From CAD (STL/STEP) through voxel to pore network - how to make it?

[pashaLyb]

I'm new to OpenPNM and I would like to test it for PTL/GDL (porous transport layer/gas diffusion layer) characterization. Usually one will use µCT scan to build the p-network. I'm wondering if it is method to import a PTL structure defined by CAD geometry or STL/STEP file? This is interesting for a specific kind of the PTL/GDL that have a relatively coarse deterministic structure like metal net and could be easily defined/modeled with CAD and exported as STL or STEP file.

I copy here the answer by @jgostick

We currently do not have that, only the other way around (export to salome, comsol, stl, etc). There are two things you're actually asking about here though. One is to generate a voxel image out of a CAD file so that a network can be extracted. Two is importing a network that is defined by sticks and balls in CAD. Both are feasible but I've never 'looked' inside a cad file to see how the info is stored. Would you be able to share one? Also, feel free to start a fresh 'discussion' for this subject so we can have the thread in one place.

I'm more interested in first thing/topic you mentioned - "generate a voxel image out of a CAD file so that a network can be extracted"
Here are couple of examples I found in WEB

• from CAD with FreeCAD
• Open Cascade
• Online Voxelizer

With the Online Voxelizer I converted a stl file I found im web to this voxel-model.

something-inside20210314-23024-1w6tujn.zip

My goal is to take STL or STEP --> transfer to any voxel format OpenPNM supports --> conduct simulations

Thanks

[pashaLyb]

maybe it's possible to utilizes this package? - p-hofmann/PyVoxelizer

[jgostick]

PyVoxelizer and FreeCAD both seem promising as they have python bindings (or are pure python). I will play with them and report back.

[pashaLyb]

one more I found
tasotasoso/STLSimpleVoxelizer

[jgostick]

Yes, I'm finding quite a few as well. I'm trying to identify a stable and mature one.

[jgostick]

The trimesh library seems to be the best available. I have confirmed that it has the ability to take an stl and convert to voxels, but the voxel conversion seems to be pretty slow.

[jgostick]

OK, here is a script that works. I had to play a bit because the m.voxelized method of the mesh objects in trimesh only do the 'surface' mesh not the total volume.

import porespy as ps
from skimage.measure import marching_cubes
import scipy.ndimage as spim
import trimesh
import numpy as np


# Generate a basic image for testing
im1 = np.zeros(shape=[60, 60, 60], dtype=bool)
im1[tuple((slice(20, 40, None), slice(20, 40, None), slice(20, 40, None)))] = 1
im1[tuple((slice(30, 50, None), slice(30, 50, None), slice(30, 50, None)))] = 1

# Obtain a mesh from the boolean image.  
im1 = np.pad(im1, 3, mode='constant', constant_values=0)
im2 = spim.convolve(im1*1.0, weights=ps.tools.ps_round(1, im1.ndim, smooth=False))
im2 = im2/im2.max()
verts, faces, normals, values = marching_cubes(im2, level=0.5,
                                               gradient_direction='ascent')
m = trimesh.Trimesh(vertices=verts, faces=faces, vertex_normals=normals)

# If you have an stl file already, you could use:
# m = trimesh.load('filename.stl', file_type='stl')

# Finally, use the contains method of the mesh to see which points land within it
s = 1  # Increase this spacing to reduce image resolution
args = [range(0, im1.shape[i], s) for i in range(im1.ndim)]
pts = np.meshgrid(*args)
coords = np.vstack([pts[i].flatten() for i in range(im1.ndim)]).T
b = m.contains(coords)  # This step is not fast!!
b = b.reshape((np.array(im1.shape)/s).astype(int))