Discussion Forum

Hi

I can only warn you that ANSYS and COMSOL uses different normalisation methods for their eigenmode calculations, therefore the eigenmodes seem very "large" in COMSOL when you compare them to ANSYS or NASTRAN.
This is as such no error but a question of which "norm" to use. Unfortunately going from one norm to the other is not trivial w.r.t matrix extracting/handling/processing issues.

I have given up and hope that V4 will give us more flexibility, with COMSOL and its current norm we loose the simple "equivalent mass" or mass participation factor analysis of our eigenmodes.

Now, normally for harmonic analysis you define yourself the excitation load, so the normalisation for the eigenmodes, as I describe above, should in principle not bother you directly, and you should get correct results.

Good luck
Ivar

Hi Kerem,

I too am modeling cmuts and comparing with ANSYS simulations. I have also noticed differences in the magnitudes for comparing harmonic analyses between ANSYS and Comsol. Comsol seems to produce amplitudes nearly 10x as high as ANSYS. Have you made any more progress on this problem?

Also, for your electrostatic analysis, I'd be interested to hear about the methodology you are employing. I am currently using a variation of the Maxwell Stress Tensor method, but I have yet to obtain results that match what I would expect. I have posted on here before but no one in this forum seems to have any suggestions on the correct way to model them. I hate to say it, but I may have to switch back to ANSYS for modeling since its ability to work with electrostatic forces seems to be much more developed...

Regards,

Mike

Hi

I have also experienced some difficulties with the maxwell stress tensor, but it was mainly because of insufficient mesh density, or as important assymetric meshing, to catch the field gradients correctly, and that this method only works if all the sourrounding area is "air/vacuum" or some low mu material. see also
www.comsol.com/community/exchange/71/

The virtual work approach is more precise, but slightly heavier to implement.

Finally, I havent noticed any discrepancies in harmonic analysis so far, only in eigenmode analysis typically for structural, as I said above, but this is clearly due to a normalisation issue and as such is no "bug". There is a couple of other discussion on this and how to renormalise (if your model is light enough to be exported to matlab in "full"), but I do not believe this normalsation would show up in a harmonic study. But I haven't done any systematic tests of ANSYS versus Comsol on such a case, and we have given up the use of the former.

On the other hand are we sure that ANSYS is right ? Do you clearly know ALL the underaying hypothesis ?

I have done quire some comparisons between NASTRAN and COMSOL but for thermal and structural, apart for the same eigenfrequency normalisation, we have found no significant difference, therefore we are now running only Comsol.

Anyhow, when it comes to too specific issues as what you are referring to, I use COMSOL "support", they are doing an excellent job.

Good luck
Ivar

Hi Ivar,

Thanks for your insight. And don't get me wrong, I would definitely like to stay with Comsol, but I have been struggling with this problem for a while now and have yet to find a solution. Comsol support has been quick to respond with suggestions, but none of their solutions have seemed to give me accurate results.

Regarding the Maxwell Stress Tensor, I have looked at your model and read through all of the relevant discussions in this forum to ensure that my model is set up correctly. I have ran convergence studies to ensure the mesh density is fine enough (it converges, but to the wrong solution). I have ensured a symmetric mesh because my model is a membrane that employs a symmetry boundary condition. I believe my issue is a combination of two factors:

1: It seems that all electrostatic force examples simply apply the force between two electrodes with no internal dielectrics other than a vacuum. In my case, I have a vacuum gap that is sandwiched between 2 nitride isolation layers, and the electrodes are on either side of this dielectric stack. Thus the question arose as to where to apply the force...on the electrode internal boundary, or on the nitride surrounding the air gap. Since the stress tensor only works if the sub-domain is surrounded by air, it seems I am forced to apply it to the nitride surrounding the air gap, but then the effective gap is reduced and the electrostatic force is greater than it should be. Comsol support suggested that I simply integrate the tensor over the air-gap domain and then apply the negative of that force to the surrounding nitride boundaries. This method seemed to give more reasonable results, but I am still not convinced that this is the correct approach.

2: My membrane is not surrounded by air. And although for pull-in calculations I could easily try to surround it with an air domain, this will not be feasible for my other models that require the membrane to radiate pressure waves into a fluid. Thus, it seems I might need an alternative method for calculating the electrostatic force if this is the reason that my model is not correct. I had considered using the virtual work method, but was unsure from the Comsol documentation whether this approach also required that the surrounding area be a vacuum or air. Perhaps you could shed some light on this...

I have contacted Comsol support again to see if they have any other suggestions for me. I will post any information that proves useful.

Finally, regarding the harmonic analysis magnitudes. It is the strangest thing. The overall frequency response behavior is nearly identical between ANSYS and Comsol (as it should be), but as mentioned earlier, the magnitudes (i.e. pressure values) are much higher in Comsol. Most of our ANSYS simulations have been compared with experimental results, and this is why I am using it as a benchmark. ...But I am with you on this one, I see no reason why the magnitudes should differ between the 2 programs. However, I have no explanation for this discrepancy.

Thanks again for offering your thoughts on this issue. Any other suggestions/comments would be appreciated.

Mike

Hi again

Well from my knowledge the virtual work is rather universal and does not require any low ur surroundings as the current Maxwell tensor case. I have also understood that COMSOL is improving on these ACDC Force calculation approaches for their new version 4, to come soon, I hope they will manage.

On the other side, I have also a few request for MEMS simulation waiting for an opening in my availability schedule, and I have already decided to go for the virtual work appraoch for these cases, but currently I'm blocked on traditional large structural, thermal, PZT and wind load study for several weeks.

Will report back if I find out more.

By the way, I have noticed that when I get stuck in a multi-phsics case, I often find a solution by restarting the issue with a very simple model with one then two.. physics and building it up in parallel, analytically (often with help from Maple and MaplSim) and with Comsol.
My second source of inspiration is the forum, by searching to understand the questions and playing with comsol on these questis I have learned quite a lot

Good luck, and doot give up to quickly :)

Ivar

Hi Ivar,

In response to your last post, I found a reference in the AC/DC user's guide that seems to imply that even the virtual work method requires the domain to be surrounded by air. If you look on page 62, it says:

"...Therefore you are often limited to compute the total force by integrating the stress tensor or using the method of virtual work. Note that none of these methods allows you to compute and visualize the force distribution inside a domain, but only to compute the total force and torque in situations where the device is surrounded by air (or when this is a good approximation)"

Your thoughts?

Also, are there any model examples where the virtual work method calculates the force and then applies that to the structure? I am trying to do a parametric sweep of increasing biases until the membrane collapses, so I need to apply the resultant force to my structure somehow with each parameter step. The only available models seem to calculate electrostatic force after the solver is finished. Furthermore, it is always calculated as a total force rather than a force density as with the Maxwell stress tensor. Thus, I am also confused as to how this force can be applied appropriately to a particular electrode surface.

Thanks in advance,

Mike

Hi

yes, I see, but I thought this applied essentiually to the MAxwell stress tensor ? could be to both then.

I would propose that you send a request to comsol support for this issue a its rather generic and important

would need some testing to find out

Good luck
Ivar