Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Beam frequency response

Please login with a confirmed email address before reporting spam

Hello everybody!

I need to make dynamic simulation of silicon beam vibration using a mini shaker. For this I need to use frequency analysis. However in COMSOL I can't find any acceleration as a load. Since I define acceleration for mini shaker can I find it somewhere in COMSOL for my model???
If the answer is 'no', is it possible to make it through force (F=ma)??? If 'yes', how can I find mass of my beam in COMSOL, I guess it has to calculate mass automatically.
And last question. I have huge problem with mesh visualization. In the beam surface after meshing I have totally black surface, while on the back and sides everything is fine (see picture in attachment). I have no idea what to do, may be somebody has some ideas???

I hope for you reply,
thank you in advance,
Vitaliy


5 Replies Last Post 7 juin 2010, 01:54 UTC−4
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 1 juin 2010, 01:51 UTC−4
Hi

First of all, is it 3.5 or 4 you are using ? your image looks like 3.5 so I'll assume that.

For the mesh, it's black, but how many surface elements do you have? probably some hundred tousands, and they are so tightly packed that you cannot see through, you need a hughe multicore and GB ram PC to get that calculated without ram limitations, but that is another issue.

Some tricks:
to better visualise thin structures such as MEMS, you can tun "off" the orthonormal scaling of the graphics window, on the lower border of your image (graphical window) you have an "EQUAL": if you double click on it and then hit the rescale (arrows n-s-e-w icon, top border) you wil get a distorted view, but its much easier to pick items like that.

you have also the "options visualisation settings", to turn on and off items such as mesh etc to ease the graphics handling

Then I see that your volume is cut in smaller parts, probably respecting the symmetry you have (I like that if the volumes are truely cut, not just the surface) this means that you can select only 1/4 or 1/8 to use symmetry if you have RAM problems, which is very often the case for such models).

Then for the mesh, I would start with a far coarser mesh, select all upper surfaces and mesh the boundary (I would start with "tets", but "quads" will give some even more precise results probably but they require far more time and RAM to solve, that's typically for the last run, as mesh quality check). And you sweep them through your volume. I would use minimum 3 per layer thickness, if possible more.

I would also run first a static respons with a "body force load" of constant G=9.81[m/s^2] and Fz= -G*rho_smsld [N/m^3] to see the sag under static gravity. Then you can record the maximum deflexion value and estimate the first piston eigenmode "f"[Hz] by: 2*pi*f = sqrt(k/m)=sqrt(k*Dz/(m*Dz))=sqrt(m*G/(m*Dz))=sqrt(G/Dz) so your frequency is not only linked (to a first approximation) to the stiffness "k" over mass "m" ratio, but also to the gravity "G" (or body load) and the maximum deflection "Dz" ratio.

Then I would run an eigenfrequency, with perhaps 9 modes (you can leave the body load as this is ignored in eigenmode calculations), it's good to check that you find your piston or Dz mode at a frequency with an error of maximum some 10-20% from your estimated "f" value.

And finally I would do an harmonic sweep with a parametric "freq" value running around some of the interesting modes you have, you need to use a rather fine step to catch the peaks as by default you have litle damping in Comsol (some numerical damping is in there to avoid NAN's). I prefer to look at these graphs in log/log mode, they are easier to interprete as you also see the anti-resonances.

Then only you can start to add stress levels to see stress stiffening, in 3.5 you need to use matlab as you must first run a stationary run to set the stress level and then to save it and restart from there, in V4 you can get it done via the solver sequencing.

Last thing: if you have thin layers (metal layers) these tend to add many elements and use a lot of ram, you should consider to add them onto your model as "surface layers" (note that using "shells" and 3D structural is not obvious as you need to write all the physics to link the variables, in particular the rotations you have in shell but not in 3D, this is rather heavy), for surface layers see the doc.

There is a lot about all this in the MEMS doc and the structural doc.

