Ideal mesh for LES fluid flow simulation

  • 769 Views
  • Last Post 26 June 2018
peteroznewman posted this 17 June 2018

My friend Mantovani wants to build a Backward Facing Step flow simulation that includes an LES model.

Here is some guidance on meshing for LES that Mantovani found:

 

 

Here I applied the concept of Adapted Mesh that preserves aspect ratio to the backward facing step problem.
The first image shows the length of the inlet to the step. The length to the outlet from the step is much larger.

The second image shows the mesh details around the step.

While the element sizes for this illustration may be far too large for a good result, they can be easily reduced to a proper size.

Any comments from the community are welcome.

Attached is an ANSYS 19.0 archive.

Attached Files

  • Liked by
  • José Mantovani
  • raul.raghav
Order By: Standard | Newest | Votes
José Mantovani posted this 18 June 2018

It's very very good Peter! I will run simulations with this mesh and compare the results! 

Soon I post the results here! 

Thank's so much!

  • Liked by
  • vganore
vganore posted this 21 June 2018

Please post some results. Next step could be to see how the results are changing by generating structured mesh with edge sizing function (conventional method). 

Vishal Ganore, ansys.com/student

  • Liked by
  • José Mantovani
José Mantovani posted this 22 June 2018

Hello vganore! Thanks for attention and help!

I'm not using this mesh made by Peter, he used a schematic used in (like) structural analysis with contact zones and this did not work well on FLUENT. I am doing it by the conventional method, imposing functions of size in the limits.

Keep calm, I have not forgotten here, I am performing the simulations and collecting the data using RANS methods. Next week I hope to do this for LES modeling.

I've also changed the basis for the simulation, now I'm using now the Jovic and Driver experiment and not the Driver and Seegmiller experiment. I'm in the last week of school so I need to pay attention to my course too, next week I'll be on vacation. Soon I hope to share the results here and how I did to help the community. I'm writing an article about it, I hope to publish soon.

Hugs,

Mantovani.

raul.raghav posted this 23 June 2018

Adding to the discussion, LES is a little tricky when it comes to modeling. Following are a few points you should look into more carefully:

There is no grid independence in LES (implicitly filtered due to the grid). A grid independent LES simulation is essentially DNS because with finer grid you are resolving smaller and smaller turbulent scales and you can keep on going till you resolve the Kolmogorov scales which makes it DNS. In LES, you try to resolve the larger energy containing eddies and model the smaller "sub-grid scale (sgs)" eddies. So its essentially a filtering process (subfilter or sub-grid filtering) and as the filter size goes to zero, you're resolving all the turbulent scale and it becomes DNS. One of the main advantages and reasons of using LES is the fact that it provides an appropriate middle ground between numerical accuracy and computational cost. Good numerical accuracy demands larger computational costs.

Ten questions concerning the large-eddy simulation of turbulent flows

So the grid independence study that RANS models satisfy is not the same for LES models.

In LES, the total turbulent energy of the fluid is the sum of the resolved LES energy and unresolved sgs energy. So, you should aim to have a mesh with the y+ < 1 and as Peter pointed out, the aspect ratio needs to be carefully taken care of. And the grid resolution in general for LES should resolve more than 80% of the total TKE (Turbulent Flows by Pope, 2006).

A Dynamic Procedure for Calculating the Turbulent Kinetic Energy

The Large Eddy Simulation Model (LES)

 

 

 

Rahul

  • Liked by
  • José Mantovani
José Mantovani posted this 26 June 2018

It's very nice Rahul. I will study about it, thanks for the suggestions Maybe the LES approach, If I want do it, I need a better computer because the results with a low time flow is not good. As I said, I make first, the RANS approach in way to converge the experimental data with my numerical results from a BFS simulation with geometry and boundary conditions by the Jovic and Driver experimental study. I have some doubts about the reference values, my results of Cf and Cp did not converge with experimental data.

I will create a thread here in ANSYS Community in somehours with my doubts, because this can help me and other user's to how validate a CFD simulation using RANS approach. I'm write this...

I hope your help guys!

Thanks.

Close