Mesh - Latest Discussions - COMSOL Forums

Meshing of Disproportionate Geometries

Thu, 09 Feb 2012 16:01:13 +0000

Hi,

Small structures are often disproportionate Geometries with orders of magnitude difference in height to width. E.g. There is a thin film 10nm high (y) and 0.5cm(x) long.

There was a thread on a similar topic in 3D which suggested the use of sweep meshing.
http://www.comsol.de/community/forums/mesh/thread/19108/
However I am usually working in 2D since that simplifies the model and there this isn't possible.
In the upper example when creating a standard normal mesh I get very many meshpoints even when there are only 4 "layers" of mesh points. So perhaps longer a meshpoint could be longer in x direction and smaller in y direction. However how does this influence the accuracy of the simulation.

I also found some other approaches used.
(1) "scaling the problem": E.g. in an electric only problem: A diode which is only about 100nm thick is scaled to 1 or 10um with 10 or 100 times less resistance.

(2) "cutting the problem" Since sometimes most of the lenght of the "layer" is not important another geometry is created which just simulates the interesting part and the other part is taken with a boundary condition in account.

Especially I am looking for some literature about handling handling disproportionate kind of geometries in kind of meshing. Perhaps were such (simple) methods as the upper 2 are used and put on a scientific ground.

Cheers
Robert

edit:
I found a thin film paper which describes the problem and the solution (1) as follows:
http://de.arxiv.org/abs/1001.4447v2
To confirm the scenario discussed above, we calculated
the spatial distribution of the electrical potential and the
current in ring-shaped MTJs using finite element method
(FEM) simulations. The geometry of the ring-shaped junc-
tions is shown in Fig. 6. The in-plane conductivities used in
the simulations for the various layers of the junction stack
are summarized in Table 3. Since the simulation tool did
not allow to represent the real dimensions of the sample be-
cause of its large aspect ratio we scaled the geometry and
the conductivity in the out-of-plane direction by a factor of
100.

3D Swept mesh for Thin domain with ununiform faces

Mon, 06 Feb 2012 18:48:47 +0000

how should i mesh the thin SS layer by swept mesh ... i dont get it at all .. ! even after reading the documentation ... the problem is because of un even faces ..

3d - Rf Simulation: PML mesh

Mon, 06 Feb 2012 07:52:50 +0000

Hi people,
I must simulate a block of lossy material (10x10x15cm) in the space and a plane wave that hits it. This is the extention of the problem I had posted some time ago ("http://www.comsol.com/community/forums/general/thread/24889/").
With your help I have set the space around the box 1.5 lambda times thick and the PML zone wide 1 lambda.
Now the new question is:
how make the PML mesh?
How many plane waves can I set toghether? Their experession become very big: is possible to semplify thei expression?

*****
My settings are:
for the electric field (in Electromagnetics waves tab)

KEx*exp(i*emw.k0*(VerNx*x+VerNy*y+VerNz*z))
KEy*exp(i*emw.k0*(VerNx*x+VerNy*y+VerNz*z))
KEz*exp(i*emw.k0*(VerNx*x+VerNy*y+VerNz*z))

and with this parameters:

Fi 0 Longitude
Teta 0 Latitude
Alfa 0 polarization
Emax 1000 Electric field amplitude
GrToRad pi/180 Degrees to radiant conversion
KEn Emax*cos(GrToRad*Alfa) Normal field
KEt Emax*sin(GrToRad*Alfa) Tangential field
KEx KEt*sin(GrToRad*Fi)-KEn*cos(GrToRad*Teta)*cos(GrToRad*Fi)
KEy -KEt*cos(GrToRad*Fi)-KEn*cos(GrToRad*Teta)*sin(GrToRad*Fi)
KEz KEn*sin(GrToRad*Teta)
VerNx cos(GrToRad*Fi)*sin(GrToRad*Teta)
VerNy sin(GrToRad*Fi)*sin(GrToRad*Teta)
VerNz cos(GrToRad*Teta)
*****

Thank you for attention.

Bye, Ale.

Meshing 3D geometry

Sun, 05 Feb 2012 19:53:40 +0000

Hello,

I am trying to figure out the correct way to mesh a 3D geometry that consisted of a chamber containing several cells as shown in the attached figure. When I applied the free tetrahedral mesh to the entire geometry I was able to model the diffusion of my analyte of interest but my domain point probe plot do not show the concentration vs. time for the two points I have set as 60 um above the bottom of the chamber (hence on the same level as the cells). This makes me think that meshing is just a surface mesh and not a volume mesh as evident by a very low mesh volume in the statistic menu.

