Engineering Drawings and Your Chinese Supplier: Revision Control, Tolerances and the Sample That Isn't the Product

The first thing to understand is that it isn't really a document problem

A purchase order for 5,000 units of a thing and its engineering drawing are supposed to say the same thing. In practice they usually disagree in ways nobody notices until the tooling is cut. I've been the person on the other end of that discovery four times now, and each time the fix cost between two hundred and eight hundred dollars plus two weeks. The fifth time, I stopped issuing POs with attachments and started issuing them with a print pack.

What follows is the order of operations that has actually worked, and where the money gets lost if you skip a step.

Drawing, revision, and the letter that goes at the end of the filename

Every drawing carries a revision letter. The trap is that a supplier will open the drawing, work from it, then quote and produce against a print they pulled from a folder six months ago, because it's the one the sales manager has bookmarked. You can't fix that with goodwill. You fix it by putting the revision on the PO, in the line item, and by asking the supplier to send back a signed copy of the exact print they are going to use with the revision circled.

Do that once and you will learn something. About a third of the time the signed copy comes back with different content from what you sent. Not maliciously — someone "cleaned up" a dimension, or converted a PDF to DWG and lost a tolerance block, or opened the .dwg in a version that resolved a hatching layer incorrectly. All of these are real things that have happened to me.

The institutional version of this is a drawing number plus revision in the file name and in the title block, and a written rule that no order is released without a revision letter. The small-company version is a single PDF, flattened, with a big revision letter in the corner, and a signed return. I use the small-company version and it works.

Tolerances: the ones you actually mean versus the ones you inherited

Most prints are copied from an older print. That's how a ±0.2mm general tolerance survives into a part that is now moulded instead of machined, and how a datum label hangs off a face that got remodelled two revisions ago. The consequence is that the factory quotes to the inherited tolerance, and if they hit it, you don't have a claim — even if the part doesn't work.

The fix is a short tolerance review before the first RFQ, done on the dimensions that matter and left alone on the ones that don't. For an injection-moulded part, shrink varies by material and by wall thickness, and a tolerance on a plastic part that is inside half a percent of the nominal is a tolerance that either costs money or gets waived. For a machined part, the difference between ±0.05 and ±0.01 is often a second operation, and a second operation is a second setup, and a second setup is where your lead time goes.

Ask the supplier to mark up which dimensions they consider critical and which they consider negotiated. Good ones will. Bad ones will tell you all of them are critical, which is a way of saying they don't know.

Datums, GD&T and the drawing that a CAM programmer will read literally

Geometric tolerancing is not decoration. If a position tolerance is referenced to datum A which is a small hole on the edge of the part, and the part rests on datum B during manufacture, then the inspection and the manufacture are measuring two different things. This produces the specific, familiar pattern where the first article passes at the factory, fails at your end, and both reports are technically correct.

I've started asking for the datum scheme to be visible in the first article report. If the report references a datum that isn't on the print, or uses a datum numbering that doesn't match, I stop and fix the print before I book the volume run. It's a ten-minute check and it has saved me a full remanufacture once.

Material callouts: the ones that get swapped quietly

"Stainless steel, 304" on a print is a material family with a chemical range and a hardness range. Two suppliers can both deliver 304 and give you parts with noticeably different machinability, and one of them can deliver 304 with a nickel content at the bottom of the range, which is still 304. If the part has to be welded, or passivated, or hold a thread, the bottom of the range behaves differently.

What I write into the print now is the standard and the grade — AISI 304, ASTM A240, with a requirement for a mill certificate per lot, matched to the part. That last bit matters: a mill certificate for 3 tonnes of sheet that got split across four subcontractors tells you the sheet was 304, not which part is 304.

For plastics it's worse, because a resin grade is a trade name and a data sheet, not a chemical formula. If the part needs UV stability or a specific flame rating, the grade goes on the print with the manufacturer's name, and the print says "no substitutions without written approval". Otherwise you'll get a generic equivalent that reads fine on paper and yellows in a year.

Where the sample sits in all this

The signed sample is the reference point for everything downstream, and the most common failure I see is that the signed sample was made by hand, by a toolmaker, on a bench, from the CAD — and the production part is made in a mould with a different gate location. They are not the same artefact. It's not dishonest, it's how first articles get made when you want to move fast.

The only version of this that holds up is a sample made from the production tool, run under production settings, at the production cycle time, inspected and signed. That's what you should be buying when you pay for a "T1 sample". If a supplier can't provide one before the volume run, you're signing off a prototype and calling it a pre-production sample, and you should price that risk accordingly — usually in a second smaller volume order before the big one.

Packaging, because it's on the drawing on page seven and nobody reads page seven

Packaging specs get bolted onto the print as an appendix and then ignored. The two things that matter and cost real money: the unit load (how many per inner bag, per carton, per pallet, and the stack height) and the protection level (which is a written requirement, not "good export packing"). I've had a shipment of 4,000 machined brackets arrive with surface rust because the factory used normal plastic bags with no VCI paper, and the print said "export packing" on the appendix. My fault, not theirs.

Write the packaging spec as its own numbered clause with measurable items: bag material, VCI presence, carton test grade, pallet type, wrapping, and a photo of a signed-off carton. The photo of the signed-off carton is the cheapest insurance in the whole process.

The checklist I actually run

Before the PO goes out, I confirm: the print has a revision letter and a date; the supplier has returned a signed copy of that exact revision; the tolerance block has been reviewed for inherited nonsense; datums are consistent between manufacture and inspection; material grades are named with a standard and a mill certificate requirement; the sample is tool-made, not bench-made; packaging is a numbered clause with a photo. That's seven things and it takes about forty minutes. The alternative has cost me, in aggregate over five years, something in the region of eleven thousand dollars and a lot of weekends.

None of this requires a quality department. It requires that you treat the drawing as a negotiating document rather than a picture, and that you get the supplier to say in writing which version of it they are about to build.