First article inspection: what FAI proves and what it does not

A bracket with sixty numbered characteristics comes back from a first article with fifty-eight results and two entries that read "reference". The two are the hole positions that the assembly fixture uses as datums. Everything else is measured to three decimal places and signed, and the report still cannot answer the only question the customer asked.

First article inspection is a complete record of every drawing characteristic on parts made with production tooling, at the production location, by the production process. It is a strong document when it is complete and a misleading one when it is a sample dressed as a record.

The standards and compliance hub covers the other documents that travel with a first lot, including mill certificates and weld records.

What does a first article inspection actually record?

It records four things: that the part number and revision are correct, that the materials and special processes are the specified ones, and that every characteristic on the drawing has been measured with a result inside tolerance. The fourth record is the signature that ties the three together.

AS9102C, published by SAE in 2023, is the aerospace standard and the most useful template even outside aerospace, because it splits the record into three forms:

• Form 1 — part number accountability. The part, its revision, the serial or lot, and any sub-assemblies that make it up.
• Form 2 — product accountability. The raw material, its specification and certificate, and each special process such as heat treatment, plating or coating, with the processor's approval reference.
• Form 3 — characteristic accountability. Every characteristic, numbered, with its nominal, tolerance, measured result and pass or fail.

The standard does not set a sample size. It requires parts from the first production run, which is the phrase that most disputes turn on.

The characteristics a sheet metal FAI has to cover

A drawing carries more than dimensions, and the report has to reflect that. The table maps the characteristic types on a typical fabricated part to how they are recorded. The last column is what Ganesh puts into the pack.

Two failure patterns show up here. The first is a report that measures dimensions and skips notes, which is where coating thickness and thread class usually hide. The second is a report that measures the first part thoroughly and the second and third only at the features that were difficult, which turns the third part into a sample.

How long a first article takes, and why the drawing decides

Work the time for the bracket above: 60 characteristics, three parts. A CMM programme needs the part set up on a granite table with the datums established, and the first part absorbs 90–120 minutes of programming and setup when no programme exists. After that, 60 measured points at roughly 25 seconds each is about 25 minutes per part on the machine.

Manual features are slower per feature but need no programme. A height gauge, a bore gauge and a thread plug take about 90 seconds each, so 15 manual features add 22 minutes per part. Three parts therefore land between 2.5 and 4 hours of direct inspection, before anyone writes the report.

The report is the underestimated part. Sixty characteristics across three parts is 180 result lines, each needing a nominal, a tolerance, an actual, an instrument and a pass or fail. Two to three hours is realistic, and it is where transcription errors enter.

Three things move the number. Size is the first: a 2.4 m frame will not fit a small CMM, so it has to be measured in place with a laser tracker or on a large-bed machine, and that is a different quotation. Tolerance is the second: a ±0.05 mm callout cannot be settled with a caliper, and the instrument choice multiplies the time. Surface is the third: dry film thickness, adhesion and gloss each need their own instrument and their own record.

That arithmetic is worth doing before the purchase order, because a first article on a complex assembly can cost more than the first lot of parts.

FAI, sample inspection and capability: three different claims

These three get used interchangeably in quotations and they answer different questions.

An FAI answers "did we make this part to this drawing, with this process, once". A sample inspection answers "are the parts in this lot behaving like the last lot". A capability study answers "will this process hold the tolerance over thousands of parts".

The statistics are worth knowing because they set the expectation. If thirty consecutive parts pass a dimensional check, the 95% upper confidence limit on the defect rate is 9.5%. That is not a strong statement about a tolerance, and it is the strongest statement thirty parts can make.

To demonstrate a defect rate below 2% with the same confidence, the run has to reach 149 consecutive good parts. Automotive PPAP practice goes further and asks for an initial process study over a production run, with a minimum of 300 consecutive parts unless the customer agrees otherwise.

The second consequence is that an FAI on three parts says nothing about drift. A progressive die that produces one good part in three cavities, a brake that springs back differently on the first stroke of a shift, and a weld cell that moves with fixture wear all pass a three-part FAI and fail in month three.

Measurement uncertainty decides whether a result means anything

A measured value is not the same as a true value. ISO 14253-1 sets the approach for proving conformity with uncertainty in play: where the measured value sits inside the uncertainty band around a limit, the conformance decision is indeterminate and has to be resolved by agreement or by a better measurement.

The practical rule is that the instrument should resolve to roughly one tenth of the tolerance band, and no worse than a quarter. On a ±0.5 mm general tolerance, a 0.02 mm CMM uncertainty is 4% of the 1.0 mm band and is comfortable. On a ±0.1 mm callout the same uncertainty is 20% of the band, and a result at 0.09 mm is not evidence of conformity.

