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 The information provided may be out of date.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.
ver. 3.5a, When to know when a Solution is Valid?
Posted Jan 5, 2011, 10:46 p.m. EST Studies & Solvers, Structural Mechanics Version 3.5a 7 Replies
Please login with a confirmed email address before reporting spam
Hello,
I appologize in advance for the lengthy question, but I feel it necessary to illustrate my concern.
I've been working with a 3D model in ver. 3.5a without the MEMS application mode which utilizes the following applications: solid, stress-strain (smsld), electrostatics (es) and moving mesh (ale).
This model is an off-shoot of the 3D Cantilever beam. A very thin membrane (~0.34 nm) is suspended by two nodes above an air gap (250 nm in depth) which is above a substrate (depth = 50 nm) above a surface where I apply a swept voltage (1,45) V. The model is 7.2 microns in length (however deflection membrane is only 3.21 microns in length).
I've been told before that my aspect ratio (about 20,000) is much too large for this specific combination of Physics applications, however, I've compiled a model which solves.
Here is my concern: I previously had difficulty solving past 22 Volts, as the solution would quit there. I had used the following solver settings (usually with a quad mesh):
es, ale: Direct (SPOOLES)
smsld: Conjugate gradients w/ Algebraic Multigrid
-or-
es, ale, smsld: Direct (SPOOLES)
[note: I was instructed once by support to us Conj. gradients for the (es) application, however, this does not work and immediately gives the error upon attempting to solve: zero on diagonal of matrix.]
The problem is that I need to solve to at least 30 V, as specifically required by my research advisor. Now, I went back and changed this model in the following two ways:
1) I made subdomains 1,2, 9 and 14 (1,2,14 are substrate regions) "Active in this domain" in the solid, stress-strain Physics Subdomain settings, whereas before only subdomain 9 was active. [note: this change then caused boundaries 1-5, 7,8,10,18,19,21,22, 24,25,51,53-56, 58,70 and 71 to become active, which I made all of them fixed (I allowed boundaries 32, 33, 36 and 42 to remain free because they are the thin membrane boundaries which must deflect)].
2) In the moving mesh subdomain settings, I set subdomains 3-8, 10-13, 15 and 16 to "No Displacement" (which are all "air" subdomains), whereas before all of these subdomains were kept at "Free displacement." [Note: this caused all of the moving mesh boundaries to be grayed-out in the boundary settings dialog box, whereas before over half of the boundaries where not grayed-out, where I had set them all to dx = dy = dz = 0, no displacement, except for the thin membrane boundaries which were held at dx = u, dy = v, dz = w, physics induced displacement)].
The first step seemed logical because the substrate blocks should remain fixed for the second step seemed to free up memory for this model.
These two changes allow me to solve past 100 V (applied to conductor at the bottom of the model). These changes also changed the solver settings, so that instead of specifying a different solver for each application, I'm only able to specify one linear system solver for the parametric solution, which I chose to be Direct (SPOOLES).
--------------------------
My question is this: with these changes, is the solution that I acquire valid for this theoretical model? It is important to know whether these changes have compromised the integrity of the theory behind the model (which is like I said an off-shoot of the 3D cantilever beam with an extreme aspect ratio).
One thing to note, I've solved the same model parameters in 2D (same model length of membrane = 3.21, air = 250 nm thick, substrate = 50 nm thick) and I've received much different results, where, for example, maximum deflection of the membrane has the following values:
2D model
V = 33 V
D_max = 128.5 nm
3D model
V = 33 V
D_max = 102.5 nm
Lastly, to note, the 2D model fails at 33.9 Volts while the 3D model goes past 100 Volts, which is certainly peculiar and worrisome. [Note: The changes made in Steps 1 and 2 for the 3D model seem to agree with how the 2D model was constructed].
Please advise whether Steps 1 and 2 provide an accurate or faulty solution, and any other changes that may be necessary to ensure an accurate solution. I've used a triangular mesh for this particular model, using a quad mesh doesn't seem to help.
