Scan-based first article inspection in aerospace

A first article inspection is not a measurement task that happens to produce paperwork; it is an accounting task that happens to require measurement. Every characteristic on the drawing — every dimension, every profile callout, every flag note about edge break and surface treatment — gets a number, a requirement, a result, and a method, and the report is only done when nothing is unaccounted for. On a simple bracket that is an afternoon. On an airfoil, a housing, or an integrally bladed rotor, the accounting itself becomes the dominant cost — which is exactly why 3D scanning keeps being proposed for FAI, and exactly why it needs to be proposed carefully.

How this report was built. Unlike the forging & casting report, which drew on a dedicated research corpus, the findings here mostly synthesize standards documents and established industry practice, with a handful of platform-specific items drawn from that same corpus and marked as such in the text: statements about what AS9102 and its neighbors require paraphrase the standards listed at the end, claims marked reportedly or vendor-sourced come from vendor case material I could not independently verify, and where my support is partial I say so in the text. Consult the standard itself — and your customer's supplement to it — before acting on any paraphrase here. The 3D models are AI reconstructions for illustration; they are not scans of real parts, and no dimensional claim rests on them.

1 · What a first article inspection legally is

Start with what the document is for, because everything else follows from it. AS9102 — the SAE standard that aerospace customers flow down for first article inspection — exists to produce objective evidence that a production process, run the way it will actually run, yields a part meeting every design characteristic and specification requirement: understood, accounted for, verified, and documented. Three words in that sentence carry the legal weight. Production means the first article must come from the real process — production tooling, production methods, production sequence — not a model-shop favor that happens to share a part number. Every means complete accounting: each characteristic on the drawing or model, flag notes included, gets a unique number and a verdict. Documented means the first article inspection report — the FAIR — is the deliverable; the part itself is almost incidental.

The FAIR has a canonical shape: three forms. Form 1 (Part Number Accountability) identifies the part and ties an assembly-level FAI to the detail-part and sub-assembly FAIs beneath it. Form 2 (Product Accountability) lists materials, special processes, and functional tests against their specifications and certifications — the paper trail behind heat treat, coatings, NDT. Form 3 (Characteristic Accountability, Verification and Compatibility Evaluation) is the heart of the exercise and the subject of this report: every design characteristic, uniquely numbered, with its requirement, its measured result, the tooling or method that produced the result, and — if it failed — a nonconformance reference. The standard permits equivalent formats; what it does not permit is missing content. And a characteristic that fails does not get quietly annotated: it gets a documented nonconformance, and the FAI stays open until the disposition is closed and the affected characteristics are re-verified.

When does an FAI trigger? A new part number's first production run takes a full FAI; afterward, changes trigger partial FAIs scoped to the affected characteristics. The re-accomplishment rules, paraphrased from the standard (its wording, and your customer's supplement, governs):

The enforcement mechanism is contractual. AS9100's production-process-verification requirement (clause 8.5.1.3 in revision D) obliges you to verify, using a representative item from the first production run, that the process, documentation, and tooling can produce conforming product — and AS9102 is the standard method of showing it. AS9145 folds the FAI into the aerospace version of PPAP as one element of the approval package. Customer supplements then tighten whatever the standards left loose. If you are doing FAI, it is because a contract says so, in a chain that usually starts at an airframer or engine OEM and flows down through every tier — which is also why the report format is so rigid: your Form 3 may be read three tiers above you by someone who will never see your shop.

2 · The characteristic-accounting problem

Trigger	Typical scope
New part number, first production run	Full FAI
Design change affecting fit, form, or function	Partial — affected characteristics
Change in manufacturing source, process, inspection method, location, tooling, or materials	Partial, where the change can affect fit, form, or function
Change to an NC program, or its translation to another medium	Partial — affected characteristics
An event that adversely affects the manufacturing process	Partial or full, per impact
Lapse in production — commonly cited as two years, and the customer can specify otherwise	Full

The practice everyone knows by its nickname: ballooning. Each characteristic on the drawing gets a numbered balloon (some shops say bubbling), and the balloon numbers key the drawing to the Form 3 rows one-to-one. The point is auditability — a customer quality engineer must be able to pick any requirement on the drawing and find its result in the first article inspection report without interpretation, and pick any row on Form 3 and find its requirement on the drawing.

