is the mesh, another is the vortex (extracted by Q method.)

Any ideas? Thanks


to assess if your simulation is correct you must control the variables that are actually solved... the Q method does not say much....

sail June 5, 2012 08:51

It is hard to judge by your picture, but my best guess would be something related to the mesh.

from the first picture I see some discontinuities of the cell volume in certain areas, probably at the intersections within the various blocks.

I'm not an LES expert, but maybe this have something to do with the sub grid scale modelling? have you tried using some Dynamic SGS model?

another idea could be to increase the ortogonality corrector of the solvers or change the number of the multigrid scheme (assuming you are using GAMG).

just my 0.02€

edit: i'm assuming you are using OpenFoam.

lakeat June 5, 2012 09:11

Thanks for your input!
There is no discontinuity in the mesh, it is totally multi-blk structured mesh. It is just paraview display issue.

And I met this before when I did LES simulations, where in front of the obstacle there is always velocity wriggling, it is not just simply velocity gradient, because the wriggling is always on those mesh cells with high aspect ratio. (You can imagine that kind of mesh, since it is a multi-block mesh, so the near wall mesh will project to the far field, so to have a very high aspect ratio cells in the far field.)

I've been asked that why there is vortex there, I dont know how to get rid of this.
I hope there would be a solution not on the mesh, but on the setting of the diffsion scheme or something.

I do use non-orthogonality correction, (3 in OpenFOAM), but to no avail. And I was using filteredLinear for the convection of U.


Any experience?


EDIT
-------
Sometimes I was wondering if this is due to the delta definition in LES, which I have tried both cube-root of cell volume, or the smoothed delta, but still there is wriggling.

FMDenaro June 5, 2012 10:26

Basing on my experience on LES I suggest:

- perform a simulation on the same grid without any SGS model (LES no-model), check the solution by means of statistics, then compare the statistic with the modelled solution
- perform a grid refinement and check the new solutions, do that for 3 different grids

Of course I am assuming all the stability criteria are verified.

lakeat June 5, 2012 10:41

Will do. Thank you very much.

The 2nd suggestion would be difficult, but I would try your 1st suggestion first. May I know your point for the 1st suggestion is? What if something happen, or what if something does not happen, then your point is?

Btw, whether Lambda or Q criterion, I found the threshold is in general vague to set, I'd like to hear how would you identify the vortex. Even though it might be of little impact on the Statistic results.

FMDenaro June 5, 2012 10:55

Quote:

Originally Posted by lakeat (Post 364857)
Will do. Thank you very much.

The 2nd suggestion would be difficult, but I would try your 1st suggestion first. May I know your point for the 1st suggestion is? What if something happen, or what if something does not happen, then your point is?

Btw, whether Lambda or Q criterion, I found the threshold is in general vague to set, I'd like to hear how would you identify the vortex. Even though it might be of little impact on the Statistic results.


you can check the effect of the SGS model, I expect that the no-model simulation is stable but with possibly energy pile-up at the Nyquist frequency, that means you should see small vortical structures (of the order of some cell size) that are only of numerical nature, not physically relevant. The simulation with the SGS model (I assume you have an eddy viscosity model) should show such structures no longer present. Check rms and possibly spectra.

If you use the lambda-2 criterion, then a small negative value should be used as threshold

lakeat June 5, 2012 11:42

Okay, will do. Thanks for suggestion.

Generally, my experience on the bluff body simulation shows the statistical results is usually not bad and make sense.

And concerning these vortex identification methods, it seems to become an problem, I mean they seems to be "strongly physically defined" but not quite "numerical flexible".

This is usually little to do with the statistical results.

For example, I'd like to hear your suggestion too on this.
For instance, not everyone will go wall resolved LES, many are now go hybrid, and even in hybrid, not everytime you will have an ideal ~cubic cells in the wake region, actually many cells will be with high aspect ratio inevitably. The SGS model responds to these non-ideal grids with different manner, some works good, others not, other prefer even using max(delta_x, delta_y, delta_z) as filter..

I would hope there is a more powerful/flexible vortex identification to suit for hybird approach, high strectching meshes etc...

FMDenaro June 5, 2012 11:52

Quote:

Originally Posted by lakeat (Post 364867)
Okay, will do. Thanks for suggestion.

Generally, my experience on the bluff body simulation shows the statistical results is usually not bad and make sense.

And concerning these vortex identification methods, it seems to become an problem, I mean they seems to be "strongly physically defined" but not quite "numerical flexible".

This is usually little to do with the statistical results.

For example, I'd like to hear your suggestion too on this.
For instance, not everyone will go wall resolved LES, many are now go hybrid, and even in hybrid, not everytime you will have an ideal ~cubic cells in the wake region, actually many cells will be with high aspect ratio inevitably. The SGS model responds to these non-ideal grids with different manner, some works good, others not, other prefer even using max(delta_x, delta_y, delta_z) as filter..

I would hope there is a more powerful/flexible vortex identification to suit for hybird approach, high strectching meshes etc...

My opinion is that it exists much confusion in LES... physical modelling, mathematical filtering, numerical methods are not independent each other... for example, using implicit filtering both the grid and the numerical scheme defines the filter shape, inevitably you have a built-in smooth transfer function and a smoothing of the energy at resolved scales that is not du to the SGS model. Therefore, small vortical structures are "filtered" implicitly by the numerics and are then dissipated further by the SGS model. What you see in the small (resolved) vortical structures in LES can be very different from what you can see in a DNS at the same grid level. That does not mean that is wrong what you see in LES. Remember that in LES you resolve for a different variable, filtering effects can be relevant

michael1023 December 27, 2012 01:47

Hi,老魏

I investigated the same case (including same grid system and LES model and so on) both in OpenFoam and Fluent. The phenomenon of velocity wriggling was only found in OpenFoam. Do you find any solutions to get rid of this phenomenon.
Thank you.

Zhou Qiang

michael1023 December 27, 2012 03:31

Hi, 老魏。

I doubt this non-physical wiggle may cause by the divergence schemes. At first, I use the linear scheme (general scheme) which is always considered as the reason to cause the non-physical wiggles. Now I change the scheme to NVD scheme. But I have not got the new result so that I am not sure it can be effective.

michael1023 December 28, 2012 02:23

Hi, 老魏.

I think the unrealistic velocity is caused by the divergence scheme that you applied, but not caused by the grid system.

As we known, the central differencing scheme (the same as the linear scheme in OpenFoam) may cause the instability and non-physical wiggles. Thus I have applied the SFCD scheme (a kind of NVD scheme) and the non-physical wiggles have been avoided.

In addition, the SFCD scheme is not suitable for all kinds of cases since the vortex may be partly suppressed by this scheme in some cases. You can try to use this kind of scheme, please tell me if it works.

Zhou Qiang

immortality February 23, 2013 12:47

hi
whats the formula of SFCD and SFCDV?


All times are GMT -4. The time now is 03:50.