**I am unable to attach either the 2D or 3D models to this thread as the file sizes are too large (~9 Mb), only allowed (mph) files can be attached, so I can't attach zipped files. Is there any other way to share these files with the discussion board?
Thank you for your help and time.
Best regards,
Kevin Myhro
I appologize in advance for the lengthy question, but I feel it necessary to illustrate my concern.
I've been working with a 3D model in ver. 3.5a without the MEMS application mode which utilizes the following applications: solid, stress-strain (smsld), electrostatics (es) and moving mesh (ale).
This model is an off-shoot of the 3D Cantilever beam. A very thin membrane (~0.34 nm) is suspended by two nodes above an air gap (250 nm in depth) which is above a substrate (depth = 50 nm) above a surface where I apply a swept voltage (1,45) V. The model is 7.2 microns in length (however deflection membrane is only 3.21 microns in length).
I've been told before that my aspect ratio (about 20,000) is much too large for this specific combination of Physics applications, however, I've compiled a model which solves.
Here is my concern: I previously had difficulty solving past 22 Volts, as the solution would quit there. I had used the following solver settings (usually with a quad mesh):
es, ale: Direct (SPOOLES)
smsld: Conjugate gradients w/ Algebraic Multigrid
-or-
es, ale, smsld: Direct (SPOOLES)
[note: I was instructed once by support to us Conj. gradients for the (es) application, however, this does not work and immediately gives the error upon attempting to solve: zero on diagonal of matrix.]
The problem is that I need to solve to at least 30 V, as specifically required by my research advisor. Now, I went back and changed this model in the following two ways:
1) I made subdomains 1,2, 9 and 14 (1,2,14 are substrate regions) "Active in this domain" in the solid, stress-strain Physics Subdomain settings, whereas before only subdomain 9 was active. [note: this change then caused boundaries 1-5, 7,8,10,18,19,21,22, 24,25,51,53-56, 58,70 and 71 to become active, which I made all of them fixed (I allowed boundaries 32, 33, 36 and 42 to remain free because they are the thin membrane boundaries which must deflect)].
2) In the moving mesh subdomain settings, I set subdomains 3-8, 10-13, 15 and 16 to "No Displacement" (which are all "air" subdomains), whereas before all of these subdomains were kept at "Free displacement." [Note: this caused all of the moving mesh boundaries to be grayed-out in the boundary settings dialog box, whereas before over half of the boundaries where not grayed-out, where I had set them all to dx = dy = dz = 0, no displacement, except for the thin membrane boundaries which were held at dx = u, dy = v, dz = w, physics induced displacement)].
The first step seemed logical because the substrate blocks should remain fixed for the second step seemed to free up memory for this model.
These two changes allow me to solve past 100 V (applied to conductor at the bottom of the model). These changes also changed the solver settings, so that instead of specifying a different solver for each application, I'm only able to specify one linear system solver for the parametric solution, which I chose to be Direct (SPOOLES).
--------------------------
My question is this: with these changes, is the solution that I acquire valid for this theoretical model? It is important to know whether these changes have compromised the integrity of the theory behind the model (which is like I said an off-shoot of the 3D cantilever beam with an extreme aspect ratio).
One thing to note, I've solved the same model parameters in 2D (same model length of membrane = 3.21, air = 250 nm thick, substrate = 50 nm thick) and I've received much different results, where, for example, maximum deflection of the membrane has the following values:
2D model
V = 33 V
D_max = 128.5 nm
3D model
V = 33 V
D_max = 102.5 nm
Lastly, to note, the 2D model fails at 33.9 Volts while the 3D model goes past 100 Volts, which is certainly peculiar and worrisome. [Note: The changes made in Steps 1 and 2 for the 3D model seem to agree with how the 2D model was constructed].
Please advise whether Steps 1 and 2 provide an accurate or faulty solution, and any other changes that may be necessary to ensure an accurate solution. I've used a triangular mesh for this particular model, using a quad mesh doesn't seem to help.
**I am unable to attach either the 2D or 3D models to this thread as the file sizes are too large (~9 Mb), only allowed (mph) files can be attached, so I can't attach zipped files. Is there any other way to share these files with the discussion board?
