search

KratosMultiphysics / Kratos

Kratos Multiphysics (A.K.A Kratos) is a framework for building parallel multi-disciplinary simulation software. Modularity, extensibility and HPC are the main objectives. Kratos has BSD license and is written in C++ with extensive Python interface.
https://kratosmultiphysics.github.io/Kratos/
Other
1k stars 243 forks source link

Eigenvector-matrix on nodes: strange values #6054

Open veiguf opened 4 years ago

veiguf commented 4 years ago

Hi, It seems to me that there are some strange results on the eigenvector-matrix. I get them with the following code:

for node in main_model_part.Nodes:
    print(node.GetValue(StructuralMechanics.EIGENVECTOR_MATRIX))

Using the example of the beam_eigenvalue_analysis, the values for the 1st, the 8th and the 22th frequency (index 0, 7 and 21 in Python) and the 3rd (index 2 in Python) degree of freedom have a completely different order of magnitude compared to the other values. Can this be correct? How are the eigenvectors normalized?

pooyan-dadvand commented 4 years ago

@KratosMultiphysics/structural-mechanics

loumalouomega commented 4 years ago

The cantilever beam has an analytical solution for eigen modes: https://www.comsol.fr/multiphysics/eigenfrequency-analysis image

loumalouomega commented 4 years ago

may be related with the torsional modes?

e-dub commented 4 years ago

Torsional modes might be a good tip! We saw it on another model and saw the same "problem" with the example from the repository.

What we are seeing in this case is that the 8th and 22nd modes have values of the eigenvector a magnitude higher than the other modes. Something with the angles? Scaling with the mass matrix? Rad deg? Or do we have a mistake in our thinking?

The problem came up when using the eigenvectors for calculating the analytical sensitivities of the eigenfrequencies and assumed that the eigenvectors were scaled to the mass matrix ("normalize_eigenvectors": true in JSON). Has anyone use analytical gradients when optimizing beams considering eigenfrequencies?

philbucher commented 4 years ago

I had similar experiences

@armingeiser maybe you have an idea?

armingeiser commented 4 years ago

Hm the material/cross section parameters for this example do not seem to be "physical". I suggest to run the analysis using values for moment of inertia and area that are calculated from a "real" cross section e.g. rectangular or circular.

I had the experience that if these values do not have proper relations to each other, the behaviour of beams can be quite unexpected.

@veiguf: The third degree of freedom of each node or globally?

veiguf commented 4 years ago

I checked again and it is not only the third but also the forth and the fifth degrees of freedom of each node. As an example, I attach a screenshot of the EIGENVECTOR_MATRIX of the first node. Bildschirmfoto vom 2019-12-12 15-29-57

philbucher commented 4 years ago

this might be related to #6434

e-dub commented 4 years ago

We saw the problem described in this issue with beams and shells, though not with volume elements. Therefore, we are thinking that it has to do with the rotational degrees of freedom. We had a couple ideas where we want to look, but have not had time yet. Next couple weeks...

I have not been following issue 6434 in detail. Is that problem with all element types?

philbucher commented 4 years ago

I think it depends on the eigensolver and hence I would expect it to be more or less independent of the element type, but not sure

e-dub commented 4 years ago

The eigenvalues appear to be good, some degrees of freedom of the eigenvectors seem to be strange, but only beams and shells.

What I would like to look at the coming weeks: