Research & development

Engineering Drawing & CAD Analysis

Research on turning point clouds into room models and 2D plans, alongside mechanical CAD projection and drawing analysis.

Building prototype: saved point-cloud input, a 3D rendering of the inferred semantic room model, and its actual SVG floor plan. This is one synthetic room test; the 3D view renders semantic JSON, not the experimental IFC export.

Overview

My engineering-geometry work at Riverstart covers two distinct problems. The building prototype infers a room model from a point cloud and exports floor plans. The mechanical-part research evaluates scans against reference CAD and projects supplied STEP models into engineering views. The examples below show the actual data and geometry behind each track.

Building point cloud -> room model -> floor plan

The building prototype reads XYZ or NumPy point clouds with declared units, estimates rectangular room bounds and detects door/window openings from gaps in the wall points. These are geometric methods; wall thickness is a declared input. A semantic model records the room, walls and openings, supplying both the 3D visualization and 2D floor-plan exports. The figures use a saved synthetic fixture, with its original point cloud, geometry JSON and SVG plan available below.

Mechanical scans, reference CAD and engineering views

The mechanical track evaluates scan-to-reference registration separately from STEP-to-drawing projection. It compares Open Cascade, CadQuery and build123d routes, checks visible and hidden edges, and compares generated views with supplied drawings. The through-bore example below isolates the projection stage with known analytic geometry. A reference STEP used for registration is an input, not a model reconstructed from the scan.

Current stage

The room workflow has passed a synthetic software-path test; real-building reconstruction still needs reference data and engineering acceptance. It currently assumes a single axis-aligned room. IFC export is experimental, with cross-format opening placement under review. Mechanical results establish registration and projection baselines; arbitrary scan-to-parametric CAD and manufacturing-ready drawing generation remain separate research goals.

Project figures

Detailed comparison of the synthetic room point cloud and inferred walls with a door and window at matching positions — Building prototype, enlarged: the input points and inferred room geometry share the same coordinates and viewpoint. Transparency reveals the interior and openings.

Actual STEP rendering of a synthetic through-bore block, shown in an orthographic 3D view with millimeter axes — Separate mechanical CAD track: a synthetic through-bore STEP fixture for projection and hidden-line tests. This solid is a supplied test input.

XY and XZ projections of the same synthetic block, with solid visible edges and dashed hidden bore edges — Mechanical CAD output: generated XY and XZ views of the same through-bore fixture, with visible edges in solid lines and hidden edges dashed.

What It Covers

Infer axis-aligned room boundaries, walls and rectangular openings from XYZ point clouds
Represent inferred geometry as a semantic room model and export 2D DXF/SVG floor plans
Evaluate mechanical scan registration against supplied CAD references
Generate visible and hidden 2D edges from STEP models using Open Cascade
Preserve source geometry and outputs for comparison and engineering review

Stack And Topics

Python
Open Cascade
CadQuery
build123d
NumPy
SciPy
IfcOpenShell
ezdxf
Geometry Processing