Gap automeshing #2390
Gap automeshing #2390
Conversation
|
one more point:
|
|
Thanks for this great feature!
Agree that we can set 1 as default if ti works quite well already. Then we can add an option like
Right now, we don't have good conformal scheme for thin strips, so probably no need right now.
Not too concerned, but I wonder that it's not hard to store them?
My understanding is that metallic structures are more of a concern. So we probably only need to distinguish between metallic (PEC, lossyMetal) and dielectric. |
dbochkov-flexcompute
force-pushed
the
daniil/gap-meshing
branch
from
3665737 to
f5a5d78
Compare
| "The underlying algorithm detects gaps contained in a single cell and places a snapping plane at the gaps's centers.", | ||
| ) | ||
|
|
||
| dl_min_from_gap_width: bool = pd.Field( |
There was a problem hiding this comment.
Now that we have dl_min from quite a few places: gap refinement, corner refinement, thickness refinement, and minimal feature, shall we allow controling them individually and combine them into a new class?
| # reuse the same command but flip dimensions | ||
| h_cells_ij, h_cells_dx = self._find_vertical_intersections(y, x, vertices[:, [1, 0]]) | ||
| if len(h_cells_ij) > 0: | ||
| h_cells_ij = np.roll(h_cells_ij, axis=1, shift=1) |
There was a problem hiding this comment.
more explanation here?
dbochkov-flexcompute
force-pushed
the
daniil/gap-meshing
branch
from
f5a5d78 to
395f617
Compare
|
sprinkled around some more comments in the code, had to recall myself what I was doing 😅 |
dbochkov-flexcompute
force-pushed
the
daniil/gap-meshing
branch
from
395f617 to
ff7ac19
Compare
There was a problem hiding this comment.
Another great feature! I found a couple of small things in the associated comments. The only real sticking point for me is that it would be nice to get a symmetric grid, when the structure itself is symmetric.
I think it came up in the meeting and maybe there is no easy solution. But for the structure below it seems to always give an asymmetric grid. I thought maybe part of the reason could be due to how intersections are found with grid boundaries, which does not make use of a tolerance. The grid boundary and the polygon vertex might be very close.
And here is the original grid without gap refinement. (setting gap_meshing_iters=0)
| If `None`, recomputes internal override structures. | ||
| internal_snapping_points : List[CoordinateOptional] | ||
| If `None`, recomputes internal snapping points. | ||
|
|
There was a problem hiding this comment.
missing some of the parameters in the docstring
There was a problem hiding this comment.
looks like missing parameters in the original function's docstring as well
There was a problem hiding this comment.
periodic and min_dl_from_gaps
|
|
||
| # loop over all shapes | ||
| for mat, shapes in merged_geos: | ||
| if not mat.is_pec: |
There was a problem hiding this comment.
We should support LossyMetalMedium as well, I need to make the same check in my next PR. Maybe we can add is_metal_conductor to AbstractMedium
| @@ -1497,6 +1534,253 @@ def _override_structures_along_axis( | |||
| override_structures += refinement_structures | |||
| return override_structures | |||
|
|
|||
| def _find_vertical_intersections(self, x, y, vertices): | |||
There was a problem hiding this comment.
add some type hints for the return value of these methods
|
|
||
| # generate horizontal snapping lines | ||
| snapping_lines_y = [] | ||
| if len(v_cells_ij) > 0: |
There was a problem hiding this comment.
Seems possible to avoid duplicating the code for the horizontal and vertical lines (as you did for _find_vertical_intersections
| cells_dy = [] | ||
|
|
||
| # for each polygon vertex find the index of the first grid line on the right | ||
| grid_lines_on_right = np.argmax(x[:, None] > vertices[None, :, 0], axis=0) |
There was a problem hiding this comment.
It might make sense to use a small tolerance, so that we can distinguish between almost exactly coincident with the cell boundary and certainly greater than.
There was a problem hiding this comment.
That might minimize the number of false positives, could also result in some of the asymmetry present in the resulting grids.
Integrated gap meshing in an iterative way. Now
LayerRefinementSpechas two additional parameters:gap_meshing_iters: NonNegativeInt = 1to set the number of iteration to perform for attempting resolving all gapsdl_min_from_gap_width: bool = Truewhether or not to reducedl_minreduce to the minimal detected gap width.A few of points to discuss:
examples of meshing thin strips and gaps:
