Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Different results of Simulation from Solid Mechanics and shell

Please login with a confirmed email address before reporting spam

Hello, I have a problem when doing a simulation with a bend pipe meshed by shell element. The results of principal strains are one time larger than meshed by solid mechanics. But the results of principal stresses from both shell and solid models are the same.

The parameter of the bend pipe, Position for analysis and line graphs for principal strains are uploaded as Attachment. In my Simulation, the pipe is loaded by 300N downwards on cross section A (see photos), cross section B is fixed. The material is structural steel. The position for plotting graph is an outside circle on bend pipe (position C in photo). There are two photos show the principal strains from shell element and solid mechanics. That's obvious to observe the big difference between two graphs.The results of principal strains by shell element are one time larger than meshed by solid mechanics.

And I have also simulated with the shell and solid models in software Autodesk Simulation, the results of stains and stresses are the same as the one from solid model in Comsol . So it means the result of solid model in Comsol is correct,the problem now is just the inaccuracy of principal strain by shell model in Comsol. Is it a mistake from setting and operation by shell element, or maybe a bug in Comsol?

So I ask for your help here and look forward to a suggestion from some of you!

Thank you,
Alick



6 Replies Last Post Aug 10, 2013, 4:53 a.m. EDT

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 11, 2013, 12:00 p.m. EDT
Hi Alick,

I simulated your problem with shell interface, and got same results as yours. I modeled a thin-walled tube before with wall thickness 1/20 of the tube radius, and my results are identical to some published data. Your tube is relatively thick, therefore thin shell theory may not be very accurate. It is interesting that autodesk simulation gives more accurate results. I am guessing they are using a different formulation for the thin shell theory equations that also support thick wall structure.

You can try reduce the wall thickness see if COMSOL can give you good results and find out the critical ratio between wall thickness and local curvature (for tubes, it is the radius), and avoid using the shell interface when the thickness to local curvature ratio is larger than the critical value.

Let me know when you have new information or figure out the problem.

Thanks,

Shuping
Hi Alick, I simulated your problem with shell interface, and got same results as yours. I modeled a thin-walled tube before with wall thickness 1/20 of the tube radius, and my results are identical to some published data. Your tube is relatively thick, therefore thin shell theory may not be very accurate. It is interesting that autodesk simulation gives more accurate results. I am guessing they are using a different formulation for the thin shell theory equations that also support thick wall structure. You can try reduce the wall thickness see if COMSOL can give you good results and find out the critical ratio between wall thickness and local curvature (for tubes, it is the radius), and avoid using the shell interface when the thickness to local curvature ratio is larger than the critical value. Let me know when you have new information or figure out the problem. Thanks, Shuping

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 15, 2013, 9:29 a.m. EDT
Hi Shuping,

Thank you for your reply, I have simulated by two models with different wall thickness (as shown in the table), but unfortunately it seems that, although the ratio between wall thickness and radius of the tube is already smaller than 1/20, but the inaccuracy is still ridiculously large. Will you get the same result as me?

Wall thickness[mm]..........Wall thickness/.........Max. principal strain.........Max. principal strain.........Inaccuracy
........................................radius of the tube..........by Shell [μm/m]...............by Solid [μm/m]
............4.2...............................1/6.7...............................232................................108...........................+114.8%
............2..................................1/14................................913................................482...........................+89.4%
............1,4...............................1/20................................1700...............................938...........................+81,2%
............1..................................1/28...............................3100...............................1550..........................+100%
............0.5...............................1/56...............................10400.............................5100..........................+103.9%

With respect!
Hi Shuping, Thank you for your reply, I have simulated by two models with different wall thickness (as shown in the table), but unfortunately it seems that, although the ratio between wall thickness and radius of the tube is already smaller than 1/20, but the inaccuracy is still ridiculously large. Will you get the same result as me? Wall thickness[mm]..........Wall thickness/.........Max. principal strain.........Max. principal strain.........Inaccuracy ........................................radius of the tube..........by Shell [μm/m]...............by Solid [μm/m] ............4.2...............................1/6.7...............................232................................108...........................+114.8% ............2..................................1/14................................913................................482...........................+89.4% ............1,4...............................1/20................................1700...............................938...........................+81,2% ............1..................................1/28...............................3100...............................1550..........................+100% ............0.5...............................1/56...............................10400.............................5100..........................+103.9% With respect!

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 16, 2013, 12:30 a.m. EDT
Hi Alick,

Regarding this problem, my understanding is that the thin-shell theory essentially assumes that the strain in the transverse direction is zero. You can check the magnitude of the transverse strain from the results of 3D solid element relative to the in-plane strain, and see if neglecting it is a reasonable assumption. If the transverse strain is large, then the discrepancy of the results from two approaches are due to the limitation of thin shell theory itself. Let me know what you find.

I looked at the documents for autodesk simulation about their shell element:

wikihelp.autodesk.com/Simulation_Mechanical/enu/2014/Help/0328-User_s_G328/0572-Setting_572/0607-Analysis607/0656-Nonlinea656/0657-Element_657/0672-Planar_E672/0678-Shell_El678

I don't see too much difference, although I am not an expect in FEA nor in thin-shell theory. I think both software use a relaxed thin-shell theory that allows the rotation of local normal direction on the reference surface at original state while undergoing deformation. In other words, this type of elements should also work for relative thick shells.

An expert in solid mechanics should easily give us a clear answer. I am more of a thermofluids guy, so the above is almost all I know about shells.

Thanks,


Shuping
Hi Alick, Regarding this problem, my understanding is that the thin-shell theory essentially assumes that the strain in the transverse direction is zero. You can check the magnitude of the transverse strain from the results of 3D solid element relative to the in-plane strain, and see if neglecting it is a reasonable assumption. If the transverse strain is large, then the discrepancy of the results from two approaches are due to the limitation of thin shell theory itself. Let me know what you find. I looked at the documents for autodesk simulation about their shell element: http://wikihelp.autodesk.com/Simulation_Mechanical/enu/2014/Help/0328-User_s_G328/0572-Setting_572/0607-Analysis607/0656-Nonlinea656/0657-Element_657/0672-Planar_E672/0678-Shell_El678 I don't see too much difference, although I am not an expect in FEA nor in thin-shell theory. I think both software use a relaxed thin-shell theory that allows the rotation of local normal direction on the reference surface at original state while undergoing deformation. In other words, this type of elements should also work for relative thick shells. An expert in solid mechanics should easily give us a clear answer. I am more of a thermofluids guy, so the above is almost all I know about shells. Thanks, Shuping

Henrik Sönnerlind COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 30, 2013, 4:03 p.m. EDT
Hi Alick,

Thanks for reporting this bug. This will be fixed for the next version scheduled to be released in October 2013.

It is a problem in a postprocessing variable, and does not affect any other aspects of the shell results than the bending contribution to the total strain. The underlying results (the membrane, bending and shear strains) are not affected.

Regards,
Henrik
Hi Alick, Thanks for reporting this bug. This will be fixed for the next version scheduled to be released in October 2013. It is a problem in a postprocessing variable, and does not affect any other aspects of the shell results than the bending contribution to the total strain. The underlying results (the membrane, bending and shear strains) are not affected. Regards, Henrik

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Aug 10, 2013, 4:46 a.m. EDT
It is clear now, thanks for your reply, Henrik!
It is clear now, thanks for your reply, Henrik!

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Aug 10, 2013, 4:53 a.m. EDT
Hello Shuping, in the end we get the answer, thank you for your effort.

Regards,
Alick
Hello Shuping, in the end we get the answer, thank you for your effort. Regards, Alick

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.