Setting up a DES

max3.2 · May 4, 2011, 03:47

Hi,

Im about to set up a DES from a acceptably converged (after approx 150 It.) SST Solution. Calculation of Total Time and Timesteps is no the problem, but whats a good value to choose for Iterations? Do i have to calculate the 150Loops every Timestep or is it sufficient to choose a lower value (around 40, there's not much RES left...)

Regards
Max

ghorrocks · May 4, 2011, 07:26

Transient simulations usually run best on about 3-5 coeff loops per iteration. Multiphase or other tricky physics (eg chemistry) might go up to 10. This is all described in the documentation.

max3.2 · May 4, 2011, 07:40

Allright, thanks. Ill take a deeper look.
Another thing - even if i said i knew how to set up timestepping, i was wrong. DES Blending Function wont go anywhere near 0. Am I right to assume my timestep is to big? I cant find anything in the Documentation (wheres in DES-chapter cited appendix 55?) and it seems the things i found were wrong. I have to admit, i just lookes around the net

max3.2 · May 4, 2011, 14:37

I now tried to subsequently reduce the timestep, and so im getting larger areas with small blending values. but still i do not reach 0. is this an error or maybe connected to the model? is there a possibility to increase the threshhold so cfx will blend earlier to LES?

ghorrocks · May 4, 2011, 19:25

The DES model is an advanced turbulence model. Unless you are familiar with LES models (things like dissipation, filtering functions and matching it to turbulent spectrums) then I would not recommend using the DES model.

ANSYS does advanced turbulence training courses which describe the use of these models, or you can do it the old fashion way of reading the literature.

The 2-equation turbulence models (k-e, SST, k-w) are much simpler to use. This means you are much more likely to correctly apply it. Why not use one of them?

max3.2 · May 5, 2011, 05:23

Quote:

Originally Posted by ghorrocks

The 2-equation turbulence models (k-e, SST, k-w) are much simpler to use. This means you are much more likely to correctly apply it. Why not use one of them?

Thats already done.
I trie to model Vortex movement and turbulent structure over time, and im writing a paper about it done with LES/DES. So its okay if i have to read some literature, ive got plety of time.
So I've done mainly about LES, but most of it was about the math, and not how to find the known parameters in Pre.
I will go on try to find some more, seems like it is enough for 3 months of work

ghorrocks · May 5, 2011, 07:30

That sounds good. There is lots of background knowledge you need before you start on a LES model. Make sure you understand Kolmogorov length scales, Taylor microscales, turbulence energy cascades and all that sort of stuff as a background. Then do some LES models where you attempt to correctly model turbulence energy cascades. Once you can correctly model a turbulent energy cascade you are ready to proceed.

max3.2 · May 5, 2011, 08:58

Allright i will check those.
Is there a book or something you'd say is a "must-read"? I've got the Peric, Ferzinger CFD-Book, and some other german stuff about LES-Modelling, it should be enough, but if there's something outstanding, I'll check that too.

ghorrocks · May 5, 2011, 18:40

Peric and Ferzinger are good. I would also recommend "Turbulence Modelling for CFD" by Wilcox. It is my reference for turbulence modelling.

max3.2 · May 10, 2011, 12:41

Ok, I've done some research, there's still more to come, but i have a few question...
If I get it right, CFX Docu says boundary layer grid spacing may not fall under 0.1 boundary layer thickness.
Now I also found in "Fröhlich,J: LES Simulation [2006]" (german text translated): Upper Value for Grid-Spacing delta < 0.1 boundary layer, which is the exact opposite. Whats right?
At this time, I'm not sure if there's an adavantage in comparison to URANS, since the Mesh is very fine and some Vortex structures are already resolved. Theres some Vortex shedding at Blade-PS down to Blade-SS of the next blade and at the TE of the blade. Those Vortices are then carried out of the domain. (Q-Crit)

Still I am trying to get the Blending function to 0, bur I'm not finding any information on this function. I'm trying timesteps around 1ms (which is still to small for the problem, but way over Kolmogorov. But since im not doing a DNS, it should work...) and getting RMS Courant Numbers around 150. This is said (again, Documentation and Peric, Ferzinger) to be possible. If I go up with my step, Courant also goes up. How far can i go?

Since i got the Mesh to work with, I think i have to do some changes there. This again leads to the question above, if I need to coarsen or fine it.

I will continue doing research, but I'd be more than happy for any advice

Best Reagrds
Max

ghorrocks · May 10, 2011, 18:47

Often LES is set up to use the taylor microscale as the time step size, or some multiple of it. This means you will be resolving the large eddies but modelling the small ones - of course this is the whole point of LES.

How does your time step compare to the taylor microscale?

max3.2 · May 11, 2011, 04:56

Im not sure if i do the Taylor scale calculation right.
Im getting a taylor scale (a lenght) which is approx 1/4 of my timestep (a time?)

ghorrocks · May 11, 2011, 06:45

You cannot compare a time and a distance meaningfully. You need to get an eddy turnover time to get the equivalent time scale.

max3.2 · May 13, 2011, 09:44

Hmm im not getting a appropriate turnover time. until now, i've only found some correlations and so on...
still im looking for ansys tech report on des, cited as [55]...someone got this?

