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.
meshing metals intelligently in rf problems
Posted Apr 13, 2015, 11:38 p.m. EDT Mesh 8 Replies
Please login with a confirmed email address before reporting spam
From what I’ve been able to figure out, COMSOL doesn’t really know how to mesh bulk metals (where the minimum dimensions are much greater than the rf skin depth) intelligently. There is no rf field inside bulk metals, so it shouldn’t be necessary to mesh the metals – other than their surfaces.
Yes, it is in principle possible to exclude all of the interior of bulk metals from the problem, mesh the rest of the problem, define the boundaries between the bulk metals and the rest of the problem as impedance boundaries, and get it to work. I have gotten this to work on simple problems, where there may be a dozen or so domains and only a few simple bulk metal parts to be excluded from the problem. But there doesn’t seem to be a practical method of doing this where there are hundreds of complex metal parts, thousands of surfaces, many ports, etc.
I’ve tried boundary layer meshing on some complex parts, but this feature seems to be directed at CFD problems, and seems much worse than free tetrahedral for bulk metals in rf problems. Sweeping surfaces also seems to be prohibitively complex when dealing with a large number of very complex geometries.
My real problem is that I’m trying to substantially reduce the number of mesh elements in an extremely complex rf/microwave problem. I’m currently at ~10M DOFs, and it may grow to more than 50M DOFs before I’m through with it if I can’t figure out an easy way to exclude all of the metal parts from the problem.
Our geometry includes a lot of various shaped small metal and ceramic parts, surrounded by air (or vacuum). The only way we’ve gotten satisfactory meshes thus far has been to first mesh all the small parts (in some carefully chosen magical order) and then mesh the surrounding vacuum space.
It would be great if there was a simple method of telling COMSOL to just mesh the boundaries of a selected domain with a triangular mesh (with suitable controls), make those surfaces impedance boundaries, and exclude that domain from the emw solver. That would probably allow us to get working meshes with at least 30% fewer mesh elements, and that is a huge deal when we’re talking about more than 5M mesh elements.
Is there some way to do this, that I’ve missed? If not, it seems this is a capability COMSOL should be working on adding.
David
8 Replies Last Post Aug 20, 2015, 1:24 p.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
You can use either the impedance and/or transition boundary conditions. Provided the skin depth is less than metal thickness these should work.
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
On rare occasions in simple problems, I have gotten that to work. But usually COMSOL won't let you assign the needed boundary conditions, and even in those cases where you can apply impedance boundaries, I haven't been able to figure out how to control the boundary meshes.
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Any chance that you could upload a sample file of what you want to do? Does not have to be proprietary but something to help see the problem a bit more.
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hello David,
The appropriate approach here is to only solve for the electromagnetic wave equations in the air domain. Generally the easiest way to set up such a model would be to only select your air domains when you are selecting the domains for which the electromagnetic waves physics interface is being used. That is, never introduce the metallic domains into the problem. You can then apply a single impedance boundary condition on all of the metal surfaces. (Of course you will need to assign different material properties to the different surface in the materials branch.)
Boundary layer meshing should not be used unless the skin depth is a significant fraction of the part dimensions. You will always want to use the impedance boundary condition if the skin depth is small (~10x or less) of the characteristic size of your metallic objects.
Also, with the automatic meshing available in version 5.0, the domains in which you are not solving the electromagnetic waves physics will not be meshed at all unless you are also solving other physics such as heat transfer on these domains.
Best Regards,
Walter
The appropriate approach here is to only solve for the electromagnetic wave equations in the air domain. Generally the easiest way to set up such a model would be to only select your air domains when you are selecting the domains for which the electromagnetic waves physics interface is being used. That is, never introduce the metallic domains into the problem. You can then apply a single impedance boundary condition on all of the metal surfaces. (Of course you will need to assign different material properties to the different surface in the materials branch.)
Boundary layer meshing should not be used unless the skin depth is a significant fraction of the part dimensions. You will always want to use the impedance boundary condition if the skin depth is small (~10x or less) of the characteristic size of your metallic objects.
