AQL Sampling and Defect Classes for Corrugated QA

You cannot inspect 100,000 boxes. So quality acceptance rests on a statistical bargain: inspect a defined sample, apply a defined limit, and accept or reject the whole lot on that basis. Done properly it is objective and both sides can live with it. Done vaguely it becomes an argument every delivery.

The four terms that do the work

• Lot (N) - the quantity presented for inspection at one time.
• Sample (n) - how many units you draw from it.
• Acceptance number (c) - the largest number of defective units you may find and still accept.
• AQL - the acceptable quality limit: the defect level the sampling plan is built around, at a stated confidence.

ISO 2859 is the standard that ties them together. You do not have to invent any of this, and you should not: an in-house scheme means renegotiating the rules whenever there is a dispute.

Not all defects are equal

The single most useful thing you can do is stop treating "defect" as one category. Classify by consequence:

Critical Can endanger health, breach a legal requirement, destroy the packed goods, or halt a filling line. Major Causes the pack to fail its function, triggers customer complaints, or costs real production efficiency. Minor A general reduction in quality with no significant consequence.

Those consequence classes then map onto usability classes and their AQL values. Broadly: a defect that makes the packaging unusable carries no acceptance number at all - it is grounds to reject the delivery. Defects that seriously impair usability sit at the tight end, around AQL 0.65 to 1.0. Defects that impair usability to a lesser extent sit around 1.5 to 4.0. Defects that are noticeable but do not meaningfully affect how the packaging performs sit at 6.5.

The point of the tiering is that it puts the argument where it belongs - at specification time, not at the receiving dock. A scuffed liner and a mis-scored crease are not the same event and should never share a tolerance.

A worked lot

A delivery of N = 20,000 folding boxes arrives. The plan calls for a sample of n = 80, drawn completely at random.

Within that 80, the tolerances by class come out as:

Unusable (class 1) no defects permitted Seriously impaired (2A) 1 defect at AQL 0.65 2 defects at AQL 1.0 Impaired to a lesser extent (2B) 3 defects Noticeable only (class 3) 8 defects

Two rules people get wrong:

• One box, one defect. If a single box has four things wrong with it, you evaluate only the most serious. You are counting defective units, not defects.
• Spread the sample. Drawing all 80 from the top of one pallet is not a random sample of a 20,000-unit run - it is a sample of the end of one shift. Pull from the beginning, the middle and the end of the delivery.

For very large orders, break the delivery into sensible lots - one loading unit, say - rather than treating a whole month's production as a single batch. A lot that big means one bad shift can reject an enormous quantity, and the sampling plan will not be sensitive to it anyway.

Condition the samples, or the numbers mean nothing

Corrugated is hygroscopic. Its strength - and some of its dimensional behaviour - moves with moisture. Any arbitration test therefore has to happen in a defined climate, and the standard reference is 23 C and 50% relative humidity. If your incoming inspection measures a box that has spent two days on a humid dock and compares it against a supplier's conditioned lab value, you are not comparing like with like, and the disagreement that follows is unresolvable because neither number is wrong.

Dimensions: what tolerance is reasonable

Dimensional disputes are common and mostly avoidable, because there are published expectations. Measured between two creasing centrelines, or between a crease and the associated outer edge, a workable tolerance ladder is:

up to 300 mm +/- 3 mm over 300 to 700 mm +/- 4 mm over 700 to 1,200 mm +/- 0.6% over 1,200 mm by agreement

If your box is machine-erected, say so at quotation. Automatic converting frequently needs tighter tolerances than the general ladder, and a tolerance you never specified is not a tolerance your supplier owes you.

Warp is the other recurring fight, and it is measurable rather than aesthetic: lay the sheet on a flat surface with the bend upward, divide the maximum elevation by the distance between the supporting edges, and express it as a percentage. Now it is a number, and a number can have a limit.

What to put in the purchase order

Most acceptance disputes trace back to a specification that never existed. At minimum, agree in writing:

• The defect classes and which specific defects fall in each.
• The AQL per class, and the sampling standard you are both working to.
• Lot size definition - how a delivery gets broken into batches.
• Dimensional tolerances, including any tighter ones your line needs.
• The conditioning climate for any arbitration test.
• Who pays for what when a lot is rejected.

None of this is exotic. It is a page. But it converts "these boxes look bad" into "this lot exceeded the class 2A acceptance number at AQL 0.65" - and only one of those two sentences can be resolved without a meeting.