Any help is much appreciated,
CDuong

compressing parallel plate capacitor and measuring capacitance

Sat, 04 Feb 2012 23:23:37 +0000

Hi all,

I have a simple parallel plate structure with a rubber material in between. I want to compress the structure and measure the capacitance as it is compressed. I can successfully calculate the capacitance using the electrostatics physics, but unsure how I can add a compression force (keeping one plate fixed and applying a force on the other) and see the deformation and calculate the capacitance again.

I read in the forums that I need to use ALE to couple the two physics together. I am not sure how I can do that. Is there a tutorial somewhere that I can look at?

Droplet breakup in a T-Junction in a smaller channel

Sat, 04 Feb 2012 23:09:38 +0000

Hi all,

I have a specific set of parameters for wich I am trying to simulate the breakup of droplets in a T-junction. Entering these parameters for the geometry from the example MEMS module[1] gives fine results (droplets form).
However, if I slash the channel size in half (50x50x20 mum) I get no droplets but a coflow regime, where the dispersed phase occupies the upper part of the channel.
Now I have a real life system that has exactly the same dimensions (50x50x20 mum) and I can generate droplets for the same set of parameters in that.
So, I am somewhat stumped on how to proceed from here. I tried doubling the meshsize and increased the timeframe of the simulation and even fiddled around with the parameters, but it always gives me the cowflow regime.

Any help appreciated.

[1] The geometry in the example module is a 100x100x50 mum channel.

Count Elements on Surface

Fri, 03 Feb 2012 16:48:40 +0000

When meshing, comsol spits out the total number of elements contained in the mesh. Is there any way to extract the number of elements on a single surface? I tried to look into the post-processing "Derived Values" section but couldn't find anything.

"Using linear geometry shape order for mesh element near coordinates..."

Fri, 03 Feb 2012 16:00:58 +0000

What does it mean when I get dozens of messages saying "Using linear geometry shape order for mesh element near coordinates...". Is there something wrong or inefficient about my meshing or my geometry? I'm looking for ways to optimize it.

working with Spice software

Thu, 02 Feb 2012 16:51:23 +0000

Hi

I am new to Comsol and would like to model and design some 3D components like transformers and inductors in AC/DC module. However, I would eventually want to simulate the full circuit through importing the circuit from the SPICE program for full sensitivity and performance analysis. I have few questions here and would appreciate if you could answer them.

(1) I am going to use two different versions of spice program here, initially using the free LTSpice from Linear Technology and later PSPICE from Cadence. I would like to check if COMSOL could support both version of Spice?

(2) Does Comsol work by importing the circuit information from the SPICE software through the *.cir file? Do both versions of the Spice programs able to generate the *.cir format? Does Comsol working with other format or will it prefomance the same function?

(3) Is Comsol only import the circuit network layout and components information from the SPICE and have an internal solver to solve the complete circuit entirely or it will also need to send command to use the SPICE solver for final solutions?

(4) How well Comsol support the Spice program, particularly the Cadence PSpice? Does it support all components and parameters such as coupling coefficient, k from the SPICE? Any limitation when working with these two programs.

Thanks in advance,
Chek Yang

Problem with the meshing of a structure

Thu, 02 Feb 2012 15:52:49 +0000

I am trying to mesh a structure. I have a block and objects inside that block. The blocks gets meshed insanly coarse, while the objects inside get meshed rather fine. The block has elements that are in the size of it's diameter, this just cannot be. Where do I make a mistake.

Prescribed mesh velocity

Wed, 01 Feb 2012 07:35:28 +0000

Hi everyone,

in a prescribed mesh velocity, if i want that a boundary has a free prescribed mesh velocity in x direction i have two options:

vx no write nothing

or

√ vx u

if i change the coordinate system, how can i to do the same?

One way will be

vt no write nothing

but the other?

√ vt ??????

Thanks you

JL

avoided 2783 inverted elements by using linear geometry shape order

Tue, 31 Jan 2012 17:43:02 +0000

Hello all;
when i start running my i get this massage what it should mean and how i can avoid this error?
Best regard

Single element mesh

Mon, 30 Jan 2012 20:35:42 +0000

How do I make a mesh consisting of exactly "one element"?

Time dependent and non linear model takes (too?) long to solve...

Sun, 29 Jan 2012 21:52:09 +0000

Hello everyone,

I am simulating electrical machines within the rotating machinery interface in the ACDC module.

