sdr does not converge

Deranda · January 20, 2016, 03:44

Hello,

in my simulation the sdr residuum does not converge.
When I look at the normalized residuals, everything looks quiet good, but when I turn off the normalization, the it looks really really bad. Something around 0.8 to 2.... (sdr starts at 500 then drops to 2 after 30 iterations,

My question is, is there any other possibility to achieve better results instead of increasing the inner iterations?

Im using the k-w Menter SST turbulence model for modelling a cyclone separator.

best regards
Nils

fluid23 · January 20, 2016, 12:03

What are your wall y+ values?

Deranda · January 20, 2016, 15:42

in the area of the main body the values have a range from 0.3 to 20 (only 6% above 5) with an all y+ approach.
Then there are two areas, which are extruded with a large grid, just to get a homogeneous outlet and inlet stream...

meanwhile I have not normalized residuals of about 0.3 (with 10 iteration as well as with 20)

best regards
Nils

fluid23 · January 20, 2016, 15:59

Is this k-e or k-w?

fluid23 · January 20, 2016, 16:02

Wait... I just caught the whole w/o normalization part, sorry. If your normalized residuals drop 3 orders of magnitude and flatten out, then you are golden. I wouldn't even worry about non-normalized if your normalized residuals are OK.

Deranda · January 21, 2016, 03:20

Hi,

in my normalized Residuals, the turbulent kinetic energy does not converge ^^ It drops from 1 to 0,2 even if I use 20 inner iterations.

Well sdr does converge, because its reference should be one of the first iterations, which were totally bad.

best regards Nils

fluid23 · January 21, 2016, 09:47

what turbulence model?

Deranda · January 21, 2016, 10:32

k-w menter sst with an all y+ approach
time step: 0.0002s (Implicit Unsteady)
Inner Iterations: 10

fluid23 · January 21, 2016, 10:47

My first thought is that you may need more inner iterations. I usually can't get by with only 10. How are you initializing your model? The first few time steps often require more attention.

I would also try to refine your mesh a little. For all wall y+ in k-w SST you should really stay below 1 or between 30 and 60. It sounds like you are in the the interpolated region for a lot of the wall boundary cells. If you have low Reynolds number then I would just stick with low wall treatment and target <1.

https://steve.cd-adapco.com/articles/en_US/FAQ/RD-5-273

Deranda · January 21, 2016, 10:50

My y+ Values stay 99.1971% below 1 und 100% below 5,
after simulating 0.15 seconds

fluid23 · January 21, 2016, 11:24

Ok, well it doesn't sound like y+ is your issue. Unfortunately, convergence problems are the hardest to resolve. There are so many different factors that can contribute to this that it can start to feel like you are chasing your tail a bit.

Possibility 1: Your boundary conditions are inappropriate. Either they are too close, of the wrong type, or something to that effect. I run into issues quite often doing helicopter hover cases. When this happens I run the mesh diagnostic plugin. This allows you to see not only mesh quality, skewness, aspect ratio, etc... But also will show cells with high residuals. If you run this and find that high residuals are near boundary conditions, then you can usually just ignore the convergence plot (assuming you cannot resolve the issue by correcting boundary conditions in the first place). If you find them more in the middle of your domain near your geometry of interest then you may need to define a mesh refinement scheme or volumetric control, something to further subdivide the high residual cells.

http://macrohut.cd-adapco.com/phpBB3...ilit=mesh#p776

Possiblilty 2: Your time step is actually too small. In this case your round off errors can become large and the resulting noise makes convergence problematic.

Possibility 3: Your inner iterations need to be increased. I know you are trying to avoid this, but it is what it is. I would at least trying increasing to see what happens.

Best of luck!

Deranda · January 22, 2016, 03:44

I didnt found the area with Cells residuals in the diagnostics, but I displayed my skewness angle.
In my volume of interest, 0,00107771% of the cells have a skewness angle above 85° and 0,00246333% falls below the VolumeChange kriteria.
In the Outlet-Area, which is not of high interest and separated by a poroues media, are more skewness angles above 85° at the boundaries.
maybe that causes the sdr resdiuals...
The last night, I simulated again with 30 inner iterations and its almost the same result regarding the residuals, so that should not be the problem...
The next step will be correcting the skewness angles, maybe I'll find a way, dont know yet^^

I will tell you my results

January 20, 2016, 03:44	sdr does not converge	#1
Deranda Member Nils Join Date: Nov 2015 Posts: 59 Rep Power: 11	Hello, in my simulation the sdr residuum does not converge. When I look at the normalized residuals, everything looks quiet good, but when I turn off the normalization, the it looks really really bad. Something around 0.8 to 2.... (sdr starts at 500 then drops to 2 after 30 iterations, My question is, is there any other possibility to achieve better results instead of increasing the inner iterations? Im using the k-w Menter SST turbulence model for modelling a cyclone separator. best regards Nils

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
can't converge in FLUENT	ok___ko	FLUENT	1	April 30, 2013 03:47
k and epsilon were hard to converge in multiphase model of Fluent	Yanlong Li	ANSYS	0	January 2, 2013 06:25
HELP !In relaxtion factor converge is taken or not	MANOJ KUMAR	FLUENT	5	September 22, 2005 05:16
Converge problem for multiphase flow	Jen	FLUENT	2	September 8, 2005 09:47
Converge problem for multiphase flow	Jen	FLUENT	4	July 20, 2005 17:52

January 20, 2016, 12:03		#2
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	What are your wall y+ values?

