City (DWG) Import best workflow for 3D printing

Problem: CityGML → DWG → Rhino → Non-printable geometry

I am working with a city model exported from CityGML (exported as .dwg).
When I import it into Rhinoceros 3D, the buildings are organized into multiple layers (e.g. Building, RoofSurface, WallSurface, GroundSurface, etc.).

Goal

I want to create a clean, closed solid in order to extract a 200 × 200 mm section and prepare it for SLA 3D printing.


Problem

The buildings are not clean, closed polysurfaces, but instead consist of many separate surfaces.
There are numerous naked edges and small geometric inconsistencies.

As a result, the following does not work reliably:

  • Boolean operations

  • Cutting out a solid volume (e.g. a cube)

  • Preparing the model for 3D printing


What I have tried so far

I have already attempted the following steps:

  1. MergeAllCoplanarFaces
    → to simplify adjacent planar surfaces

  2. Join / general repair attempts
    → to create closed polysurfaces

  3. Repair Mesh
    → to fix geometry issues

  4. ShrinkWrap (at the end)
    → to generate a closed volume


Result

  • The model remains messy and fragmented

  • Many naked edges persist

  • Boolean operations are unreliable or fail

  • The model is not suitable for SLA printing


Question

  • Is there a better workflow for CityGML → Rhino?

  • Should I convert to mesh earlier in the process?

  • Is there a robust way to unify/clean this type of geometry?

    Muensterhuegel_ohne_Dachueberstaende.dwg (3.8 MB)

The model is a bit of a mess. This means it is possibel to get there, be ready for a challenge.

Not sure what the best way to bring CITYGML into Rhino. Make sure it does not include the STL file format. Any other format might be better.

Here are the step I took:

  1. Explodeblock command to explode all the blocks. Those are not needed.
  2. Then I selected the ground and using Gumball, extruded the mesh to create ground thickness.
  3. Then select the points at the bottom of the extruded ground and use SetPT to sett all the points to the same Z value. This will create a flat base.
  4. Then shrinkwrap it all with a edge length of 0.1.
  5. The result probably still needs work, but it is closer:


I can send you the file in and email if needed.

Thank you for the reply!
Indeed the workflow works more or less good. I will have several files like these to do.
Waiting right now for the printing result. Lets see if the mesh is fine enough for a good smooth print.
The aim is to do a translucent SLA print with a certain thickness.

cheers
Arkadius

Dear community,

I am still preparing a detailed city DWG in Rhino 8 for SLA printing. The full model consists of 25 puzzle-like tiles, scaled to 1:1250. The print supplier only needs the outer A-surface as STL; they will add wall thickness themselves.

My current workflow is:

  1. Scale the model to 1:1250.
  2. Move the full assembly close to 0,0,0 while preserving the relative position of all 25 tiles.
  3. Run ShrinkWrap.
  4. Use SplitDisjointMesh and delete the small floating artifacts.
  5. Check the remaining main mesh.

With:

Target Edge Length: 0.15 mm
Polygon Optimization: 0%
Smoothing: 0
Offset: 0

I get a completely clean result:
valid, closed mesh, 0 naked edges, 0 non-manifold edges, 0 self-intersections, 0 reversed faces
about 46.4 million faces per tile

The problem is file complexity.

I tested several ways of reducing it:

Method Faces Self-intersections
ShrinkWrap 0.15 / Opt 0% 46.4 M 0
ShrinkWrap 0.15 / Opt 1% 10.4 M 124
ShrinkWrap 0.15 / Opt 10% 8.4 M 122
ShrinkWrap 0.15 / Opt 25% 7.5 M 130
ShrinkWrap 0.15 / Opt 50% 6.6 M 169
ReduceMesh ~50% 31.9 M 195
ReduceMesh Planar Only 13.8 M 1,954

MeshSelfIntersect showed that the intersections in the optimized versions are spread throughout the model, not concentrated around a few bad objects.

I also tested Target Edge Length = 0.20 mm, but that loses too much architectural detail.

My questions

  • Has anyone experienced ShrinkWrap Polygon Optimization introducing self-intersections even at very low values such as 1%?
  • Is there a better method for simplifying a clean, very dense architectural mesh without creating self-intersections?
  • Would you simply keep the clean 46M-face mesh if the downstream software can handle it?
  • Since I only need the A-surface, would removing the flat underside before STL export be the best next step for reducing the file?
  • Are there Rhino plug-ins or external mesh tools that work better than ReduceMesh for this kind of dense city geometry?

I wonder if moving the the object to the origin first, then scaling might be a better method to reduce error.

You might try Balancer nPro. for the decimation. It does a fantastic job of reducing the # of triangles and preserving small details due to a tolerance setting that allows a unit value to be entered. You’ll probably need to check to be sure the object is still error free after the reduction.