Can you create separate meshes in FiPy by reading "physical names" from Gmsh?

Question

I have the following code representing a .geo file for Gmsh that I want to load in FiPy for process simulation. It contains: (i) Oxide, (ii) Silicon and (iii) Gas as physical entities that needs to identified as separate meshes.

SetFactory("OpenCASCADE");
//
ns = 1e-1;
ns2 = 1e-2;
//
x1 = 0;
x2 = 1;
y1 = 0;
y2 = 0.5;
Point(1) = {x1, y1, 0, ns};
Point(2) = {x2, y1, 0, ns};
Point(3) = {x2, y2, 0, ns2};
Point(4) = {x1, y2, 0, ns2};
Line(1) = {1, 2};
Line(2) = {2, 3};
Line(3) = {3, 4};
Line(4) = {4, 1};
Curve Loop(1) = {1, 2, 3, 4};
Plane Surface(1) = {1};
Physical Surface("Oxide") = {1};
//
y1 = 0.5;
y2 = 1.0;
shiftX = 0.4;
Point(51) = {x1, y1, 0, ns2};
Point(61) = {x1+shiftX, y1, 0, ns2};
Point(71) = {x1+shiftX, y2, 0, ns2};
Point(81) = {x1, y2, 0, ns2};
Line(51) = {51, 61};
Line(61) = {61, 71};
Line(71) = {71, 81};
Line(81) = {81, 51};
Curve Loop(21) = {51, 61, 71, 81};
Plane Surface(21) = {21};
Physical Surface("Silicon") = {21};
//
Point(52) = {x2-shiftX, y1, 0, ns2};
Point(62) = {x2, y1, 0, ns2};
Point(72) = {x2, y2, 0, ns2};
Point(82) = {x2-shiftX, y2, 0, ns2};
Line(52) = {52, 62};
Line(62) = {62, 72};
Line(72) = {72, 82};
Line(82) = {82, 52};
Curve Loop(22) = {52, 62, 72, 82};
Plane Surface(22) = {22};
Physical Surface("Silicon") += {22};
//
y1 = 1.0;
y2 = 1.5;
shiftX = 0.4;
shiftY = 0.5;
Point(9) = {x1, y1, 0, ns2};
Point(91) = {x1+shiftX, y1, 0, ns2};
Point(92) = {x1+shiftX, y1-shiftY, 0, ns2};
Point(101) = {x2-shiftX, y1-shiftY, 0, ns2};
Point(102) = {x2-shiftX, y1, 0, ns2};
Point(10) = {x2, y1, 0, ns2};
Point(11) = {x2, y2, 0, ns};
Point(12) = {x1, y2, 0, ns};
Line(9) = {9, 91};
Line(91) = {91, 92};
Line(92) = {92, 101};
Line(101) = {101, 102};
Line(102) = {102, 10};
Line(10) = {10, 11};
Line(11) = {11, 12};
Line(12) = {12, 9};
Curve Loop(3) = {9, 91, 92, 101, 102, 10, 11, 12};
Plane Surface(3) = {3};
Physical Surface("Gas") = {3};
//

Is there a way to read this in FiPy and create 3 meshes with the given physical name?

jeguyer · Accepted Answer · 2021-08-12T15:47:47.990

Here is a refinement on the brute-force solution I initially offered:

import fipy as fp

mesh = fp.Gmsh2D('''
SetFactory("OpenCASCADE");
//
ns = 1e-1;
ns2 = 1e-2;
//
x1 = 0;
x2 = 1;
y1 = 0;
y2 = 0.5;
Point(1) = {x1, y1, 0, ns};
Point(2) = {x2, y1, 0, ns};
Point(3) = {x2, y2, 0, ns2};
Point(4) = {x1, y2, 0, ns2};
Line(1) = {1, 2};
Line(2) = {2, 3};
Line(3) = {3, 4};
Line(4) = {4, 1};
Curve Loop(1) = {1, 2, 3, 4};
Plane Surface(1) = {1};
Physical Surface("Oxide") = {1};
//
y1 = 0.5;
y2 = 1.0;
shiftX = 0.4;
Point(51) = {x1, y1, 0, ns2};
Point(61) = {x1+shiftX, y1, 0, ns2};
Point(71) = {x1+shiftX, y2, 0, ns2};
Point(81) = {x1, y2, 0, ns2};
Line(51) = {51, 61};
Line(61) = {61, 71};
Line(71) = {71, 81};
Line(81) = {81, 51};
Curve Loop(21) = {51, 61, 71, 81};
Plane Surface(21) = {21};
Physical Surface("Silicon") = {21};
//
Point(52) = {x2-shiftX, y1, 0, ns2};
Point(62) = {x2, y1, 0, ns2};
Point(72) = {x2, y2, 0, ns2};
Point(82) = {x2-shiftX, y2, 0, ns2};
Line(52) = {52, 62};
Line(62) = {62, 72};
Line(72) = {72, 82};
Line(82) = {82, 52};
Curve Loop(22) = {52, 62, 72, 82};
Plane Surface(22) = {22};
Physical Surface("Silicon") += {22};
//
y1 = 1.0;
y2 = 1.5;
shiftX = 0.4;
shiftY = 0.5;
Point(9) = {x1, y1, 0, ns2};
Point(91) = {x1+shiftX, y1, 0, ns2};
Point(92) = {x1+shiftX, y1-shiftY, 0, ns2};
Point(101) = {x2-shiftX, y1-shiftY, 0, ns2};
Point(102) = {x2-shiftX, y1, 0, ns2};
Point(10) = {x2, y1, 0, ns2};
Point(11) = {x2, y2, 0, ns};
Point(12) = {x1, y2, 0, ns};
Line(9) = {9, 91};
Line(91) = {91, 92};
Line(92) = {92, 101};
Line(101) = {101, 102};
Line(102) = {102, 10};
Line(10) = {10, 11};
Line(11) = {11, 12};
Line(12) = {12, 9};
Curve Loop(3) = {9, 91, 92, 101, 102, 10, 11, 12};
Plane Surface(3) = {3};
Physical Surface("Gas") = {3};
//''')

