split_bodies() generates a body that has 0 volume about pyvista HOT 6 OPEN

Looks like threshold is the likely culprit:

dataset.volume  # 81672476.45755768
dataset.connectivity().volume  # 81672476.45755768
dataset.connectivity().threshold([-1,1]).volume  # 0.0

I'm not sure what the cause is from threshold, perhaps it's a pyvista bug or a bug with the underlying vtk method.

But, as a workaround, the connectivity filter itself can be used to perform a similar operation as threshold, e.g. replace this line

def split_bodies(mesh, label=False, progress_bar=False):
    conn = mesh.connectivity('all')
    conn_labels = conn.cell_data['RegionId']
    bodies = pyvista.MultiBlock()
    for vid in np.unique(conn_labels):
        # Now extract it:
        b = mesh.connectivity('specified', vid,  progress_bar=progress_bar)
        if not label:
            # strange behavior:
            # must use this method rather than deleting from the point_data
            # or else object is collected.
            b.cell_data.remove('RegionId')
            b.point_data.remove('RegionId')
        bodies.append(b)
    return bodies

calling split_bodies(dataset)[0].volume will return the correct (non-zero) volume

from pyvista.

@user27182 yes - this is about the result of split_bodies having 0 volume while it should be non-zero.

from pyvista.

I believe the issue stems from two things:

1. The input mesh is polydata, but threshold returns an unstructured grid
2. Volume for unstructured grids is calculated as the sum of the size of each cells. But triangle cells are 2D, which have zero volume, so the total volume of any unstructured grid with triangle cells is always zero. Volume for PolyData is therefore calculated differently.

So, there is no bug with the volume property or threshold filter in the sense that I believe they are working correctly. The issue is just that any PolyData -> UnstructuredGrid operation will result in non-zero volume meshes having zero-volume thereafter. Therefore, to preserve the volume we need a PolyData -> PolyData filter.

The reason workaround function above works is because the connectivity filter handles polydata as a special case and is therefore a PolyData -> PolyData filter.

Since threshold and split_bodies return unstructured grids, @kretes you can try calling extract_surface after split_bodies, as suggested here #3009 (comment) . Alternatively, you can use the function suggested above instead of dataset.split_bodies.

from pyvista.

#5950 adds a new method param to split_bodies to allow using the connectivity filter internally. The issue of zero-volume can then be resolved with, e.g.:

import pyvista
from pyvista import examples
dataset = examples.download_lucy()
print(dataset.volume)  # 81672476.45755768

split_dataset = dataset.split_bodies(method='connectivity')
print(split_dataset.volume)  # 81672476.45755768

@kretes does this work for you?

from pyvista.

Interesting. Checking the actual number of bodies with n_blocks shows that there is only one dataset in the multiblock.

import pyvista
from pyvista import examples
dataset = examples.download_lucy()
bodies = dataset.split_bodies()
print(bodies.n_blocks)  # 1

Also, under the hood this filter uses connectivity to label the regions before splitting. If we do:

np.unique(dataset.connectivity()['RegionId'])

We get [0], which confirms there is only one mesh returned by that function.

We can further confirm that the mesh returned by split_bodies() is that same as the dataset.

np.array_equal(dataset.connectivity().points, dataset.split_bodies()[0].points)  # True

So, it seems that there is definitely just one mesh at the input and output. Perhaps there is a bug with the __contains__ method for pv.MultiBlock?

from pyvista.

I think I misunderstood the bug, thinking that split_bodies was returning two meshes, one of which had zero volume. But the bug is that mesh.volume returns zero for a single mesh which is expected to have non-zero volume? Is that correct @kretes ?

Edited for clarity.

from pyvista.