Quality control

AQL inspection levels explained

When a supplier writes "AQL 2.5" on a quote, it tells you almost nothing until you know which inspection level and which defect class. This is how to specify it properly.

"AQL 2.5" on a factory quotation is close to meaningless on its own. It does not say how many parts will be checked, which defects count toward the 2.5, or whether the same rule applies to a cosmetic scratch and a bore that will not accept its shaft. Two factories can both quote "AQL 2.5" and run completely different inspections.

For machined components that gap matters more than for consumer goods, because sampling assumes defects are scattered randomly through the lot. Dimensional errors that follow tool wear are not. Here is what the standard says, and where it must be overridden.

What ISO 2859-1 actually defines

AQL stands for acceptable quality limit. The governing standard for attribute sampling is ISO 2859-1, the international equivalent of MIL-STD-105E. It is a set of acceptance sampling tables, not a quality system. Given a lot size, an inspection level and an AQL value, it returns a sample size and an acceptance number.

General Inspection Levels I, II and III

The three general levels change the sample size for the same lot at the same AQL, by a factor of roughly 1.6 between adjacent levels:

• Level I — about 40% of the Level II sample. Cheaper, weaker discrimination. For large lots of mature parts with a proven process.
• Level II — the default, and the basis of almost every published table. Use it for normal releases and new suppliers.
• Level III — about 1.6 times the Level II sample. For critical components, first production runs, or a supplier returning after a rejection.

Special levels S-1 to S-4 use much smaller samples and are meant for destructive or very expensive tests. They are unsuitable for dimensional acceptance on a machined lot.

The AQL number is a defect rate, not a permission slip

AQL 2.5 means a lot with 2.5% defective pieces has a high probability of being accepted — roughly 95% under the standard's operating characteristics. It is the level the scheme is designed to accept, not the level you should expect to receive.

• Accepting a lot does not mean zero defects. At AQL 2.5 you can receive a lot with a few percent defective and have no claim. If the assembly cannot tolerate that, tighten the AQL or check the feature 100%.
• AQL 0 does not exist in the general tables. "Zero defects" is not an AQL value the standard supports. Write it as 100% inspection of a defined characteristic.

Sample size: Level II in practice

Sample size depends only on the lot size band; the AQL value changes the acceptance number, not the sample.

Lot size	Sample size (Level II)	Ac / Re at AQL 1.0	Ac / Re at AQL 2.5
91–150	20	0 / 1	1 / 2
151–280	32	0 / 1	2 / 3
281–500	50	1 / 2	3 / 4
501–1,200	80	2 / 3	5 / 6
1,201–3,200	125	3 / 4	7 / 8
3,201–10,000	200	5 / 6	10 / 11
10,001–35,000	315	7 / 8	14 / 15

Above 35,000 units the sample size steps up to 500 and then caps at 800. The acceptance number does not keep scaling with lot size, because the sample itself stops growing at the same rate. Check those two rows against the exact edition of ISO 2859-1 your supplier claims to follow before writing them into a purchase order — some cells were amended between editions.

Read the 501–1,200 row as an example. A lot of 1,000 parts at Level II means 80 parts inspected. At AQL 2.5 you accept with 5 defectives in that sample and reject at 6, which is a stricter practical screen than the label suggests.

Critical, major and minor defects

ISO 2859-1 does not define defect classes. They are a layer you define, and each class should carry its own AQL:

• Critical — a hazard, a regulatory breach, or a part unusable in the assembly: wrong material grade, a thread that will not engage, a missing operation. Normally zero acceptance in the sample.
• Major — functional failure or a line stop downstream: out-of-tolerance bore, wrong thread depth, a burr that interferes with assembly. Often AQL 0.65–1.0 on machined parts.
• Minor — cosmetic or non-functional: light tool marks, a discoloured patch, a scratch on a hidden face. Often AQL 2.5–4.0.

Common practice is critical zero, major 1.0, minor 2.5. Consistency across suppliers matters more than the exact numbers: if the same defect is a major at one factory and a minor at another, the reports are not comparable.

Why machined parts break the sampling model

Sampling assumes a homogeneous lot. Machining violates that in predictable ways. A dimension that is good at the start of a run can drift out over 500 parts as the tool wears. A warm spindle moves every part cut in the first hour. A fixture that slipped at part 400 makes a block of 30 bad parts. A random 80-piece sample can pass a lot that is out on one dimension throughout.

The practical resolution is a split plan:

• Critical-to-function dimensions: 100% inspection or in-process gauging. Sampling does not apply.
• General dimensions: AQL sampling, usually Level II.
• Variable process data: reviewed, not sampled. Readings every 10th or 25th part show the drift a final sample cannot.
• Material and heat-treatment certificates: checked document by document.

For a machined part, the first-article report and the in-process measurement sheet matter more than the final sample. The AQL gate proves the lot passed; it does not prove the process was in control.

What to write in the purchase order

A workable quality clause names four things: the standard and edition, the inspection level, the AQL per defect class, and which characteristics are exempt because they are 100% checked. Add the gauge used and the measurement conditions. If the supplier cannot restate that plan back to you, the report you receive at the end will be a form, not evidence.