how can use the LES filter? (laplace filter and anistropic filter)

ethan oh · January 3, 2012, 10:20

Our thease is about filtration in LES.
We study LESfilters codes (LESfilter,NewFilter,Simple,Laplace,anistropic) but we have following problem about them:
1-Algorithm and Formulation of filtration that used in these codes.
2-According to the discription part of laplacefilter.H and anisotropicfilter.H :

Description
Kernel as filter as Test filter with ratio 2
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~
Box filter: g = delta2/24 -> g = delta2/6
Spherical box filter: g = delta2/64 -> g = delta2/16
Gaussian filter: g = delta2/24 -> g = delta2/6

what is "g" parameter and where is the effect of "g" in these codes.
3-What is "widthcoeff_" that used in these codes?

We REALLY need the answer of these question.thanks for your help.

Bernhard · January 3, 2012, 11:14

The implementation of the filter can be found for example in de laplacaFilter.C file. It is defined in the operator() for different field types. This is smart, since if you somewhere define

Code:

laplaceFilter filter_;

You can operate use it as filter_( anyField ) as a type of function (this is what is done in the LES models).

Now if you follow the coeff_ in this piece of code, you will find how widthCoeff_ is used.

With respect to your second question, I am not so familiar with the Laplace filter, so I cannot help you there.

ehsan · February 26, 2012, 06:25

Hello,

Could I kindly ask you which filter type is used in LES of incompressible flows OpenFoam? In fact, we don't set any filter in LESproprties for interPhaseChangeFoam.

Regards

Djub · July 18, 2012, 12:16

Hi Ehsan,

I think the available filters in Foam (simple, laplace, anisotropic,...) are in fact the filters for the dynamic models (synSmagorinsky, dynOneEqEddy, maybe dynLagrangian,...). In fact, these are the filter for the test level, not the grid level. With oneEqEddy or Smagorinsky, there is only one level, so only one filtering, chosen by delta (PrandtlDelta, cubeRootVolDelta, smoothDelta, maxDeltaxyz).

Please, correct me if I am wrong!

Sincerely,

Djub

arunsmec · November 7, 2012, 01:09

Hi Djub
I am trying to simulate a roundjet. I am using Smagorisnsky model and wish to use a Gaussian filter.
According to you the filter is dictated by delta. So how can i implement a Gaussian filter for my problem? Which 'delta' will give a fairly Gaussian filter?

Thank you

Djub · November 7, 2012, 05:11

hi arunsmec,

OK, it seems I was not exact. Not wrong neither, but not exact...

delta is the SIZE of the filter, not his SHAPE.

The LESfilter can act on the both levels (subgrid and dynamic):
- simpleFilter acts only on the dynamic level, with a simple top-hat filter
- laplaceFilter and anisotropicFilter can affect both levels:

Kernel as filter as Test filter with ratio 2
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~
Box filter: g = delta2/24 -> g = delta2/6
Spherical box filter: g = delta2/64 -> g = delta2/16
Gaussian filter: g = delta2/24 -> g = delta2/6

Thus, you should be able to use a Gaussian filter, changing the kernel.

but I don't know how...

francesco_capuano · November 28, 2012, 17:04

For LES, OpenFOAM adopts the so-called implicit filtering technique, i.e., the equations are implicitly filtered by the discretization schemes and computational grid: no explicit filtering is applied. In such technique, the filter width (which appears in the SGS models) is typically related to a certain measure of the local grid cell size, which can be chosen via the 'delta' parameter; a pretty usual choice is to take the cube root volume of the local cell.

However, some SGS models, for instance those belonging to the dynamic family, require two levels of filtering. When using those models, the first level of filtering is implicit (as usual), whilst the second one (the so-called test filter) has to be performed explicitly by means of one of the discrete forms of filters provided.

aylalisa · February 7, 2014, 17:13

Hello francesco_capuano,, Hello Djub!