from fipy.meshes.mesh2D import Mesh2D

def extract_mesh(mesh, mask):
    cellFaceIDs = mesh.cellFaceIDs[..., mask]
    faceIDs = numerix.unique(cellFaceIDs.flatten())
    facemap = numerix.zeros(mesh.faceVertexIDs.shape[1], dtype=int)
    facemap[faceIDs] = faceIDs.argsort()
    
    faceVertexIDs = mesh.faceVertexIDs[..., faceIDs]
    vertIDs = numerix.unique(faceVertexIDs.flatten())
    vertmap = numerix.zeros(mesh.vertexCoords.shape[1], dtype=int)
    vertmap[vertIDs] = vertIDs.argsort()

    return Mesh2D(mesh.vertexCoords[..., vertIDs],
                  vertmap[faceVertexIDs],
                  facemap[cellFaceIDs])
    
mesh_gas = extract_mesh(mesh=mesh, mask=mesh.physicalCells["Gas"])
mesh_silicon = extract_mesh(mesh=mesh, mask=mesh.physicalCells["Silicon"])
mesh_oxide = extract_mesh(mesh=mesh, mask=mesh.physicalCells["Oxide"])

The derived meshes have compact vertex and face lists and reasonable numbers of exterior faces.

nearest still won't do what you want, because the vast majority of the faces (exterior or otherwise) of the gas mesh are not anywhere near the oxide mesh. You only want the faces that are coincident. To see how complicated that is, you need about 98% of this code.

I think a much better answer is to define the boundaries you are interested in using the Gmsh GEO script, with Physical Line definitions. These can then be extracted as mesh.physicalFaces (some further work would need to be done to extract_mesh(), but I think it's tractable).

Thanks. Some issue - I can't find the vertices (nodes) at the gas/oxide interface with the following code: `near_ids = nearest(mesh_gas.vertexCoords, mesh_oxide.vertexCoords)` Apparently the vertexCoords are not filtered like cellFaceIDs. Any ideas on that? — mindful_machina, Aug 11 '21 at 23:29
That's not surprising. The meshes created as I suggest have all vertices and faces in common; they just have no cells that are defined by most of those faces or vertices. One result is that, as far as FiPy can tell, they have enormous numbers of "exterior" faces. I'll see if I can improve on this, but I think the best answer is to define the boundary you're interested in with the Gmsh geo script. FiPy probably already has too much mesh manipulation code and is unlikely to get more; better to use tools designed for the purpose, like Gmsh, meshio, shapely, ... — jeguyer, Aug 12 '21 at 11:52

score 0 · Answer 2 · answered Sep 28 '21 at 14:22

Thanks jeguyer. I was able to create separate mesh with the physicalCell identifier. In regards to finding the vertices at gas/oxide interface, I was able to trace the interface contour using surfCoords from the following code:

gasFaces = gasMesh.exteriorFaces
gasVertices = numerix.unique(gasMesh.faceVertexIDs[..., gasFaces].flatten())
gasVertexCoords = gasMesh.vertexCoords[..., gasVertices]

oxideFaces = oxideMesh.exteriorFaces
oxideVertices = numerix.unique(oxideMesh.faceVertexIDs[..., oxideFaces].flatten())
oxideVertexCoords = oxideMesh.vertexCoords[..., oxideVertices]

surfVertexID = numerix.nearest(gasVertexCoords, oxideVertexCoords)
surfCoords = gasVertexCoords[..., numerix.unique(surfVertexID)]

Can you create separate meshes in FiPy by reading "physical names" from Gmsh?

2 Answers2

Linked