Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

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.

Singular matrice associated with moving mesh

Please login with a confirmed email address before reporting spam

Hello everybody,

first, I am very new to Consol :)

I built a 2D-model of a silicon membrane. I defined a point load "F" that stresses the membrane which gets deformed subsequently. An electrode under the membrane forms a capacitor together with a ground-electrode. Deformation of the membrane results in a change of the capacitance.

What I want to do is solving for the value of capacitance with a parametric solver and varying "F".

I get a solution for the deflection of the membrane so far. But the capacitance stays constant and doesn't change for different "F".
So I read I have to use a moving mesh, that allows Comsol to compute the electric field for a changing geometry.

I tried to upgrade my model with such a moving mesh, but I can't get a solution. I used values "u" and "v", that describe the deflection of the membrane, for the prescribed deformation of the moving mesh. But all I get is an error message, that tells me something about a singular matrix together with the variables "u" and "v"... Tried hard, but no idea.

Perhaps somebody has some minutes to check my model. It's Comsol 4.

Thanks a lot in advance!
Chris


7 Replies Last Post 2011年1月4日 GMT-5 05:28
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 2011年1月3日 GMT-5 02:25
Hi

I'm not sure if it is some limitation of v4.0 (I have the latest 4.1 that is much smoother).

The only change I'm doing is to group your 3 solver cases into 1 direct solver sequence (you can also use a segregated one as the solid is driving the ale and then the electrostatic) and it solves OK.

I cannot really see any error, even if I agree the displacement seems blocked to the last laod case but I suspect it lis linked to the solver sequence you have selected, use rather one sequence and possibly segregated sequences.
By the way in v4 you do not really need ALE for such a case, you can mesh the "air" as a solid with nu=0, rho=1 and E=1E5 which is about the stiffness of "air". This is because in v4 the default is that solid deformation is driving the spatial frame (in 3.5 you needed to turn the "ref frame" on in the physics to do this)

For some reason (most pobably linked to 4.0 versus 4.1 I cannot calculate the capacitance I get a strange error message, this is not the case for freshly v4.1 build models.

So my recommandations are : get hand on the latest version ;) (call your rep if you do not already have the CD)
and so far as possible solve it in 1 step with all 3 physics (or add a segregfated step manually)

One thing though, if I calculate the capacitance of a 20um gap, 3.8mm^2 area capacitance I get half the one given here by COMSOL

C = 8.854E-12[F/m]*1*(3.8[mm])^2/20[um] = 6.4[nF] for me (8.8E-2 is "epsilon0_const" in COMSOL

I have remade a simular case in v4.1 hereby, where I get a better capacitive value (closer to the above value). Do not forget to define the thickness of the solid volume, I have assumed 10mm


--
Good luck
Ivar
Hi I'm not sure if it is some limitation of v4.0 (I have the latest 4.1 that is much smoother). The only change I'm doing is to group your 3 solver cases into 1 direct solver sequence (you can also use a segregated one as the solid is driving the ale and then the electrostatic) and it solves OK. I cannot really see any error, even if I agree the displacement seems blocked to the last laod case but I suspect it lis linked to the solver sequence you have selected, use rather one sequence and possibly segregated sequences. By the way in v4 you do not really need ALE for such a case, you can mesh the "air" as a solid with nu=0, rho=1 and E=1E5 which is about the stiffness of "air". This is because in v4 the default is that solid deformation is driving the spatial frame (in 3.5 you needed to turn the "ref frame" on in the physics to do this) For some reason (most pobably linked to 4.0 versus 4.1 I cannot calculate the capacitance I get a strange error message, this is not the case for freshly v4.1 build models. So my recommandations are : get hand on the latest version ;) (call your rep if you do not already have the CD) and so far as possible solve it in 1 step with all 3 physics (or add a segregfated step manually) One thing though, if I calculate the capacitance of a 20um gap, 3.8mm^2 area capacitance I get half the one given here by COMSOL C = 8.854E-12[F/m]*1*(3.8[mm])^2/20[um] = 6.4[nF] for me (8.8E-2 is "epsilon0_const" in COMSOL I have remade a simular case in v4.1 hereby, where I get a better capacitive value (closer to the above value). Do not forget to define the thickness of the solid volume, I have assumed 10mm -- Good luck Ivar


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 2011年1月3日 GMT-5 07:00
Hi Ivar,

I meshed the space between the electrodes as air and combined all steps in one solver sequence - and it works! Thanks a ot for this advice!! I get suitable values for the capacitance. For no membrane load and a subsequent d=20µm Comsol gives a value of about 6.4pF. This value you calculated analytically I see. So far so good. But some more questions:

1.) The electrodes of the capacitor shall be circular shaped, not square. Is there any way of doing this in 2D? Or does anybody have an idea how to transform the solution for the square-shaped electrodes into one for the circular-shaped?
2.) I don't understand why in some Comsol-examples solving is done in several steps if it also works in one combined step as you proposed?

