6
$\begingroup$

I came across puzzling results for certain 3D meshes, which I reduced to a minimal case. I made a 1-2-3 cuboid in Blender, exported it to this OBJ file, then imported it to Mathematica, which generates a plausible result:

(rawBlenderCuboid = 
 Import["https://filebin.ca/3TcNMJskChjW/one_two_three_cuboid.obj"]) // InputForm

MeshRegion[{{1., -3., -2.}, {1., -3., 2.}, {-1., -3., 2.}, {-1., -3., 
 -2.0000009536743164}, {1., 3., -1.9999990463256836}, {-1., 3., -2.}, 
 {-1., 3., 1.9999990463256836}, {0.9999989867210388, 3., 
 2.0000009536743164}}, {Polygon[{{1, 3, 4}, {5, 7, 8}, {1, 8, 2}, {2, 7, 
   3}, {3, 6, 4}, {5, 4, 6}, {3, 1, 2}, {7, 5, 6}, {8, 1, 5}, {7, 2, 8}, 
   {6, 3, 7}, {4, 5, 1}}]}]

This has no defects according to FindMeshDefects. It has the expected area of $88=2*(2*4+4*6+6*2)$:

Area[rawBlenderCuboid]

Its volume, however, is zero

Volume[rawBlenderCuboid]

0

it should be $48$, as is found for the bounding region:

Volume[BoundingRegion[rawBlenderCuboid]]

48.

Its moment of inertia is inexplicable

MomentOfInertia[rawBlenderCuboid]

{{541.333, 7.56979*10^-6, -0.0000177622}, 
 {7.56979*10^-6, 242.667, -0.0000184774}, 
 {-0.0000177622, -0.0000184774, 421.333}}

A hand calculation with unit density leads one to expect $(208, 80, 160)$ for the principal components of moment of inertia, which is indeed found for the bounding region:

MomentOfInertia[BoundingRegion[rawBlenderCuboid]]

{{208., -7.10543*10^-15, -5.32907*10^-15}, 
 {-7.10543*10^-15, 80.0001, 1.15463*10^-14}, 
 {-5.32907*10^-15, 1.15463*10^-14, 160.}}

Bounding regions seem always to be cuboids, so they are not acceptable for my real examples, which are not convex. Therefore, the rather nice solutions from this SE post about convex hulls would not serve.

A clue comes from the fact that Mathematica does not find the mesh to be solid: 

SolidRegionQ[rawBlenderCuboid]

False

whereas the bounding region is reckoned as solid:

SolidRegionQ[BoundingRegion[rawBlenderCuboid]]

True

Both the bounding region and the raw region are both valid regions according to MeshRegionQ, and they're both embedded in 3 dimensions:

RegionEmbeddingDimension[rawBlenderCuboid]

3

RegionEmbeddingDimension[BoundingRegion[rawBlenderCuboid]]

3

so it's not easy to see the more profound differences between them.

These examples from Wolfram's curated collections also have zero volume and are not solid.

I realize from this SE post that I can fit tetrahedra and get to a volume that way, or that I could implement integrals inside polyhedra using a source like this one. Either approach is considerable work, and I wonder whether I am just missing some really simple answer already done in Mathematica. I also think that the approaches documented here for parametric and procedurally generated meshes would not be of direct help. This is my first foray into the new mesh functions, so I am a long way from mastery.

|improve this question
$\endgroup$
  • $\begingroup$ I've never worked with meshes in MMA before, but could it be due to your surface not being closed (hence, having 0 volume)? $\endgroup$ – anderstood Jul 17 '17 at 0:24
  • $\begingroup$ My surfaces are closed and watertight according to MeshMixer and trimesh respectively. I precondition them before sending them to Mathematica. I want to use Mathematica to cross-check the volume and moment-of-inertia calcs from trimesh. MeshMixer from Autodesk is very strong; I tend to believe it, but it doesn't do Volume and MI. The papers from geometrictools (cited above) document the algos in trimesh. They're applications of Stoke's theorem, so very believable, but also very involved, so deuced difficult to code correctly. That's why I want to cross-check against something independent. $\endgroup$ – Reb.Cabin Jul 17 '17 at 0:59
8
$\begingroup$

When Mathematica gives correct surface area and volume of zero, what you have is 2-manifold embedded in 3D space, not a solid. It is like the difference between a Sphere and Ball.

