Hi Paul. Matplotkib : excellent tool! It is also what I use.
Yes the script is based on modal superposition feed by mode shape of a simply supported plate for now. If the method is interesting; intarface with my previous FDM script is easy
Much more important than DML! enjoy!
Christian
Clear. No problem. This is probably the easiest parameter to choose... and easy to adapt. I will move to that.
@pway Hello Paul,
I change for the data you used in #275. Possible remaining difference are on the damping coeff (0.05), force (I set to 2.83V in 4Ohms and Bl=4NA so 2.83N in current driving mode).
The LF cut off in my simulation is not visible neither (I have it in the previous post with a smaller plate - 400x600mm and higher E=10MPa)
In the conditions I used (simply supported, exciter not included in the mode calculation), the 1st mode is very low (13Hz) so below my range of frequency which starts at 20Hz. This not really surprising has the Young modulus is low (EPS?) and in consequence the bending stiffness also (1.23Nm).
For testing, I would suggest to come to a stiffer material to have a 1st mode let say 50 to 80Hz?
Last image is our 2 FR in lin frequency scale. Can't say less than : some differences! At least we have some approaches that start working. Need more debugging!
Simulation :
The plate is Lx = 0.5 m by Ly = 1.0 m
The mesh is Nx = 26 cells by Ny = 51 cells
with a grid dx = 20 mm by dy = 20 mm
Bending stiffness B = 1.23 Nm, Areal density µ = 0.39 kg/m²
Driving Point at 0.4 0.4
Exciter : Re = 4.0 Ohm, Le = 0.1 mH, Bl = 4.0 N/A
Point Of Observation at 3.0 m
I change for the data you used in #275. Possible remaining difference are on the damping coeff (0.05), force (I set to 2.83V in 4Ohms and Bl=4NA so 2.83N in current driving mode).
The LF cut off in my simulation is not visible neither (I have it in the previous post with a smaller plate - 400x600mm and higher E=10MPa)
In the conditions I used (simply supported, exciter not included in the mode calculation), the 1st mode is very low (13Hz) so below my range of frequency which starts at 20Hz. This not really surprising has the Young modulus is low (EPS?) and in consequence the bending stiffness also (1.23Nm).
For testing, I would suggest to come to a stiffer material to have a 1st mode let say 50 to 80Hz?
Last image is our 2 FR in lin frequency scale. Can't say less than : some differences! At least we have some approaches that start working. Need more debugging!
Simulation :
The plate is Lx = 0.5 m by Ly = 1.0 m
The mesh is Nx = 26 cells by Ny = 51 cells
with a grid dx = 20 mm by dy = 20 mm
Bending stiffness B = 1.23 Nm, Areal density µ = 0.39 kg/m²
Driving Point at 0.4 0.4
Exciter : Re = 4.0 Ohm, Le = 0.1 mH, Bl = 4.0 N/A
Point Of Observation at 3.0 m
Hi Christian
I get eigenfrequencies very close to yours.
The big differences seem to be that
Unfortunately it seems that in Elmer, the damping is not fully implemented. I have checked Rayleigh damping alpha (mass proportional, giving high damping at low frequency) and it works as expected. I wanted to use a high value for Rayleigh damping beta, to see if I could get results close to yours, but unfortunately beta seems to be unimplemented.
Modal superposition has advantages, including faster calculation, and fine control over damping (you can select a different damping constant for each mode, if you want). I may move to that too. Are you using a formula similar to the Thompson paper we were discussing for Rayleigh integral (last 2 formulae in post 170?) What range of modes are you summing over at any particular frequency (formula says to sum ALL contributions, but I think we can assume just a few nearby modes will contribute significantly at any given frequency)? Can you show some of your calculated mode shapes at frequencies between the eigenmodes? We can compare to results from the 'brute force' direct harmonic method of Elmer.
I get eigenfrequencies very close to yours.
The big differences seem to be that
- You have higher values overall
- You have a very high value of damping, smoothing the modal peaks, and increasing with frequency.
Unfortunately it seems that in Elmer, the damping is not fully implemented. I have checked Rayleigh damping alpha (mass proportional, giving high damping at low frequency) and it works as expected. I wanted to use a high value for Rayleigh damping beta, to see if I could get results close to yours, but unfortunately beta seems to be unimplemented.
