AQLAcceptable Quality Level

AQL — Acceptance Quality Limit — is the worst tolerable process average defect rate, expressed as a percent (e.g. AQL 1.0 means 1.0 % defective). It is not a guarantee that 1.0 % defective is what you'll get; it is the input you feed into ISO 2859-1 / ANSI Z1.4 to look up the right sample size and accept/reject numbers. Plans are designed so a process running at the AQL has a ~95 % probability of producing a lot that gets accepted, and processes much worse than the AQL have a steeply falling probability of acceptance.

AQL plans are widely used for incoming inspection of components, finished-device QC, packaging-line inspection, and any situation where 100 % inspection is impractical and process capability data is unavailable or insufficient. Medical-device inspection (FDA QSR / ISO 13485) and pharma packaging inspection (EU GMP Chapter 5 / USP <1207> for container-closure integrity) both routinely reference ANSI Z1.4 / ISO 2859-1 sampling.

Inputs: lot size N, inspection level (typically II — General), AQL value (e.g. 1.0 %). The standard yields a sample-size code letter (a function of N and the inspection level), then for that letter and that AQL the plan gives a sample size n, an accept number Ac and a reject number Re. You draw n units at random from the lot, count defectives, accept if defectives ≤ Ac, reject if defectives ≥ Re.

Note: a higher AQL means a more permissive plan (smaller sample, larger Ac). AQL 0.10 is much tighter than AQL 4.0 — at the same lot size, AQL 0.10 might allow 0 defects in 200, while AQL 4.0 might allow 14 in 200.

The most under-used part of the standard. Lots are not all sampled under the same plan — the inspection state depends on recent supplier history. The default is Normal. If two of the five most recent lots are rejected under Normal, the plan switches to Tightened (tighter Ac/Re for the same sample size). If five consecutive lots are accepted under Tightened, switch back to Normal. If ten consecutive lots are accepted under Normal and other supplier-stability criteria are met, the plan may switch to Reduced (smaller sample for the same protection — economical reward for good performance). If a single Reduced lot is rejected or fails the switching score, switch back to Normal.

The Operating Characteristic (OC) curve plots, for a given plan (n, Ac), the probability of accepting a lot vs the true defect rate of the lot. At p = AQL the curve gives the Producer's Risk (typically α ≈ 5 % — a 95 % probability the lot is accepted when the process is at AQL). At a much higher defect rate — the Lot Tolerance Percent Defective (LTPD) or Rejectable Quality Level (RQL) — the curve gives the Consumer's Risk (typically β ≈ 10 % — only a 10 % probability such a bad lot is accepted).

The shape of the curve depends almost entirely on n. Doubling the sample size makes the curve much steeper — better discrimination between good and bad lots. Increasing Ac at constant n shifts the curve right (more permissive) without changing the slope. A plan with n=20, Ac=0 (the F/AQL-1.0 plan above) has a notoriously shallow curve and is often a poor real-world discriminator — when a deep distinction matters, larger n is required.

AQL selection is not arbitrary. Common practice: critical defects (safety/regulatory impact) AQL 0.10 % or sample-100 %; major defects (functional, fit-for-purpose) AQL 0.65 to 1.5 %; minor defects (cosmetic) AQL 2.5 to 4.0 %. Defect classification must come first — and the classification rationale must be documented. Sampling for critical defects at AQL 1.0 is a finding waiting to happen.

The deeper question is whether to sample at all. ANSI Z1.4 is appropriate when (a) process capability is unknown or insufficient, (b) destructive testing makes 100 % inspection impossible, or (c) the cost of inspection per unit is much lower than the cost of a defect reaching the customer. When Cpk is high and stable and the process is in control, in-process SPC + reduced AQL sampling beats heavy acceptance sampling on both economics and discrimination. When destructive testing dominates and Cpk is unknown, ISO 3951 variables sampling is more efficient than ISO 2859 attributes sampling.

1. AQL value chosen without defect classification — same AQL applied to safety-critical and cosmetic defects.
2. Switching rules never applied — supplier sat on Normal for years despite repeated rejects (or sat on Reduced after a single accept streak with no monitoring).
3. Sampling 'plan' is 'inspect 10 units' regardless of lot size — not anchored to ISO 2859-1 / Z1.4 at all.
4. Sample taken from the top of the pallet (convenience) rather than random — destroys the statistical basis of the plan.
5. Plan documented for receiving inspection but actual inspector takes a smaller sample because the lot is small (no calculation of code letter F → 20 units, just '5 looks fine for 100').
6. Critical defect AQL of 0.10 used but the n=200 / Ac=0 OC curve was never reviewed — supplier at 0.5 % defective still gets 37 % acceptance probability.
7. AQL plan accepted a lot containing a critical defect — and the response was to re-sample 'to confirm', diluting the original sample's statistical meaning.
8. Sampling plan never re-evaluated when the lot size, defect classification or process capability changed.

• Material/part master holds defect classification (critical/major/minor) and the AQL per class; the sampling plan is generated automatically from lot size, inspection level and current switching state.
• Switching state is tracked per supplier per part — recent lot history drives Normal → Tightened → Reduced transitions automatically per the standard's rules, with the active state visible on the receiving record.
• Inspector worksheet shows code letter, sample size, accept/reject numbers and the random-sampling instructions; entering a sample size that doesn't match the standard is blocked.
• Disposition is computed from inspector defect count and Ac/Re — manual override requires QA second e-signature and a justification that is itself trended.
• OC curve and AOQ table are available per plan for the QA reviewer; switching to Tightened or back to Normal is shown with the rationale.
• Supplier scorecard rolls up AQL pass/fail and switching-state history alongside on-time delivery, CoA timeliness and CAPA closure — one supplier health score, not a folder of spreadsheets.