In case I would like to simulate transition from laminar to turbulent flow I guess I have to choose a dynamical sgs model like dynSmagorinsky.
That way I ensure that in different flow regimes the appropriate value of the coefficient (SmagorinskyCoefficient C

) is used, instead of some damping function (van Driest). condition: fine grid (in the direction I want to do explicit filtering).

Does this mean in case the grid spacing is fine enough I don't have to care neither for wall functions nor for damping functions?
And in case I have a coarse grid I don't have to care for damping functions but I have to care for wall functions?

Do I have to set one of the availabe filters ( laplaceFilter, simpleFilter or anisotropicFilter) only in the case I use a dynamic LESmodel (dynSmagorinsky)?

@Djub
You've written that

is the size of the filter. I've seen in an example case ( that the LESModel and delta are defined in the LESProperties file with help of appropriate keywords and in the same tutorial case the defined solver is pisoFoam.

How does OpenFoam account for $\tau_{ij}^R$ in the Navier-Stokes equations which results from analytical procedure of filtering?
How is the filtered nonlinear convective term of momentum equation incorporated in the solution procedure? pisoFoam is a solver for transient incompressible flow, initially without containing an additional shear stress term that accounts for turbulence effects?

Are the stress components of the tensor $\tau_{ij}^R$ computed with help of Smagorinsky's closure idea:
$\tau_{ij}^r =-2 \nu_r \overline{S}_{ij}$
$\nu_r = l^2_S \overline{S} = (C_S \Delta )^2 \overline{S}$ ?

...and somehow add (?) to the discretized equations?

In case of high Re:

= constant
- no dynamik model
- close to wall: damping function necessary
- if mesh is coarse --> wall functions necessary

In case of low Re + wall treatment + high Re:

= computed tensor
- dynamik model necessary (to account for all situations)
- therefore no damping function necessary
- wall function --> depending on mesh resolution (yes if mesh=coarse, no if mesh=fine)
????

To what extent do the discretization schemes in fvSchemes influence the choosen LESmodel???

How exactly is the filter technique defined?
I thought the transfer function of the filter function is determined by the energy spectrum of the flow. So I thought depending on the spectrum you receive a transfer function and this one leads to the Coefficient C respectively to the filter width $\Delta$ ???
But in OpenFoam the user is prompted to decide the filter size $\Delta$ .
How is the user defined filter size processed in OpenFoam?
Is the spectrum determined by the choice of the LESmodel?

I really hope for your expert support!

Thanks.
Aylalisa

ehsan · August 27, 2015, 17:16

Hello
Could any one says what is the purpose of:

pow(mesh.V(), 2.0/3.0)/widthCoeff_

in laplcaFilter.C.

Regards

turbu · July 12, 2017, 15:44

A few questions about LES filtering in OpenFOAM

Mahdi2010 · January 15, 2018, 10:53

Does anybody have idea about how to distinguish between Gaussian filter and Box filter in OpenFOAM?

Because according to the source code, there is no difference between Gaussian and Box filter. Both are implemented under the name of "Laplace filter" and both get the same coefficient (either 24 or 6) as the input. For sure the discrete formulation of these filters must be different, but I cannot see the difference in implementation.

zhangyan · December 26, 2018, 03:37

Quote:

Originally Posted by Mahdi2010

Does anybody have idea about how to distinguish between Gaussian filter and Box filter in OpenFOAM?

Because according to the source code, there is no difference between Gaussian and Box filter. Both are implemented under the name of "Laplace filter" and both get the same coefficient (either 24 or 6) as the input. For sure the discrete formulation of these filters must be different, but I cannot see the difference in implementation.

Hi
Have you got an answer now?

