The realistic expectation
You will move the shapes. You will largely not move the design intent. Plan the migration around that fact — translate the archive, remodel what is actively developed — and it becomes a manageable project rather than a disappointment.
CAD migrations are usually driven by something other than CAD: an acquisition, a licensing change, a customer mandate, or consolidating two engineering teams. Whatever the trigger, the technical reality is the same.
What actually survives
| Element | Neutral (STEP) | Direct translator |
|---|---|---|
| Solid geometry | Usually good | Good |
| Surfaces | Usually good | Good |
| Assembly structure | Yes | Yes |
| Component positions | Yes | Yes |
| Mates / constraints | No | Sometimes partial |
| Feature history | No | Rarely, and imperfectly |
| Sketch constraints | No | Rarely |
| Configurations / variants | No | Sometimes one at a time |
| PMI / tolerances | Newer STEP APs | Varies |
| Drawings | No | Sometimes, degraded |
| Metadata / properties | Partial | Usually yes |
The row that causes the disappointment is feature history. After migration a part is a dumb solid — accurate, but with no extrude to edit, no fillet radius to change, no pattern to re-count. Engineers expecting to open a translated model and edit it parametrically will be frustrated on day one, so set that expectation before the project starts, not after.
Segment your data before you migrate it
Treating every part identically is what makes migrations expensive. Sort first:
- Archive — never changing again Obsolete parts, superseded revisions, reference geometry. Translate as static solids. This is usually the largest group by far and the cheapest to handle.
- Reference — occasionally consulted, rarely edited Translate, and remodel only if someone actually needs to change one.
- Active — under development These need genuine parametric models. Budget to remodel them, because translation will not deliver what your engineers need.
- Purchased and standard parts Do not migrate. Re-download from the supplier or your standard library in the new system's native format.
Most migration budgets are blown by remodelling parts nobody was ever going to edit. The segmentation exercise usually saves more than any tooling decision.
Feature recognition — useful, oversold
Most CAD systems offer feature recognition, which infers features from imported geometry — detecting that a set of faces forms a hole, a fillet, or a pocket.
- Works reasonably on prismatic mechanical parts with standard features.
- Struggles with freeform surfacing, complex blends, and unusual construction.
- Produces a flat feature list, not the structured, intent-carrying tree the original had.
- Requires review — a mis-recognised feature is worse than none, because it behaves unexpectedly when edited.
It is genuinely useful for making translated parts somewhat editable. It is not parametric recovery, and vendors describing it that way are overselling.
Validation is not optional
Translated geometry can be subtly wrong in ways nobody notices until a part is manufactured:
- Compare mass properties — volume, surface area, centre of mass. Differences beyond tolerance indicate a translation problem.
- Check for invalid topology — gaps, non-manifold edges, faces that failed to stitch.
- Verify assembly positions, since transform errors are easy to miss visually.
- Spot-check critical dimensions on parts that matter.
- Automate it. Manually validating thousands of parts will not happen; a script comparing mass properties across the whole set will.
Volume comparison is the single most effective automated check. It is cheap to compute on both sides and catches most serious translation failures — missing faces, failed stitching, unit errors — without any human inspection.
What gets forgotten
- Drawings — often more numerous than models and much harder to migrate. Frequently the honest answer is to keep them as PDFs and redraw only what changes.
- PDM data — revisions, approvals, where-used relationships, ECO history. Frequently the harder half of the project.
- Custom tooling — macros, templates, title blocks, automation scripts built up over years.
- Standards and libraries — company part libraries and drawing standards must be rebuilt in the new system.
- Training — engineer productivity drops for months. Budget for it honestly.
A workable sequence
- Inventory and segment — this determines the entire budget.
- Pilot on a representative subset, including your most awkward parts, and measure translation quality.
- Build automated validation before bulk translation.
- Translate the archive in bulk, validating automatically.
- Remodel active parts — usually the longest phase.
- Run both systems in parallel until confidence is established. Do not switch off the old system on the same day.
Planning a CAD migration or building translation tooling? Tell us the systems and roughly how many parts. See our CAD engineering service and why CAD interop is hard.