[Thromkar]

Hello everyone,

I am Simulating a heattransfer problem with Ansys Fluent as seen on the attached picture.

Setup.jpg

I'm Using K-Omega-SST Turbulencemodel with "Coupled" Pressure-Velocity Coupling and the Pseudo-Transient formulation. Deactivating the Pseudo-Transient Solver leeds to the same results with slower convergence.

When performing a mesh independence study, i experienced a weird behaviour. As i increased my cell count, the pressure drop increased. The picture below shows the pressure drop over a mesh parameter. Doubling the mesh parameter will leed to 8 times more cells.

MeshRefinement.png

As we can see the pressure drop is increasing, but i never reach mesh independence. Refining the mesh even more is an option i want to avoid, since i allready got 6 million Elements in a 2x5x15 mm domain.

The Mesh is as good as it can get. Max. y+ Values are at 0,8 for most of the meshes. The y+ of the finest meshes are even lower.

I experienced the same behaviour with larger Setups, which contain more of the periodic geometries and using fixed velocity inlet and pressure outlet.

My question is, is there something wrong with my setup? Can i use symmetry boundary conditions (since the geometry is symmetrical I assume I can)?

[acalado]

Are you using symmetry to essentially "double" your domain?

Which wall treatment functions are you using? I'm assuming near wall if your y+ is near 1.

And your results don't look too different. Between first and last mesh there is about 1% relative difference, which I would think is a good enough solution.

Cheers

[LuckyTran]

Looks good. Forget the phrase mesh independence, it's mesh sensitivity.


However, I am skeptical your results for any given mesh are that accurate. For example you're saying that, for a fixed mesh, as you iterate your pressure drop stays exactly 622.5 +/- 0 and doesn't oscillate at all? Put numerical error bars on this plot. I wouldn't be surprised to see 622.5 +/- 5.

[Thromkar]

Are you using symmetry to essentially "double" your domain?

Which wall treatment functions are you using? I'm assuming near wall if your y+ is near 1.

And your results don't look too different. Between first and last mesh there is about 1% relative difference, which I would think is a good enough solution.

Cheers

I am using symmetry on both sides to essentially make a infinite domain.

When I understand it correctly, the SST model is switching automatically between Wall Functions and near wall treatment. Since i'm at y+<1 i assume, that im using near wall treatment. Am I wrong here?
Do i need to use the "Low-Re Correction" to enable near wall treatment?

Looks good. Forget the phrase mesh independence, it's mesh sensitivity.


However, I am skeptical your results for any given mesh are that accurate. For example you're saying that, for a fixed mesh, as you iterate your pressure drop stays exactly 622.5 +/- 0 and doesn't oscillate at all? Put numerical error bars on this plot. I wouldn't be surprised to see 622.5 +/- 5.

Thanks you. In the future i will use the phrase "mesh sensitivity".

Sorry i didn't mention my convergence criterias. I defined my own convergence criteria by monitoring the "Periodic Pressure Gradient" and the "Bulk Temperature Ratio". My simulation is converged when both variables differ less than 1e-6 in the last 50 iterations. So the results aren't differ in 6th significant digits. I am very sure that my Results are not oscillating anymore. BTW all of my residuals are way below 1e-10.

[LuckyTran]

Originally Posted by Thromkar

Sorry i didn't mention my convergence criterias. I defined my own convergence criteria by monitoring the "Periodic Pressure Gradient" and the "Bulk Temperature Ratio". My simulation is converged when both variables differ less than 1e-6 in the last 50 iterations. So the results aren't differ in 6th significant digits. I am very sure that my Results are not oscillating anymore. BTW all of my residuals are way below 1e-10.

In my experience if you run longer say over 100,000 iterations you'll find some slow gradual fluctuations in the periodic pressure gradient and bulk temperature ratio. And these can be quite significant.


But regardless, I think you've got some nice results.