Modal superposition has advantages, including faster calculation, and fine control over damping (you can select a different damping constant for each mode, if you want). I may move to that too. Are you using a formula similar to the Thompson paper we were discussing for Rayleigh integral (last 2 formulae in post 170?) What range of modes are you summing over at any particular frequency (formula says to sum ALL contributions, but I think we can assume just a few nearby modes will contribute significantly at any given frequency)? Can you show some of your calculated mode shapes at frequencies between the eigenmodes? We can compare to results from the 'brute force' direct harmonic method of Elmer.
Last edited:
Yes, I am wondering if they are not a bit too high...(it won't be the 1st time I introduce an error of a factor or whatever in a formula... ;-) ). Check needed
Damping was set to 0.05 (need to start somewhere...) but constant with frequency. I saw the effect you detected after posting. The reason is not in the damping. Also to be checked (maybe not enough modes involved in the higher portion of the frequency range, only 50 modes used in this simulation)
Yes similar. I get them from A Putra's thesis. As previously, the script is organized to export a pdf to explain the content. I should be able to post it very soon
For now, at each point of frequency, the sum is made over all the modes available. Some optimization is probably possible as when the difference between the frequency of interest and the mode is important, the mode contribution might be low. Not implemented for now.
In this script, I used the word "mode shape" in the meaning of eigenvector (phi in the formulas) giving the shape of the displacement at the natural frequencies. For each frequency, the script compute what I call speed shape which the speed of the points of the plate at the considered frequency, for the exciting point used. So it is the result in speed of the modes superposition under the effect of the damping. When modes are closed, because of the damping or because of the exciter position, we don't get the nice mode shape then. But this available. For now, I plot them for some natural frequencies. Which frequencies would you like in between?
Christian
Hi again Christian
Thanks for the code you sent by DM. I reply here to keep the continuity in the discussion.
Yes it was around 10C cooler by the sea. A nice weekend
. We came back after dark to avoid the heat.
I like your py2pdf usage, there is a long history of literate programming with a similar combination of text and code. My personal preference would be to make the code as readable as possible (eg to use more advanced markdown variants), because I find the formatting code in the py file distracting.
I do have an interest in publishing however, having worked in developing educational resources. I have not used latex much personally, but in the past I have some interest in Latex and alternatives like ConText, Sile, Lout. I have mainly been involved in publishing to HTML, XSL, ePub. I like the idea of markdown and extensions (like asciidoc), and Pandoc/asciidoc seems to be the smoothest path from human-readable markup into published format.
It will take me time to go through your code, but some initial thoughts.
I will read through your code with interest, thanks!
Paul
Thanks for the code you sent by DM. I reply here to keep the continuity in the discussion.
Yes it was around 10C cooler by the sea. A nice weekend
. We came back after dark to avoid the heat.I like your py2pdf usage, there is a long history of literate programming with a similar combination of text and code. My personal preference would be to make the code as readable as possible (eg to use more advanced markdown variants), because I find the formatting code in the py file distracting.
I do have an interest in publishing however, having worked in developing educational resources. I have not used latex much personally, but in the past I have some interest in Latex and alternatives like ConText, Sile, Lout. I have mainly been involved in publishing to HTML, XSL, ePub. I like the idea of markdown and extensions (like asciidoc), and Pandoc/asciidoc seems to be the smoothest path from human-readable markup into published format.
It will take me time to go through your code, but some initial thoughts.
- I will be using Elmer at least to the eigen analysis stage, because it support plates of any shape. I realised that I can still apply a 'constant' damping ratio across frequencies by just recalculating the mass damping (alpha) term at each frequency, so I may stick with that.
- I would like to separate the effect of exciter location from the effect of plate shape, so I think that when investigating plate shape, I will apply the excitation to the whole plate during harmonic analysis (ie no specific location). Then investigate exciter location as a separate stage. Otherwise the interpretation of plate shape is complicated by the arbitrary exciter location accentuating or hiding different mode shapes with changing frequency.
- I look forward to see plot including higher freqs. I would remove the 1/6 octave smoothing in the first instance, because it has a similar effect to damping so confuses the issue. Is it possible that you need to use a lower value of damping as f increases?
- I still view it as an open question for us whether the Rayleigh integral gives sensible results. Ive read that is can be unstable, but perhaps it's OK the way we are using it, for far-field only.
- The intermediate mode shapes would be of interest because we can compare with Elmer's harmonic analysis, to see if it looks similar, as a sort of validation of the method.
I will read through your code with interest, thanks!
Paul
Last edited:
@pway
Hello Paul,
I am currently coming back to the original topic of the thread doing some tests on a material to get its characteristics (tap tests).
Do you have already an elmer script for a free edge rectangular orthotropic plate that you could share please?
