CAD in a manufacturing business: the model as a production input, not a picture

Downstream readers judge a model differently from its author

A model that renders correctly on screen may still be unusable to a programmer: surfaces that do not close, imported bodies with no feature history, sketches driven by dimensions nobody can safely change, duplicate geometry hidden in a suppressed configuration. The estimator sees only the shape and prices it wrong; the programmer discovers the problem when the toolpath fails; the inspector finds it when a scan will not compare. Manufacturers who close this loop do something simple — they let the people downstream reject a model back to design with a reason, and they track how often it happens rather than treating each case as a personality clash.

Where the tolerance lives decides who is accountable

If the contractual definition is a drawing, the model is a convenience and the drawing governs. If the definition is an annotated model, the drawing may not exist and every downstream reader must be able to see the annotations in whatever they open it with. The failure mode is a hybrid: a model that carries some tolerance information, a drawing that carries the rest, and a supplier quoting from one without the other. Deciding which is the governing document, stating it on the purchase order, and keeping that decision consistent across the product range removes an entire category of dispute after parts arrive.

Exchange formats drop exactly what you needed

Neutral formats move geometry reliably and lose most of what surrounds it: feature history, annotations, material and finish data, assembly constraints. Native exchange preserves more but requires the receiving party to run compatible software, which is a commercial constraint on your supply base rather than a technical one. Version differences add a second trap, since a file saved in a newer release may not open in an older one that a supplier still runs. Establishing what format you send, what you expect back, and who checks that a translated body is still watertight prevents most of the silent geometry damage that shows up much later.

Design intent that no machine can reach

Internal corners with a radius smaller than any practical cutter, pockets deeper than a tool can go without chatter, features on faces that cannot both be presented in one setup, tolerances specified on surfaces that will move when the part is unclamped — these are not modelling errors, they are manufacturability errors that a design tool will happily accept. Every one of them costs least to correct while the model is still open on a designer screen. That means the process planner or an experienced machinist needs to see designs before release, in a review that is scheduled rather than dependent on someone being asked a favour.

Modelling standards are an argument worth having once

How a company builds models sounds like a matter of taste until somebody has to modify a colleague's work. Origin placement, unit settings, layer and naming conventions, template files, the use of library components, whether configurations or separate files represent variants: each decision affects how reusable the data is years later. Writing these down and applying them to new work, rather than retrofitting the archive, is the pragmatic route. The test of whether the standard is real is simple — hand a model to a designer who did not create it and see how long it takes them to make a change confidently.

Frequently asked questions

Should we send suppliers models, drawings, or both?

Send both, and be explicit about which one governs if they disagree. Models let a supplier programme and quote quickly; the governing definition determines what you can reject a part against. Where you rely on an annotated model with no drawing, confirm the supplier can actually read the annotations in their software rather than assuming it, because a supplier who opens geometry only will quote as though every dimension carries a general tolerance.

How do we stop suppliers working from superseded models?

Control the issue rather than the copy. Send files from a controlled release point that records what went where, mark the revision inside the file name and on the model itself, and reference the exact revision on the purchase order so the shop floor at the supplier has a check. Sending files by personal email attachment defeats all of this, because nobody can later reconstruct which version a supplier was actually holding when they cut metal.

Is model-based definition worth adopting for a mid-sized plant?

It pays off where inspection is done by scanning or by probing against the model, where the same data feeds programming directly, and where drawing production is a genuine bottleneck. It costs more where suppliers, inspectors or customers still work from paper, since you end up maintaining two definitions. A sensible route is to adopt it for new product families and internal work first, and to keep drawings for parts bought from a supply base that is not equipped for it.