Mesh dependency - possible that refinement = worse results?

crisgc · November 14, 2012, 07:46

Hi everyone,

I am simulating a simple jet breakup, laminar flow, issuing from a capillary nozzle applying an oscillatory boundary condition to achieve breakup. The study is 2D axisymmetric. It seems that I obtain good results with a fine mesh (aprox. 40000 elements) when I compare them with experimental cases but if I refine the grid even more (100000 elements or further) the results are worse and I cannot even produce the breakup and the interface seems quite different to what it should look like (tiny ripples appear).

Any suggestion? Is it possible that refining the mesh led to worse results?

Thank you for your help!

RodriguezFatz · November 14, 2012, 08:35

Yes, this is possible. You must generally be careful if your result changes when you refine the grid. This is a problem of "solving the equations right".
When your results - compared to experiments - become worse due to refinement, this can be a problem of "solving the right equations". In your case, it is possible that your results are "accidentally correct". The numerical error cancels the model error. This does not mean, that the results are good. It is just some kind of random incident.

crisgc · November 14, 2012, 08:43

Wow! Curious!

Thank you RodriguezFatz for your reply! One question, what do you mean by "the numerical error cancels the model error"? That I am using the wrong solver / algorithms but the error associated to the numerical method used makes the solution to seem right?

What makes me wonder (and become a bit nervious) about how to solve it! I am using a VOF (usually + Level Set) and I wouldn't know what other model to use for that kind of problem. Could you recommend me anything?

Thank you once more!

RodriguezFatz · November 14, 2012, 08:53

More or less: I meant, that a bad grid (as a source of numerical error) can cancel the error that the model (as a source of error compared to reality) produces. Then, the solution seems to be right, but isn't.

Can you afford even more refinement? If it is possible, you should try at least two stages of refinement of the grid (allways!). Now have a look at how your relevant values change over the three grids. If you are allready in a convergent region of the grid, all relevant values should change monotonic (over the refinement). The difference between the coarse and medium grid has to be larger than the difference between the medium and fine grid.
Do you have a copy of http://www.cfd-online.com/Books/show_book.php?book_id=4
?
Everything is explained in there!

crisgc · November 14, 2012, 12:16

Hmm.. OK, I will try to check again with three grids at different refinement stages as you've recommended me and let's see what happens.

I will also try to obtain a copy of the book.

Thank you for your help!

November 14, 2012, 07:46	Mesh dependency - possible that refinement = worse results?	#1
crisgc New Member Join Date: Nov 2011 Posts: 3 Rep Power: 14	Hi everyone, I am simulating a simple jet breakup, laminar flow, issuing from a capillary nozzle applying an oscillatory boundary condition to achieve breakup. The study is 2D axisymmetric. It seems that I obtain good results with a fine mesh (aprox. 40000 elements) when I compare them with experimental cases but if I refine the grid even more (100000 elements or further) the results are worse and I cannot even produce the breakup and the interface seems quite different to what it should look like (tiny ripples appear). Any suggestion? Is it possible that refining the mesh led to worse results? Thank you for your help!

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[snappyHexMesh] How to Do External Mesh for Airfoil sHM	msuaeronautics	OpenFOAM Meshing & Mesh Conversion	1	September 23, 2012 04:00
[snappyHexMesh] Layers:problem with curvature	giulio.topazio	OpenFOAM Meshing & Mesh Conversion	10	August 22, 2012 09:03
[snappyHexMesh] external flow with snappyHexMesh	chelvistero	OpenFOAM Meshing & Mesh Conversion	11	January 15, 2010 19:43
Convergence moving mesh	lr103476	OpenFOAM Running, Solving & CFD	30	November 19, 2007 14:09
basic of mesh refinement	arya	CFX	4	June 19, 2007 12:21

November 14, 2012, 08:35		#2
RodriguezFatz Senior Member Philipp Join Date: Jun 2011 Location: Germany Posts: 1,297 Rep Power: 26	Yes, this is possible. You must generally be careful if your result changes when you refine the grid. This is a problem of "solving the equations right". When your results - compared to experiments - become worse due to refinement, this can be a problem of "solving the right equations". In your case, it is possible that your results are "accidentally correct". The numerical error cancels the model error. This does not mean, that the results are good. It is just some kind of random incident.

November 14, 2012, 08:43		#3
crisgc New Member Join Date: Nov 2011 Posts: 3 Rep Power: 14	Wow! Curious! Thank you RodriguezFatz for your reply! One question, what do you mean by "the numerical error cancels the model error"? That I am using the wrong solver / algorithms but the error associated to the numerical method used makes the solution to seem right? What makes me wonder (and become a bit nervious) about how to solve it! I am using a VOF (usually + Level Set) and I wouldn't know what other model to use for that kind of problem. Could you recommend me anything? Thank you once more!

November 14, 2012, 08:53		#4
RodriguezFatz Senior Member Philipp Join Date: Jun 2011 Location: Germany Posts: 1,297 Rep Power: 26	More or less: I meant, that a bad grid (as a source of numerical error) can cancel the error that the model (as a source of error compared to reality) produces. Then, the solution seems to be right, but isn't. Can you afford even more refinement? If it is possible, you should try at least two stages of refinement of the grid (allways!). Now have a look at how your relevant values change over the three grids. If you are allready in a convergent region of the grid, all relevant values should change monotonic (over the refinement). The difference between the coarse and medium grid has to be larger than the difference between the medium and fine grid. Do you have a copy of http://www.cfd-online.com/Books/show_book.php?book_id=4 ? Everything is explained in there!

November 14, 2012, 12:16		#5
crisgc New Member Join Date: Nov 2011 Posts: 3 Rep Power: 14	Hmm.. OK, I will try to check again with three grids at different refinement stages as you've recommended me and let's see what happens. I will also try to obtain a copy of the book. Thank you for your help!