CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > Main CFD Forum

solver algorithm for LES

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

Like Tree1Likes
  • 1 Post By arjun

Reply
 
LinkBack Thread Tools Display Modes
Old   January 2, 2017, 08:42
Default solver algorithm for LES
  #1
Member
 
raunak jung pandey
Join Date: Jun 2016
Posts: 64
Rep Power: 2
raunakjung is on a distinguished road
Hello I am using Fluent 15.0 to carry out my LES simulations.

Which solver algorithm is appropriate for fast and accuracte results ?
ITA/NITA/FSM/PISO/SIMPLEC

My simulations run normally and converge when using SIMPLE/SIMPLEC and Coupled but takes long time


My simulation shows follwoing error when using PISO and NITA
Pressure Correction = 0.3
Moment Correction = 0.7
" divergence detected in AMG solver : pressure correction"
raunakjung is offline   Reply With Quote

Old   January 2, 2017, 10:38
Default
  #2
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 3,081
Rep Power: 35
FMDenaro will become famous soon enoughFMDenaro will become famous soon enough
Try non-iterative fractional step setting second order time and space accuracy
FMDenaro is offline   Reply With Quote

Old   January 4, 2017, 09:29
Default
  #3
Member
 
raunak jung pandey
Join Date: Jun 2016
Posts: 64
Rep Power: 2
raunakjung is on a distinguished road
[QUOTE=FMDenaro;631794]Try non-iterative fractional step setting second order time and space accuracy[/QUOTE

Thank you for your answer. I am still facing convergence problem while using PISO and NITA. THis doesnt exist when I am using SIMPLE/SIMPLEC or Coupled and second order Implicit transeint formulation.

WHat maybe the reasons for this ?
raunakjung is offline   Reply With Quote

Old   January 4, 2017, 09:32
Default
  #4
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 3,081
Rep Power: 35
FMDenaro will become famous soon enoughFMDenaro will become famous soon enough
are you using the dynamic SGS model? Instability can happen for the eccessive value of the time step.
Could you plot the total kinetic energy versus the time?
FMDenaro is offline   Reply With Quote

Old   January 4, 2017, 10:34
Default
  #5
Member
 
raunak jung pandey
Join Date: Jun 2016
Posts: 64
Rep Power: 2
raunakjung is on a distinguished road
Quote:
Originally Posted by FMDenaro View Post
are you using the dynamic SGS model? Instability can happen for the eccessive value of the time step.
Could you plot the total kinetic energy versus the time?
No I couldnt plot total kinetic energy vs time.
I am using LES WALE and Smagorinsky model.
My time step size is 0.00001 sec giving courant no under 1.
raunakjung is offline   Reply With Quote

Old   January 4, 2017, 12:24
Default
  #6
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 3,081
Rep Power: 35
FMDenaro will become famous soon enoughFMDenaro will become famous soon enough
why you can not? have you tried to reduce the CFL number of one order?
FMDenaro is offline   Reply With Quote

Old   January 11, 2017, 07:59
Default
  #7
Member
 
Join Date: Jun 2016
Posts: 63
Rep Power: 3
Zbynek is on a distinguished road
If you use PISO in combination with NITA, you should employ 2-3 neighbor corrections. Besides, a high-quality mesh is desirable. NITA is naturally less stable than ITA since the coupling among equations is actually not enforced - there are only inner iterations, no outer iterations. Which implies that it is not applicable to all situations. The equation "faster computational method = the same stability & accuracy as the slower computational method" simply does not work

@FMDenaro - I am not aware of any simple method how to plot the turbulence kinetic energy in LES. Since it is defined via velocity fluctuations, one actually has to get there through instantaneous velocities and averaged velocities. To get averaged velocities means to save the instantaneous velocities at many time steps and to compute the mean and fluctuations after the calculation has finished (or have data from a previous calculation). Correct me if there is an easier way to get k.
Zbynek is offline   Reply With Quote

