First article versus sample inspection: what each proves

A first article inspection answers one question: can this tool, on this machine, with this fixture, make a part that meets every characteristic on the drawing? Sample inspection answers a different one: is this lot consistent with the defect rate we agreed? Buyers accept one document as evidence for the other, and that is where releases go wrong.

The gap is measurable. A 24-characteristic bracket inspected once tells you nothing about the 4,800 parts behind it. An 80-piece sample at AQL 1.0 tells you nothing about whether the die will hold a 0.3 mm hole position at 40,000 strokes. Each record covers a different failure.

What follows sets out what each inspection proves, the sample sizes and acceptance numbers that go with each, the record fields that survive an audit, and the point where a first article has to be repeated.

What a first article inspection proves, and what it cannot

A first article measures every characteristic on the drawing, on one part from the first production run, and records nominal, tolerance, actual and pass or fail against each. The drawing is ballooned so that characteristic 14 on the report is characteristic 14 on the print, and the report is tied to one revision. Material certificates, the finish specification and the tool or fixture reference sit alongside it.

That covers a lot of ground. It proves the blank size is right, the bend allowance in the flat pattern produces the drawn flange, the hole-to-bend distance clears the tooling, the weld fixture locates the part where the drawing says, and the coating build lands inside the callout. It is the only document that touches every characteristic at once.

It proves those things for one part at one moment. A first article reading 12.50 mm on a 12.50 ±0.50 mm dimension tells you the setup is centred. It does not tell you the process is centred, because a single part carries no information about spread. If the true standard deviation were 0.15 mm, one reading in the middle of the band is entirely ordinary, and that process would put roughly 0.09% of parts outside the limits.

To convert a first article into evidence of capability you need a sample, not a part. Thirty pieces is the usual floor for a short-term index. On a 1.0 mm tolerance band, a Ppk of 1.67 requires:

σ ≤ T / (6 × Ppk) = 1.0 / (6 × 1.67) = 0.10 mm

So the question a first article cannot answer is whether σ is under 0.10 mm. That is a capability study, and it belongs in the PPAP, not in the first article report.

The second thing a first article proves is narrower than buyers assume: it proves the part meets the drawing, not that the drawing is right. A flange that clears a tooling die in a trial bend can still foul the fixture at the next station, and a hole that measures true on the bench can close by 0.2 mm after the weld. The first article is a check of the manufactured part against the print, and it will not catch a print that cannot be made repeatably. That check is the manufacturability review, and it has to happen before the tool is cut.

When a first article has to be repeated

A first article has a trigger list, and the list is the part of the procedure that gets argued about. The obvious triggers are a new tool, a drawing revision and a material change. The expensive ones are quieter: a tool repair that touches a forming surface, a rebuild of a weld fixture, a transfer of the job between presses, or a production gap long enough that the tool is stripped, stored and re-set with a new shim pack.

A repair that welds up a worn draw radius and re-polishes it changes friction in the blank holder. The drawing has not changed, the material has not changed, and the first twenty parts will pass. The drift shows up at 5,000 strokes. Treat any repair to a forming surface as a first-article trigger, and re-measure the two or three characteristics that the repaired surface controls rather than the full balloon.

Where the customer has no trigger list, the working compromise is an annual re-qualification on the highest-runner parts plus a first article on every tool event. That is not a standard requirement; it is a contract term, and it belongs in the purchase order so that both sides know who pays for the measurement.

What sample inspection proves, and what an AQL hides

Sample inspection takes a defined number of pieces from a lot, checks them against a list of attributes, and accepts or rejects the lot on the count of nonconforming pieces. ISO 2859-1 sets the plans: lot size and inspection level give a code letter, the code letter gives a sample size, and the AQL gives an acceptance number and a rejection number.

The AQL is the most misread number in the purchase order. It is the quality level at which the plan accepts 95 times in 100. It is not the defect rate the plan rejects, and it is not a guarantee that the lot is at or below that rate. The arithmetic is worth doing once, because it changes how a buyer reads a pass.

Take a lot of 800 pieces at general inspection level II. The code letter is J, the sample is 80 pieces, and at AQL 1.0 the acceptance number is 2 and the rejection number is 3. Acceptance probability for a process running at a defect rate p is the binomial probability of 0, 1 or 2 defectives in 80:

• At p = 1.0%: P(accept) = 0.9535, which is the 95% the AQL defines.
• At p = 2.0%: P(accept) = 0.784.
• At p = 5.0%: P(accept) = 0.231.

A 2%-defective lot passes almost four times in five. A 5%-defective lot still passes nearly one time in four. That is not a defect in the standard; it is what a sample of 80 can resolve. If the consequence of a bad part is a field failure, a sample plan is the wrong instrument and the drawing needs a 100% check or a poka-yoke.