Hope this helps
Have fun Comsoling
Ivar
Hi First of all, is it 3.5 or 4 you are using ? your image looks like 3.5 so I'll assume that. For the mesh, it's black, but how many surface elements do you have? probably some hundred tousands, and they are so tightly packed that you cannot see through, you need a hughe multicore and GB ram PC to get that calculated without ram limitations, but that is another issue. Some tricks: to better visualise thin structures such as MEMS, you can tun "off" the orthonormal scaling of the graphics window, on the lower border of your image (graphical window) you have an "EQUAL": if you double click on it and then hit the rescale (arrows n-s-e-w icon, top border) you wil get a distorted view, but its much easier to pick items like that. you have also the "options visualisation settings", to turn on and off items such as mesh etc to ease the graphics handling Then I see that your volume is cut in smaller parts, probably respecting the symmetry you have (I like that if the volumes are truely cut, not just the surface) this means that you can select only 1/4 or 1/8 to use symmetry if you have RAM problems, which is very often the case for such models). Then for the mesh, I would start with a far coarser mesh, select all upper surfaces and mesh the boundary (I would start with "tets", but "quads" will give some even more precise results probably but they require far more time and RAM to solve, that's typically for the last run, as mesh quality check). And you sweep them through your volume. I would use minimum 3 per layer thickness, if possible more. I would also run first a static respons with a "body force load" of constant G=9.81[m/s^2] and Fz= -G*rho_smsld [N/m^3] to see the sag under static gravity. Then you can record the maximum deflexion value and estimate the first piston eigenmode "f"[Hz] by: 2*pi*f = sqrt(k/m)=sqrt(k*Dz/(m*Dz))=sqrt(m*G/(m*Dz))=sqrt(G/Dz) so your frequency is not only linked (to a first approximation) to the stiffness "k" over mass "m" ratio, but also to the gravity "G" (or body load) and the maximum deflection "Dz" ratio. Then I would run an eigenfrequency, with perhaps 9 modes (you can leave the body load as this is ignored in eigenmode calculations), it's good to check that you find your piston or Dz mode at a frequency with an error of maximum some 10-20% from your estimated "f" value. And finally I would do an harmonic sweep with a parametric "freq" value running around some of the interesting modes you have, you need to use a rather fine step to catch the peaks as by default you have litle damping in Comsol (some numerical damping is in there to avoid NAN's). I prefer to look at these graphs in log/log mode, they are easier to interprete as you also see the anti-resonances. Then only you can start to add stress levels to see stress stiffening, in 3.5 you need to use matlab as you must first run a stationary run to set the stress level and then to save it and restart from there, in V4 you can get it done via the solver sequencing. Last thing: if you have thin layers (metal layers) these tend to add many elements and use a lot of ram, you should consider to add them onto your model as "surface layers" (note that using "shells" and 3D structural is not obvious as you need to write all the physics to link the variables, in particular the rotations you have in shell but not in 3D, this is rather heavy), for surface layers see the doc. There is a lot about all this in the MEMS doc and the structural doc. Hope this helps Have fun Comsoling Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 1 juin 2010, 04:10 UTC−4
Thank you, Ivar for you previous letter! It was very helpful! However I still have some questions left

1) you asked me which version do I use. I use 3.5
2) the 'black area' problem is not solved at the moment. I used really coarse mesh (713 elements) and it was still black. Moreover after simulation this black area remained. However on the back area and on sides everything is fine, the problem is just on the top surface. Is there something I can do with it? Or is there a problem with my PC?
3) as far as I understood from your approach for my frequency analysis I should do following:
- make a static analysis
- then check whether it is ok or not by comparing calculated and modeled eigenfrequencies (using modal analysis)
- if it is ok, apply the frequency range of my interest
Then if I want to change my acceleration I have to go to the static analysis, change load and repeat the loop. Am I understanding things right???
4) Is it possible to apply air damping in such situation? How to do it?

Thank you in advance,
regards,
Vitaliy
Thank you, Ivar for you previous letter! It was very helpful! However I still have some questions left 1) you asked me which version do I use. I use 3.5 2) the 'black area' problem is not solved at the moment. I used really coarse mesh (713 elements) and it was still black. Moreover after simulation this black area remained. However on the back area and on sides everything is fine, the problem is just on the top surface. Is there something I can do with it? Or is there a problem with my PC? 3) as far as I understood from your approach for my frequency analysis I should do following: - make a static analysis - then check whether it is ok or not by comparing calculated and modeled eigenfrequencies (using modal analysis) - if it is ok, apply the frequency range of my interest Then if I want to change my acceleration I have to go to the static analysis, change load and repeat the loop. Am I understanding things right??? 4) Is it possible to apply air damping in such situation? How to do it? Thank you in advance, regards, Vitaliy

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 1 juin 2010, 05:57 UTC−4
Hi

for me the "black area looks like a fine mesh (typically what I get with fine meshes) or is it linked to extra points showing up?
Check your settings in View Visualisation ...

The static analysis is not reuired, but I usually do it with a gravity load to check that my model behaves as I expect it to.

