Engineering Answers / Production & Quality
AQL tells you how many parts to look at and how many failures to accept. It does not tell you what a defect is — and on silicone, the definition is where the argument lives.
Silicone parts are sampled under ISO 2859-1 at general inspection level II: lot size sets a code letter, the code letter sets the sample size, and the AQL level sets the acceptance number. The practical work is defect classification — critical at AQL 0 with no acceptance, major at 0.65 or 1.0 for function, minor at 2.5 for cosmetics. Sampling cannot reliably catch a low-rate sealing failure, so any characteristic that leaks or tears needs a 100% functional check instead.
AQL sampling follows ISO 2859-1. Three inputs produce one decision: the lot size, the inspection level, and the AQL.
The lot size gives a code letter. The code letter, at general inspection level II, gives the sample size. The AQL gives the acceptance number — how many nonconforming units the sample may contain before the lot is rejected.
| Lot size | Code | Sample size (level II) |
|---|---|---|
| 91–150 | F | 20 |
| 151–280 | G | 32 |
| 281–500 | H | 50 |
| 501–1,200 | J | 80 |
| 1,201–3,200 | K | 125 |
| 3,201–10,000 | L | 200 |
| 10,001–35,000 | M | 315 |
| 35,001–150,000 | N | 500 |
The acceptance numbers come from the standard's table and change with both sample size and AQL. The one number worth remembering without the table is that critical defects are Ac 0 — one occurrence rejects the lot.
AQL is arithmetic. Classification is judgement, and it is where a silicone inspection plan earns or loses its value.
Critical. A defect that creates a safety risk or makes the part unusable in the application: a tear through a sealing lip, a part that leaks at the specified pressure, a component that detaches under load on a baby or medical product. AQL 0. No acceptance, regardless of lot size.
Major. The part does not perform its function reliably but does not create a safety risk: a dimension out of tolerance that affects assembly, hardness outside the specified band, an incomplete cure, flash across a sealing face.
Minor. The part works and looks wrong: a cosmetic blemish on a non-functional face, a colour outside the ΔE limit, a trimming mark, a slight parting-line witness.
| Defect class | Typical AQL | Example on a silicone part |
|---|---|---|
| Critical | 0 | Leak at rated pressure; tear at a sealing lip |
| Major | 0.65–1.0 | Undercure; hardness out of band; flash on a seal |
| Minor | 2.5 | Cosmetic blemish; trimming mark; slight colour drift |
The most common mistake is a single line in the purchase order — "AQL 2.5" — with no classification. That sentence means a leaking seal costs the supplier no more than a scuff on the back of the part, and it means the inspector decides which is which at the bench.
1. Functional characteristics with a low failure rate. Sampling estimates a defect rate; it cannot certify that no unit leaks. Anything that fails by leaking, tearing or detaching needs a 100% functional test — a pressure decay check, a pull test, a visual under magnification at the sealing surface. Sampling is for the characteristics where a small escape rate is tolerable.
2. Soft-part measurement. A critical dimension measured with a calliper on a 40 Shore A part is not a measurement, it is an opinion, and the AQL arithmetic is being applied to noise. The dimension needs a defined method — non-contact, low-force, or a potted cross-section — before it can go into a sampling plan at all.
3. Cosmetic judgement without a boundary. Two inspectors, one part, two verdicts. Cosmetic limits have to be a physical boundary sample set that both sides have signed, not an adjective in a document.
Critical characteristics get a 100% check with a recorded result. Major characteristics get level II sampling at AQL 1.0. Cosmetic gets level II at 2.5 against a signed boundary set. Every dimension that appears in the plan has a named measurement method and a named instrument. If a customer sends their own plan, we map it against ours and flag the characteristics where the two would disagree before the first lot runs.
That last item is what turns sampling from a formality into evidence. A lot that was inspected without retained samples cannot be defended six months later, and on a silicone part the parts themselves are the only proof of what the flash, the colour and the hardness actually were.
It depends on what the part does, not on what it is made of. Medical and baby products typically run AQL 0.65 for major defects and 2.5 for minor, with critical defects at zero acceptance. Industrial gaskets often run 1.0 and 4.0. The level should be written into the drawing or the quality agreement, because a supplier who chooses it unilaterally will choose the one that passes.
Because the failure rate is lower than the sample can resolve. At a lot of 3,200 parts and a sample of 200, an AQL of 1.0 accepts up to a handful of defects, which means a process running at 0.5% leak rate can pass sampling indefinitely and still ship sixteen leakers in that lot. A leak is a functional characteristic: it needs a 100% test on the parts that matter, and sampling is for the characteristics that are too expensive to check completely.
No, and mixing them is the most common cause of a rejected lot that should have passed. Cosmetic limits belong on a boundary sample set and in a separate, looser AQL — typically 2.5 or 4.0. Functional defects belong at 0.65 or 1.0. When both are pooled into one number, either the cosmetic standard becomes impossibly strict or the functional one becomes useless.