Thank you for your help and time.
Best regards,
Kevin Myhro
7 Replies Last Post Jan 16, 2011, 6:14 a.m. EST
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Kevin
I see you are pushing 3.5a far ;)
Now at a certain voltage you should get to the "pull-in" limit and your solution has a bifurcation, no ?
and COMSOL will anyhow have trouble in the vicinity of this point. HAve you checked where the pull-in is for your case ?
For the file transfer, using V4 allows to save the meshing strategy so there is then no need to deposit the full model. In v3.5a you might clear the solution and leave it to us to resolve, this might help on the file size
I'm not sure it helps but if you have rather "small" displacements you might be able to avoid the ALE, mesh the air/vacuum gap as a solid with nu=0, rho=1, E=100 or so and turn on the "reference frame in the physics. This should normally allow you to use the mesh of the cavity for the ACDC. This works fine in V4 where the "reference frame is "n" by default in Structural.
These are just some hints, sorry for not having time to look deeper into your issue, it looks interesting (and challenging)
--
Good luck
Ivar
I see you are pushing 3.5a far ;)
Now at a certain voltage you should get to the "pull-in" limit and your solution has a bifurcation, no ?
and COMSOL will anyhow have trouble in the vicinity of this point. HAve you checked where the pull-in is for your case ?
For the file transfer, using V4 allows to save the meshing strategy so there is then no need to deposit the full model. In v3.5a you might clear the solution and leave it to us to resolve, this might help on the file size
I'm not sure it helps but if you have rather "small" displacements you might be able to avoid the ALE, mesh the air/vacuum gap as a solid with nu=0, rho=1, E=100 or so and turn on the "reference frame in the physics. This should normally allow you to use the mesh of the cavity for the ACDC. This works fine in V4 where the "reference frame is "n" by default in Structural.
These are just some hints, sorry for not having time to look deeper into your issue, it looks interesting (and challenging)
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hey Ivar,
I've learned that my 3D solution was invalid because I had set air subdomains to zero displacement, thereby negating the ALE application. Apparently, solving this model without the moving mesh application in use is only valid in the small deformation approximation. However, I have a thin membrane (0.34 nm) which is deflecting on the order of 100 nm, so indeed, it is a Large deformation (as I've specified in the smsld application physics properties).
So the aforementioned 3D model was invalid (which I had difficulty uploading).
Your advice has been helpful once again: to upload an unsolved model to reduce the file size--- a simple, albeit obvious fix to my problem (which I didn't realize myself).
I'm back to my current dilemna: Solve this 3D model (which is an off-shoot of the 3D cantilever beam with different geometries) to at least 30V, which is imperative as instructed by my research advisor.
I've attached the unsolved 3D model, and I assure you that upon solving it with the given initial conditions it will stop at 23 V.
My question is this: how my I alter my solver settings or mesh to push this solution further to 30V? (as you say, I'm pushing 3.5a as far as I can). I've tried different meshes, including quad meshes with more swept layers, but that usually takes up too much memory (although I understand it aids in convergence). Furthermore, I've tried many different combinations of solvers and I'm having much difficulty.
Thanks again, as always. We appreciate your help, Ivar.
Best regards,
Kevin
I've learned that my 3D solution was invalid because I had set air subdomains to zero displacement, thereby negating the ALE application. Apparently, solving this model without the moving mesh application in use is only valid in the small deformation approximation. However, I have a thin membrane (0.34 nm) which is deflecting on the order of 100 nm, so indeed, it is a Large deformation (as I've specified in the smsld application physics properties).
So the aforementioned 3D model was invalid (which I had difficulty uploading).
Your advice has been helpful once again: to upload an unsolved model to reduce the file size--- a simple, albeit obvious fix to my problem (which I didn't realize myself).
I'm back to my current dilemna: Solve this 3D model (which is an off-shoot of the 3D cantilever beam with different geometries) to at least 30V, which is imperative as instructed by my research advisor.
I've attached the unsolved 3D model, and I assure you that upon solving it with the given initial conditions it will stop at 23 V.
