The report is the deliverable
A first article inspection (FAI) is usually treated as a gate to pass before shipping. On prototypes it is something more useful: the first quantitative proof that the drawing and the process agree. A CMM report with forty rows of numbers and no stated measurement uncertainty proves very little. A report with twelve rows and a stated instrument, environment and datum scheme proves a great deal.
The difference matters most on parts with tolerances tight enough to matter — the ±0.05 mm bores, the 0.02 mm flatness callouts, the true position on a bolt circle. Those are exactly where the measurement method decides the answer.
Balloon the drawing before touching the machine
Number every characteristic on the drawing: dimensions, GD&T frames, notes, thread callouts, surface finish symbols, material and heat treat specs. Each balloon becomes one line in the report. This is the AS9102 approach, and the discipline catches the most common FAI failure: a characteristic nobody listed, so nobody measured.
Two things get skipped most often: notes (deburr, marking, cleaning) and reference dimensions that turned into inspection dimensions by accident. If a number is in parentheses, it is reference and should not be reported as pass or fail; if a supplier is failing you on a reference dimension, the drawing is ambiguous and should be fixed.
Pick the instrument from the tolerance, not from habit
The useful rule of thumb is that measurement uncertainty should be at most one fifth of the tolerance band. In practice:
- ±0.5 mm or looser: calipers and height gauges are fine. A good caliper still has roughly 0.02–0.03 mm of operator influence on an outside diameter.
- ±0.05 mm: a bridge CMM with 2–3 µm volumetric accuracy, or a micrometer in the hands of someone who knows the part. This is the range where most CNC machining prototype work lives.
- ±0.01 mm or tighter: temperature-controlled CMM at 20 ± 1 °C, with the part soaked 4–8 hours. Not optional.
The temperature point is worth doing the arithmetic on. Aluminum expands about 23 µm per metre per °C. On a 100 mm part, a 5 °C drift between the gauge and the drawing reference moves every dimension by 11.5 µm — the entire tolerance band on a ±0.01 mm callout. Steel at roughly 11 µm/m·°C is more forgiving, but a shop that measures aluminum parts at 26 °C and reports them against a 20 °C drawing is reporting noise.
Fixturing moves the number more than the machine does
A CMM measures the part as it is held, not as it is used. Support the part on its three primary datum points, clamp directly over the supports, and use the minimum force that keeps it from moving. Clamp in the middle of a 1.5 mm wall and it will deflect 0.05–0.1 mm under surprisingly light pressure, which then shows up as a flatness failure that disappears when the part is unclamped.
For thin or flexible parts, report the clamping scheme in the report. Better yet, measure the feature in the free state with an optical or scanning method and say so.
Datum alignment: the choice that changes the verdict
Align to the datum reference frame exactly as the drawing defines it. Best-fit alignment is legitimate only when the drawing invokes it, usually through a profile callout on a free-form surface. On a normal prismatic part, a best-fit that distributes error across all surfaces can turn a genuine 0.06 mm profile deviation into a comfortable 0.03 mm — and the part still will not assemble.
Two details that trip people up:
- Datum shift at MMB. If the drawing shows a datum at maximum material boundary, the reported position must account for the allowed shift. Reporting the unshifted value over-rejects good parts.
- True position as a diameter. Position tolerance is diametrical. If your CMM outputs a radial deviation, double it before comparing to the frame, or you will pass parts at twice the allowed error.
What a usable report contains
Whether it is one page or twenty, the structure should be the same:
- Balloon number, feature description, nominal, tolerance, actual, deviation, pass or fail
- Instrument used per characteristic, with its stated uncertainty
- Ambient temperature at start and end, and part soak time
- Datum scheme and alignment method
- CMM program revision, operator, date, and the drawing revision inspected
- An explicit out-of-tolerance table with disposition — rework, concession, or scrap — not a number buried in row 27
A single flagged dimension with a stated disposition gets a decision in an hour. The same dimension highlighted in yellow somewhere in a forty-row spreadsheet gets discovered after the parts have shipped.
Four traps that produce false results
Measuring a burr. A rolled edge on a milled slot reads 0.05 mm oversize. Deburr first, then measure, and record that the part was deburred.
Threads by CMM. A CMM can locate a thread's pitch diameter approximately, but a go/no-go gauge is the acceptance method for thread fit. Report the gauge result, not a CMM-derived diameter.
Measuring before finishing. Type II anodize adds roughly 5–15 µm per surface; hard anodize can add 25–50 µm. A 10.00 mm pin becomes 10.02–10.03 mm after coating. Inspect critical fits after finishing, or state the pre-finish dimension and the expected growth.
Claiming surface finish without a trace. Ra is a statistical parameter from a traced profile. Quoting a single Ra number from a visual comparison to a comparator plate is not a measurement. If Ra 0.8 µm is on the drawing, ask for the trace and the cutoff length used.
Sampling and cost: why the characteristic list should be short
FAI means 100 % of the listed characteristics on the first part, and typically on the first three parts for a new process. After that, AQL sampling is reasonable — but only for characteristics you have already proven capable.
Cost scales with characteristic count, not part size. On a prototype with 60 dimensions, inspecting all 60 can cost more than the part. The pragmatic split is ten to twenty key product characteristics — the fits, the sealing faces, the mounting pattern — inspected fully, plus a general dimensions check at a coarser level. That gets you a defensible report at a price that still makes sense for a 20-piece run.
The same logic applies to castings. A die casting FAI usually inspects the machining datums and the sealing surfaces fully, then handles the rest of the as-cast geometry by profile at a few section planes. Checking every rib thickness to ±0.1 mm on a first article tells you less than checking the three faces the part will actually be located on.
What good looks like
A first article report should let an engineer who has never seen the part answer three questions in under five minutes: which dimensions are critical, did they pass, and if not, what happens next. If the report cannot do that, the inspection happened but the information did not survive it.
Ask for the report before the parts ship, and ask for the instrument list with it. It is a cheap question, and the answer separates a supplier who measures from one who checks boxes.
