Help needed for a shell model analysis

[giorgio88]

Hi everybody,

I need to obtain the stiffness and the mass matrices from 2 different FE models, one is meshed with solid elements and the other one with shell elements.

I created the 2 meshes with virtualab, exported as a .dat file, performed an analysis with nastran and, with a matlab routine, I red the matrices from the .pch files.
To validate this procedure I am simply calculating the natural frequencies of the two models both with virtual lab and with matlab.

My problem is that the matching between the frequencies of the shell model, when calculated in matlab and in virtualab, are completely different: in matlab I obtain only 3 rigid body modes (instead of 6), the first flexible mode is at a very low frequencies (10 times smaller than the frequencies obtained in VL), and the other frequencies only have the same order of magnitude, but the error could be very important (from 50 to 100%).

Just to make everything more clear, I modified the exported (from virtualab) .dat file adding before "BEGIN BULK DATA":
SOL 103
PARAM,COUPMASS,1
PARAM,EXTOUT,DMIGPCH
The shell elements are CQUAD4, in the shell property PSHELL I have membrane effect, bending effect and transverse shear (as well as the thickness of the element, obviously); material is steel.

I read the .pch files from the 2 models with the same matlab routine.

In matlab, I tried to calculate eigenvalues in different way, just to be sure, and results are always the same.

I've already looked for some solutions in the forum but I didn't find anything... Does anybody know how to solve this problem?

 

[Zadrobu]

in matlab I obtain only 3 rigid body modes (instead of 6), the first flexible mode is at a very low frequencies (10 times smaller than the frequencies obtained in VL), and the other frequencies only have the same order of magnitude, but the error could be very important (from 50 to 100%).

Does your Matlab routine calculate normal modes for a free-free situation (i.e. no model constraints)? If yes, then having only 3 rigid body modes would be an indicator that this calculation is not fully correct. There should be 6 rigid body modes.

 

[BlasMolero]

Hello!,
The normal modes/Eigenvalue (SOL103) solution should be the same, not matter you use VL or NX NASTRAN, or any other FEA code, this is a basic FE analysis, then some options regarding mass considerations (lumped or coupled), or the eigenvalue extraction method used, etc.. could be the reason of the differences in eigenvalues results.

Simply define a test problem of a cantilever beam, post here the details (dimmeensions, materias, BCs, etc..), and I can run the problem using FEMAP & NX NASTRAN and post the results here, OK?.
Or post your nastran input deck (*.dat) here and we can take a look to it.
Best regards,
Blas.

~~~~~~~~~~~~~~~~~~~~~~
Blas Molero Hidalgo
Ingeniero Industrial
Director

IBERISA
48004 BILBAO (SPAIN)
WEB:

http://www.iberisa.com

Blog de FEMAP & NX Nastran:

http://iberisa.wordpress.com/

 

[giorgio88]

Thank you for answer.

I have already tried to change options about mass, not about the solution methods (I'm always using default options in nastran, which should be the Lanczos algorithm for a real solutions); by the way, I also used another solver (samcef), and results are the same obtained with nastran.

At this point I think that the problem is on my matlab code, also because I'm calculating eigenvalues in free-free conditions; I really don't know if it is due to the inversion of the mass matrix or to the calculation of eigenvalues, so I think I will write from zero the algorithms (maybe matlab functions "eig", "eigs", "inv" ecc. are optimized in a way that can affect my result).

Let me know if you have any other ideas please (anyway, if I won't forget it, tomorrow I will post my nastran input deck).

 

[giorgio88]

Hello again,

attached you can find 2 files: the .dat file used to start the analysis with nastran, and the .pch imported in matlab.

In these files there isn't my model, but a simple one that I'm using for tests: a plate clamped at one end. Furthermore, they refer to a static analysis performed just to see if, at least in this case, there would have been matching (as you can imagine, result completely wrong).

Today I tried to solve the eigenvalues problem with other algorithms, but it always converges to the same wrong results.

Maybe, the problem is that I have to do some pre-conditioning before to use the matrices, whose condition numbers are not good at all (but I have no idea where to start), or, alternatively, that I have to modify the .pch file before to import it in matlab(even if it's strange it works perfectly for the solid elements model).

Thanks in advance