My question is this: how my I alter my solver settings or mesh to push this solution further to 30V? (as you say, I'm pushing 3.5a as far as I can). I've tried different meshes, including quad meshes with more swept layers, but that usually takes up too much memory (although I understand it aids in convergence). Furthermore, I've tried many different combinations of solvers and I'm having much difficulty.
Thanks again, as always. We appreciate your help, Ivar.
Best regards,
Kevin
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
your model is not evident to catch and 3.5 is getting old. I believe I have got the essential, so a few questions:
why simulate all that volume ?, you should have little fringe effects (at least as first order) so why not "just" model your membrane and the air gap underneath, and leave out completely your high epsilon material ?
If this is acceptable, then your model becomes a 2D model, made up by 2 rectangles, the thin one for the film, and the air (possibly a third rectangle if you believe you need the layer underneath as the voltage is applied through this material
Then I would try to turn on the deformed frame in the solid part, and use the deformed frame for the ES and forget about the ALE, one step less. For the meshing I would use at least 3 elements in the thickness of the membrane, if possible a few more.
Now I have no way to validate if this method gives the same results as a full ALE, but so long the air mesh is not getting distorted it should be OK. tst it out on something known to be sure
--
Good luck
Ivar
your model is not evident to catch and 3.5 is getting old. I believe I have got the essential, so a few questions:
why simulate all that volume ?, you should have little fringe effects (at least as first order) so why not "just" model your membrane and the air gap underneath, and leave out completely your high epsilon material ?
If this is acceptable, then your model becomes a 2D model, made up by 2 rectangles, the thin one for the film, and the air (possibly a third rectangle if you believe you need the layer underneath as the voltage is applied through this material
Then I would try to turn on the deformed frame in the solid part, and use the deformed frame for the ES and forget about the ALE, one step less. For the meshing I would use at least 3 elements in the thickness of the membrane, if possible a few more.
Now I have no way to validate if this method gives the same results as a full ALE, but so long the air mesh is not getting distorted it should be OK. tst it out on something known to be sure
--
Good luck
Ivar
Attachments:
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi again ;)
One thing with my rapid approach without ALE, as I have put a rather high E for the "air 1E5, this might interact with the beam, so perhaps one should use 1 or 1e-3, at least check the effect. Because I see only some 40 nm deflection @ 30 V as you were talking about 100 nm. and a pull-in-distance is, if I remember right around 1/3 of the gap, so for you it should be around 100 nm
But what is the pull down voltage ? Have you checked with i.e Analysis and Design Principles of MEMS Devices of Minhang Bao ? isbn 0 444 51616 6 Elsevier, 2005 (see p180-190)
You could also try to simulate in 3D with a shell/plate beam (provided large displacement is there, perhaps not in 3.5)
--
Good luck
Ivar
One thing with my rapid approach without ALE, as I have put a rather high E for the "air 1E5, this might interact with the beam, so perhaps one should use 1 or 1e-3, at least check the effect. Because I see only some 40 nm deflection @ 30 V as you were talking about 100 nm. and a pull-in-distance is, if I remember right around 1/3 of the gap, so for you it should be around 100 nm
But what is the pull down voltage ? Have you checked with i.e Analysis and Design Principles of MEMS Devices of Minhang Bao ? isbn 0 444 51616 6 Elsevier, 2005 (see p180-190)
You could also try to simulate in 3D with a shell/plate beam (provided large displacement is there, perhaps not in 3.5)
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hello Ivar,
I agree that you certainly understand the essentials of my model. Yes, 3.5a is becoming obsolete, but I'm stuck with it until I can convince my research advisor to upgrade to 4.1. To answer your question about why I would simulate all of that volume in the 3D model I showed you, including the small fringe effects, is because my research advisor wants to see a simulation which accounts for these fringe effects with a very detailed 3D geometry. If these fringe effects really are neglible, however, then I can reduce my 3D model to a smaller size, maybe even all the way down to a 2D model (but we'd like to have both 2D and 3D model data).