The arithmetic is what makes this brutal. A simple machined bracket might carry a few dozen characteristics. A housing, a manifold, or an airfoil casting easily runs into the hundreds before you count the note-driven characteristics — break edges, identification marking, surface treatment coverage — that also require rows. Done manually, the workflow is: mark up a drawing PDF balloon by balloon, build the characteristic list in a spreadsheet, measure everything, transcribe results row by row, then reconcile. Every step has a characteristic failure mode: missed notes, transposed digits, balloon numbers that drift out of sync with the spreadsheet after a drawing revision, results pasted against the wrong row. None of these are measurement errors — the part may be perfect — but each is a defect in the objective evidence, and customer FAI rejections for documentation defects rather than nonconforming metal are a familiar frustration in the supply chain (I know of no reliable public statistics on the split, so I won't invent one).

Practitioners commonly report that on complex parts the documentation effort rivals or exceeds the measurement effort — stated here as general industry experience, not a measured figure. A small software industry exists precisely because of this: tools that extract characteristics from drawings by OCR, or harvest them directly from model annotations, and manage the balloon-to-form linkage. Whatever the tooling, the structural fact remains: Form 3 is per-characteristic accounting, so anything that lowers the marginal cost of one verified, traceable result compounds across hundreds of rows. That is the lens through which to evaluate scanning — not "is the scanner accurate," but "what does it do to the cost and integrity of a row."

3 · Where scanning genuinely helps — and where it doesn't

The honest case for structured-light or laser scanning in first article inspection rests on three mechanisms.

Full-field evidence for full-field callouts. A profile-of-a-surface tolerance applies to the whole toleranced surface, but a tactile CMM samples it — a probe path is a decision about where the truth lives. A dense scan evaluates the callout against the surface as a whole, and the deviation colormap is supplementary objective evidence of a kind a point report cannot give: it shows where the profile condition is consumed, not just that its worst point passed. For drawings that lean on profile and surface callouts — castings, airfoils, sheet and composite contours — this is the difference between sampling a requirement and actually inspecting it.

One setup, many rows. A single digitization captures the accessible geometry once; hundreds of surface-borne characteristics are then evaluated from the same dataset, each keyed to its balloon number in the inspection plan. The evidence isn't automatic — every characteristic still needs its own evaluation with the drawing's datum logic, not a lazy best-fit — but the marginal cost of the next row collapses, which is precisely the economics section 2 asked for.

Cheap re-accomplishment. Partial FAIs are where scanning quietly earns most. When the trigger is an NC-program change on one operation, a parametric inspection project re-run on a new scan reproduces the affected characteristics — same plan, same datum logic, same report format — at marginal cost. The re-FAI stops being a project and becomes a run.

Now the other column, because the failure mode in aerospace is overclaiming. Scanning does not replace the hard gauge or the tactile CMM on the characteristics they exist for. Small tight-tolerance diameters are the documented weak point: in the tactile-versus-optical comparison study cited in my forging report, structured-light step heights read 23–36 µm high against a CMM and small-hole diameters erred by up to ~125 µm, improving with hole size — bias, not noise. Threads are verified by gauging, not surface reconstruction. Surface texture callouts need a texture instrument. Bores that serve as datum features for tight positional schemes deserve tactile measurement, because the datum frame's uncertainty propagates into every characteristic referenced to it. Internal geometry — cooling passages, cored cavities — is CT territory, with its own standards ecosystem (ISO 15708, ASTM E1441 and E1570, VDI/VDE 2630 for dimensional CT). The doctrine that regulated supply chains actually run, and the one I argued for on the forging side, is hybrid inspection: tactile holds the datums and the tight primitives, full-field optical owns the freeform surfaces and the profile callouts, CT owns the inside.

Form 3 is built for exactly this. The form asks each characteristic for its result and its method — it was never premised on one instrument doing everything. A well-planned scan-based FAI is really a method map: every balloon number assigned in advance to scan, CMM, functional gauge, or certificate, with the assignment agreed before anything is measured. The question is never "can we FAI this part with the scanner"; it is "which rows should the scanner own."

