AS9103, Variation Management of Key Characteristics, is the aerospace industry's SPC standard, published by SAE for the International Aerospace Quality Group and sitting alongside AS9100 (quality management), AS9102 (first article inspection) and AS9145 (APQP/PPAP for aerospace).
Its subject is narrower and sharper than "do SPC". It is about key characteristics — the features whose variation significantly affects fit, performance, service life or manufacturability — and it asks that their variation be managed: understood, monitored, controlled and reduced across the life of the programme.
What it expects, in practice
- Key characteristics are identified and flowed down. From the design authority, through the drawing and the contract, into your control plan and your suppliers' control plans. If a KC arrives on a drawing and does not appear on a control plan, the flow-down has broken.
- Measurement capability is established first. A chart drawn with a gauge that cannot resolve the variation is a chart of the gauge. MSA / gauge R&R is a precondition, not an optional extra.
- The process is shown to be stable before it is shown to be capable. In that order. A capability index computed on an unstable process is a number describing a process that does not exist, because there is no single distribution to describe.
- Data is collected and charted during production, not only at first article. AS9102 answers "was this part right"; AS9103 answers "is this process predictable".
- Out-of-control conditions get a documented reaction. Cause, action, verification — the same loop as everywhere, with the aerospace expectation that the record survives audit years later.
- Variation is reduced over time. The standard is not satisfied by a stable chart forever; the point of watching is to act on what you learn.
The low-volume problem the standard does not solve for you
AS9103 was written for an industry that frequently makes eleven of something a month. That creates real methodological difficulty:
- Three-sigma limits estimated from twenty individual readings are imprecise, and twenty readings can be two months of production.
- Setup-to-setup variation usually dominates, so a baseline taken inside one setup produces limits that flag every subsequent setup — a chart that will be ignored within a month.
- Capability indices on a handful of parts carry enormous uncertainty. A Cpk of 1.4 from 25 readings has a lower 95% confidence bound near 1.05, which is a different conversation with a customer. Our Cpk confidence calculator will show you the interval for your own n.
The practical answers — baseline across setups rather than within one, chart the machine rather than the part number by normalising to deviation from nominal, and re-baseline once deliberately rather than continuously — are covered on the aerospace industry page, with a worked chart of a key characteristic on a 28-part run.
Where this platform helps, and where it does not
Helps:
- Individuals (I-MR) charts as a first-class chart type, which is what low-volume KC monitoring actually needs.
- Limit sets versioned with their baseline window, author and timestamp, so a re-baseline is a defensible decision with a record attached.
- Signals with a lifecycle — acknowledged, cause assigned, closed by a named person — which is the documented reaction the standard asks for.
- Capability alongside the chart, computed by an engine whose tests reproduce published worked examples.
- A share link so a customer's supplier-quality engineer can see six months of a KC without a licence.
Does not:
- No MSA or gauge R&R. The precondition above is not something we do.
- No AS9102 first-article forms, no ballooning, no bubble numbers.
- No non-normal capability. Position, flatness, roundness and runout are bounded at zero, are not normal, and a normal-assumption Cpk on them will mislead you. This is a real gap for aerospace KCs.
- No CUSUM or EWMA for slow drift on high-value parts.
- No serial-number genealogy or configuration management.
The document
AS9103 is sold by SAE and through the IAQG member associations. As always, this page is our reading of what it asks for; the standard is the authority.