The model I calculate right now solves. But the way it solves is not satisfying in my oppinion. As soon as I choose the iron to be nonlinear, the solving takes days to calculate just 50ms with a time step of 100us.

I'm not sure if I'm allowed to upload the model in this forum, since the Geometry is not meant to be public.

The Geometry is a fourth of the motor and periodic boundary conditions are used for this. The current, stamped into the slots, is calculated by time dependent equations which involve the derivation of the flux field linkage. So comsol has to solve the equation for the current and vector potential simultaneously.

I already tried to improve the model this way:
http://www.comsol.com/support/knowledgebase/1127/
But it didn't really increase the speed. Actually with it's help the model could first be solved nonlinear...

Does anybody have general advices to get nonlinear time dependent models solved faster?

Thanks in advance.

automatic remeshing

Thu, 26 Jan 2012 17:01:35 +0000

Hi

What is the definition of "mesh quality"?

Is it possible for an initial mesh to be of poor quality?

In my model automatic remeshing threw me out at t=0 if I asked for a minimum mesh quality of 0.01

but ran OK with 0.001(this is only step 1 and the mesh has not moved yet.

How can I improve the quality of the initial mesh?

What is the lowest quality I can live with to keep the model running with intermittent remeshings?

thanks

David

Mesh sensitivity Test

Sun, 25 Dec 2011 09:46:01 +0000

Hi all,

I had carried out mesh sensitivity test on my 2-D asymmetrical model, using stress-strain axial symmetry module to calculate the maximum von Mises stress. However, the more mesh I refine, the result seem does not convergence.The stress value getting bigger and bigger.

Appreciate for expert comment.

Merry Christmas and happy new year.

Extract mesh information in Matlab

Wed, 19 Jan 2011 21:34:34 +0000

Hello,

I'm new to Comsol. Can anybody explain how to extract surface mesh info from Comsol? I'm using version 3.5.
The following is what I did:
(1) load a 2D curve y=f(x) into Comsol.
(2) Revolve and generate an axi-symmatric 3D object.
(3) Use init mesh to generate mesh. (But not sure it is the volume mesh or surface mesh)
(4) Export mesh to fem structure.

The information I need from the exported fem structure includes:
(1) x,y,z coordinates of nodes on the surface only.
(2) element connection.
(3) edge list.

Now I can see that "fem" is in my matlab workspace. But I don't know how to go further to dig out more. I
thought there should be something called "fem.mesh.p" or "fem.mesh.t". But it seems in my fem structure,
there is only something called "fem.xmesh".

Another problem is I don't know whether the mesh generated in comsol is volume mesh or surface mesh.
I only need surface mesh.

Thanks for your help.

Importing Boundary Conditions Using a .nas File

Mon, 17 Jan 2011 12:36:19 +0000

Is there any way to import boundary conditions into comsol using nastran file .nas mesh import.

Many thanks

Calculation of normal vectors on surface

Wed, 22 Dec 2010 14:10:11 +0000

Hi,

I need to calculate normal vectors on a 3D geometry surface to calculate boundary conditions from that (it's a rather complex matlab code). I took the postinterp help to get a starting point and tried to calculate the normal vectors in the center of each element. I used postinterp and tried to determine nx, ny and nz, but, whatever I do, I only get NaNs. In the postinterp help, a parameter s was used to do that, this also works fine in 3D, but I am not sure how I would transform certain points to that parameter... Here's my code:

% Create sphere surface geometry
clear fem
fem.geom=geomcoerce('face',sphere3(1,'pos',[0,0,0]));
geomplot(fem,'edgelabels','on'), axis equal
% Generate mesh
fem.mesh = meshinit(fem);
fem.shape = 2;
fem.xmesh = meshextend(fem);
% Get element triangulation
a=get(fem.mesh,'el');
tri=a{3}.elem;
p=fem.mesh.p;
% Calculate center point of each triangle
pn=(p(:,tri(1,:))+p(:,tri(2,:))+p(:,tri(3,:)))/3;
% Calculate normal vectors
nx=postinterp(fem,'nx',pn)
% this gives only NaNs, why?

Any ideas? Also the interpolation in the mesh points does not work,

nx=postinterp(fem,'nx',fem.mesh.p)

gives me the same NaNs...

Nils

Mesh element volume

Wed, 01 Jul 2009 19:07:48 +0000

Dear all,

I have a material property having a different value in a domain and
it is function of individual volume of deformed mesh element and undeformed mesh
element.
Could you tell me how to get the volume of the mesh element?
In 2D, this may be the area of the each element. (Am I right?)

I appreciate your reply.

Thank you.

Yoon