If you want to apply pre-stress while having an eigenmode analysis, as one often have in MEMS with thin layers, this can be done by first a static analysis, or by defining directly an initial stress value then restart the eignemode. But it's far more complex than just to run an eigenmode analysis, so to start with a simple eigemode, would be my approach.

normally you get a series of modes, that you can look at what I would call piston modes (translation along Z) or tip/tilt modes (rotations around X,Y) and then higher modes as beam modes, first order, second order etc.

I would expect the piston mode to be the lowest (>0) and its should be close (10-20%) to the one you can estimate from the maximum gravity sag Dz.

Then the harmonic parametric sweep wil lgive you the peaks, but their shapes, in reality will depend strongly on the damping you have (and this is also delicate to inter into COMSOL, as it could be gas film damping etc, complex often non-linear physics.

If you want to do as if you shake the system (with a known acceleration PSD) you need to modulate the force amplitude = acceleration you apply with the frequency. Typically for such devices the amplitudes (max peak to half widths) are very damping dependent, but qualitatively they agree mostly well

I believe the air damping is treated as examples in the mems doc, and certainly discussed several times in the conference papers, try a search on the forum, the knowledge base and the model library. Personally, I have not applied squeezed film damping to such structures, neve have had the time and bucks to do that ;)

Have fun Comsoling
Ivar
Hi for me the "black area looks like a fine mesh (typically what I get with fine meshes) or is it linked to extra points showing up? Check your settings in View Visualisation ... The static analysis is not reuired, but I usually do it with a gravity load to check that my model behaves as I expect it to. If you want to apply pre-stress while having an eigenmode analysis, as one often have in MEMS with thin layers, this can be done by first a static analysis, or by defining directly an initial stress value then restart the eignemode. But it's far more complex than just to run an eigenmode analysis, so to start with a simple eigemode, would be my approach. normally you get a series of modes, that you can look at what I would call piston modes (translation along Z) or tip/tilt modes (rotations around X,Y) and then higher modes as beam modes, first order, second order etc. I would expect the piston mode to be the lowest (>0) and its should be close (10-20%) to the one you can estimate from the maximum gravity sag Dz. Then the harmonic parametric sweep wil lgive you the peaks, but their shapes, in reality will depend strongly on the damping you have (and this is also delicate to inter into COMSOL, as it could be gas film damping etc, complex often non-linear physics. If you want to do as if you shake the system (with a known acceleration PSD) you need to modulate the force amplitude = acceleration you apply with the frequency. Typically for such devices the amplitudes (max peak to half widths) are very damping dependent, but qualitatively they agree mostly well I believe the air damping is treated as examples in the mems doc, and certainly discussed several times in the conference papers, try a search on the forum, the knowledge base and the model library. Personally, I have not applied squeezed film damping to such structures, neve have had the time and bucks to do that ;) Have fun Comsoling Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 6 juin 2010, 05:19 UTC−4
Hello, Ivar!

You were right, my black spot problem was solved after some playing with visualization properties. I just activated the 'headlight' mode with white color and the problem was solved. However only in 3D. For 2D I still can't find any solution... But I don't really need it now, 3D is what I need.
Up to now I've done frequency response for different geometries and accelerations. The results are trustful as were compared with ANSYS results. But now I am facing with damping problem.
In the previous letter you said that air damping problem was discussed several times in MEMS tutorial and forum. Unfortunatelly I coudn't find any information how to implement air for my problem. If it is possible can you give me a small hint. May be I can neglect air influence since I have permanent frequency and acceleration done by shaker???

thank you in advance,
regards,
Vitaliy
Hello, Ivar! You were right, my black spot problem was solved after some playing with visualization properties. I just activated the 'headlight' mode with white color and the problem was solved. However only in 3D. For 2D I still can't find any solution... But I don't really need it now, 3D is what I need. Up to now I've done frequency response for different geometries and accelerations. The results are trustful as were compared with ANSYS results. But now I am facing with damping problem. In the previous letter you said that air damping problem was discussed several times in MEMS tutorial and forum. Unfortunatelly I coudn't find any information how to implement air for my problem. If it is possible can you give me a small hint. May be I can neglect air influence since I have permanent frequency and acceleration done by shaker??? thank you in advance, regards, Vitaliy

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 7 juin 2010, 01:54 UTC−4
Hi

I was thinking of:

Squeezed-Film Gas Damping in an Accelerometer p130 memsmodlib.pdf

But you might have a different case, if you have a plate just below and above its relevant, and will influence independently of your "shaking"

Good Luck
Ivar
Hi I was thinking of: Squeezed-Film Gas Damping in an Accelerometer p130 memsmodlib.pdf But you might have a different case, if you have a plate just below and above its relevant, and will influence independently of your "shaking" Good Luck Ivar

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.