STL to STEP File Converter

STL files are simply lists of triangles: raw, minimal information to approximate a shape.

solid Mesh
  facet normal 0.5 0.5 -0.707107
    outer loop
      vertex -5.20417e-15 17.9605 0
      vertex 17.9605 6.07153e-15 0
      vertex -6.93889e-15 3.46945e-15 -12.7
    endloop
  endfacet
  facet normal -0.5 0.5 0.707107
    outer loop
      vertex -17.9605 4.33681e-15 0
      vertex -3.46945e-15 6.93889e-15 12.7
      vertex -5.20417e-15 17.9605 0
    endloop
  endfacet
  facet normal -0.5 0.5 -0.707107
    outer loop
      vertex -17.9605 4.33681e-15 0
      vertex -5.20417e-15 17.9605 0
      vertex -6.93889e-15 3.46945e-15 -12.7
    endloop
  endfacet
endsolid Mesh

You can see how this raw geometric information would be enough to 3d print a part, but not right to adjust the mug handle or simulate it's strength.

Given those points, and triangle normal vectors, we need to convert these triangles into solid boundary geometry.

To do that, we search triangles for those that are adjacent (share an edge) and coplanar (normal directions equal, within tolerance).

If triangles meet this criteria, we merge them, and ultimately build robust 2d shapes from their outlines.

STL files do not store curvature information. Thankfully, using common creation algorithms, we can reverse engineer them.

Generally, parts that have a 25-gon (or similar), are very likely to be circles, especially if they can be built tangent to existing faces or create through holes.

When we see patterns that those 25-gons have the same area and incremental angle, after merging triangles and within reasonable tolerances, we can assume that the design intent is rounded faces. By analyzing the total angle and identifying patterns in the points involved, we can build robust curvature that blends with adjacent faces for ultimately watertight geometry.

STEP files are standardized in their hierarchical structure, almost like a Table of Contents, which made it straightforward to build and validate robust geometry. STEP files are built from:

CLOSED_SHELLs (including units/tolerances/metadata)
ADVANCED_FACES
 PLANEs
  AXIS(s)
   DIRECTIONs
 FACE_BOUNDs
  EDGE_LOOPs
   ORIENTED_EDGEs
    LINEs
     VERTEX_POINTs
      CARTESIAN_POINTs

Given this structure, once we understood the desired geometry, it was straightforward to build manifold geometry.

After writing the code in Python, with the help of Numpy and Pygame (for STL visualization), we needed to host it.

• The first step was building an AWS Lambda Function for computing power to handle the conversion. Here I learned the importance of O(n) time complexity, at scale, and needed to restructure some key functions (or pay the price in student AWS credits).
• The Lambda function needed two S3 buckets, one to POST the files for conversion and one to GET the files back to the user. Of course, the files are encrypted, and deleted immediately once converted.
• Lastly, a simple AWS Amplify frontend was built to host the process and share how it works.

The code does not work perfectly nor universally. Different STL export algorithms run through different processes, and leave random artifacts that I could not catalog let alone incorporate or assign computing power to catch.

The code could be further improved to handle more complex curvature, guess units based on values (a feature at 25.4 is probably an indicator of mm/in file size), and have better error visualization/validation.