May 4, 2011, 03:47	Setting up a DES	#1
max3.2 Member Max Join Date: Apr 2011 Posts: 47 Rep Power: 15	Hi, Im about to set up a DES from a acceptably converged (after approx 150 It.) SST Solution. Calculation of Total Time and Timesteps is no the problem, but whats a good value to choose for Iterations? Do i have to calculate the 150Loops every Timestep or is it sufficient to choose a lower value (around 40, there's not much RES left...) Regards Max

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Cells with t below lower limit	Purushothama	Siemens	2	May 31, 2010 21:58
How to use DES well?	Daniel	Main CFD Forum	0	October 26, 2008 05:59
DES ( SA+LES) vs DES ( kwSST+LES) for bluff body	Phil	Main CFD Forum	0	January 8, 2008 09:04
Problem with DES model	mehul	FLUENT	0	April 12, 2006 07:59
Warning 097-	AB	Siemens	6	November 15, 2004 04:41

May 4, 2011, 07:26		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Transient simulations usually run best on about 3-5 coeff loops per iteration. Multiphase or other tricky physics (eg chemistry) might go up to 10. This is all described in the documentation.

May 4, 2011, 07:40		#3
max3.2 Member Max Join Date: Apr 2011 Posts: 47 Rep Power: 15	Allright, thanks. Ill take a deeper look. Another thing - even if i said i knew how to set up timestepping, i was wrong. DES Blending Function wont go anywhere near 0. Am I right to assume my timestep is to big? I cant find anything in the Documentation (wheres in DES-chapter cited appendix 55?) and it seems the things i found were wrong. I have to admit, i just lookes around the net

May 4, 2011, 14:37		#4
max3.2 Member Max Join Date: Apr 2011 Posts: 47 Rep Power: 15	I now tried to subsequently reduce the timestep, and so im getting larger areas with small blending values. but still i do not reach 0. is this an error or maybe connected to the model? is there a possibility to increase the threshhold so cfx will blend earlier to LES?

May 4, 2011, 19:25		#5
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	The DES model is an advanced turbulence model. Unless you are familiar with LES models (things like dissipation, filtering functions and matching it to turbulent spectrums) then I would not recommend using the DES model. ANSYS does advanced turbulence training courses which describe the use of these models, or you can do it the old fashion way of reading the literature. The 2-equation turbulence models (k-e, SST, k-w) are much simpler to use. This means you are much more likely to correctly apply it. Why not use one of them?

May 5, 2011, 07:30		#7
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	That sounds good. There is lots of background knowledge you need before you start on a LES model. Make sure you understand Kolmogorov length scales, Taylor microscales, turbulence energy cascades and all that sort of stuff as a background. Then do some LES models where you attempt to correctly model turbulence energy cascades. Once you can correctly model a turbulent energy cascade you are ready to proceed.

May 5, 2011, 08:58		#8
max3.2 Member Max Join Date: Apr 2011 Posts: 47 Rep Power: 15	Allright i will check those. Is there a book or something you'd say is a "must-read"? I've got the Peric, Ferzinger CFD-Book, and some other german stuff about LES-Modelling, it should be enough, but if there's something outstanding, I'll check that too.

May 5, 2011, 18:40		#9
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Peric and Ferzinger are good. I would also recommend "Turbulence Modelling for CFD" by Wilcox. It is my reference for turbulence modelling.

May 10, 2011, 12:41		#10
max3.2 Member Max Join Date: Apr 2011 Posts: 47 Rep Power: 15	Ok, I've done some research, there's still more to come, but i have a few question... If I get it right, CFX Docu says boundary layer grid spacing may not fall under 0.1 boundary layer thickness. Now I also found in "Fröhlich,J: LES Simulation [2006]" (german text translated): Upper Value for Grid-Spacing delta < 0.1 boundary layer, which is the exact opposite. Whats right? At this time, I'm not sure if there's an adavantage in comparison to URANS, since the Mesh is very fine and some Vortex structures are already resolved. Theres some Vortex shedding at Blade-PS down to Blade-SS of the next blade and at the TE of the blade. Those Vortices are then carried out of the domain. (Q-Crit) Still I am trying to get the Blending function to 0, bur I'm not finding any information on this function. I'm trying timesteps around 1ms (which is still to small for the problem, but way over Kolmogorov. But since im not doing a DNS, it should work...) and getting RMS Courant Numbers around 150. This is said (again, Documentation and Peric, Ferzinger) to be possible. If I go up with my step, Courant also goes up. How far can i go? Since i got the Mesh to work with, I think i have to do some changes there. This again leads to the question above, if I need to coarsen or fine it. I will continue doing research, but I'd be more than happy for any advice Best Reagrds Max

May 10, 2011, 18:47		#11
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Often LES is set up to use the taylor microscale as the time step size, or some multiple of it. This means you will be resolving the large eddies but modelling the small ones - of course this is the whole point of LES. How does your time step compare to the taylor microscale?

May 11, 2011, 04:56		#12
max3.2 Member Max Join Date: Apr 2011 Posts: 47 Rep Power: 15	Im not sure if i do the Taylor scale calculation right. Im getting a taylor scale (a lenght) which is approx 1/4 of my timestep (a time?)

May 11, 2011, 06:45		#13
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	You cannot compare a time and a distance meaningfully. You need to get an eddy turnover time to get the equivalent time scale.

May 13, 2011, 09:44		#14
max3.2 Member Max Join Date: Apr 2011 Posts: 47 Rep Power: 15	Hmm im not getting a appropriate turnover time. until now, i've only found some correlations and so on... still im looking for ansys tech report on des, cited as [55]...someone got this?