Differences between CFX and OpenFOAM regarding convergence and robustness!

magjohan · February 26, 2015, 02:56

PLEASE FOCUS ON DIFFERENCES BETWEEN OPENFOAM AND CFX HERE!

I need help to quantify and specify the differences between the commercial software (mainly CFX and Fluent) and OpenFoam. I have experience of both “sides” but have seen benefits by using the commercial software compared to OpenFoam even if I prefer the foamy-side.

EX:
Let’s take one case that I runned yesterday!

- It’s a 4.5M cells case, mesh produced in Ansys DM, this is not enough for this case but that’s my PCs limitations for the moment.
- Same mesh are being used for OF and for CFX
- Same input parameters are being set regarding boundary conditions.
- SST model in CFX and kwSST in OpenFoam

Differences, due to low yPlus at wall my experience says that no “omegaWallFunction” or “kqRWallFunction”, zero gradient are being used.

I run checkMesh and see some issues of “Number of severely non-orthogonal face”, CFX doesn’t complain at all.

When I run the case, the OpenFoam case never reaches lower residuals than 1e-2 where the CFX chase goes down below 1e-3 (MAX residuals), RMS residuals are of course much lower. CFX "converges" after 300 iterations, OpenFoam never converges with this mesh.

What are casing CFX to be so stable compared to OpenFoam, I think cases like this forces people to use commercial software rather than for ex OpenFoam even if it’s against peoples will.

My thoughts are that CFX are using more stabilizing functions for some reasons especially taking care of the near-wall-treatment in a much more robust way.

Keep the discussion going, help me find my way back to OpenFoam here!

Andrew_Sm · February 26, 2015, 03:41

Quote:

Originally Posted by magjohan

PLEASE FOCUS ON DIFFERENCES BETWEEN OPENFOAM AND CFX HERE!

I need help to quantify and specify the differences between the commercial software (mainly CFX and Fluent) and OpenFoam. I have experience of both “sides” but have seen benefits by using the commercial software compared to OpenFoam even if I prefer the foamy-side.

EX:
Let’s take one case that I runned yesterday!

- It’s a 4.5M cells case, mesh produced in Ansys DM, this is not enough for this case but that’s my PCs limitations for the moment.
- Same mesh are being used for OF and for CFX
- Same input parameters are being set regarding boundary conditions.
- SST model in CFX and kwSST in OpenFoam

Differences, due to low yPlus at wall my experience says that no “omegaWallFunction” or “kqRWallFunction”, zero gradient are being used.

I run checkMesh and see some issues of “Number of severely non-orthogonal face”, CFX doesn’t complain at all.

When I run the case, the OpenFoam case never reaches lower residuals than 1e-2 where the CFX chase goes down below 1e-3 (MAX residuals), RMS residuals are of course much lower. CFX "converges" after 300 iterations, OpenFoam never converges with this mesh.

What are casing CFX to be so stable compared to OpenFoam, I think cases like this forces people to use commercial software rather than for ex OpenFoam even if it’s against peoples will.

My thoughts are that CFX are using more stabilizing functions for some reasons especially taking care of the near-wall-treatment in a much more robust way.

Keep the discussion going, help me find my way back to OpenFoam here!

Without proof you cannot question the credibility of these software. Do you think you are only smart out here? lol

Can you prove this by uploading your case for both Cfx and OF?
OR any good journal article you have to prove your thoughts?

You sound more like a frustrated OF user

I can't just throw a piece of stone and say anything

I guess you got my point

magjohan · February 26, 2015, 06:17

Quote:

Originally Posted by Andrew_Sm

Without proof you cannot question the credibility of these software. Do you think you are only smart out here? lol

Can you prove this by uploading your case for both Cfx and OF?
OR any good journal article you have to prove your thoughts?

You sound more like a frustrated OF user

I can't just throw a piece of stone and say anything

I guess you got my point

Thx for the reply,

This is for the moment more on my on desk with my own comparisons. I have been using OpenFoam for industrial cases before with great success.

I have also seen a couple of comparisons, thesis works and papers, between OF and commercial software without any differences in results so I’m not questioning the strength of OF.

Put it in this way, I know that the mesh isn’t fine enough, residuals does not go down to a proper level, I’ usually aiming for res_max below 1e-4. Does CFX handling this mesh better than OF, if so, why?

I will try to upload residual data later when I’m back at the office!

dkxls · February 26, 2015, 06:32

As Andrew said, there is not much point in discussing this any further without the cases setup and results!

However, a general note concerning the robustness of numerical schemes:
The more "accurate" your scheme is the more you get into trouble concerning it's "robustness". Meaning, first order schemes are generally very robust and will give you some results in most cases, however the question immediately arises on how accurate these results are!

To obtain accurate results, higher order methods are necessary. However, higher order schemes (and even already 2nd order centered schemes) tend to produce wiggles/oscillation and become unstable. Hence, there is a vast amount of stabilization techniques and a good starting point for further reading are the TVD schemes (please consult any CFD text book, e.g. Ferziger or Hirsch, for more details).

So, when comparing CFD codes, one has to specify the numerical methods in detail! There is no point to any further discussion otherwise.
Ideally one would compare CFD codes for canonical cases and validated them against experiments or accepted DNS results (examples here would be channel flows or turbulent boundary layers).

