In another post (https://forum.freecadweb.org/viewtopic. ... &start=100) I raised a concern about the accuracy of the results obtained with the CalculiX 3-node beam element. Here some further notes and considerations.
I analysed the problem of a simply supported beam under uniform load and found that for a coarse mesh (1 three-noded element) the mid-span displacement is 20% lower than that obtained with classical beam theory (i.e. 59.5393 mm versus a theoretical value of 74.4048 mm):
It is only with a very fine mesh that a good agreement with theory is achieved (73.7716 mm, which is 0.9% below the theoretical value).
My initial suspicion was that the uniform load was inconsistently applied and upon close inspection I found that there is indeed a difference between the nodal load values exported to INP file and what is obtained by integrating the Hermitian shape functions for a three-node beam (e.g. https://pdfs.semanticscholar.org/12be/d ... 87f271.pdf):
However, manual correction of the *CLOAD card did not give any improvement of the results.
To demonstrate that a normal 3-node Hermitian element (i.e. with displacement and rotational degrees of freedom) should give very accurate results, I generated the stiffness matrix for a single element and multiplied its inverse with the above consistent load vector:
This shows that a single three-noded element can reproduce the theoretical value for the maximum displacement of a simply supported beam under uniform load to within 0.0000512%.
It is therefore worth further investigating why the CCX beam gives such large errors for this simple case.
CalculiX 3-node Beam Element
Moderator: bernd
Forum rules
and Helpful information for the FEM forum
and Helpful information for the FEM forum
CalculiX 3-node Beam Element
Last edited by HarryvL on Sat Apr 07, 2018 3:08 pm, edited 1 time in total.
Re: CalculiX 3-node Beam Element
check this: https://github.com/FreeCAD/FreeCAD/blob ... #L820-L863
should we change them to your values? I do not know any of such FEM back gound in detail. I just used the values from the book mentioned in the source code.
bernd
should we change them to your values? I do not know any of such FEM back gound in detail. I just used the values from the book mentioned in the source code.
bernd
Re: CalculiX 3-node Beam Element
ahh and in the other topic where I copied all the messages from guido and martin I should add some more. They had another conversation on the mailing list in the regard of beam examples.
BTW: Might be worth for you to sign ccx mailinglist. There is only one or two posts each day, but the developer guido and one of the advanved users martin kraska reads and writes regulary over there.
BTW Do you know martin kraskas ccx examples on github? To recode them in python in FreeCAD has been on my todo for years, but there has been any time some more impotant dev to do
BTW: Might be worth for you to sign ccx mailinglist. There is only one or two posts each day, but the developer guido and one of the advanved users martin kraska reads and writes regulary over there.
BTW Do you know martin kraskas ccx examples on github? To recode them in python in FreeCAD has been on my todo for years, but there has been any time some more impotant dev to do
Re: CalculiX 3-node Beam Element
No need to change them Bernd. The shape functions I used are for a classical 3-node Hermitian beam but in CCX this is translated into a volume element. So the manual change I made to *CLOAD had no effect. Also, I now think I know why the CCX 3-Node element behaves so poorly. It's probably shear locking. That's what you can get when you use volume elements for beams, plates and shells. The same element with reduced integration (B32R) should perform much better for this example. I will try when I am at my computer. However good results with reduced integration are not guaranteed for all situations. You can get funny zero energy patterns (called hourglassing).
Finally, the traditional beam element with 3 nodes has 6 degrees of freedom and can therefore represent a 5th degree polynomial displacement exactly. This implies that up to linear loads can be calculated without theoretical error. The small error I got was due to numerical rounding.
So what do I learn from all of this? 1) The classical beam element is still by far the best for elastic analysis. 2) However, for plasticity you need an element of CCX type. 3) Use reduced integration, but always check for hourglassing.
Finally, the traditional beam element with 3 nodes has 6 degrees of freedom and can therefore represent a 5th degree polynomial displacement exactly. This implies that up to linear loads can be calculated without theoretical error. The small error I got was due to numerical rounding.
So what do I learn from all of this? 1) The classical beam element is still by far the best for elastic analysis. 2) However, for plasticity you need an element of CCX type. 3) Use reduced integration, but always check for hourglassing.
Re: CalculiX 3-node Beam Element
I know the examples. They are really cool and show the strength of Calculix.
They show the workflow for many practical analyses and therefore give inspiration on where FEM workbench could go in the future ?! I would be keen to help map out those multi-step work flows for a few typical cases ... for example how to automate what I did by hand for the knee joint example. The whole design workflow for a practical structural engineering problem, so to say.
Re: CalculiX 3-node Beam Element
Well, I tried and it's not shear locking. The results for the B32R element are exactly the same - 20% in error. I read CCX 2.13 manual and see that similar examples/issues are discussed there. The manual says displacements are quadratic (2nd order polynomial), which implies that displacement and rotations are both interpolated independently to 2nd order, unlike in the classical beam element.HarryvL wrote: ↑Sat Apr 07, 2018 7:59 am ... Also, I now think I know why the CCX 3-Node element behaves so poorly. It's probably shear locking. That's what you can get when you use volume elements for beams, plates and shells. The same element with reduced integration (B32R) should perform much better for this example.
To verify that the order of interpolation leads to the 20% error, I derived the stiffness matrix for a quadratic interpolation of displacements, inverted it and multiplied this with the consistent load vector for quadratic interpolation:
The displacement thus obtained is indeed very close to the CCX result.
Conclusion: to obtain accurate results with the CCX beam element you need to use fine meshes
Re: CalculiX 3-node Beam Element
Note: I corrected an error in figures "Spreadsheet4.png" and "Spreadsheet5.png"