Discussion Forum

Hi

first of all line loads on 3D models is to avoided as far as possible, as these represents singularities and can (but does not necesarily lead to) give convergence issues.
It's much better to define boundary loads, or to restrict them spatially with a Gaussian load shape provided you have enough mesh elements across the 2/3 of the load region.

When you add domains and boundaries it is normally for defining BC or loads, so by default COMSOl will line up mesh nodes on the boundaries, edges and points you define. You can now, in the latest versions, add lines or regions specifically for the mesh concentration assistance. without creating specific entities for BC, but I'm not sure it works the other way around.

So fr I have avoided using beams and shells, because of the extended work to knit togther the BC at the nodes, for complex models it becomes quickly lengthy. We need some scripting and good automated methodology to speed up things. Therefore I, so far use mainly full solids, even f its by far slower to solve, and not necesarily more precise

--
Good luck
Ivar

Thanks for your reply already. So ill try to create a good model for my case. In fact i dont want to use a constant area load, even more id like to initiate the load with 24 stamps (circular arrays in the picture) wich are closer to the real load. At the end i want to make a cut in the middle area to visualize the tensions beginning at the stamps. So i just created the pattern with a new workplane and now im trying to make a good mesh.

Wich mesh would be best for a couple of thin layers? The heights of the layers varies from 3.2 mm to 0.2 mm, wich element size should i take? I thought about aspect rations of 1/20 or 1/50 element size so id take maximum size elements for the 3.2 mm layer of 160 and for the 0.2 mm layer of 10.

My problem is when i make a free quad-net and try to extrude it from top surface where the layer with the stamps is on to the botton surface i usually get errors telling me the extruded net couldnt be created for the area because the origin and target area do not fit together.

By now i have like zero experiences with modeling thin layers, im really happy to get some support here ;)
Btw. what is BC?

I attached a picture of the model. The created mesh isnt very fine by now, but at first i want to make it work.

With special greetings

Hi

you have several questions here, a few answers:

I generally cut my volumes along internal boundaries where I want to study the result (stress displacement) the fact that a given uniform domain is cut up in smaller parts adds some RAM requirement to cope for the continuity on the boundary, but this is a little cost today compared to the ease to postprocess. On the other side COMSOL has the Data Set has all the cut plane /line ... features to do the same at postprocessing level, these I use too and are as precise, more a question when I think about what to do ;)

for meshing if you have physical thickness (volumes) and not surface physics, a thin layer should have enough elelemnts in the thickness to resolve the gradients in the layer (if any), but this implies also that the lateral sizes are not that much larger than the depth of the mesh element => a very dense mesh which expoldes the RAM requirement and slow down the solving for week-end type runs (give you time t go skiing though with good consience;)
So here there is a balance to make. The surface thin physics is then easier as it assumes little or constant gradient across the thickness and ou have by far less elements.

Finally for the sweep mesh you have to respect the topology 1) often its better to have the same patter on source and destination, but this is not a strict rule, if you mesh them manually, mesh the source boundary that is cut up once and copy the mesh identically to the destination, then sweep by telling source and detination (often one can let COMSOL mesh the destination but always chekc manually the source for complex parts. BUT the lateral sides MUST be of same shape, if one is cut by a adjacent domain, then you should cut your sweep volume such that all lateral sides have the same type of boundaries

And do not forget that the sweep mesh creates brick type elements, if you want to continue in tets, you neeed to convert the lateral BOUNDARIES before you apply a tet mesh to the lateral adjacent domains.

It's worth to go to the COSMOL meshing course, it gives you the fundamentals and advanced tricks, as meshing is moe an "art", you will oly manage well by expericing, and understanding the basic rules and limitation of the different techniques.
Do not forget a mesh is a discretization, its like sampling music (this is well understood today I have noticed) if you do not use a high enough sampling rate = dense mesh yo will not catch the high frequencies = resolve the high steep gradients and your results will be less precise, but no reason to over do it neither ;)

--
Good luck
Ivar

Thank you for the advices. Actually I'd love to go on a meshing course but there is no time right now for it, only a couple of weeks to go to the end of my thesis. Guess ill do it some time since FEM is really fun to me ... when it is doing what i want ;)

Actually i got a few problems still. i got 5 layers and the same pattern on top layer, botton layer and mid layer. Actually I just want to have this pattern at the third layer, but without also creating it on the mid layer i cant sweep it with the line wich is created by the rubber. I can sweep it until the last layer but there it says that each target area must be unique to one or more source areas. Even i have the same pattern on top and botton surface i cant copy it to the botton surface and mesh it with the existing swept mesh on the top side of it.
Id like to sweep the net down on the last layer and finally do tets on the frame and the rubber in between the laminate and the frame.

Right now im kinda clueless even I tried to mesh the layers from both sides and sweep it. Id be curious how you would solve it and give you the model just to put a mesh on that fifth layer and mesh the frame+ rubber so i could understand where my problem is but this model is a little big for uploading (30 mb). But if you would do it, id be really happy and send it to you via e-mail or something.

But i feel im not that bad by now for studying this programm myself at evenings. I attached the file with some textmarks inside for better understanding.

With special greetings.

*but without also creating it on the TOP layer i cant sweep it with the line wich is created by the rubber. (you see this in the picture)

Ok now i created all layers with the same geometry (was a lot of work somehow) but now he tells me that all source domains must be connected.

Somehow he doesnt like this, but i dont know why.

Hi

you are right the sweep mesher requires that all your layers are structured the same way, which means often you need to be clever to use the domain selections to reconstruct your patterns, either during geometry definition and /or after meshing and during material, physis and BC definitions. Try to use the advanced features such as tangeant recognition etc, it saves often quite some time ;)

Creating a new Definition - View and using the hide entity feature often helps to clear the screen for unwanted details or structures, just as extensive se of views with or without the axis preserve scale ratio on and off.

sometimes the sweep mesh works better if you do 1 or 2 layers at the time rather than all in one

--
Good luck
Ivar