[raunakjung]

I am studying internal and external flow field of a fluidic oscillator.
This is my first attempt at using LES simulations in Ansys CFX. I am working in Reynolds no 10000 to 60000. I hope to use Tachycon supercomputer for my calculations. ( 3200 nodes , 25408 cores , Rpeak 300 tflops )

Is the Reynolds no range to high to carry out classical LES simulations ?

How can I calculate total time required for the simulation ?

Is classical LES too computationally expensive at such range ? (I am considering using WLES model.)

How is the grid requirement for such a calculation determined ? (I am calculating Kolmogorov length scale to determine grid points )

All constructive comments are requested.
Thank You

[mprinkey]

The textbook answer is to refine the mesh to fully capture the inertial range in all portions of the flow field that can exhibit relevant coherent structures. Use the recommended convection interpolation schemes and sub-grid models per the Ansys Fluent documentation. But, bare in mind that the 2nd or 3rd order schemes in Ansys will not resolve length scales at the order of the cell sizes. You will need several cells to cover a resolved eddy. When that it taken into account, the meshing requirements often become nearly absurd. While I can't recommend this, I do know that common practice is to mesh as finely as you can reasonably hope to get a solution in the allotted computer time and "see what you get."

By the way, meshing down to the Kolmogorov scale will be a DNS, not LES. Of course, the same caveats about required mesh spacing being much smaller than the relevant length scale still applies, though you shouldn't need the sub-grid model if you try to do that.

Just one other point--in the past, adaptive meshing has not really helped because of upsets to load balancing and the overhead from remeshing, but that situation may have changed with more recents Fluent releases...I haven't really stayed current.

[lcarasik]

Originally Posted by raunakjung

I am studying internal and external flow field of a fluidic oscillator.
This is my first attempt at using LES simulations in Ansys CFX. I am working in Reynolds no 10000 to 60000. I hope to use Tachycon supercomputer for my calculations. ( 3200 nodes , 25408 cores , Rpeak 300 tflops )

Is the Reynolds no range to high to carry out classical LES simulations ?

How can I calculate total time required for the simulation ?

Is classical LES too computationally expensive at such range ? (I am considering using WLES model.)

How is the grid requirement for such a calculation determined ? (I am calculating Kolmogorov length scale to determine grid points )

All constructive comments are requested.
Thank You

Please stop posting the same questions in different topics.

[raunakjung]

Originally Posted by mprinkey

The textbook answer is to refine the mesh to fully capture the inertial range in all portions of the flow field that can exhibit relevant coherent structures. Use the recommended convection interpolation schemes and sub-grid models per the Ansys Fluent documentation. But, bare in mind that the 2nd or 3rd order schemes in Ansys will not resolve length scales at the order of the cell sizes. You will need several cells to cover a resolved eddy. When that it taken into account, the meshing requirements often become nearly absurd. While I can't recommend this, I do know that common practice is to mesh as finely as you can reasonably hope to get a solution in the allotted computer time and "see what you get."

By the way, meshing down to the Kolmogorov scale will be a DNS, not LES. Of course, the same caveats about required mesh spacing being much smaller than the relevant length scale still applies, though you shouldn't need the sub-grid model if you try to do that.

Just one other point--in the past, adaptive meshing has not really helped because of upsets to load balancing and the overhead from remeshing, but that situation may have changed with more recents Fluent releases...I haven't really stayed current.

thank you for pointing out about meshing requirements. I think taylor length scale is appropriate than using the kolmogorov scale.

[FMDenaro]

http://scitation.aip.org/content/aip...1063/1.3676783

[raunakjung]

Originally Posted by FMDenaro

http://scitation.aip.org/content/aip...1063/1.3676783

I have been using RANS simulation to kind k and e which I would use to find turbulent length scale and taylor micro scale. Is this the proper approach to find mesh for LES ?

[FMDenaro]

I don't think that RANS can give you infos up to the Taylor micro-scale lenght. However, the LES mesh does not need to be so fine, the filter is at the level of the inertial range of the energy spectrum

[raunakjung]

After finding the turbulent kinetic energy and turbulent eddy dissipation. I am using the following to find turbulent energy length scale and taylor micro scale

Turbulent length scale L=k^(3/2)/∈
Taylor micro scale λ=√(10kν/ε)
Kolmogorov scale η=L*Re^((-3)/4)
Turbulence Reynolds number Re*L=k^2/εν

v = kinematic viscosity
I am using the following estimator to calculate the mesh resolution
Δ=max⁡( λ,L/10)

[raunakjung]

Originally Posted by FMDenaro

I don't think that RANS can give you infos up to the Taylor micro-scale lenght. However, the LES mesh does not need to be so fine, the filter is at the level of the inertial range of the energy spectrum

After finding the turbulent kinetic energy and turbulent eddy dissipation. I am using the following to find turbulent energy length scale and taylor micro scale

Turbulent length scale L=k^(3/2)/∈
Taylor micro scale λ=√(10kν/ε)
Kolmogorov scale η=L*Re^((-3)/4)
Turbulence Reynolds number Re*L=k^2/εν

v = kinematic viscosity
I am using the following estimator to calculate the mesh resolution
Δ=max⁡( λ,L/10)

[FMDenaro]

have a look here
http://www.springer.com/us/book/9783540263449