Old   January 11, 2017, 10:35
Default
  #8
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 3,081
Rep Power: 35
FMDenaro will become famous soon enoughFMDenaro will become famous soon enough
Quote:
Originally Posted by Zbynek View Post
If you use PISO in combination with NITA, you should employ 2-3 neighbor corrections. Besides, a high-quality mesh is desirable. NITA is naturally less stable than ITA since the coupling among equations is actually not enforced - there are only inner iterations, no outer iterations. Which implies that it is not applicable to all situations. The equation "faster computational method = the same stability & accuracy as the slower computational method" simply does not work

@FMDenaro - I am not aware of any simple method how to plot the turbulence kinetic energy in LES. Since it is defined via velocity fluctuations, one actually has to get there through instantaneous velocities and averaged velocities. To get averaged velocities means to save the instantaneous velocities at many time steps and to compute the mean and fluctuations after the calculation has finished (or have data from a previous calculation). Correct me if there is an easier way to get k.

Just to be clear: I wrote total kinetic energy that is available from a LES solution.
If the flow is incompressible then the solution v_bar allows to compute k_LES =v_bar . v_bar. Of course, that is not equivalent to k_bar. However, k_LES can be used to assess a statistical steady state or an energy pile-up leading to numerical instability.


what do you mean for:
NITA is naturally less stable than ITA since the coupling among equations is actually not enforced - there are only inner iterations, no outer iterations. Which implies that it is not applicable to all situations.
FMDenaro is offline   Reply With Quote

Old   January 12, 2017, 06:14
Default
  #9
Member
 
Join Date: Jun 2016
Posts: 63
Rep Power: 3
Zbynek is on a distinguished road
That is a pretty loose statement. I should have rather said that NITA is not suitable for every application. Obviously, strong coupling between equations would have adverse effects on convergence. I can imagine that to achieve a good convergence in multiphase flow modeling may be challenging using NITA. And just to quote the Fluent guide: the NITA solver is not recommended for highly viscous fluid flow.
Zbynek is offline   Reply With Quote

Old   January 12, 2017, 06:30
Default
  #10
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 3,081
Rep Power: 35
FMDenaro will become famous soon enoughFMDenaro will become famous soon enough
Despite of any high order accurate time-integration, the fractional step method has an intrinsic error due to the time-splitting. I don't know much about multiphase flows but I do not understand the warning about viscous flows...
FMDenaro is offline   Reply With Quote

Old   January 12, 2017, 08:17
Default
  #11
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 627
Rep Power: 19
arjun will become famous soon enough
Quote:
Originally Posted by FMDenaro View Post
Despite of any high order accurate time-integration, the fractional step method has an intrinsic error due to the time-splitting. I don't know much about multiphase flows but I do not understand the warning about viscous flows...
Once to compare these solvers Prof. Peric (Milovan Peric) did some tests and demostrated that even with time step as small as 5 times of iterative method PISO could not produce the results same as iterative method.

I created a fractional solver where I could include this correction and could demostrate that when including this non-linear correction I could achieve results like iterative scheme for the test cases similar to what Milovan used (ie vortex shedding from cylinders). By switching this term off got results same as PISO.

So yes, what you say has real effect and could be demostrated.


This is why I created hybrid solver for multiphase flows where courant determines times step and using iterative scheme is expensive while NITA is unstable.

I have been using this in FVUS for VOF calculations to cut the calculation cost by at least half.
arjun is offline   Reply With Quote

Old   January 12, 2017, 09:31
Default
  #12
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 3,081
Rep Power: 35
FMDenaro will become famous soon enoughFMDenaro will become famous soon enough
Quote:
Originally Posted by arjun View Post
Once to compare these solvers Prof. Peric (Milovan Peric) did some tests and demostrated that even with time step as small as 5 times of iterative method PISO could not produce the results same as iterative method.

I created a fractional solver where I could include this correction and could demostrate that when including this non-linear correction I could achieve results like iterative scheme for the test cases similar to what Milovan used (ie vortex shedding from cylinders). By switching this term off got results same as PISO.

