Urgent; convergence problem in MRF simulation

Mansureh · January 31, 2011, 00:31

Hi everybody
I'm trying to simulate a Multiple Reference Frame (MRF) case with Fluent 12. rotational speed is about 0-1600 rpm and I should simulate the case in every 200 rpm, I mean 0 , 200, 400, ... , 1600 rpm. At low rotational speed It converges but with gradual increase of rotational speed convergence become more difficult.
For judgment about convergence I monitor total pressure at inlet and outlet boundaries, because my gold is earning total pressure drop. And now, in 800rpm, residuals had been decreased but total pressures are oscillating in a wide range, I mean 20 Pa for inlet boundary , and about 5 Pa for outlet boundary. I dont know what the reason should be.

physical models and mesh features are:
relative velocity formulation, RNG k-epsilon model, enhanced wall treatment, 2700000 structured hexa elements , SIMPLE algorithm, PRESTO, first order upwind for all equations, velocity inlet and pressure outlet BCs.
fluid is air at 25 C.
By the way, the simulation has been done via double precision and parallel processing.

I think maybe problem is due to Algebraic Multigrid methods and their parameters. Can any body tell me what the problem is?
what's your idea?

thanks for any help

scott · February 1, 2011, 21:38

Hi Mansureh,

Have you tried to view the results and see if there is any obvious reason for this problem? Maybe the mesh is not refined enough around the mesh interface and you are getting a discontinuity across it? Its hard to say without seeing the mesh and results. I would expect it to get harder to converge as the speed goes up if this was the case.

Also, you could try Second Order discretisation schemes, they tend to be more accurate.

My only other comment is, make a more refined mesh and test this. You may find that it is just a function of grid resolution.

Regards,

Scott

Mansureh · February 2, 2011, 06:46

Hi Scott

thanks for reply.
I think there's no promlem with mesh, because the mesh is fine enough. It consists of 2,700,000 hexahedral element such that more than 98% of them have skewness below 0.1. I think the reason is something other than this, but I can't find it out.

regards
Mansureh

scott · February 2, 2011, 06:57

Hi Mansureh,

Ok, I guess it depends on the physical size of the geometry, I regularly use 30 million cells, and the minimum I will generally use is around 4 million, but then I have a relatively large domain.

Maybe try reducing the pseudo timestep size or modify the relaxation factors.

Regards,

Scott

scott · February 2, 2011, 07:00

Hi Mansureh,

Also definitely try second order if you think the mesh is fine enough, generally it should run very well in first order though, so I doubt this would help convergence, generally I find cell count the best method, especially as gradients start to increase as would be the case as you increase rpm.

Regards,

Scott

January 31, 2011, 00:31	Urgent; convergence problem in MRF simulation	#1
Mansureh New Member Join Date: Jul 2010 Posts: 22 Rep Power: 16	Hi everybody I'm trying to simulate a Multiple Reference Frame (MRF) case with Fluent 12. rotational speed is about 0-1600 rpm and I should simulate the case in every 200 rpm, I mean 0 , 200, 400, ... , 1600 rpm. At low rotational speed It converges but with gradual increase of rotational speed convergence become more difficult. For judgment about convergence I monitor total pressure at inlet and outlet boundaries, because my gold is earning total pressure drop. And now, in 800rpm, residuals had been decreased but total pressures are oscillating in a wide range, I mean 20 Pa for inlet boundary , and about 5 Pa for outlet boundary. I dont know what the reason should be. physical models and mesh features are: relative velocity formulation, RNG k-epsilon model, enhanced wall treatment, 2700000 structured hexa elements , SIMPLE algorithm, PRESTO, first order upwind for all equations, velocity inlet and pressure outlet BCs. fluid is air at 25 C. By the way, the simulation has been done via double precision and parallel processing. I think maybe problem is due to Algebraic Multigrid methods and their parameters. Can any body tell me what the problem is? what's your idea? thanks for any help Last edited by Mansureh; January 31, 2011 at 01:23.

February 2, 2011, 06:46	thanks for reply	#3
Mansureh New Member Join Date: Jul 2010 Posts: 22 Rep Power: 16	Hi Scott thanks for reply. I think there's no promlem with mesh, because the mesh is fine enough. It consists of 2,700,000 hexahedral element such that more than 98% of them have skewness below 0.1. I think the reason is something other than this, but I can't find it out. regards Mansureh

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
convergence problem	commonyue	Main CFD Forum	1	December 1, 2009 04:54
Submerged fin, Convergence problem	supermouniette	FLUENT	10	July 6, 2009 11:47
Convergence of CFX field in FSI analysis	nasdak	CFX	2	June 29, 2009 02:17
multiphase mixing Problem with MRF model in MixSim	Srinivas	FLUENT	0	October 17, 2005 07:35
convergence problem with SIMPLER	NURAY KAYAKOL	Main CFD Forum	1	February 24, 1999 14:43

February 1, 2011, 21:38		#2
scott Member Scott Join Date: Sep 2009 Posts: 44 Rep Power: 17	Hi Mansureh, Have you tried to view the results and see if there is any obvious reason for this problem? Maybe the mesh is not refined enough around the mesh interface and you are getting a discontinuity across it? Its hard to say without seeing the mesh and results. I would expect it to get harder to converge as the speed goes up if this was the case. Also, you could try Second Order discretisation schemes, they tend to be more accurate. My only other comment is, make a more refined mesh and test this. You may find that it is just a function of grid resolution. Regards, Scott

February 2, 2011, 06:57		#4
scott Member Scott Join Date: Sep 2009 Posts: 44 Rep Power: 17	Hi Mansureh, Ok, I guess it depends on the physical size of the geometry, I regularly use 30 million cells, and the minimum I will generally use is around 4 million, but then I have a relatively large domain. Maybe try reducing the pseudo timestep size or modify the relaxation factors. Regards, Scott

February 2, 2011, 07:00		#5
scott Member Scott Join Date: Sep 2009 Posts: 44 Rep Power: 17	Hi Mansureh, Also definitely try second order if you think the mesh is fine enough, generally it should run very well in first order though, so I doubt this would help convergence, generally I find cell count the best method, especially as gradients start to increase as would be the case as you increase rpm. Regards, Scott