Also, with the automatic meshing available in version 5.0, the domains in which you are not solving the electromagnetic waves physics will not be meshed at all unless you are also solving other physics such as heat transfer on these domains.
Best Regards,
Walter
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hello Walter,
Yes, thanks. I've made this work on cases with relatively simple geometries - where there are only a few hundred metal boundaries, and they're not too hard to select. Just did it for a case involving several different types of dielectrics and several different types of metals - but with only about a hundred metal surfaces, and none of them too hard to get to. It's a nice little fully parameterized shielded solenoid, tuned and matched to 50-ohms, in this case at 50 MHz. (Was going to attach it, thinking it might be a useful demo, but couldn't figure out how to make that work.)
There are two problems. First of all, I don't know how to control the meshing near the surface of various metal parts independently that are not part of the emw solution and thus are not meshed. There is only a single set of settings for meshing the air in contact with many different metal parts, which have different requirements.
Secondly, it becomes prohibitively difficult in a case where there are 500 domains and 6000 surfaces, most of which are very difficult to select. It seems like COMSOL should be able to do this automatically - with a default surface conductivity the same as that of the adjacent unmeshed metal domain.
Since we can't get our 4-socket 512GB 32-core computer to run rf problems more than about half as fast as our single-processor 8-core 64GB computer (where we are at memory limits), we might have no choice but to try to manually select thousands of surfaces deep inside a complex system so we can avoid meshing the metal parts. But I'm very worried (from experience thus far) that without the ability to control the meshing independently at all these metal surfaces, we won't be able to get it to build a mesh in the vacuum.
Best regards,
David Doty
Yes, thanks. I've made this work on cases with relatively simple geometries - where there are only a few hundred metal boundaries, and they're not too hard to select. Just did it for a case involving several different types of dielectrics and several different types of metals - but with only about a hundred metal surfaces, and none of them too hard to get to. It's a nice little fully parameterized shielded solenoid, tuned and matched to 50-ohms, in this case at 50 MHz. (Was going to attach it, thinking it might be a useful demo, but couldn't figure out how to make that work.)
There are two problems. First of all, I don't know how to control the meshing near the surface of various metal parts independently that are not part of the emw solution and thus are not meshed. There is only a single set of settings for meshing the air in contact with many different metal parts, which have different requirements.
Secondly, it becomes prohibitively difficult in a case where there are 500 domains and 6000 surfaces, most of which are very difficult to select. It seems like COMSOL should be able to do this automatically - with a default surface conductivity the same as that of the adjacent unmeshed metal domain.
Since we can't get our 4-socket 512GB 32-core computer to run rf problems more than about half as fast as our single-processor 8-core 64GB computer (where we are at memory limits), we might have no choice but to try to manually select thousands of surfaces deep inside a complex system so we can avoid meshing the metal parts. But I'm very worried (from experience thus far) that without the ability to control the meshing independently at all these metal surfaces, we won't be able to get it to build a mesh in the vacuum.
Best regards,
David Doty
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
How are you building your geometry? If you can build / import your metal parts as separate steps in the geometry, then you can create selections as they are built so you don't have to select things by hand after the fact.
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Thanks. We haven't used that feature yet. Sounds like it should be helpful.
David
David
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hi every body,
I have an issue with meshing my geometry, 3 cubic blocks putted each on the other and they have different heights with the same contact area, the upper and bottom surfaces of the first and last blocks includes channels, serpentine ones. I am still getting weird results from solving the problem and once I stop running it I got a lot of error messeges related to the mesh at either domains, surfaces or points, specific x y z dimension points.
Any idea or help of how to mesh like this kind of geometries will be really appreciated!
Thanks in advance.
Mohammed
I have an issue with meshing my geometry, 3 cubic blocks putted each on the other and they have different heights with the same contact area, the upper and bottom surfaces of the first and last blocks includes channels, serpentine ones. I am still getting weird results from solving the problem and once I stop running it I got a lot of error messeges related to the mesh at either domains, surfaces or points, specific x y z dimension points.
Any idea or help of how to mesh like this kind of geometries will be really appreciated!
Thanks in advance.
Mohammed