So yes, what you say has real effect and could be demostrated.


This is why I created hybrid solver for multiphase flows where courant determines times step and using iterative scheme is expensive while NITA is unstable.

I have been using this in FVUS for VOF calculations to cut the calculation cost by at least half.

What type of correction have you implemented? Some years ago I deeply analysed the errors appearing in the fractional step method and the sources were highlighted in the splitting but also in the bc.s.
Thus, some remedies were proposed to get full high order accuracy

https://www.researchgate.net/publica...ary_conditions
FMDenaro is offline   Reply With Quote

Old   January 12, 2017, 10:00
Default
  #13
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 627
Rep Power: 19
arjun will become famous soon enough
Quote:
Originally Posted by FMDenaro View Post
What type of correction have you implemented? Some years ago I deeply analysed the errors appearing in the fractional step method and the sources were highlighted in the splitting but also in the bc.s.
Thus, some remedies were proposed to get full high order accuracy

https://www.researchgate.net/publica...ary_conditions


What I did was that I stored the momentum matrix and once the velocity correction was available I corrected flux and added another correction to source to momentum eqn. Then solved again (rather than setting up whole thing).

I wrote down outline in a ppt, I try to find it and attach here. (it was long time ago so don't remember the steps exactly).

Give me a bit of time to find that ppt.
arjun is offline   Reply With Quote

Old   January 12, 2017, 10:11
Default
  #14
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 627
Rep Power: 19
arjun will become famous soon enough
Okay found the pdf.
slide 10 first chart is from Prof. Milovan's previous test.

Second and third from my tests with on and off of this correction.

This outlines the algo that I tried.


Quote:
Originally Posted by FMDenaro View Post
What type of correction have you implemented? Some years ago I deeply analysed the errors appearing in the fractional step method and the sources were highlighted in the splitting but also in the bc.s.
Thus, some remedies were proposed to get full high order accuracy

https://www.researchgate.net/publica...ary_conditions
Attached Files
File Type: gz DesignOfCorrectiveSIMPLE.pdf.tar.gz (191.1 KB, 2 views)
FMDenaro likes this.
arjun is offline   Reply With Quote

Old   January 12, 2017, 10:48
Default
  #15
Senior Member
 
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 3,081
Rep Power: 35
FMDenaro will become famous soon enoughFMDenaro will become famous soon enough
No specific corrections in the BC.s? Have you checked the convergence slope on some analytical test?
FMDenaro is offline   Reply With Quote

Old   January 12, 2017, 11:19
Default
  #16
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 627
Rep Power: 19
arjun will become famous soon enough
Quote:
Originally Posted by FMDenaro View Post
No specific corrections in the BC.s? Have you checked the convergence slope on some analytical test?
For BCs I chose to keep it in two forms Dirichlet or correction = 0 on fixed value types (inlet etc) or Neumann or 0 grad for correction. I took care while updating source (I think) but I feel it is worth investigating. (i don't remember details well).

Anyway did not work much on it but on a provided test case from industry it beat PISO of starCD by factor of 2 or so. (105 seconds vs 265 seconds of PISO in that test case).

All in all I feel it is worth exploring but I have no time. May be in future I try it in FVUS, as for now there is lot to do there so no time.
arjun is offline   Reply With Quote

Reply

Tags
les model

Thread Tools
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 On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Floating Point Exception Error nyox FLUENT 8 January 22, 2016 08:53
Quarter Burner mesh with periosic condition SamCanuck FLUENT 2 August 31, 2011 11:34
Working directory via command line Luiz CFX 4 March 6, 2011 21:02
n-s equation solver using simple algorithm paresh.halder FLUENT 0 December 3, 2010 17:00
The correction on pressure equation of SIMPLE algorithm in MRFSimpleFOAM solver renyun0511 OpenFOAM Running, Solving & CFD 0 November 10, 2010 02:47


All times are GMT -4. The time now is 04:00.