January 3, 2012, 10:20	how can use the LES filter? (laplace filter and anistropic filter)	#1
ethan oh New Member ethan oh Join Date: Feb 2011 Posts: 9 Rep Power: 15	Our thease is about filtration in LES. We study LESfilters codes (LESfilter,NewFilter,Simple,Laplace,anistropic) but we have following problem about them: 1-Algorithm and Formulation of filtration that used in these codes. 2-According to the discription part of laplacefilter.H and anisotropicfilter.H : Description Kernel as filter as Test filter with ratio 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~ Box filter: g = delta2/24 -> g = delta2/6 Spherical box filter: g = delta2/64 -> g = delta2/16 Gaussian filter: g = delta2/24 -> g = delta2/6 what is "g" parameter and where is the effect of "g" in these codes. 3-What is "widthcoeff_" that used in these codes? We REALLY need the answer of these question.thanks for your help. ivfuente likes this.

January 3, 2012, 11:14		#2
Bernhard Senior Member Bernhard Join Date: Sep 2009 Location: Delft Posts: 790 Rep Power: 22	The implementation of the filter can be found for example in de laplacaFilter.C file. It is defined in the operator() for different field types. This is smart, since if you somewhere define Code: laplaceFilter filter_; You can operate use it as filter_( anyField ) as a type of function (this is what is done in the LES models). Now if you follow the coeff_ in this piece of code, you will find how widthCoeff_ is used. With respect to your second question, I am not so familiar with the Laplace filter, so I cannot help you there.

July 18, 2012, 12:16		#4
Djub Senior Member Julien Join Date: Jun 2012 Location: France Posts: 152 Rep Power: 14	Hi Ehsan, I think the available filters in Foam (simple, laplace, anisotropic,...) are in fact the filters for the dynamic models (synSmagorinsky, dynOneEqEddy, maybe dynLagrangian,...). In fact, these are the filter for the test level, not the grid level. With oneEqEddy or Smagorinsky, there is only one level, so only one filtering, chosen by delta (PrandtlDelta, cubeRootVolDelta, smoothDelta, maxDeltaxyz). Please, correct me if I am wrong! Sincerely, Djub oshubham likes this.

November 7, 2012, 01:09	Filter for Smagorinsky model	#5
arunsmec Member Aru Join Date: Feb 2012 Location: Chennai, India Posts: 40 Rep Power: 14	Hi Djub I am trying to simulate a roundjet. I am using Smagorisnsky model and wish to use a Gaussian filter. According to you the filter is dictated by delta. So how can i implement a Gaussian filter for my problem? Which 'delta' will give a fairly Gaussian filter? Thank you

November 7, 2012, 05:11	Trying to help	#6
Djub Senior Member Julien Join Date: Jun 2012 Location: France Posts: 152 Rep Power: 14	hi arunsmec, OK, it seems I was not exact. Not wrong neither, but not exact... delta is the SIZE of the filter, not his SHAPE. The LESfilter can act on the both levels (subgrid and dynamic): - simpleFilter acts only on the dynamic level, with a simple top-hat filter - laplaceFilter and anisotropicFilter can affect both levels: Kernel as filter as Test filter with ratio 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~ Box filter: g = delta2/24 -> g = delta2/6 Spherical box filter: g = delta2/64 -> g = delta2/16 Gaussian filter: g = delta2/24 -> g = delta2/6 Thus, you should be able to use a Gaussian filter, changing the kernel. but I don't know how... sungda likes this.

February 26, 2012, 06:25		#3
ehsan Senior Member Ehsan Join Date: Mar 2009 Posts: 112 Rep Power: 17	Hello, Could I kindly ask you which filter type is used in LES of incompressible flows OpenFoam? In fact, we don't set any filter in LESproprties for interPhaseChangeFoam. Regards

November 28, 2012, 17:04		#7
francesco_capuano Member Francesco Capuano Join Date: May 2010 Posts: 81 Rep Power: 16	For LES, OpenFOAM adopts the so-called implicit filtering technique, i.e., the equations are implicitly filtered by the discretization schemes and computational grid: no explicit filtering is applied. In such technique, the filter width (which appears in the SGS models) is typically related to a certain measure of the local grid cell size, which can be chosen via the 'delta' parameter; a pretty usual choice is to take the cube root volume of the local cell. However, some SGS models, for instance those belonging to the dynamic family, require two levels of filtering. When using those models, the first level of filtering is implicit (as usual), whilst the second one (the so-called test filter) has to be performed explicitly by means of one of the discrete forms of filters provided. Xinze, ruedi, abhi22 and 5 others like this.