I've noticed in the model you uploaded you split the model into left and right halves, as mentioned in the months past. Also, you eliminated the ALE application and turned on the deform frame in the ES, as discussed. Also, there is a maximum deformation of about 42 nm at 30 V, as discussed. However, you used a thin film thickness of 3.4 nm, while I require a thin film thickness of 0.34 nm.
I've made the appropriate change (and also used a Parametric solution with UMFPACK instead of a Parametric segregated soln) and now I'm receiving a displacement of 95.7 nm at 23 V which agrees well with prior 2D simulation data (I'm still awaiting the comparison of this simulation data to the real deflection data of the actual MEMS device used in my Condensed Matter research group). I've also used the following Mapped Mesh Parameters:
Boundaries: 1, 8, 15 (mesh elements: 10)
Bound. 2, 4, 6, 7, 9, 11, 13, 14 (elements: 20)
Bound. 3, 10, 16 (elements: 50)
Bound. 5, 12, 17 (elements: 5)
So I'm once again stuck with a problem: This 2D solution stops at 23 V, when using a Young's Modulus of 1 or 1e-3 or 1e-6 (using 1e5 allows solution to run to 26 V, however, this is bad data because the air is resisting the deformation of the thin membrane too much).
I would like to get this 2D model up to 30V, that way I'll have a 2D simulation model to compare with experimental data and maybe I can use what I learned in the 2D model to expand to a working 3D model once again for a more detail solution including edge effects.
Three quick questions:
1) Why did you chose to use a Parametric Segregated solution; is that generally more precise than a typical Parametric solution?
2) You've used Maxwell Stress tensor variables: dnTx2_es, dnTy2_es, while I've always used: dnTEx_es, dnTEy_es. Is there any significant difference between these?
3) In the model you uploaded, you had the thin membrane overlap the air region (so it had a longer length). I adjusted this so that the lengths were equal and so that the subdomains would be flush and allow easier mesh generation. Is there a specific design reason why you chose to overlap the thin membrane like that?
Please take a look at the 2D model that I uploaded (and adjusted from yours) and give advice on how I might be able to now push this solution to 30 V (it is certainly peculiar that both my 2D and 3D models stop at 23 V now).
Thanks for the reference to Min hang Bao's MEMS book. No, I haven't yet cross-checked with that book, but I will look into it. Unfortunately, they do not carry it at any of my University libraries (UC Riverside) so I will have to look elsewhere.
Thanks again for all of your time and help, Ivar. You have certainly been instrumental in helping me better understand COMSOL.
Best regards,
Kevin
I agree that you certainly understand the essentials of my model. Yes, 3.5a is becoming obsolete, but I'm stuck with it until I can convince my research advisor to upgrade to 4.1. To answer your question about why I would simulate all of that volume in the 3D model I showed you, including the small fringe effects, is because my research advisor wants to see a simulation which accounts for these fringe effects with a very detailed 3D geometry. If these fringe effects really are neglible, however, then I can reduce my 3D model to a smaller size, maybe even all the way down to a 2D model (but we'd like to have both 2D and 3D model data).
I've noticed in the model you uploaded you split the model into left and right halves, as mentioned in the months past. Also, you eliminated the ALE application and turned on the deform frame in the ES, as discussed. Also, there is a maximum deformation of about 42 nm at 30 V, as discussed. However, you used a thin film thickness of 3.4 nm, while I require a thin film thickness of 0.34 nm.
I've made the appropriate change (and also used a Parametric solution with UMFPACK instead of a Parametric segregated soln) and now I'm receiving a displacement of 95.7 nm at 23 V which agrees well with prior 2D simulation data (I'm still awaiting the comparison of this simulation data to the real deflection data of the actual MEMS device used in my Condensed Matter research group). I've also used the following Mapped Mesh Parameters:
Boundaries: 1, 8, 15 (mesh elements: 10)
Bound. 2, 4, 6, 7, 9, 11, 13, 14 (elements: 20)
Bound. 3, 10, 16 (elements: 50)
Bound. 5, 12, 17 (elements: 5)
So I'm once again stuck with a problem: This 2D solution stops at 23 V, when using a Young's Modulus of 1 or 1e-3 or 1e-6 (using 1e5 allows solution to run to 26 V, however, this is bad data because the air is resisting the deformation of the thin membrane too much).