Area[Sphere[]]

4 π

Volume[Sphere[]]

0

Area[Ball[]]

∞

Volume[Ball[]]

(4 π)/3

Area[RegionBoundary[Ball[]]]

4 π

Update

box = 
  MeshRegion[
    {{1., -3., -2.}, {1., -3., 2.}, {-1., -3., 2.}, {-1., -3., -2.}, 
     {1., 3., -2.}, {-1., 3., -2.}, {-1., 3., 2.}, {1., 3., 2.}},
    {Polygon[
      {{1, 3, 4}, {5, 7, 8}, {1, 8, 2}, {2, 7, 3}, {3, 6, 4}, {5, 4, 6}, 
       {3, 1, 2}, {7, 5, 6}, {8, 1, 5}, {7, 2, 8}, {6, 3, 7}, {4, 5, 1}}]}]

 RegionDimension @ box

2

This confirms you have a hollow box; a solid brick would give 3.

2nd Update

You can generate a brick from a box with DelaunayMesh.

brick = DelaunayMesh @ MeshCoordinates[box]

RegionDimension @ brick

3

Volume[brick]

48

MomentOfInertia[brick] // Chop

{{208., 0, 0}, {0, 80., 0}, {0, 0, 160.}}

|improve this answer
$\endgroup$
  • $\begingroup$ Any ideas about a quick way to calculate the volume and moment of inertia of the 3-volume enclosed by the 2-manifold? I have the references above to the Tet-filling method and to the Stokes-theorem method but I'm (candidly) looking to avoid work :) $\endgroup$ – Reb.Cabin Jul 17 '17 at 3:29
  • $\begingroup$ In a fresh Kernel, with version 11.1.1.0 on Mac OS X, copying your exact example for box and brick, I get Volume[brick] == 52. What a mystery! $\endgroup$ – Reb.Cabin Jul 17 '17 at 14:53
  • $\begingroup$ I found it. You have a {1., 3., -3.} in the paste above which should be {1., 3., -2.}. You probably have the correct coordinates in your actual notebook. That settles it and I'm accepting this as answer. Thanks again. $\endgroup$ – Reb.Cabin Jul 17 '17 at 15:02
  • $\begingroup$ @Reb.Cabin. Thanks for pointing out the error. I have corrected it. $\endgroup$ – m_goldberg Jul 17 '17 at 15:26
5
$\begingroup$

If the mesh is a triangle mesh, the "Stokes theorem" can be applied in the following way:

M = Import["https://filebin.ca/3TcNMJskChjW/one_two_three_cuboid.obj"];
With[{triangles = Partition[MeshCoordinates[M][[Flatten[MeshCells[M, 2][[All, 1]]]]], 3]},
 Total[Det /@ triangles]/6.
]
(* 48. *)

But the easiest way will be to transform your mesh into a BoundaryMeshRegion.

B = BoundaryMeshRegion[MeshCoordinates[M], MeshCells[M, 2]];
Volume[B]
MomentOfInertia[B]
(* 48. *)
(* {{208., 3.74847*10^-6, -3.57628*10^-6}, {3.74847*10^-6,80., -6.49028*10^-6}, {-3.57628*10^-6, -6.49028*10^-6, 160.}} *)
|improve this answer
$\endgroup$
  • 1
    $\begingroup$ the BoundaryMeshRegion does a great job for non-convex regions, and is more broadly applicable than DelaunayMesh in @m_goldberg's answer above. For example, try DelaunayMesh[MeshCoordinates[DiscretizeGraphics[ExampleData[{"Geometry3D", "Torus"}]]]] and BoundaryMeshRegion[ With[{t = DiscretizeGraphics[ExampleData[{"Geometry3D", "Torus"}]]}, BoundaryMeshRegion[MeshCoordinates[t], MeshCells[t, 2]]]]. Both of you have provided useful info and I wish I could accept them both as answers. $\endgroup$ – Reb.Cabin Jul 17 '17 at 17:18
  • $\begingroup$ You're welcome. $\endgroup$ – Henrik Schumacher Jul 17 '17 at 17:32
2
$\begingroup$

You can import OBJ files directly as BoundaryMeshRegion's

In[5]:= RegionDimension@Import["https://filebin.ca/3TcNMJskChjW/one_two_three_cuboid.obj",
  "BoundaryMeshRegion"]
Out[5]= 3

This is documented on ref/format/OBJ.

|improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.