CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > Software User Forums > OpenFOAM > OpenFOAM Pre-Processing

Which LES model and discretization schemes to choose?

Register Blogs Members List Search Today's Posts Mark Forums Read

Like Tree2Likes
  • 2 Post By Andy_bm

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   April 8, 2017, 07:49
Default Which LES model and discretization schemes to choose?
  #1
New Member
 
Lukas Lebovitz
Join Date: Mar 2017
Location: Zürich, Switzerland
Posts: 25
Rep Power: 6
lukas.lebo is on a distinguished road
Hello CFD Online

I'm new to openFoam and using version 2.4.0. I'm interested in studying windload on buildings and I will be using a turbulent inlet generator developed by Marc Immer (Turbulent Inflow Generator for LES).

I'm a bit baffled by the number of different LES models and discretization schemes out there.

Are there any guidelines on which models to choose for specific cases? Or a composition of Pro's and Con's for different models and schemes? Most LES models let you set their model constants which would be another question mark for me.

Or is anybody brave enough to give some reasoned recommendations on which LES model and discretization schemes to use?

Any help is greatly appreciated!
lukas.lebo is offline   Reply With Quote

Old   April 8, 2017, 10:29
Default
  #2
Senior Member
 
piu58's Avatar
 
Uwe Pilz
Join Date: Feb 2017
Location: Leipzig, Germany
Posts: 681
Rep Power: 11
piu58 is on a distinguished road
Some words to the discredidation schemes:

CFD calculates the flow values (most important p and U), but not their derivatives. The derivatives must me calculated suing differences.

If you have an orthogonal mesh this process is straightforward. Skewed cells need a correction for geometry. If you calculate more correct values the results may give very high derivatives which makes the result numerical unstable. To overcome this you give some of the accuracy and change the calculation for more stability. One way is to limit the result.
For high accuracy it is a fine idea to make much effort with the mesh and use more accurate schemes.
__________________
Uwe Pilz
--
Die der Hauptbewegung überlagerte Schwankungsbewegung ist in ihren Einzelheiten so hoffnungslos kompliziert, daß ihre theoretische Berechnung aussichtslos erscheint. (Hermann Schlichting, 1950)
piu58 is offline   Reply With Quote

Old   April 8, 2017, 14:54
Default
  #3
New Member
 
Lukas Lebovitz
Join Date: Mar 2017
Location: Zürich, Switzerland
Posts: 25
Rep Power: 6
lukas.lebo is on a distinguished road
Thank you Uwe for taking the time to help me out. I haven't thought about it from this perspective yet.

Now I understand that you would choose the scheme depending on the quality of your mesh. I heard that for LES a scheme with high accuracy i.e. central differencing is beneficial. In return this means you basically want a structured mesh for LES.

Can we evaluate the stability-accuracy-relation on a few common schemes in this post? (as this might be helpful for others too)

I haven't engaged this topic too deeply but as far as I understand a linear scheme would provide best accuracy for worst stability. Whereas stability can be retained by introducing limiters as in limitedLinear (for a tradeoff in accuracy). Where do linearUpwind and the gamma scheme fit into this? And how do they compare regarding computational efficiency (if there is any significant difference at all)?
lukas.lebo is offline   Reply With Quote

Old   April 8, 2017, 16:25
Default
  #4
Senior Member
 
piu58's Avatar
 
Uwe Pilz
Join Date: Feb 2017
Location: Leipzig, Germany
Posts: 681
Rep Power: 11
piu58 is on a distinguished road
All in all , we should use second order schemes, so I discourage the use of upwind (without linear), because it is a first order scheme.

LinearUpwind is a combination of linear and upwind and stabilises the solution, as always at the expense of accuracy. Tobi evaluated this en détail:
http://www.holzmann-cfd.de/index.php...s-linearupwind


For starting it may be best to use the Gauss linear limited schemes. You may set the limitation value to the one which you need to stabilize your solution. I always start wit 1 which means not limited at all. a limitation of 0.7 or even 0.5 is not that worse. If you don't reach with this value I recommend the look at the mesh first.

May be somebody else has another recommendation, but for my (not too large) experience linear limited gives a lot of flexibility.
__________________
Uwe Pilz
--
Die der Hauptbewegung überlagerte Schwankungsbewegung ist in ihren Einzelheiten so hoffnungslos kompliziert, daß ihre theoretische Berechnung aussichtslos erscheint. (Hermann Schlichting, 1950)
piu58 is offline   Reply With Quote

Old   April 8, 2017, 16:54
Default
  #5
New Member
 
Lukas Lebovitz
Join Date: Mar 2017
Location: Zürich, Switzerland
Posts: 25
Rep Power: 6
lukas.lebo is on a distinguished road
Thanks Uwe, you really helped me. The evaluation of Tobi is very interesting.

