3D scan to CAD: how a scanned part becomes an editable model

TL;DR: 3D scan to CAD is the process of rebuilding a scanned mesh as an editable CAD model with defined features and dimensions. It runs in four stages: align the scan to sensible datums, extract features, model them, then check the result against the scan. This post covers each stage, which reverse engineering software does what, and why a STEP file isn't the whole deliverable.

Most descriptions of 3D scan to CAD make it sound like a conversion. Load the mesh, press a button, receive a model. Software vendors are happy to leave that impression standing.

It isn't a conversion. It's a rebuild. The scan is the reference, and somebody re-draws the part on top of it, deciding at every step which measurements represent the designer's intention and which are wear, moulding marks or noise.

That's why it takes as long as it does, and why two people can produce very different models from the same scan file. One traces the surface faithfully and hands you a lumpy solid. The other works out the part's logic and hands you something you can change.

Here's the process we run, stage by stage, and the parts of it that decide how good the model turns out.

What does 3D scan to CAD mean?

3D scan to CAD means rebuilding scan data as a proper CAD model with defined features and editable dimensions. The scan gives you a mesh, which is a fixed skin of triangles. The CAD model gives you geometry with logic behind it: holes that are holes, planes that are flat, and dimensions you can change.

The gap between those two things is structural. Scans are unstructured data. CAD models are stored as boundary representations, meaning a defined set of faces, edges and the relationships between them, which is the format STEP files use and the thing scan data has to be parsed into. Nothing in a mesh says where one face ends and the next begins. Somebody has to draw that line.

This is the core of any reverse engineering job, and it's where the hours go. The capture takes an afternoon. The rebuild takes the rest of the week.

The step nobody talks about: setting the datums

Before any modelling happens, the scan has to be placed in a coordinate system that suits the part. Get this right and everything downstream is easier. Get it wrong and you fight the geometry for days.

A scan arrives in whatever orientation the scanner left it in, which is usually nothing to do with how the part works. The job is to pick reference features and lock the part down. Three datums between them control all six degrees of freedom, and flat faces, corners and cylindrical holes make the best candidates.

Choosing them means understanding the part first. One reverse engineering guide is blunt about the stakes: the initial alignment either speeds up the entire process or causes a massive slowdown, depending on which features you pick. Pick the mounting face and the main bore on a bracket and every later sketch falls onto a sensible plane. Pick a cosmetic curve and nothing lines up with anything.

The same discipline applies earlier in the chain. A well-run scan starts with calibrating the scanner and verifying that calibration before capture, then cleaning the data and aligning it to a global coordinate system before the modelling starts. This is why our 3D scanning service asks what the part bolts to. It's not small talk.

How do you turn a mesh into features?

By splitting the mesh into regions that share a shape, then fitting geometry to each one. Software groups triangles into planes, cylinders, cones and freeform patches. From there you either fit surfaces over them automatically, or draw sketches on cross-sections through the mesh and rebuild the part the way it was made in the first place.

Those two routes give very different results. Auto-surfacing lays a quilt of surfaces across the mesh. It's fast, it follows organic shapes well, and it produces a model that looks right and edits badly, because there are no features to edit.

The feature-based route is slower. You take cross-sections through the mesh, trace 2D profiles over it and build up the part with extrudes and revolves, using the scan as a guide underneath. What you get is a history tree: a list of operations that can be reopened and changed.

We use both, and the part decides. A styled housing with compound curves gets surfaced. A machined bracket gets rebuilt feature by feature, because that's how it was made and that's how you'll want to modify it. Our product design team makes that call at the start, and tells you which one you're getting.

What reverse engineering software do you need?

Something that can hold a large mesh and fit geometry to it. The options fall into three groups: dedicated reverse engineering packages, plugins that add scan tools to CAD you already run, and the mesh tools built into mainstream CAD. Cost runs from included to five figures.

Dedicated packages. Geomagic Design X and PolyWorks Modeler sit at the top. Full parametric reconstruction, automated feature recognition, deep alignment tools, and prices to match.

Plugins. Mesh2Surface and QuickSurface add reverse engineering into Rhino, SolidWorks and SpaceClaim. Independent comparisons put the Rhino version at roughly 1,495 euros perpetual or around 595 euros a year, which is a different order of spend from the dedicated tools.

Built-in tools. SolidWorks ScanTo3D and the Fusion 360 mesh tools are already in the box if you own the CAD. They're worth knowing about and worth being realistic about: Fusion's mesh tools are suited to light referencing and simple surface extraction rather than production reverse engineering of complex parts.

The software matters less than the person driving it. A good engineer in Rhino beats a beginner in Design X on every part we've seen.

How do you know the model is right?

You lay the finished model back over the original scan and measure the gap between them. The software produces a colour map showing where the model sits proud of the scan and where it sits under, against a tolerance you choose. Green is inside, red is outside. It's the only real proof the model matches the part.

The process has several names, including deviation analysis and computer-aided verification, and it works by aligning the model and the scan data in one coordinate system, then generating a colour-coded map of the differences. You set the tolerance band yourself, so the colours tell you which areas fall above and below what you're willing to accept. Serious suppliers treat it as the final proof of work at the end of every reverse engineering project.

One nuance worth understanding. Red doesn't automatically mean wrong. If we've deliberately made a worn face flat again, or thickened a section that kept snapping, the model should show a deviation there. The colour map tells you where you've departed from the scan. You still have to decide, feature by feature, which departures were on purpose.

Ask any supplier for the deviation report. If they can't produce one, nobody checked.

What you get at the end, and what a STEP file loses

The usual deliverable is a STEP file. It's the neutral exchange standard, defined by ISO 10303, and it holds exact geometry, topology and assembly structure without tying you to one CAD package. The AP242 profile has now replaced IGES for anything carrying tolerancing and manufacturing information, and it's what most of the industry archives in.

STEP carries the shape. It doesn't carry the recipe. Feature history doesn't survive the export, so edits get made on the imported solid rather than on a parametric tree. Open a STEP file and you get a finished lump of geometry, not the sequence of decisions that produced it.

Two practical consequences. Ask for the native CAD file alongside the STEP if you want to change the part later. And ask directly whether the model has a usable feature tree at all, because an auto-surfaced model and a feature-built one look identical in STEP until the day you try to move a hole.

Alongside those, expect a mesh file for printing, and the deviation report. From there the part can go into low volume manufacturing or straight onto a printer through industrial 3D printing without touching another supplier.

Three things to take away. The datum alignment at the start does more for the final model than the software does. Auto-surfaced and feature-built models cost different amounts and behave differently, so agree which one you're buying. And a model with no deviation report behind it is a claim rather than a measurement.

Send us a scan you already have, or the part itself, and tell us what you want to do with the model afterwards. That answer changes how we build it. Email sales@nexform.co.uk or use the contact form with a CAD file, a sketch or the part, and we'll come back with a quote.