Turbulent flow at high Reynolds number - Convergence problem

Numerical methods and mathematical models of Elmer
Paul2015
Posts: 9
Joined: 07 Apr 2018, 15:43
Antispam: Yes

Turbulent flow at high Reynolds number - Convergence problem

Post by Paul2015 » 13 Apr 2018, 10:48

Dear All,

First of all, thanks a lot to Elmer developers and community members for this efficient tool.

I am a CFD user since more than 25 years and I have a significant experience in OpenFoam. My goal, by using Elmer, is to simulate conjugate heat transfer or acoustics, in view to complete OpenFoam in these domains.
At first, I tried to simulate (not so) high Reynolds flows and, after a lot of tests and search on this forum or in the documentation, I am still unable to obtain a converging solution.
I came back to the tutorials and try to increase the inlet velocity, and I have the same problems.

I am using Elmer 8.2.

For instance, with the tutorial "Turbulent Incompressible Step Flow", if I increase the inlet velocity, the simulation diverges after a few iterations (about 3-5).
I have tested all the advices that I found (no Newton method, at first, of course wall law, etc...) but I am always facing the same problem.
I also tested, on another case, the increase in mesh (which was already fine...) and there's no changes...
By comparison, a coarser mesh, in the same conditions, on the same model, converges very quiclky with OpenFoam...

My question :
- Do you have the method to reach very high Reynolds number (for example, in the "Turbulent Incompressible Step Flow" case, to converge with an inlet velocity > 2 m/s, at least (my objective is to reach higher inlet velocity, of course) ?
May be it dépends on the solvers options, etc... ? So, if someone has already carried out high velocity simulations ?...
- In addition, I noticed that an increase in mesh refinement leads to a very important increase in memory size, which will strongly limit the use of Elmer to small cases (in contrary of OpenFoam that I use on very large models). Is there a method to avoid that ?

Thank you very much in advance for your answers...

Best Regards,

P.S. : I don't join the case file, as we can discuss on an existing tutorial

Paul2015
Posts: 9
Joined: 07 Apr 2018, 15:43
Antispam: Yes

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by Paul2015 » 16 Apr 2018, 10:56

Hello,

To extend my question most generally, is there some limitations for high Reynolds simulation with Elmer ?
If not, did someone of you already carry out some simulations at high Reynolds (typically in air at high velocity) and what were the specificities of your model/solver parameters to converge in a usual case (3D, ...) ?

Thank you very much in advance,
Best Regards,

raback
Site Admin
Posts: 3326
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by raback » 17 Apr 2018, 23:34

Hi Paul

In your introduction you more or less outline the problem. Elmer development is guided by the projects where we are involved in. We try to pick them synergetically but in some areas there are less opportunities. Now in turbulent flows OpenFOAM dominates the open source community. This means that there is not much request for Elmer in this application area. It has to do also with the formulations. OpenFOAM uses FVM which is very suited for parabolic equations and Elmer uses FEM which is ideal for elliptic equations. An CFD example very fitting for Elmer is the highly non-newtonian flow of glaciers. Also computational electromagnetism, acoustics, fsi, lubrication etc. are areas where Elmer is quite strong.

In principle we have a lot of models for CFD also in high-Re regime. Unfortunately they are not very robust. The equations for (k,e) or (k,omega) don't seem to behave too well if the initial guess is poor. I guess the reason is that FEM is not conservative and does not ensure monotonicity of the solution. There are numerical tricks also in FEM that could fix these (e.g. Flux Corrected Transport) but we just haven't worked too much in that area.

CFD was in fact among the first application areas of Elmer. We still feel that the stabilized versions are pretty ok. They basically introduce an implicit LES model. There is also a stabilization method called "vms" as in Variational Multiscale Method. These models unfortunately require that the timestep is small enough. Otherwise turbulent effects will be filtered out.

Actually one way to complement these shortcomings is combination of two software. There is recent developments to couple Elmer and OpenFOAM. These are rather generic. There are two different approaches 1) file based and 2) mpi-based. You can find discussion on these on the forum, and more on eof-libracy.com (by Univ. of Latvia).

-Peter

Paul2015
Posts: 9
Joined: 07 Apr 2018, 15:43
Antispam: Yes

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by Paul2015 » 18 Apr 2018, 11:26

Dear Peter,

Thanks a lot for your reply.
Actually, this is what I thought (and what I feared...).

Anyway, I think that it is possible to simulate flows at higher Reynolds number by taking some specific precautions.
Since I initiated this thread, I've found some examples in this forum that shows that it is possible, in some conditions.
The convergence problems happend very quicly when we increase velocity (for instance from 2 to 3-5 m/s). I am still thinking that it is a solver configuration problem, as the physic of the case has not changed so much by increasing the velocity so slightly...