Now I'm hoping that someone still cares about elaborating differences of the LES models.
lukas.lebo is offline   Reply With Quote

Old   April 9, 2017, 01:55
Default
  #6
Senior Member
 
piu58's Avatar
 
Uwe Pilz
Join Date: Feb 2017
Location: Leipzig, Germany
Posts: 681
Rep Power: 11
piu58 is on a distinguished road
Quote:
Originally Posted by lukas.lebo View Post
Thanks Uwe, you really helped me. The evaluation of Tobi is very interesting.

Now I'm hoping that someone still cares about elaborating differences of the LES models.
Some words to the LES models.

The LES simulates the vortices down to a size which depends on the mesh. What happens with these large eddies is not covert by the simulation in a direct way.

In reality, energy ist transferd for the large eddies to smaller ones and from these to even smaller ones until the size is so small that the friction force ist large enough to dissolve the eddie in heat. This scale is called https://en.wikipedia.org/wiki/Kolmogorov_microscales. Problems of the real world cannot be simulated down to this scale (in most cases).

What happens downwards form the large eddies needs to be covered by a model. This model should predict the energy loss of the large eddies in course of time at least. There exist some models for it, and you have to look at the publication which describes a special model to know what the intention of the model is. Unfortunately, not all of the publications are free available.

There exist models which give a good over all balance, and others, which are more accureate in the near of boundaries.

I think, that dynamic models are more accurate, but I am not entirely sure of it (i did not try all of them). In Openfoam and for in-compressible media , there exist the often used Dynamic one equation eddy-viscosity model and the subsgrid scale with Lagrangian averaging. From my experience, the last is more accurate but tends more to oscillations / numerical instability.
__________________
Uwe Pilz
--
Die der Hauptbewegung überlagerte Schwankungsbewegung ist in ihren Einzelheiten so hoffnungslos kompliziert, daß ihre theoretische Berechnung aussichtslos erscheint. (Hermann Schlichting, 1950)
piu58 is offline   Reply With Quote

Old   April 9, 2017, 06:51
Default
  #7
New Member
 
Lukas Lebovitz
Join Date: Mar 2017
Location: Zürich, Switzerland
Posts: 25
Rep Power: 6
lukas.lebo is on a distinguished road
If we compare the dynamic Smagorinsky and the dynamic OneEqEddy model, the biggest difference is that the SGS turbulent viscosity and thus the turbulent energy is computed differently.

In the OneEqEddy model you solve a transport equation for k and in Smagorinsky model you compute it with a model equation.

Are there any known advantages of solving this additional transport equation for the SGS turbulent energy?
lukas.lebo is offline   Reply With Quote

Old   April 9, 2017, 15:12
Default
  #8
Senior Member
 
piu58's Avatar
 
Uwe Pilz
Join Date: Feb 2017
Location: Leipzig, Germany
Posts: 681
Rep Power: 11
piu58 is on a distinguished road
To the one eqaution model:

As far as I understand this model, a new value, the subgrid-scale kinetic energy, ist introduced. This you may imagine as the kinetic energy trapped in an eddy. The energy moves with the flow and dissipates with the time. The model looks reasonable for me, as long as we look at eddies in the free flow. In the near of boundaries, the transport equation is not fully applicable.

There is a short wiki page of the model

https://www.cfd-online.com/Wiki/Kine...id-scale_model

The reference is

Kim, W and Menon, S. (1995), "A new dynamic one-equation subgrid-scale model for large eddy simulation", In 33rd Aerospace Sciences Meeting and Exhibit, Reno, NV, 1995.
__________________
Uwe Pilz
--
Die der Hauptbewegung überlagerte Schwankungsbewegung ist in ihren Einzelheiten so hoffnungslos kompliziert, daß ihre theoretische Berechnung aussichtslos erscheint. (Hermann Schlichting, 1950)
piu58 is offline   Reply With Quote

Old   July 14, 2017, 03:42
Default
  #9
Senior Member
 
akidess's Avatar
 
Anton Kidess
Join Date: May 2009
Location: Germany
Posts: 1,379
Rep Power: 27
akidess will become famous soon enough
Quote:
Originally Posted by piu58 View Post
LinearUpwind is a combination of linear and upwind and stabilises the solution, as always at the expense of accuracy. Tobi evaluated this en détail:
http://www.holzmann-cfd.de/index.php...s-linearupwind
No, it's not. linearUpwind is also known as second order upwind, and does not blend the with the linear scheme (also known as central differencing).
__________________
*On twitter @akidTwit
*Spend as much time formulating your questions as you expect people to spend on their answer.
akidess is offline   Reply With Quote