Chris
Hi Ivar, I meshed the space between the electrodes as air and combined all steps in one solver sequence - and it works! Thanks a ot for this advice!! I get suitable values for the capacitance. For no membrane load and a subsequent d=20µm Comsol gives a value of about 6.4pF. This value you calculated analytically I see. So far so good. But some more questions: 1.) The electrodes of the capacitor shall be circular shaped, not square. Is there any way of doing this in 2D? Or does anybody have an idea how to transform the solution for the square-shaped electrodes into one for the circular-shaped? 2.) I don't understand why in some Comsol-examples solving is done in several steps if it also works in one combined step as you proposed? Chris

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 2011年1月3日 GMT-5 07:18
Hi

in 2D-axi you can solve a circular shape, or you need to go to 3D, but that is very computational expensive.

By the way try to avoid "point" loads, use rather small area/boundaries instead.

For the multiple steps, its always cheaper, computation and timewise to segregate the variables (solve one after the other sequentially, thenn loop around) as the global model matrix to resolve is smaller for each step. But this means that your physics is such hat you might decide a driver/master and a slave physics. In your case it's easy, the load is driving the structure thant deformes the gap, that gives an electric ACDC physics change, so you can gain (time and sometimes convergence + less frustrations) by segregating your solver sequ3ence into this ordr. But you must decide and do it manually, COMSOL cannot guess everything (luckily we remain still, the masters ;)

--
Good luck
Ivar
Hi in 2D-axi you can solve a circular shape, or you need to go to 3D, but that is very computational expensive. By the way try to avoid "point" loads, use rather small area/boundaries instead. For the multiple steps, its always cheaper, computation and timewise to segregate the variables (solve one after the other sequentially, thenn loop around) as the global model matrix to resolve is smaller for each step. But this means that your physics is such hat you might decide a driver/master and a slave physics. In your case it's easy, the load is driving the structure thant deformes the gap, that gives an electric ACDC physics change, so you can gain (time and sometimes convergence + less frustrations) by segregating your solver sequ3ence into this ordr. But you must decide and do it manually, COMSOL cannot guess everything (luckily we remain still, the masters ;) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 2011年1月3日 GMT-5 09:00
Hi,

by now I don't want to go to 2D if not necessary. But I don't understand how I can tell Comsol that the electrodes should be circular shaped within my existing geometry?

Chris
Hi, by now I don't want to go to 2D if not necessary. But I don't understand how I can tell Comsol that the electrodes should be circular shaped within my existing geometry? Chris

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 2011年1月3日 GMT-5 09:36
Hi

in standard 2D the shape is normally infinite, or per 1[m] in the depth "z" or out of paper direction. So you have only the 2D axi method, then you solve even half the 2D model it's even simpler.

Only one thing, in 2D you need to define the depth if you want absolute values (as else they are all per [m] depth (slightly physics dependent), while in 2D axi you need to usee the "loop length" = 2*pi*r to go from line values to surface values in the cylindrical axi-symmetric view. (note that asymetric loading cannot be obtained in 2D-axi.

It's a question of habit and it is so much easier in 2D than in 3D (solver much quicker, witht some limitation as always ;)

--
Good luck
Ivar
Hi in standard 2D the shape is normally infinite, or per 1[m] in the depth "z" or out of paper direction. So you have only the 2D axi method, then you solve even half the 2D model it's even simpler. Only one thing, in 2D you need to define the depth if you want absolute values (as else they are all per [m] depth (slightly physics dependent), while in 2D axi you need to usee the "loop length" = 2*pi*r to go from line values to surface values in the cylindrical axi-symmetric view. (note that asymetric loading cannot be obtained in 2D-axi. It's a question of habit and it is so much easier in 2D than in 3D (solver much quicker, witht some limitation as always ;) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 2011年1月3日 GMT-5 16:07
First: thanks a lot for your time and efforts. That's realy worth admiring!

But I'm sorry, I've still got no real imagination of how to create a cicular electrode while using 2D. Perhaps you might have a small example? Still using 4.0 for the moment :)

Best regards
Chris
First: thanks a lot for your time and efforts. That's realy worth admiring! But I'm sorry, I've still got no real imagination of how to create a cicular electrode while using 2D. Perhaps you might have a small example? Still using 4.0 for the moment :) Best regards Chris

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 2011年1月4日 GMT-5 05:28
Hi

I have made a simple 4.0a case in 2D axi hereby, hope it helps

--
Good luck
Ivar
Hi I have made a simple 4.0a case in 2D axi hereby, hope it helps -- 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.