February 7, 2014, 17:13		#8
aylalisa Senior Member Join Date: Nov 2012 Location: Bavaria Posts: 145 Rep Power: 14	Hello francesco_capuano,, Hello Djub! In case I would like to simulate transition from laminar to turbulent flow I guess I have to choose a dynamical sgs model like dynSmagorinsky. That way I ensure that in different flow regimes the appropriate value of the coefficient (SmagorinskyCoefficient C) is used, instead of some damping function (van Driest). condition: fine grid (in the direction I want to do explicit filtering). Does this mean in case the grid spacing is fine enough I don't have to care neither for wall functions nor for damping functions? And in case I have a coarse grid I don't have to care for damping functions but I have to care for wall functions? Do I have to set one of the availabe filters ( laplaceFilter, simpleFilter or anisotropicFilter) only in the case I use a dynamic LESmodel (dynSmagorinsky)? @Djub You've written that is the size of the filter. I've seen in an example case ( that the LESModel and delta are defined in the LESProperties file with help of appropriate keywords and in the same tutorial case the defined solver is pisoFoam. How does OpenFoam account for $\tau_{ij}^R$ in the Navier-Stokes equations which results from analytical procedure of filtering? How is the filtered nonlinear convective term of momentum equation incorporated in the solution procedure? pisoFoam is a solver for transient incompressible flow, initially without containing an additional shear stress term that accounts for turbulence effects? Are the stress components of the tensor $\tau_{ij}^R$ computed with help of Smagorinsky's closure idea: $\tau_{ij}^r =-2 \nu_r \overline{S}_{ij}$ $\nu_r = l^2_S \overline{S} = (C_S \Delta )^2 \overline{S}$ ? ...and somehow add (?) to the discretized equations? In case of high Re: = constant - no dynamik model - close to wall: damping function necessary - if mesh is coarse --> wall functions necessary In case of low Re + wall treatment + high Re: = computed tensor - dynamik model necessary (to account for all situations) - therefore no damping function necessary - wall function --> depending on mesh resolution (yes if mesh=coarse, no if mesh=fine) ???? To what extent do the discretization schemes in fvSchemes influence the choosen LESmodel??? How exactly is the filter technique defined? I thought the transfer function of the filter function is determined by the energy spectrum of the flow. So I thought depending on the spectrum you receive a transfer function and this one leads to the Coefficient C respectively to the filter width $\Delta$ ??? But in OpenFoam the user is prompted to decide the filter size $\Delta$ . How is the user defined filter size processed in OpenFoam? Is the spectrum determined by the choice of the LESmodel? I really hope for your expert support! Thanks. Aylalisa

August 27, 2015, 17:16		#9
ehsan Senior Member Ehsan Join Date: Mar 2009 Posts: 112 Rep Power: 17	Hello Could any one says what is the purpose of: pow(mesh.V(), 2.0/3.0)/widthCoeff_ in laplcaFilter.C. Regards

July 12, 2017, 15:44		#10
turbu New Member victor Join Date: Nov 2015 Location: pku,china Posts: 5 Rep Power: 11	A few questions about LES filtering in OpenFOAM

January 15, 2018, 10:53		#11
Mahdi2010 Member Mahdi Join Date: Jul 2012 Posts: 53 Rep Power: 14	Does anybody have idea about how to distinguish between Gaussian filter and Box filter in OpenFOAM? Because according to the source code, there is no difference between Gaussian and Box filter. Both are implemented under the name of "Laplace filter" and both get the same coefficient (either 24 or 6) as the input. For sure the discrete formulation of these filters must be different, but I cannot see the difference in implementation.