1. The problem with generative image-to-3D for engineered parts
Generative image-to-3D systems (photogrammetry and diffusion-based mesh generators) reconstruct an approximate surface optimised for perceptual plausibility. For entertainment assets this is acceptable. For an engineered profile it is not: wall thickness, fillet radius and overall depth are inferred rather than measured, and the output carries no datum, scale or dimension. The result cannot be placed in a technical data sheet (TDS) or a compliance submission.
| Method | Geometry source | Dimensional fidelity | Fit for TDS |
|---|---|---|---|
| Diffusion mesh (image-to-3D) | Inferred surface | None (no scale) | No |
| Photogrammetry | Multi-view estimate | ±1–5% typical | Marginal |
| Manual CAD + render | Modelled solid | Exact | Yes (slow) |
| Deterministic extrusion of section | Measured 2D section | Exact by construction | Yes (seconds) |
2. The deterministic principle for constant sections
A pultruded, extruded or rolled profile is a constant cross-section swept along a straight axis. Its 3D form is therefore fully determined by the 2D section polygon: no reconstruction is required, only extrusion. Every geometric and inertial property is an exact function of that polygon, computed by integrating over its boundary (Green's theorem), with closed-form expressions for circular members.
- Silhouette of the render ≡ the input section (by construction).
- Area A and second moments Iₓ, I_y integrate exactly from the polygon.
- Mass per unit length = A × ρ; no empirical correction needed.
- Occlusion resolved by a depth buffer, eliminating 2.5D clipping artefacts.
3. Standard dimension annotation set
For a doubly- or singly-symmetric structural section, the conventional callouts are the overall height h and width w, the wall or flange thicknesses, and the centroidal reference axes y–y (major) and z–z (minor). Annotations should be anchored in the section plane and re-projected to screen space each frame so they remain legible and correctly placed under rotation and zoom.
| Shape | Overall | Thicknesses | Axes |
|---|---|---|---|
| I / wide-flange | h, w | t_f (flange), t_w (web) | y–y, z–z |
| Channel (U) | h, w | t_f, t_w | y–y, z–z |
| Angle (L) | a, b | t | u–u, v–v (principal) |
| RHS / SHS | h, w | t (wall) | y–y, z–z |
| CHS / tube | OD | t (wall) | any diameter |
4. A repeatable production workflow
- Acquire the section: import DXF/CSV, or parameterise a standard shape from a photo or drawing.
- Normalise units (mm or inch) and confirm one datum dimension.
- Extrude to a fixed aspect depth; resolve visibility with a depth buffer.
- Attach the standard callout set; bind labels to the section plane.
- Render at supersampled resolution; export PNG/WebP for the manual, PDF for the TDS.
- For a range, hold camera, lighting and annotation style constant across every size.
Because the pipeline is deterministic, the same section produces the same asset every time — a prerequisite for a catalog that reads as one coherent product family and for datasheets that do not drift between revisions.