I would like to get this 2D model up to 30V, that way I'll have a 2D simulation model to compare with experimental data and maybe I can use what I learned in the 2D model to expand to a working 3D model once again for a more detail solution including edge effects.
Three quick questions:
1) Why did you chose to use a Parametric Segregated solution; is that generally more precise than a typical Parametric solution?
2) You've used Maxwell Stress tensor variables: dnTx2_es, dnTy2_es, while I've always used: dnTEx_es, dnTEy_es. Is there any significant difference between these?
3) In the model you uploaded, you had the thin membrane overlap the air region (so it had a longer length). I adjusted this so that the lengths were equal and so that the subdomains would be flush and allow easier mesh generation. Is there a specific design reason why you chose to overlap the thin membrane like that?
Please take a look at the 2D model that I uploaded (and adjusted from yours) and give advice on how I might be able to now push this solution to 30 V (it is certainly peculiar that both my 2D and 3D models stop at 23 V now).
Thanks for the reference to Min hang Bao's MEMS book. No, I haven't yet cross-checked with that book, but I will look into it. Unfortunately, they do not carry it at any of my University libraries (UC Riverside) so I will have to look elsewhere.
Thanks again for all of your time and help, Ivar. You have certainly been instrumental in helping me better understand COMSOL.
Best regards,
Kevin
Attachments:
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
Well first of all you are lining up all the good reasons why your supervisor should invest in 4.1 ;) But on the other side, if I were him, I would also insist on doing it on current verssion, as in most places in th industry, they are lagging far after and have seldom the "last" varsion (I'm, a lucky one though). And the trend of a learning process is to managed even with scarese resources.
Now this said, yes I have missed your beam thickness by a factor 10 ! a typo from my side, that is why I get a far smaller deflection, and that I'm still stable at 30V. But that is easy to change, no ?
Then I split in the middle bacause your model is fully symmetric, in this way I have a line/points to use for plotting, and postprocessing, I also force nodes on the line (by its presence) so my mesh is more symmetric, hence the results slightly more symmetric too, and I can try to model only the halve in 2D, or 1/4 in 3D (note if you are looking at the eigenfrequencies for a 1/2 model you need to do 2 runs for symmetric and antisymmetric BC's and in 3D 1/4 model you need 4 runs (2^3 for 1/8 model) for all the combinations of sym-antisym BC's. Therefore I mostly use full models for eignfrequencies. PS in the V4.1 you can normalise the eigenfrequenceies per "mass participation factors" which is important in MEMS systems where one have often many modes, with only a few higher modes of real importance (another item for you wish list) in 3.5a you need to renormalise the eigenvectors in matlab, but for a "real" model often the vector and matrices ecome too large, atleast for my models matlab crashed systematically. So now for x,y,z in 4.1 you have it internally in COMSOL, but not yet for Rx,Ry,Rz.
Your model looks very symmetric, you should be able to model 1/4 in 3D I believe, and get all fringe effects. You should perhaps also check the true size of the surroundings, yours are perhaps somwhat too large, or have you done it ?
Then you are finally probably right (because of the large deformations) too that you need the ALE then the "mesh stiffness" is not changing your physics, for small deformations this "deformation frame of "air" is working as good, and is quicker to set up.
For the mesh, if you like structured mesh, OK to use Quads (easy for such cubic structures). But you should check if Quad or Thets (obtained by converting your quads) is solving quicker, as the matrices might be easier to diagonalise in thets, there I'm not sure it makes a large difference, to be checked
For the maxwell tensor names isnt the 2 coming from the dependent variable name, I might have use 2 application modes and suppressed one, I stole the name from the "equation" as I'm forgetting these names from v3.5a, so I need to search them. No other reason
For tha air differece, it might have been again a typo in the block building, that I missed.
