Discussion Forum

Hi

I see 2 ways:

a) harmonic solver for a constant steady state sinus type oscillation with average values given as results, with the possibilit to easily scan the frequencyrange

b) or a time series with fine time step and a ramp up, i.e. define a "step1()" function with a smooting time constant of 1-2 sec, and an offset of half the smoothing time, that multiplies your sinus function. With this, amplitude, velocity and acceleration (even the jerk) will increase smoothly from "0" and you should excite few eigenmodes so probably some 4-6 full amplitude oscillations would then give you a stable output.
You could also consider to add some structural damping.
Do not forget to set the time stepping to "intermediate or strict", and perhaps even consider a "general-alpha" method, instead of the BDF (I mostly start with COMSOL defaults, apart the strict or free for harminic BC resp. diffusion case, then I restart with another solver and or other tweakings to see the effects)

For structural in both cases above I would mostly consider "0" as reasonable initial conditions, provided you can ramp the BC laods from "0"

--
Good luck
Ivar

Hi,

I'm modelling nearly the same problem as Masoud: a rectangle (250 um long, 30 um high) driven at resonance frequency (70 kHz, amplitude 1 nm) at a boundary, free everywhere else.

After reading Ivar's posts I've changed the solver to 'strict', multiplied my displacement BC with a step function (which gives me an unit error that I'm ignoring) and setting the relative tolerance to 1e-5

My problem is that as t-> inf (in my case 5 ms) the free end of the rectangle oscillates with amplitudes of 1e6 nm (the rectangle is 250 um long). Clearly, it isn't damped. However, if I add damping, the problem remains. I'm currently using isotropic damping with a magnitude of 1e50 (I'm desperate...) but it has no effect. What am I missing?
I didn't atach my project: 'file size error'. (it's 5 MB, is that too big?)

Regards,

Laurens

PS. I also get a lot of warnings that say "Problem encountered during results while solving." Sounds quite bad, but equally uninformative.

Hi

first of all the "unit error" probably comes from the fact that most operators expect unitless inputs, so one should use "t[1/s]" and not only "t" as arguments (you might want "t[1/h]" to have COMOSL exit the time in real fractions of hours).

For the damping, check the doc, you have different damping means and you could try a frequency dependent damping behaviour. anyhow a value of 1e50 will only give numerical over/undeflows, I never exceed 1:1e6 in my scalings

--
Good luck
Ivar

Hi,

Thanks for the suggestions. I've fixed the unit problem the way you suggested. The real problem is more persistent unfortunately.

Frequency-dependent damping sounds like a refinement of a simple constant damping. If we select rayleigh damping, the documentation notes:

"In most applications, leave alpha to zero (the default value) and define damping only using the beta coefficient. Then according to Equation 2-7 linearly increasing damping is obtained. Using two parameters, a frequency f0 where damping is defined and the damping ratio xi(f0) or loss factor eta(f0) can be defined
beta = xi/pi/f0=eta/2/pi/f0"
(p. 64 structural mechanics user guide)

Literature suggests an eta of between 1e-2 (gold) and 1e-6 (PMMA). This gives a beta between 1e-10 and 1e-14.
I set beta to a lot of different values: 1, 1e6, 1e-6, 1e-10, 0, -1e-6...
The amplitude still diverges for all these values, and the lowest amplitude for the last timestep corresponds to beta=0.

I can only attach the file if I delete the results, so you will have to compute them again yourself. The only solver setting I change is the time-step. Like I mentioned before, I set it to strict.

Laurens

Hi

I agree that most material damping values are very low, most damping comes from the assembly of parts, as you are obviously in the MEMS domain you bound the items together hence you have little or no assembly damping

Now I do not fully understand your model, as you override the fixed edge by a prescribed displacement one in only 1 direction, you excitation is strange for me. Furthermore I would use a force excitation rahter than a prescribed displacement.

Then I would use a mesh with at least 4 mesh elements across (you could split your geometry with a layer to help

For the eigenfrequency, as you have a "t" variable, and t is only defined if you are running a transient sovler, you should add a Parameter t = 0[s] to have it at least defined. And so not forget that an eigenfrequency solver is linear and ignores items to the right of the physics equations

If you then look at the eigenfrequency solver (with on eend fixed, the other free (no prescribed Displacement at all), with or without damping you will get complex eigenfrequencies reflecting the presence of damping. I prefere flat isotropic loss factors then you get a imaginary part = loss factor/2 of the real part.

Then instead of running a time solver, why not scan witha frequency sweep harmonic solver, whcih imposes a sinusoidal excitation of given amplitude and initial phase for a frequency or a frequency scan, then by plotting the Bode plots you will see better the effect of your damping

By the way you should consider updating to latest patch

--
Good luck
Ivar

Hi;

I have a similar problem. I have a mechanical structure which has frequency dependent damping. I want to get the frequency response curve using "frequency domain analysis" at 4.2. To specify my problem i gave the brief damping formula below:

c=x* y* z^2*L*2*y*Frequency*H

and i am using this parameter in Raykeigh damping model as stiffness damping parameter c/k.

So I want this "Frequency" parameter to be same with the frequency at each step during frequency domain analysis. To be more clear if ihave defined the frequency range as range(50000,1000,100000) Hz, i want comsol to use 50000 as Frequency at "c" formula at first step, 51000 as Frequency at second step, 52000 at third, up to 100000.

What should i do to achieve this?

thanks in advance

Onur

Hi

check the COMSOL name for the frequency and use that instead of your Frequency (or map it over in a variable statement.
Normally its "freq", but recheck in your case (see the help and or "equation view"

--
Good luck
Ivar

Thank you very much Ivar. I have solved the problem with using the method you adviced.

Onur

Hii Ivar,

Greetings for the day.

I am trying to apply a pressure sine wave(0.23[MPa]*sin(2*pi*20000[1/s]*t) in a laminar flow boundary inlet coupled with electrochemistry. But, the wave applied is not producing any changes in the deformed geometry.

The reason may be attributed to the inaccurate application of sine wave boundary condition. In this regard, I have few queries.

1. How to apply sine wave(Is it through the waveform present in the functions under global definitions?)?
2. How to set the solver as strict/intermediate/free?
3. Is there anything I need to change for my time step( I have taken 0.00001 as time step).

Your response in this regard shall be treated with highest level of alacrity.

Regards.