Thanks for the help
Christian
Hello Paul,
I am currently coming back to the original topic of the thread doing some tests on a material to get its characteristics (tap tests).
Do you have already an elmer script for a free edge rectangular orthotropic plate that you could share please?
Thanks for the help
Christian
Thank you Paul but no need for isotropic. You already shared files with isotropic material, this works now here. I am looking for an alternative Linux so I thought to Elmer to Lisa used by Eric in order to check the material characteristics that fit with the tap test.
I just had a look... very impressive! Positive point : many things directly in French... but is it the kind of tool related to this limited need (seems to have so wide capacities)... and is it the kind of play ground I can go (FEM is far to be an area of strong knowledge for me).
@pway , @Veleric
Well... I think that opening to an other FEM tool is perhaps not the right option for now. Much better to stay on tools at least 2 of us have. So I will first explore the possibility to reactivate windows on my laptop and then the possibility of Lisa... hope the Windows is still accessible and not too old!
But on the other hand, with you being French, we may never get another opportunity to have a system that potentially everyone can use. I may spend a bit of time with Salome to see if I can have it do a simple 2d plate eigenvalue analysis. If it can, it should be a matter of taking the generated config file and examining the contents, then scripting it as a python template for automation.
Christian
Your 'speed shapes' seem to be similar to those produced by Elmer in the harmonic analysis, as far as I can tell at the moment, Are you plotting the magnitude of the complex velocity field? I have separate Real and Imaginary components and have not plotted the magnitude as ParaView does not know that they are related as components. But looking at a few shapes where one component dominates, it looks quite similar.
Your 'speed shapes' seem to be similar to those produced by Elmer in the harmonic analysis, as far as I can tell at the moment, Are you plotting the magnitude of the complex velocity field? I have separate Real and Imaginary components and have not plotted the magnitude as ParaView does not know that they are related as components. But looking at a few shapes where one component dominates, it looks quite similar.
Hi Paul,
Good thing
Good question!!! It is a point I have to focus on because I am here a bit lost.
In the current code, it displays the magnitude of the pressure calculated by the Rayleigh integral (see pcurrent, pvoltage line 659, 660 of the code, p being the result of the integral). I made this choice in the idea the split between real and imaginary is the image of the phase at the frequency of interest.
The weak point is before in the calculation in umi in §3 of the pdf, bottom of page 2 (extract below) when it comes at the magnitude of frequency filter (mode). This is what is in M Putra's thesis. I wonder wy the calculation doesn't remain fully in complex. When there are only product, I guess it doesn't matter but when come addition... line 540 in the code.
Hi Paul,
Sounds like a good opportunity yes. I had a first look to code Aster. Going through the first top document of code Aster, it seems having some key features for our application : physical fields of application (mechanical vibration, acoustics), the dimensions (2D, 3D), the software compatibility (GMSH, python...), possibility of lumped elements (mass, spring...) at nodes... Code Aster is multiplateform. I even found a lnux distribution that can be ran from an usbkey (or install after) as other Debian or Ubuntu distribution : CAE Linux that embedded many design tools.
The documentation (in French) seems wide. To see the level of the English documentation... I can't translate the full set! ;-)
Christian
@pway
I had a more deeper look to code Aster. Seems promising. Some links :
I had a more deeper look to code Aster. Seems promising. Some links :
- Code Aster main page. English button on the top right. I had a look to the documentation in French and in English. Not exactly similar but the English seems also wide. The opportunity to check in both version if needed is a plus. The point is to find the command files related to the examples (many many examples available including 2D plates)
- Code Aster on Docker . Not fully sure about the way to use, I have to ask to my son for more details but Docker is a container system to make more easy the installation on different plateform. I see here that I will probably need to arrange some room on my HDD (3GB needed)... My home linux partition is a bit to small, and the laptop old and slow (time for a new one?)
- CAE Linux : a Ubuntu based Linux distribution that embedded different tools like Code Aster to flash on a usbkey. Should run by booting on the usbkey without installation. I tested many Linux distro by this way.
@pway
Hi Paul,
I continue checking the possibility to switch to Code Aster. It seems not being really multi plateform in the way the original release is on Linux and the Windows one by third parties. Have you identify the way to get it for Windows?
With my search, for a direct installation, it seems to point here : Code Aster for Windows
Hi Paul,
I continue checking the possibility to switch to Code Aster. It seems not being really multi plateform in the way the original release is on Linux and the Windows one by third parties. Have you identify the way to get it for Windows?
With my search, for a direct installation, it seems to point here : Code Aster for Windows