That is why the request matters. A report that names the instrument but not its uncertainty, and gives three decimal places on a tolerance with two, is offering precision it has not earned. Ganesh works to ±0.5 mm as a typical sheet metal tolerance and tighter on critical features by agreement, and the agreement is the place to state the gauge and the uncertainty.

What to send, and what to expect back

The most common cause of a slow FAI is a drawing that cannot be ballooned. Reference dimensions mixed with toleranced ones, a datum scheme that changes between views, and a revision cloud over the only dimension that matters all cost time at the inspection bench.

A workable pack looks like this:

• A single drawing revision, with the revision letter on every page.
• A ballooned copy with a unique number on every characteristic, including notes.
• Datums defined on the drawing, not inferred from the assembly.
• Critical and major characteristics marked, so the inspector knows where to spend time.
• The finish specification by process and thickness, with acceptance criteria.
• The material grade and standard, and the certificate type required.
• The sample size and whether parts from each die cavity are needed.

Expect the report back with the ballooned drawing, the three forms, the certificates attached by number, and a deviation list. If there are deviations, they should appear as deviations with an approval reference, not as a result quietly rounded into tolerance.

A deviation entry needs six fields: the characteristic number, the measured value, the drawing limit, the quantity affected, the proposed disposition, and the customer's approval reference. A disposition is one of use-as-is, rework, or concession; "noted" is not a disposition. A report that shows a value outside tolerance and no disposition is not a first article, it is a non-conformance that will be discovered at goods-in with less context than the inspector had.

Partial FAI, re-FAI and the change that triggers them

A first article does not expire on a calendar. It expires on a change, and the change determines the scope of the next one.

A full FAI covers a new part, a new tool, or a move to a different manufacturing location. A partial FAI covers a change that affects a defined set of characteristics: a revised hole pattern, an added bracket, a switch from laser-cut blanks to a pierce tool. AS9102 provides for the partial form because re-measuring 60 characteristics to verify four is waste.

Three conditions make a partial FAI auditable. It names the change, usually by engineering change number and revision. It measures every characteristic the change can touch, including the ones downstream that move because of it, such as a hole position that shifts when the datum edge is re-trimmed. And it references the original report, with its date and its part revision, for every characteristic that was not re-measured, so the two documents read as one record.

A gap in production is the other trigger. Two years without a run means the tool has sat, the fixtures may have been moved, and the operators are different. Treat the restart as a first article even where the drawing has not changed.

The cost of skipping this is asymmetric. A partial FAI on four characteristics is a few hours. A launch held because the customer's auditor cannot find a record for the revision that shipped is a week.

What Ganesh runs

Ganesh produces parts from 0.8 mm to 40 mm on presses from 20 T to 300 T, a 250 T punching line, a 12 kW fibre laser with a 40 ft × 16 ft bed and a 200 T, 4,450 mm brake to 10 mm. Tooling, welding fixtures and checking fixtures are built in the in-house tool room, which matters for an FAI because the fixture that holds the part for measurement is part of the process being recorded. The stamping capability covers the press range and the tool room.

First article inspection and PPAP are available, and the works hold ISO 9001:2015 and ISO 14001:2015. Coated parts can be recorded with dry film thickness and cross-hatch adhesion results from the finishing line, and welded assemblies can be recorded against a numbered weld map.

The honest boundary is sample size and metrology depth. A three-part FAI on a 60-characteristic bracket is a day of inspection work. A request for every characteristic on every part, or for a CMM programme on a 2.4 m frame, changes the scope and the lead time, and it should be priced as inspection rather than assumed inside the piece price.

When an FAI is the wrong instrument

For a high-volume part entering an automotive programme, PPAP is the governing process and the FAI sits inside it as the dimensional element. Running a standalone FAI and calling it approval leaves the customer's quality team without the process studies they need.

For a cosmetic part, an appearance approval report against a master sample is the record that resolves disputes, and a dimensional FAI will not settle a colour argument.

For a part with a single critical feature and no other risk, a targeted inspection plan with a capability study on that feature is more informative than a sixty-line report that treats every dimension as equally important.

And for a part that will be re-tooled in three months, the FAI is a snapshot with a short life. Note the trigger in the purchase order so the next report is not a surprise.

The inspection plan for repeat production picks up where the first article stops, and the stamping tolerance guide explains how the characteristics that fill the report are chosen.

Send the drawing revision with the characteristics marked on a quote request, and the engineering team will return the ballooned drawing, the measured results and the certificates that go with them.