The other half of the plan is the switching rule, which most buyers never write into the order. Under ISO 2859-1, normal inspection moves to tightened after 2 rejected lots in 5 consecutive lots, and tightened returns to normal after 5 consecutive accepted lots. Tightened inspection keeps the sample size and drops the acceptance number: at 80 pieces and AQL 1.0, the acceptance number falls from 2 to 1. That single change removes most of the leniency, and it only happens if the supplier is keeping a switching record.

There is also a floor below which the plan stops discriminating. A lot of 40 pieces at general inspection level II falls in code letter D, which calls for 8 pieces, and at AQL 1.0 the acceptance number is 0. The probability of accepting a lot that is 5% defective is then 0.95 to the eighth power, or 0.663. Two lots in three pass at a defect rate the buyer would reject if they could see it. On short runs, inspect all 40 or accept that the sample is a formality.

Every record also carries a measurement decision that neither document states by default. If the characteristic is 12.50 ±0.50 mm and the gauge has an expanded uncertainty of 0.04 mm, a reading of 12.98 mm is not a pass under a 4:1 guard band, because the accepted zone ends at 12.98 mm only when uncertainty is ignored. ISO 14253-1 sets out the decision rules: without an agreed rule, the supplier and the buyer can disagree about the same reading and both be defensible.

What each document actually records

The rows are not interchangeable, and the cheapest mistake is to buy the last row and believe the first. A certificate of conformity is a statement about the specification. It carries no measurement, and it cannot be used to argue that a 0.3 mm hole position was held.

Where the first article usually fails first

Three characteristics account for most first-article rejections on sheet metal parts, and none of them is the one the buyer worries about.

The first is a hole that is too close to a bend. The flat pattern is correct, the punch is correct, and the hole distorts because the bend takes material from the hole edge. The fix is in the design, and there is a separate rule for the distance; the first article is where it becomes visible.

The second is a weld that pulls the assembly out of square after the fixture is released. The part measures true in the fixture and 1.5 mm out of diagonal on the bench. Fixture-controlled welding is a design and sequence problem, not an inspection problem.

The third is coating build on a threaded or masked feature. The powder goes on at the specified film and the thread no longer takes the fastener, or the earth stud no longer makes contact. Masking and plugging belong in the first article record, not in a note added after the first rejection.

Each of these is cheap to correct at the first article and expensive at lot 20. That is the commercial argument for doing the first article properly: it is the last moment when a change costs a drawing revision rather than a recall.

What Ganesh runs for a first article and a first lot

Press tools, progressive dies and checking fixtures are built in-house, so the first article is measured on the tool that will run the order rather than on a trial tool from another shop. Presses run from 20 T to 300 T with 250 T punching, and parts from 0.8 mm to 40 mm go through the same release route whether they are stamped, laser cut on the 12 kW fibre machine or formed on the 200 T brake.

The dimensional record is built from the ballooned drawing, characteristic by characteristic, with the instrument and its calibration status against each line. Typical sheet metal tolerance is ±0.5 mm, with tighter features held by agreement and called out on the drawing, so the first article report shows where the tight features sit rather than averaging them into a general note.

The system behind the record is ISO 9001:2015, and the same plant runs to ISO 14001:2015, which matters to buyers who audit the site as well as the parts. PPAP and first article inspection are available on request, and the submission is agreed at the quotation stage rather than after the tool is cut.

What to put on the drawing and the inspection plan

1. The characteristics that need first-article evidence, marked on the drawing, with the rest of the ballooned list agreed in the report.
2. The tolerance class for anything untoleranced, so the first article report and the sample inspection use the same limits.
3. The sample plan: lot size bands, inspection level, AQL, and whether tightened inspection is mandatory.
4. The critical characteristics that get 100% checking rather than sampling, and the gauge or fixture that does it.
5. The first-article trigger list: new tool, tool repair, revision change, material change, machine transfer, production gap.
6. The submission format: ballooned drawing, dimensional report, material and finish certificates, capability study where a critical characteristic exists.
7. The retention period for both records, because an audit two years later asks for the lot that shipped, not the part that was measured.

When neither document is enough

A first article and a sample plan both assume a stable process making a measurable part. Where the part is a weldment with a cosmetic face, the drawing characteristic that matters is appearance, and neither document grades it. Where the failure mode is fatigue at a bracket root, a dimensional pass says nothing, and the evidence has to come from a design check and a life test. Where the order is 40 pieces, the plan calls for 8 and accepts a 5%-defective lot two times in three, so 100% inspection is cheaper than the argument.

None of this is a reason to skip either record. It is a reason to write the inspection plan for the failure the part can actually cause. Send the drawing, the revision and the annual quantity, and the engineering team will set the first article scope and the lot plan before the price is fixed. Request a quote.

The plant and quality hub carries the rest of the inspection system, the stamped part inspection guide covers the in-process checks between first article and lot release, and the bracket tolerance guide sets the limits that both records are measured against. Stamped and pressed parts are built through the stamping capability.