Different results in coupled eigen-helmholtz analysis

Numerical methods and mathematical models of Elmer
Post Reply
flowwolf
Posts: 7
Joined: 14 Dec 2016, 01:39
Antispam: Yes

Different results in coupled eigen-helmholtz analysis

Post by flowwolf »

Dear Elmer users,


We have been trying to use Elmer to simulate the high-frequency behaviour of the soft parts assembly of an axisymmetric loudspeaker driver model.

AFAIU this is possible by using 'Hierarchical coupling of structural eigenmodes and helmholtz equation' (examples given as HelmholtzStructures).

The tests that I've made give different results for this type of analysis and I don't know whether I set it up incorrectly or maybe I'm missing something.


The models were drawn in FEMM. A self-made application converts that to a .poly file, triangle meshes it, then ElmerGrid is used to convert it to the mesh format needed by Elmer. At this point, a small code is used to reassign attributes (materials in this case) to every element, based on the .poly file.

I've attached 3 test cases (.poly file, mesh, .sif file, etc.)


3rd case:
https://www.mediafire.com/file/esiczyq5 ... 3.zip/file

There are three examples within. For each example, the exact same geometry is used, only the mesh fineness is varied at the very end of the .poly files. While the mesh is already fine and the mesh itself is based on the exact same geometry, the results are different for each 3 examples. What's even more interesting, that the mesh of the 1st and 2nd geometry is more fine than the 3rd's, yet it's the 2nd that gives quite different results.
Shouldn't all the 3 results be of about the same?
https://www.mediafire.com/view/s2ajb6tv ... 1.png/file
https://www.mediafire.com/view/f1k4iswd ... 2.png/file



2nd case:
https://www.mediafire.com/file/iz1d9ddg ... 2.zip/file

One example uses a half-circle as a non-reflective boundary while the other uses a quarter-circle. The geometry, the frequency are the same but the results are different. Is this normal (even for a very fine mesh)?
https://www.mediafire.com/view/dwyxb1il ... f.png/file
https://www.mediafire.com/view/ppg1whgn ... t.png/file


1st case:
https://www.mediafire.com/file/mckpzwll ... 1.zip/file

This is for eigenvalue analysis, there is just one example, but the eigenvalues calculated vary on the number of requested eigenvalues, i.e. if the number of requested eigenvalues is set to 12, one eigenvalue is missing from the list, it's not missing if it set to 20 for example. This might be a compilation problem as well, because this case works correctly in the Windows releases of Elmer, but not on the one I built on a Debian system.

12 requested eigenvalues:

EigenSolve: 11: 2.572423E+10 0.000000E+00
EigenSolve: 12: 3.971727E+10 0.000000E+00

20 requested eigenvalues:

EigenSolve: 11: 2.572423E+10 0.000000E+00
EigenSolve: 12: 3.303824E+10 0.000000E+00
EigenSolve: 13: 3.971727E+10 0.000000E+00

(3.303824E+10 is missing above, 'Linear System Symmetric = Logical True' made it come back, but is that normal?)




So one question would be, did I set up the boundaries correctly, or does an axisymmetric case need something special?
Also, the .sif files assign bodies and materials to non-existent elements. I didn't make a difference to correct this.


This is not very important but can ElmerGrid associate triangle-mesh attributes to Elmer-mesh attributes?


Any help would be appriaciated,

ps. apologies for the mediafire links, but the zip files are quite large, I can upload them here if it's better that way

Regards,
flowwolf
kevinarden
Posts: 2237
Joined: 25 Jan 2019, 01:28
Antispam: Yes

Re: Different results in coupled eigen-helmholtz analysis

Post by kevinarden »

I believe symmetry models can only calculate symmetric modes .
Post Reply