Discussion Forum

You're trying to enter an excitation randomized using an expected power spectrum density using some assumption of probability distribution of amplitude and phase of each discrete frequency component?

I've done this sort of thing with surface roughness (to generate a randomly "rough" surface using a given autocorrelation length). For that you need to generate a series of normally distributed amplitudes, uniformly random phases, and construct a sum of sines (or in the case of multiple dimensions, sums of products of sines).

However, I've not yet done this WITHIN Comsol since the number of terms can become very large and so it's easier to do it with loops. I have done it using an outside tool then importing the resulting structure into COMSOL. But I'm interested to see what people recommend here for an all-COMSOL solution.

Daniel, thank you for your answer!

Yes, you are right about PSD, it is well-known technique to estimate reaction of design under random vibration. That is standardized procedure required by IEC 60068-2-64 and MIL-STD-810, as example.

I tried to do this simulation and was unpleasantly surprised that I could not find to specify PSD at Comsol environment. This is a pity! Yes, there are some external methods and even other more simple CAE where it is possible to calculate, but I feel Comsol has good potential to model all in one environment, but now, it seems, is not fully for random vibration analysis (( .

Well, I strongly suspect there is a way but it may not be obvious....
Worst case would be to generate the profile in an external file with a script written in, for example, Perl or Matlab. But I'm conjecturing here. My problem is multidimensional which is more complicated. I'll end up asking that with a different question....

Hi

A standard avionic/space PSD analysis is not set up to be that easy in COMSOL, unfortunately, I have been asking for it made more streamlined, since v3.3 but it's a post-processing issue specific to Solid or Structural physics so it's like the Mass Participation Factors (linear are there, except for any Rigid Connector masses, and rotational MPF are still missing). this is an Engineering exercise and something that seem to have low priority by COMSOL.
All other FEM programs I know about have a script to add up coherently or incoherently (RMS) the different modal contributions for shock and random vibration analysis, with COMSOL it's a bit trickier to do operations on the modes.

In modal time or modal frequency sweeps there is a way to give a relative amplitude to each mode but adjusting the participation factor it's somewhere deep under the solver settings, including losses per mode if I remember right. But one must recalculate the PSD into frequency dependent amplitudes. If you try a standard frequency sweep you must also translate the PSD into equivalent frequency amplitudes.
There is a few FORUM threads about how to do this, but I could not find them back just now, and I'm out of office for a few days so I cannot search it out from my COMSOL either, due to lack of remote access ;( I believe Nagi gave a good description on how to ad specific modal contributions, a few yeas ago)

I still have not seen any script proposed by a user out here. At least now I see I'm not alone missing this feature :)

Today with the application builder it should be possible to prepare the PSD transformation too, with Matlab you can do it for simple models, but it's tedious if you start from scratch, I did it once several years ago but the versions have changed too much since then.

Perhaps you should send a suggestion to support, and add the MIL standard to be sure they see that it's not just a nice to have feature.

--
Good luck
Ivar