I will have a look at your model, but I'm rather convinced that your 23V represents the "pull-down/pull-in instability limit", specially if you are at a deformation of about 1/3 of the total air gap.
Have you tried things like
range(0,1,22) range(22.1,0.1,25)
or stop and restart your model, with the sequencer ?
--
Good luck
Ivar
Well first of all you are lining up all the good reasons why your supervisor should invest in 4.1 ;) But on the other side, if I were him, I would also insist on doing it on current verssion, as in most places in th industry, they are lagging far after and have seldom the "last" varsion (I'm, a lucky one though). And the trend of a learning process is to managed even with scarese resources.
Now this said, yes I have missed your beam thickness by a factor 10 ! a typo from my side, that is why I get a far smaller deflection, and that I'm still stable at 30V. But that is easy to change, no ?
Then I split in the middle bacause your model is fully symmetric, in this way I have a line/points to use for plotting, and postprocessing, I also force nodes on the line (by its presence) so my mesh is more symmetric, hence the results slightly more symmetric too, and I can try to model only the halve in 2D, or 1/4 in 3D (note if you are looking at the eigenfrequencies for a 1/2 model you need to do 2 runs for symmetric and antisymmetric BC's and in 3D 1/4 model you need 4 runs (2^3 for 1/8 model) for all the combinations of sym-antisym BC's. Therefore I mostly use full models for eignfrequencies. PS in the V4.1 you can normalise the eigenfrequenceies per "mass participation factors" which is important in MEMS systems where one have often many modes, with only a few higher modes of real importance (another item for you wish list) in 3.5a you need to renormalise the eigenvectors in matlab, but for a "real" model often the vector and matrices ecome too large, atleast for my models matlab crashed systematically. So now for x,y,z in 4.1 you have it internally in COMSOL, but not yet for Rx,Ry,Rz.
Your model looks very symmetric, you should be able to model 1/4 in 3D I believe, and get all fringe effects. You should perhaps also check the true size of the surroundings, yours are perhaps somwhat too large, or have you done it ?
Then you are finally probably right (because of the large deformations) too that you need the ALE then the "mesh stiffness" is not changing your physics, for small deformations this "deformation frame of "air" is working as good, and is quicker to set up.
For the mesh, if you like structured mesh, OK to use Quads (easy for such cubic structures). But you should check if Quad or Thets (obtained by converting your quads) is solving quicker, as the matrices might be easier to diagonalise in thets, there I'm not sure it makes a large difference, to be checked
For the maxwell tensor names isnt the 2 coming from the dependent variable name, I might have use 2 application modes and suppressed one, I stole the name from the "equation" as I'm forgetting these names from v3.5a, so I need to search them. No other reason
For tha air differece, it might have been again a typo in the block building, that I missed.
I will have a look at your model, but I'm rather convinced that your 23V represents the "pull-down/pull-in instability limit", specially if you are at a deformation of about 1/3 of the total air gap.
Have you tried things like
range(0,1,22) range(22.1,0.1,25)
or stop and restart your model, with the sequencer ?
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi back again.
Indeed your model works to 23.3 I'm rather convinced you are at the pull-in limit, above the voltage force is such that it runs off.
One way is to load your membrane with a pressure (parametric) ONLY and calculate the V field and then compare the Es force to the pressure load force (by leaving the physics uncoupled). Then you should see the crossing of the two forces. Then you can also check the "air" forces to see how much the air laods the membrane and changes the results, versus pure ALE changes your model. Another point, with ALE you can use the full remesh functionality.
makeit a try in 2D, then in 3D
--
Good luck
Ivar
Indeed your model works to 23.3 I'm rather convinced you are at the pull-in limit, above the voltage force is such that it runs off.
One way is to load your membrane with a pressure (parametric) ONLY and calculate the V field and then compare the Es force to the pressure load force (by leaving the physics uncoupled). Then you should see the crossing of the two forces. Then you can also check the "air" forces to see how much the air laods the membrane and changes the results, versus pure ALE changes your model. Another point, with ALE you can use the full remesh functionality.
makeit a try in 2D, then in 3D
--
Good luck
Ivar