January 20, 2016, 15:42		#3
Deranda Member Nils Join Date: Nov 2015 Posts: 59 Rep Power: 11	in the area of the main body the values have a range from 0.3 to 20 (only 6% above 5) with an all y+ approach. Then there are two areas, which are extruded with a large grid, just to get a homogeneous outlet and inlet stream... meanwhile I have not normalized residuals of about 0.3 (with 10 iteration as well as with 20) best regards Nils

January 20, 2016, 15:59		#4
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	Is this k-e or k-w?

January 20, 2016, 16:02		#5
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	Wait... I just caught the whole w/o normalization part, sorry. If your normalized residuals drop 3 orders of magnitude and flatten out, then you are golden. I wouldn't even worry about non-normalized if your normalized residuals are OK.

January 21, 2016, 03:20		#6
Deranda Member Nils Join Date: Nov 2015 Posts: 59 Rep Power: 11	Hi, in my normalized Residuals, the turbulent kinetic energy does not converge ^^ It drops from 1 to 0,2 even if I use 20 inner iterations. Well sdr does converge, because its reference should be one of the first iterations, which were totally bad. best regards Nils

January 21, 2016, 09:47		#7
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	what turbulence model?

January 21, 2016, 10:32		#8
Deranda Member Nils Join Date: Nov 2015 Posts: 59 Rep Power: 11	k-w menter sst with an all y+ approach time step: 0.0002s (Implicit Unsteady) Inner Iterations: 10

January 21, 2016, 10:47		#9
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	My first thought is that you may need more inner iterations. I usually can't get by with only 10. How are you initializing your model? The first few time steps often require more attention. I would also try to refine your mesh a little. For all wall y+ in k-w SST you should really stay below 1 or between 30 and 60. It sounds like you are in the the interpolated region for a lot of the wall boundary cells. If you have low Reynolds number then I would just stick with low wall treatment and target <1. https://steve.cd-adapco.com/articles/en_US/FAQ/RD-5-273

January 21, 2016, 10:50		#10
Deranda Member Nils Join Date: Nov 2015 Posts: 59 Rep Power: 11	My y+ Values stay 99.1971% below 1 und 100% below 5, after simulating 0.15 seconds

January 21, 2016, 11:24		#11
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	Ok, well it doesn't sound like y+ is your issue. Unfortunately, convergence problems are the hardest to resolve. There are so many different factors that can contribute to this that it can start to feel like you are chasing your tail a bit. Possibility 1: Your boundary conditions are inappropriate. Either they are too close, of the wrong type, or something to that effect. I run into issues quite often doing helicopter hover cases. When this happens I run the mesh diagnostic plugin. This allows you to see not only mesh quality, skewness, aspect ratio, etc... But also will show cells with high residuals. If you run this and find that high residuals are near boundary conditions, then you can usually just ignore the convergence plot (assuming you cannot resolve the issue by correcting boundary conditions in the first place). If you find them more in the middle of your domain near your geometry of interest then you may need to define a mesh refinement scheme or volumetric control, something to further subdivide the high residual cells. http://macrohut.cd-adapco.com/phpBB3...ilit=mesh#p776 Possiblilty 2: Your time step is actually too small. In this case your round off errors can become large and the resulting noise makes convergence problematic. Possibility 3: Your inner iterations need to be increased. I know you are trying to avoid this, but it is what it is. I would at least trying increasing to see what happens. Best of luck!

January 22, 2016, 03:44		#12
Deranda Member Nils Join Date: Nov 2015 Posts: 59 Rep Power: 11	I didnt found the area with Cells residuals in the diagnostics, but I displayed my skewness angle. In my volume of interest, 0,00107771% of the cells have a skewness angle above 85° and 0,00246333% falls below the VolumeChange kriteria. In the Outlet-Area, which is not of high interest and separated by a poroues media, are more skewness angles above 85° at the boundaries. maybe that causes the sdr resdiuals... The last night, I simulated again with 30 inner iterations and its almost the same result regarding the residuals, so that should not be the problem... The next step will be correcting the skewness angles, maybe I'll find a way, dont know yet^^ I will tell you my results