Verifying results

For my thesis project I am optimizing a folded steel plate structure in Grasshopper. For the structural calculations I use Karamba. Everything works fine and I get nice results, but when i verify these results with Diamonds (by Buildsoft) I get some different results. Mostly the stress or displacement in my structure look the same, but the values aren't really alike. I checked and double checked to have the same input values and some logical values like total weight are the same in both programs. I expected some difference in the results because Karamba probably simplifies in a slightly other way than Diamonds. But the results aren't linear, sometimes the stress in Karamba is bigger and for a slightly different model it is smaller than that of Diamonds.
Is there anyone with the same problem, of someone that knows how I can solve this problem?

Kind regards

Reinout

Replies to This Discussion

Permalink Reply by Karamba3D on December 19, 2017 at 1:59pm

Hi Reinout,

I think the problem has to do with the fact that you use pointwise supports for the structure. These act like infinitely sharp needles: When refining the mesh the results of the structure do not converge (see attached definition) because the needle like supports are punching through the structure.

In order to solve this you could integrate small pieces of beams in the shell and attach the supports there.

Best,

Clemens

Attachments:

Testfolds1205d_cp.gh, 151 KB

Permalink Reply by Reinout van Arnhem on January 7, 2018 at 1:16pm

I can understand that a pointwise support can lead too some strange results, but in my comparison in Diamonds I also used pointwise supports. Without the reinforcements you suggested, so it's strange that the programs give different results, no? I'm also trying some simpler geometry and for example with simple plates the results in Karamba and in Diamonds are (with some margins) alike. But the moment I make a tent-structure, Karamba underestimates the maximum stress. I have added some pictures and the model I used.

Kind regards

Attachments:

Stress on steel tent in Diamonds.PNG, 216 KB
Stress on steel tent in Karamba.PNG, 113 KB
Benchmark tent.gh, 41 KB

Permalink Reply by Karamba3D on January 9, 2018 at 1:36pm

Hi Reinout,

the difference in the results is due to the fact that the two programs probably use different types of shell elements. The shell elements in Karamba are based on the TRIC-element of Argyris.

In the example 'Benchmark tent.gh' you use point-loads. These cause similar problems as point-wise supports.

When you increase the mesh density the results in Karamba will change. At a value of 0.02m for the mesh resulution in the MeshBreps-component I get a maximum VanMises-stress of 31.4 [kN/cm2].

Best,

Clemens

Permalink Reply by Reinout van Arnhem on January 18, 2018 at 12:21pm

Back to the pointwise supports. So you say that the value of stress and should normally converge, but they don't because the pointloads and supports pinch through the structure. I now tried to get an insight in what you were saying. So the smaller the mesh resolution has been set the more exact the calculation is I suppose? I therefore compared some simple structures with different loads and plotted their stress and displacement corresponding to the mesh resolution. I made some graphs of this and compared them with my other calculations program. In most cases, the results are alike or are converging to about the same value, not in all though. If I would have to make a conclusion of this it would be that I have to go as small as possible for the mesh resolution, but even then the results aren't quite alike for, especially when I usen an equaly distributed load (or mesh load). I have added the files.
I know it's a lot of questions. Most important for me is if there's a way to know at which value the results should converge and if there's a way that i'm certain that my results are usable in reality. Would it be for example legally acceptable to only do calculations with Karamba?

Sorry for the trouble

Reinout

Attachments:

Permalink Reply by Karamba3D on January 19, 2018 at 1:22pm

Dear Reinout,

the convergence of the results not only depends on the number of elements but also on their shape and distribution: equilateral triangles perform best, a denser mesh is needed in places where there are steeper stress gradients.

The necessary mesh density depends on the accuracy one wants to achieve. For everyday buildings the model input (e.g. loads, material parameters,...) is random to some degree. Therefore it makes no sense to be too accurate in most cases.

When opening the 'doxc'-file with OpenOffice it was empty.

One cannot predict apriori an adequate mesh density for a given problem. It is therefore necessary to try out several variants.

Regarding the legal acceptance of a calculation with Karamba: this depends on whether you are a licensed engineer. Since a Finite Element program like Karamba is a complex matter any liability sticks with the user (see license agreement). You will not find a Finite Element program where this is not the case.

Best,

Clemens

RSS