STEP-based design analysis pipeline

[DhruvaMyakeri]

Hi,

I've been building a pipeline that analyzes STEP CAD models and generates early engineering insights such as part archetype classification, manufacturability signals, structural heuristics, and material recommendations.
The system extracts geometric primitives from the model, derives normalized geometry signatures, and runs several inference stages before producing a structured engineering review report.

Right now the pipeline runs as a standalone Python tool.
I’m exploring possible geometry engines that could support deeper geometry reasoning and operations in the future, and Manifold caught my attention.
Curious if anyone here has used Manifold beyond modeling/boolean operations for geometry interrogation or analysis workflows.

Thanks!

[elalish]

I was thinking it would be fun to try Alternating Sum of Volumes with Partitioning sometime - Manifold should have all the operations you need for that. Is that the kind of thing you're thinking of?

[DhruvaMyakeri]

Thanks, that’s interesting.
The current pipeline mostly works by extracting geometric primitives and derived descriptors from the STEP model (faces, cylinders, planes, bounding box, surface area, volume, etc) and then computing higher-level signals such as slenderness, feature density, and axis accessibility before running engineering inference stages.
Right now the system doesn’t perform partitioning of the geometry itself ,it mainly relies on extracted geometric features.
But approaches like volume partitioning or alternating volume sums could definitely be interesting for deeper geometric reasoning (for example detecting internal structures, feature volumes, or structural regions).
Does Manifold expose APIs that make this kind of geometry partitioning straightforward?

[elalish]

We just have booleans and hulls, but I think that's about all you need for volume partitioning.

[DhruvaMyakeri]

That makes sense.
I think I could use those operations to derive better geometry intelligence signals and improve the design assessments in my pipeline.
Thanks for the clarification.

[DhruvaMyakeri]

I tried converting STEP geometry to mesh using pythonOCC and verified Manifold volume computations against OpenCascade's exact volume. The results matched closely, so this seems promising for performing volumetric geometry analysis on CAD models.