Transforming heavyweight engineering models into machine-readable, web-ready geometry and semantics.
Research Question
Can engineering models be transformed from heavyweight authoring artefacts into machine-readable geometry, semantics and evidence suitable for downstream AI, web, XR and digital-twin workflows?
Problem
Building Information Models (BIM) in formats like IFC and Revit contain rich geometry and semantic information, but are designed for authoring tools rather than web delivery, AI processing, or real-time visualization. The transformation from native BIM to lightweight, structured, web-ready assets requires parsing complex specifications, preserving semantic relationships, and producing optimized outputs.
Why It Matters
- Web and mobile access to BIM data without desktop software
- Digital twin creation from existing engineering models
- AR/VR visualization of construction projects
- AI-based model analysis and comparison
- Cross-platform interoperability between BIM authoring tools
- Automated quality and compliance checking
Current Approach
The BIM Cloud Pipeline implements a REST API service for model transformation:
- IFC parsing — From-scratch ISO-10303-21 parser extracting geometry and semantics
- GLB/GLTF export — Optimized 3D asset generation with trimesh
- Metadata extraction — Structured JSON with element types, properties, relationships
- Model comparison — Side-by-side multi-model analysis
- Revit adapter — Credential-gated Autodesk APS integration
Implementation Status
- IFC native path — Implemented and live-validated with buildingSMART samples
- Autodesk APS adapter — Implemented and unit-tested; requires external credentials
- Three.js dashboard — Live with orbit controls, section planes, metadata display
- Model comparison — Implemented via REST API
Limitations
- IFC coverage is a subset of the full specification
- Autodesk APS route requires external service credentials
- No contractual BIM validation claims are made
- Complex curved geometry may lose fidelity during simplification
- Semantic preservation depends on source model quality