[backscatter]

I was running an LES case (channel with a cubical obstacle in the middle) for generating turbulent velocity field for subsequent simulations. I used turbulentInlet b.c for velocity inlet to generate some random uncorrelated noise. But to my surprise, these random perturbations seem to die out too fast and flow appears to be laminar. I tried increasing the intensity of the random fluctuating component but to no avail, as the flow ultimately turns laminar with rapidly dissipating noise.
Could it be the dissipation from the mesh causing it?
Has anyone else encountered this issue before?
Any help is appreciated!

[TomasDenk]

Have you checked what is the Reynolds number for your flow? Could it be, that it is in fact laminar and the perturbations you introduce on the inlet are unphysical and therefore die out?

[Santiago]

Originally Posted by halo

I was running an LES case (channel with a cubical obstacle in the middle) for generating turbulent velocity field for subsequent simulations. I used turbulentInlet b.c for velocity inlet to generate some random uncorrelated noise. But to my surprise, these random perturbations seem to die out too fast and flow appears to be laminar. I tried increasing the intensity of the random fluctuating component but to no avail, as the flow ultimately turns laminar with rapidly dissipating noise.
Could it be the dissipation from the mesh causing it?
Has anyone else encountered this issue before?
Any help is appreciated!

Anyway uncorrelated noise set directly on the velocity will, depending on the discretization of div(Uu), be dissipated by numerical viscosity or mess up with the pressure solver residuals. Avoid using such bc for canonical studies.

[backscatter]

Originally Posted by TomasDenk

Have you checked what is the Reynolds number for your flow? Could it be, that it is in fact laminar and the perturbations you introduce on the inlet are unphysical and therefore die out?

This case has been tried out before by researchers. Re is around 6000(based on channel half height.

[backscatter]

Originally Posted by Santiago

Anyway uncorrelated noise set directly on the velocity will, depending on the discretization of div(Uu), be dissipated by numerical viscosity or mess up with the pressure solver residuals. Avoid using such bc for canonical studies.

I was also thinking about this being the cause. I'll try changing the discretization scheme to see if that makes any difference...

[Santiago]

Originally Posted by halo

I was also thinking about this being the cause. I'll try changing the discretization scheme to see if that makes any difference...

You can do something better: just not to use it. Run a precursor simulation of a turbulent channel, with the same Re, until it reaches a statistically steady solution. Then extract vertical planes that would serve as inflow, using timeVaryingMappedFixedValue patch.

Another way is to construct a POD of a turbulent channel database to use as an inflow. This method, although approximate and relatively tricky, proves very flexible.

[cryabroad]

Seems like starting from v1606 openFOAM (the .com one) offers a B.C. of turbulentDFSEMInlet, have you tried it?

Related info can be found here: https://www.openfoam.com/releases/op...conditions.php

[backscatter]

Originally Posted by cryabroad

Seems like starting from v1606 openFOAM (the .com one) offers a B.C. of turbulentDFSEMInlet, have you tried it?

Related info can be found here: https://www.openfoam.com/releases/op...conditions.php

Thanks Ruiyan for the info. I have checked it out in past while I was still searching answers for this post..

[The King]

Hi,
Same happened to my simulations. Use a sampling plane close to your outflow to set turbulence at the inlet. If it’s far enough away, you should not get periodic effects.
A

[cryabroad]

I have been having this type of problem for a while and here is what I have found after searching a lot.

First of all, the turbulentInlet B.C. is just some white noise, as being pointed out by other posts, the perturbations easily die out downstream due to numerical viscosity. Of course you can adopt the "recycling" approach, meaning get some data downstream and feed it at the inlet over and over again, but this is not very practical at least in my view.

Second, you can check out the inflowGenerator developed at Rostock, just search inflowGenerator or LEMOS. I've tested this B.C. (no fully) for a while, and it does give you correlated velocity at the inlet and some fluctuations downstream. I also experienced the problem of perturbations dying out quickly downstream but as I refined my mesh, this problem seems to go away (still doing the test so this may not be true).

Anyway, I think the above B.C. plus appropriate mesh and numerical schemes can definitely give you way better results. For my case, which is just flow inside a cylindrical pipe, I already tried mesh with around 170,000 cells (125 cells in the axial direction, this direction is very crucial!) and the perturbations downstream are quite obvious!