4 · Model-based definition: the drawing is leaving, the accounting isn't

The quiet revolution underneath all of this is that the drawing is losing its monopoly. Under model-based definition, the 3D model with its embedded annotations — PMI, product and manufacturing information — is the design authority, governed by digital-product-definition practices standardized in ASME Y14.41 and ISO 16792. AS9102's current revision C accommodates this world explicitly: design characteristics may come from 3D model data rather than a 2D drawing, and the forms are media-neutral — electronic formats carrying the required content are as valid as paper. Ballooning survives the transition, but transformed: instead of numbering views on a PDF, you number annotations in the model, and the balloon-to-row linkage can be a data relationship instead of a clerical one. This is where scan-based FAI becomes genuinely natural, because the inspection plan, the nominal geometry, and the characteristic list all live in the same dataset the scanner's software aligns to.

But here is the part that surprises engineers who expect a measurement argument: acceptance of model-based, scan-derived FAI data is procedural, not physical. No customer signs off because the scanner's spec sheet is impressive. What they audit is process: documented procedures for handling digital product definition data, configuration control of datasets (which model revision, which derivative, who translated it), and verification of the measurement software chain. The pattern was codified by Boeing's D6-51991, the quality standard its suppliers meet for digital product definition, which includes verifying the measurement software you use against the datasets you use it on. In the same spirit, Nadcap runs a measurement and inspection accreditation family (AC7130) for exactly this class of capability; one accredited UK aerospace scanning bureau's vendor case study — reported as AC7130/4 in the material I have, vendor-sourced — is candid that the accreditation demanded heavy investment in process and equipment, not just instruments. The certificate on the scanner (VDI/VDE 2634 historically, ISO 10360-13:2021 now — the transition is covered in the forging report) bounds the instrument, not your task. What makes scan data acceptable on a Form 3 is that your process for producing it is approved, your software chain is verified, and your measurement uncertainty reasoning is available when the customer asks.

The practical consequence: get the process approval before the FAI, not during it. An FAI package is the worst possible vehicle for discovering that your customer's DPD requirements don't recognize your mesh-derived diameters.

5 · The blade and the blisk

If one part family justifies the whole argument, it is the airfoil. A blade's aerodynamic surfaces are toleranced in ways that are structurally hostile to sparse probing: profile tolerances on section curves cut at defined span stations, and section-level parameters — chord, thickness distribution, leading- and trailing-edge radii, twist, bow, sweep — that are relations across the whole section, not point measurements. (That parameter list matches the capability set of dedicated airfoil-inspection software; most of the enumeration matches ZEISS INSPECT Airfoil's documented feature list, via my research corpus, but the parameters themselves are standard turbomachinery practice and vendor-independent.) A touch probe samples a section; a scan digitizes it. For twist and bow — relations between sections — full-field data is not merely more convenient, it is the representation in which the characteristic naturally lives. This is why airfoil work was among the first aerospace niches where optical inspection became routine: an investment-casting house like Doncasters Bochum runs automated optical cells on turbine blades and reports measuring times cut two-to-three-fold (vendor-sourced case study), and engine-overhaul shops use full-field deviation data to replace subjective accept/scrap judgments on worn airfoils (vendor-sourced).

The honest caveat rides along: the leading and trailing edges — the most aerodynamically sensitive features on the part — are also the hardest optical targets on it: thin, high-curvature, prone to sparse or noisy data at glancing angles. A blade FAI plan that does not specify how edge data is filtered, how edge radii are fitted, and where tactile or optical-comparator methods take over is not yet a plan. Edge characteristics are frequent candidates for the CMM column of the method map, and which fit criterion the evaluation uses stops being academic at a 0.1 mm edge radius.

Then multiply. A blisk — an integrally bladed rotor, blades and disk machined or welded as one piece — takes every per-blade characteristic and repeats it for every blade on the wheel. Twenty-odd blades, dozens of characteristics each, plus the disk itself: the Form 3 for a blisk first article inspection report runs to a characteristic count that makes manual accounting genuinely hazardous, on a part whose scrap cost concentrates an entire stage's value in one piece of metal. This is the clearest case in aerospace for the plan-once economics of section 3: one alignment, one digitization strategy, per-blade evaluations templated and repeated by pattern, results flowing to rows by construction rather than transcription. One accounting decision deserves written customer agreement before you start: how repeated identical characteristics are recorded — a row per instance, or a summarized worst case with the instances traceable behind it. AS9102 practice varies here, and discovering the customer's preference at submission is the expensive way to learn it.

