| Author |
 Topic  |
|
|
Antoine
France
1 Posts |
 Posted - Jan 26 2017 : 16:38:05
|
I am trying to use Fastcap to infer the parasitic capacitance of large wire loops (from 1 to 5 m large with multiple turns) in a distributed fashion.
As such, I have developed an automatised meshing script on Matlab for thin piecewise linear structures.
I intend this script to be fully accessible to this forum users, when it will be quite operationnal.
Visual exemple of structures I intent to model can be found here
(1) L shaped overlapping wires :
ht ps://drive.google.com/file/d/0B-DFJgC03oGGWGNRdXFQTlg3VU0/view?usp=sharing
(2) One turn coil
ht ps://drive.google.com/file/d/0B-DFJgC03oGGdVB3Y0NZbjFhTHM/view?usp=sharing
(3) One turn coil (zoom)
ht ps://drive.google.com/file/d/0B-DFJgC03oGGTjhjaUktbWlsdlU/view?usp=sharing
I bullet-proofed my script against repeating pannels, and fastcap2 seems to accept my .qui input without problems. For very simple geometries, it sometimes quite randomly work like a charm.
When I tend to use entirely realistic coil design, almost every time I encounter the following warning:
gmres: WARNING exiting without converging.
Using my script, I have a very good control on the discretisation of the lateral pannels as well as the longitudinal discretisation (how many blocks will be used to describe one filament), but I simply cannot find a way to avoid this non-convergence as I am a rookie using fastcap2.
My questions are as follow :
(1) is the clockwise, or counterclockwise description of the nodes constituing a pannel of importance in the algorithm in a qui file format? (i didn't took too much care of it, as when testing it with a simple cube, and playing with nodes orders, it didn't seem to be a problem)
(2) Are there general meshing guidelines that can be found somewhere about where the discretization are critical ? Do you have any advices ?
(3) is fastcap2 theoretically able to handle rather distant long thin structures ?
Thanks in advance,
Best regards,
Antoine
|
|
|
Enrico
545 Posts |
Posted - Jan 26 2017 : 19:44:14
|
Hi Antonie,
first of all, thanks for your intention to contribute. I'm pretty sure it will be useful for other users as well.
Regarding your specific questions:
quote:
(1) is the clockwise, or counterclockwise description of the nodes constituing a pannel of importance in the algorithm in a qui file format? (i didn't took too much care of it, as when testing it with a simple cube, and playing with nodes orders, it didn't seem to be a problem)
No, for FastCap2 is not an issue. It may become an issue in the future, for FasterCap, as a consistent definition of the node order can be used to automatically determine where the outward normal points, and therefore define the 'in' and the 'out'
quote:
(2) Are there general meshing guidelines that can be found somewhere about where the discretization are critical ? Do you have any advices ?
In general, discretization should be finer whenever you approach corners or edges, or when you are in close proximity to other structures.
quote:
(3) is fastcap2 theoretically able to handle rather distant long thin structures ?
This may be a problem, for two possible reasons:
1) if you are also using long thin panels. FastCap2 assumes that the panels are not too much stretched. From physical perspective, you are not allowing any charge gradient over the long panel; and from a numerical perspective, as there are some integrals involved, the precision will suffer.
2) if you simply have long structures stretching almost 1D or 2D. FastCap2 discretizes the whole space in 3D cubes. But if your structure is, say, a single very long thin wire, you will force a fine discretization of a very large space (a cube with a side of the same length of the length of the wire), potentially leading to out of memory conditions. This issue is not affecting FasterCap instead - you can test it running the 5 wires sample input file.
However, you are saying that for simple geometries FastCap2 works ok (even if I'm unclear about the 'randomly'), while for real world geometries it fails, I would suggest you to try to find out the simplest structure that fails the simulation and send it to me, or attach to the forum, so I can check it.
Just a side note, one common problem is overlapping panels. FastCap2 tests for overlaps but cannot detect all possible conditions; and overlapping panels break the algorithm completely, for instance with GMRES failing to converge.
Best Regards,
Enrico
|
 |
|
| |
Topic |
|
|
|
Wire structures meshing guidelines ?