Old   March 3, 2021, 10:21
Default
  #10
Member
 
Join Date: Nov 2011
Posts: 32
Rep Power: 12
Andy_bm is on a distinguished road
I conducted an investigation of the dissipative properties and stability of most numerical schemes implemented in the OpenFOAM package.
The two benchmark problems were solved:
  1. The problem of convective transfer of the scalar φ (analyze the stability of the numerical scheme).
  2. The decay of homogeneous isotropic turbulence (the dissipative properties of numerical schemes).
The Gamma and filteredLinear schemes were improved. New schemes are available at gitHub: https://github.com/eacfd/libSchemes or https://github.com/unicfdlab/libSchemes
A comparison of the modified numerical schemes with their initial implementation is shown in the attached files.

You can try new schemes in your cases and give me feedback. They are well proven for incompressible flows.

Full research published in https://link.springer.com/article/10...70048219060024

Regards, AE
Attached Images
File Type: png newschemes.png (95.0 KB, 39 views)
clapointe and lukasf like this.
Andy_bm is offline   Reply With Quote

Old   September 9, 2021, 11:59
Default
  #11
Member
 
Lukas Fischer
Join Date: May 2018
Location: Germany, Munich
Posts: 87
Rep Power: 5
lukasf is on a distinguished road
Quote:
Originally Posted by Andy_bm View Post
I conducted an investigation of the dissipative properties and stability of most numerical schemes implemented in the OpenFOAM package.
The two benchmark problems were solved:
  1. The problem of convective transfer of the scalar φ (analyze the stability of the numerical scheme).
  2. The decay of homogeneous isotropic turbulence (the dissipative properties of numerical schemes).
The Gamma and filteredLinear schemes were improved. New schemes are available at gitHub: https://github.com/eacfd/libSchemes or https://github.com/unicfdlab/libSchemes
A comparison of the modified numerical schemes with their initial implementation is shown in the attached files.

You can try new schemes in your cases and give me feedback. They are well proven for incompressible flows.

Full research published in https://link.springer.com/article/10...70048219060024

Regards, AE

Looks very interesting. I am going to test it. You can use the library after compiling it by adding this line to your controlDict:


Code:
libs
(
"libSchemes.so"
);
I noticed using rhoPimpleFoam with Gauss linear for the divergence schemes leading to over- and undershooting of the U and T field. Do you see any issues to apply your schemes on the divergence schemes for the compressible solver and LES WALE model? The flow field is nearly incompressible though with Mach number << 0.3.

Code:
divSchemes
{
    default         none;
    div(phi,U)      Gauss filteredLinearM 1 0.9;
    div(phi,K)      Gauss filteredLinearM 1 0.9;
    div(phi,h)      Gauss filteredLinearM 1 0.9;
    div(((rho*nuEff)*dev2(T(grad(U))))) Gauss linear;
}
lukasf is offline   Reply With Quote

Old   September 20, 2021, 09:05
Default
  #12
Member
 
Join Date: Nov 2011
Posts: 32
Rep Power: 12
Andy_bm is on a distinguished road
I think that Gauss linear scheme usage will lead to numerical oscillations for compressible flows. Therefore, If the Mach number increases the betta coefficient should be decreased in order to the absence of numerical oscillations. I used these schemes for jets in subsonic compressible regions; the supersonic regions were solved with vanLeer or Koren schemes.
What is the velocity (Mach number) in your problem? If it is a compressible flow try to start from convective term Gauss GammaM 0.1/0.2 0.6 or Gauss filteredLinearM 1 0.5.

Last edited by Andy_bm; September 22, 2021 at 06:33.
Andy_bm is offline   Reply With Quote

Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
What is the discretization method of S-A model yzeyue Main CFD Forum 0 April 9, 2014 23:16
Transition SST model - convergence problem with 2nd order discretization xara FLUENT 4 May 27, 2013 14:16
Spatial vs Angular Discretization in DO radiation model renaldi FLUENT 0 June 1, 2009 11:52
Reynolds transport, turbulence model, etc Beginner Main CFD Forum 1 January 7, 2009 06:36
Angular Discretization for 2d DO Model RAO FLUENT 0 April 18, 2006 06:40


All times are GMT -4. The time now is 20:24.