6 · Plan once, template the report, automate the accounting

What does automation of scan-based FAI actually look like? Less than the brochures imply, and more useful. Being specific about the mechanism is the best defense against overclaiming, so here is the workflow as I build it:

Plan once. The inspection plan is built against the nominal model before the first article exists: every characteristic constructed with the drawing's (or PMI's) datum logic, named by its balloon number, method-mapped per section 3. In a parametric inspection environment, that plan is not a document — it is an executable object. Replace the measured data, and every characteristic recomputes.

Template the report. Modern inspection software carries report templates and table exports; the characteristic table — number, requirement, actual, deviation, method — exports to the formats quality systems ingest, which is most of what a Form 3 needs. Packaged app frameworks go further: the platform I script for ships report templates, import/export templates, and checks as installable packages, so a shop's FAIR format becomes a deployable artifact rather than tribal knowledge (capability list per vendor documentation, via my research corpus). One vendor case describes exactly this pattern in production — templated reports communicating scan results from an automated cell run overnight by non-metrologists (vendor-sourced).

Automate the accounting, not the judgment. The honest boundary: what automation removes is transcription — the results-to-rows step where documentation defects breed. What it does not remove is the engineering. Someone still defines characteristics correctly, validates that the software's evaluation matches the drawing's intent, and disposes nonconformances. And one caution from my research corpus deserves quoting because vendors rarely volunteer it: standards-aware GD&T checks may re-associate source points rather than using your construction's fit, so the inspection characteristic, project standard, and datum logic must be validated separately — characteristic by characteristic, before the template is trusted, not after the customer asks. A templated FAI project inherits the quality of that validation forever; it is the highest-leverage hour in the whole workflow. If you want to see what this looks like concretely, the worked examples on this site are built on the same discipline.

What I will not claim: that any of this "generates your FAI at the push of a button." The measurement takes as long as the measurement takes, edge cases need humans, and the customer-facing forms have fields — process references, certifications, functional test results — that no geometry pipeline fills. The realistic outcome of good automation is that the Form 3's geometric rows become a byproduct of a validated plan instead of a separate clerical project, and that partial FAIs become runs instead of projects. That is worth a great deal. It is not magic, and a first article inspection is the wrong place to discover the difference.

7 · What to decide in writing before the first scan

Design authority: drawing or model, which revision, and whose translation. If MBD, name the PMI source and the dataset configuration controls. Every balloon traces to it.
The method map: every characteristic number assigned to scan, CMM, gauge, or certificate — agreed with the customer, because hybrid inspection is the doctrine, and Form 3's method column expects it.
Datum and alignment strategy: the drawing's datum logic replicated in the evaluation software, validated characteristic by characteristic; best-fit only where the drawing says so. Fit criteria named, not defaulted.
The software chain: software and version recorded in the FAIR, verification evidence per your customer's digital-product-definition requirements, process approval secured before the FAI — not during.
Measurement capability basis: what your MSA covers (for optical work, scan strategy and meshing — not just operators) and what uncertainty statement backs the tight rows. Engine-sector customers increasingly point to AS13100 and its reference manuals for expectations here.
Recording conventions: repeated identical characteristics, note-driven characteristics, and nonconformance handling — one row per instance or summarized, and who signs the disposition.
The re-FAI plan: the parametric project maintained under revision control, so partial FAIs on process changes are runs, not projects.

Every line of that list is decided before a single fringe pattern hits the part, and every line is cheaper to decide in writing than to renegotiate inside a rejected FAIR. If your shop has the scanner and the AS9102 flow-down but the decisions above were made by defaults, that is exactly the work I offer — starting with a fixed-fee pipeline assessment: hello [at] metrologymaven [dot] io.

Scan-based first article inspection: what AS9102 actually asks for