-Armin

http://en.wikipedia.org/wiki/Total_v...on_diminishing

Ferziger, Joel H., and Milovan Peric. Computational Methods for Fluid Dynamics. 3rd ed., 2002. http://www.springer.com/materials/me...-3-540-42074-3.

Hirsch, Charles. Numerical Computation of Internal and External Flows: The Fundamentals of Computational Fluid Dynamics. 2nd ed. Oxford; Burlington, MA: Butterworth-Heinemann, 2007.

RodriguezFatz · February 26, 2015, 10:26

Johan, I can confirm your experience about robustness of Ansys (in my case Fluent) compared to openFoam if it comes to getting a quick solution. Each time I just want to have a quick look at something I get convergence with Fluent's standard settings, whereas openFoam often diverges. I know, openFoam doesn't have any "standard settings", but without any additional effort, I don't get even simple cases to convergence.

But what does that mean? Of course this is a reason for "simple" engineers to go for commercial software. But what does that mean for us? Is it a good thing or a bad thing for Fluent / CFX if they converge each time even with pretty nonsense input? By using openFoam I learned a lot about convergence, numerics, solvers, ... and I also think I got an idea of what commercial software does behind the curtain...

February 26, 2015, 02:56	Differences between CFX and OpenFOAM regarding convergence and robustness!	#1
magjohan New Member Johan Magnusson Join Date: Oct 2010 Posts: 21 Rep Power: 15	PLEASE FOCUS ON DIFFERENCES BETWEEN OPENFOAM AND CFX HERE! I need help to quantify and specify the differences between the commercial software (mainly CFX and Fluent) and OpenFoam. I have experience of both “sides” but have seen benefits by using the commercial software compared to OpenFoam even if I prefer the foamy-side. EX: Let’s take one case that I runned yesterday! - It’s a 4.5M cells case, mesh produced in Ansys DM, this is not enough for this case but that’s my PCs limitations for the moment. - Same mesh are being used for OF and for CFX - Same input parameters are being set regarding boundary conditions. - SST model in CFX and kwSST in OpenFoam Differences, due to low yPlus at wall my experience says that no “omegaWallFunction” or “kqRWallFunction”, zero gradient are being used. I run checkMesh and see some issues of “Number of severely non-orthogonal face”, CFX doesn’t complain at all. When I run the case, the OpenFoam case never reaches lower residuals than 1e-2 where the CFX chase goes down below 1e-3 (MAX residuals), RMS residuals are of course much lower. CFX "converges" after 300 iterations, OpenFoam never converges with this mesh. What are casing CFX to be so stable compared to OpenFoam, I think cases like this forces people to use commercial software rather than for ex OpenFoam even if it’s against peoples will. My thoughts are that CFX are using more stabilizing functions for some reasons especially taking care of the near-wall-treatment in a much more robust way. Keep the discussion going, help me find my way back to OpenFoam here! lpz456 likes this.

February 26, 2015, 10:26		#5
RodriguezFatz Senior Member Philipp Join Date: Jun 2011 Location: Germany Posts: 1,297 Rep Power: 26	Johan, I can confirm your experience about robustness of Ansys (in my case Fluent) compared to openFoam if it comes to getting a quick solution. Each time I just want to have a quick look at something I get convergence with Fluent's standard settings, whereas openFoam often diverges. I know, openFoam doesn't have any "standard settings", but without any additional effort, I don't get even simple cases to convergence. But what does that mean? Of course this is a reason for "simple" engineers to go for commercial software. But what does that mean for us? Is it a good thing or a bad thing for Fluent / CFX if they converge each time even with pretty nonsense input? By using openFoam I learned a lot about convergence, numerics, solvers, ... and I also think I got an idea of what commercial software does behind the curtain... jherb, magjohan and lpz456 like this. __________________ The skeleton ran out of shampoo in the shower.

February 26, 2015, 06:32		#4
dkxls Senior Member Armin Join Date: Feb 2011 Location: Helsinki, Finland Posts: 156 Rep Power: 19	As Andrew said, there is not much point in discussing this any further without the cases setup and results! However, a general note concerning the robustness of numerical schemes: The more "accurate" your scheme is the more you get into trouble concerning it's "robustness". Meaning, first order schemes are generally very robust and will give you some results in most cases, however the question immediately arises on how accurate these results are! To obtain accurate results, higher order methods are necessary. However, higher order schemes (and even already 2nd order centered schemes) tend to produce wiggles/oscillation and become unstable. Hence, there is a vast amount of stabilization techniques and a good starting point for further reading are the TVD schemes (please consult any CFD text book, e.g. Ferziger or Hirsch, for more details). So, when comparing CFD codes, one has to specify the numerical methods in detail! There is no point to any further discussion otherwise. Ideally one would compare CFD codes for canonical cases and validated them against experiments or accepted DNS results (examples here would be channel flows or turbulent boundary layers). -Armin http://en.wikipedia.org/wiki/Total_v...on_diminishing Ferziger, Joel H., and Milovan Peric. Computational Methods for Fluid Dynamics. 3rd ed., 2002. http://www.springer.com/materials/me...-3-540-42074-3. Hirsch, Charles. Numerical Computation of Internal and External Flows: The Fundamentals of Computational Fluid Dynamics. 2nd ed. Oxford; Burlington, MA: Butterworth-Heinemann, 2007.