So, I still keep interested with all information about the way to obtain a good convergence. If someone has already tested Elmer in air at quite high inlet velocity, it will be very helpful to discuss how they reached these results.

2 points, however
- The use of the GUI seems to be quite limited, as the .sif file is not always complete enough to have a convergent solution (as for k-epsilon or k-omega SST : it seems that there is always something missing, so it is not obvious, without any example, to obtain a complete .sif file);
- The use of GUI also let some "bugs" in the .sif file, as the presence of "Prandtl number" when we use viscosity model, which imposes to modify the .sif manually to use ElmerSolver.exe instead.
These points are not a problem in the case that we have some examples about what to do for turbulent flows at high Reynolds and to create .sif file manually.

Thanks again and I will really appreciate if someone has already use Elmer at high Reynolds number or at hugh inlet velocity in air, for instance.

Best Regards

mzenker
Posts: 1807
Joined: 07 Dec 2009, 11:49
Location: Germany

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by mzenker » 18 Apr 2018, 11:45

Hi,
- The use of GUI also let some "bugs" in the .sif file, as the presence of "Prandtl number" when we use viscosity model, which imposes to modify the .sif manually to use ElmerSolver.exe instead.
These points are not a problem in the case that we have some examples about what to do for turbulent flows at high Reynolds and to create .sif file manually.
I am not sure if that is what you mean, but the Turbulent Prantl Number appears when you activate the turbulent heat conduction model in the heat equation. But if you delete the contents of the field, it will not appear in the sif file.

HTH,

Matthias

Paul2015
Posts: 9
Joined: 07 Apr 2018, 15:43
Antispam: Yes

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by Paul2015 » 18 Apr 2018, 12:39

Dear Matthias,

Thank you for your reply.

The Turbulent Prandtl number appears, for instance, in "Material/SST k-omega/Properties", without activating the heat equation.
Once you activated a turbulent equation, and choose the material, it is taken into account in the model and written in the case.sif.
But it seems to be not managed then by the solver and causes an error and the simulation to stop.

Best regards,

mzenker
Posts: 1807
Joined: 07 Dec 2009, 11:49
Location: Germany

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by mzenker » 18 Apr 2018, 13:22

Hi,

if you delete the contents of the "Turbulent Prantl Number" field in the SST/k-omega tab in the materials properties, it will not appear in the sif.

HTH,

Matthias

Paul2015
Posts: 9
Joined: 07 Apr 2018, 15:43
Antispam: Yes

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by Paul2015 » 18 Apr 2018, 15:19

Dear Peter, dear all,

May be I could reformulate my question in another way.
What solver configuration do you recommend in view to treat, for example, these 2 tutorials with higher air inlet velocity (for start, we can take an air inlet at 10 m/s for example in each case) ?
- In 2D, the turbulent incompressible step flow, for instance;
- in 3D (which is the most important interest to me), the Thermal flow in curved pipe (with or without heat transfer, here, what is important is the fluid velocity)
By solver configuration, I mean solver settings, as Method, Control, Convergence parameters, Numerical Techniques, ... but also the parameters of the model setup (steady, transient, scanning, etc...).

Also, for inlet boundary condition, I noticed that a velocity profile is always used, instead of an homogeneous velocity. Is there some problems to use an homogeneous velocity (divergence problems ?...).

Dear Matthias,
Thank you for your help.
I already tested that but I noticed some problems with GUI (not actualizes the changes). I will try again before generate the .sif file a first time...

Paul2015
Posts: 9
Joined: 07 Apr 2018, 15:43
Antispam: Yes

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by Paul2015 » 18 Apr 2018, 22:11

Dear all,

2 very interesting cases :

viewtopic.php?f=3&t=539

viewtopic.php?f=3&t=4925

I already test the 2nd one and I will test the 1st one... and the test case FEM/Tests/Diffuser_SST
A lot of interesting information in these cases.

The next step will be to simulate 3D cases at higher inlet velocities...

mzenker
Posts: 1807
Joined: 07 Dec 2009, 11:49
Location: Germany

Re: Turbulent flow at high Reynolds number - Convergence problem

Post by mzenker » 19 Apr 2018, 10:18

Paul2015 wrote:Dear Peter, dear all,
Dear Matthias,
Thank you for your help.
I already tested that but I noticed some problems with GUI (not actualizes the changes). I will try again before generate the .sif file a first time...
The best in such cases is to sif->generate, then sif->edit to examine the result.
Before starting the solver, don't forget to save the project.

HTH,

Matthias

Post Reply