Home / Blog / Waterproof Technology

AQL Inspection for Dry Bags

What AQL actually means on a bag line

AQL stands for acceptable quality limit, and despite the name it is not a promise of quality. It is a sampling plan: a rule that tells the inspector how many units to pull from a shipment and how many defective units the plan tolerates before statistical inference says the whole shipment is unlikely to be representative. The standard in use across most bag programmes is the sampling scheme published by ISO as ISO 2859-1, also issued as ANSI/ASQ Z1.4, and the two are functionally equivalent for practical purposes.

The misunderstanding that costs people money is treating AQL as an upper bound on defects shipped. It is not. A plan with an acceptance number of 10 still ships roughly 2.5 percent defective product when lots sit at the limit of their accept probability. That is why smart programmes pair the standard plan with a zero-tolerance rule for defects that actually matter, which for a waterproof bag means anything compromising the seal.

Inspection is also not quality control in the broader sense. Quality control is the whole system of material approval, first-article checks, in-line monitoring and final inspection described in our quality control guide. AQL inspection is one instrument inside that system, and using it as the only instrument is the most common structural mistake in bag programmes. A single sample taken on the final day cannot correct a decision made six weeks earlier about film formulation.

How lots and sample sizes are set

Everything starts with the lot. A lot is a defined quantity of identically produced goods presented for inspection at one time, and its size determines the sample size through a code letter. Splitting one physical shipment into several small lots is a well-known way to game the statistics, because smaller lots carry smaller samples and a smaller acceptance budget in absolute terms, even though the percentage is the same.

General inspection level II is the norm for bags, giving the code letters and counts below. Special levels S-1 to S-4 exist for destructive or expensive testing, which is exactly how you should classify a destructive water immersion or a weld-peel test: run those on a small side sample rather than across every inspected unit. Keeping destructive work in a side lot also leaves the main sample saleable, which matters when every unit counts.

Notice how flat the sample growth is. Going from a 1,200-piece order to a 10,000-piece order only triples the sample size, which is why larger shipments are not proportionally more inspected and why single-lot discipline matters so much. If your routine buy is above 3,200 units, consider stepped rather than sampled inspection of critical features, discussed further below. The whole plan assumes defects are randomly distributed, and anything breaching that assumption needs a different instrument entirely.

The three defect classes and how to define them

Every plan sorts findings into critical, major and minor. Critical defects are those that make the product unsafe, illegal, or unfit for its fundamental purpose. Major defects materially reduce usability or saleability. Minor defects are workmanship imperfections that a reasonable buyer would accept. The definitions are standard, but the mapping of specific bag faults onto those definitions is what programmes get wrong.

For a dry bag, the mapping should be explicit and written down before inspection begins, because arguments at nine in the morning on the last day of a booked container are expensive. The recommended mapping puts every leak path in critical: a weld void, a pinhole in the film, a bonded seam that has not fully fused, or a closure geometry that will not hold three rolls. Those all defeat the single purpose of the product.

Reading acceptance numbers the right way

An acceptance number is not a target. On a 5,000-piece lot at AQL 2.5 the inspector pulls 200 units and accepts the shipment if she finds ten or fewer defects. Finding nine and declaring success misunderstands what happened: the plan simply failed to prove the lot was bad, it did not prove the lot was good. Operating consistently near the accept boundary guarantees that roughly one in forty bags reaching your users has something wrong with it.

This is why experienced programmes tighten the defect classes rather than the AQL number. Setting AQL 1.0 for cosmetics makes the plan expensive and slow while barely improving outcomes. Keeping AQL 2.5 for cosmetics and applying zero tolerance to an agreed critical list costs almost nothing extra and addresses the failures users actually notice. Nobody writes a review about a loose thread; everybody writes one about a bag that leaked.

Switching rules are the other underused lever. The standard defines normal, tightened and reduced inspection, with rules for moving between them based on recent history. A programme that has accepted five consecutive lots can move to reduced inspection and save real money, while two rejections in five lots should trigger tightened inspection until performance recovers. Agreeing those rules in advance also removes any suggestion that a change of level is punitive.

Waterproof checks inside the AQL sample

Not every useful test scales to 200 units. A 30-minute immersion of every sampled bag is technically possible but consumes the better part of a day, and it damages nothing, so it is simply slow rather than destructive. A faster screen that scales well is air inflation: seal the bag, inflate to a defined pressure, hold for 30 seconds, and monitor. Pressure decay above a defined threshold flags the unit for immersion testing.

A good protocol runs like this: every sampled unit gets a visual and dimensional check plus an air-inflation screen taking about 90 seconds each. Any unit flagged goes to a 30-minute water immersion at one metre equivalent, which is the confirmatory test described in our immersion testing guide. Then a small destructive subset, typically two or three units from the sample plus two from the side lot, gets a weld peel and a hydrostatic head test run to the referee methods published by ASTM.

The pressure you choose for the inflation screen matters. Too high and you pass units that would fail later; too low and you generate false flags that slow everything down. Our own line uses around 6 kPa for roll-top bags, which corresponds roughly to the pressure difference experienced at 60 cm of submersion, with a decay limit of 10 percent over 30 seconds; the step-by-step procedure is set out in our air inflation leak test.

Closure cycling: the check nobody writes into the plan

Roll-top closures fail in use long before they fail in a single test, so counting defects at time zero misses the population most likely to disappoint users. The cheap addition is a closure-cycling test: take one or two units from the sample, roll and unroll them 500 times, then run them through the standard air-inflation screen and compare the decay rate with an uncycled unit.

The failure mode this catches is crease fatigue at the fold line, which shows as a fine network of surface cracks in the film after cycling. Those cracks do not necessarily leak immediately, but they leak eventually, and they are invisible at the first-article stage. Cycling is the only practical way to detect the difference between a film formulation that tolerates folding and one that does not.

A worked comparison helps calibrate expectations. In one qualification round, two otherwise identical 0.5 mm constructions were cycled 500 times. The polyether-based film held its pressure-decay figure within 4 percent of new, while the polyester-based film lost 21 percent and showed visible crazing at the second fold. Had the programme relied on appearance alone, both would have passed. The two constructions were visually indistinguishable until roughly the twentieth cycle, which is precisely why the test pays for itself.

Hardware, prints and dimensions

Hardware failures are rare on a mature line and catastrophic when they happen, which argues for a small but genuinely destructive test rather than a wide scan. Pull-testing a sample of buckles and D-rings to failure takes fifteen minutes and tells you whether the strength requirement is genuinely being met. A typical requirement is 150 N for a lash point and 250 N for a shoulder strap anchor.

Print checks need a reference. There is no objective pass or fail without an approved colour standard and a tolerance, so insist that the first-article approval includes a signed reference print plus a stated delta E tolerance measured under D65 light. In practice a tolerance of delta E 2.0 is achievable on flat TPU film, while anything above 3.0 is visible to untrained eyes under shop lighting.

Common disputes and how to avoid them

Three arguments account for almost every bad-tempered inspection. The first is lot splitting, described above, and it is solved contractually: state in the purchase terms that the lot size equals the total quantity presented for that shipment, not what the line happens to have completed. The second is whether the inspector may open retail packaging, which should also be settled in advance and is usually resolved by allowing repacking at the production partner’s cost.

The third and largest dispute is defect classification. A loose thread is a minor fault to a maker repairing ten units and a critical fault to a brand whose entire positioning rests on finish. There is no universal answer, which is exactly why the classification table must be agreed and signed before bulk begins rather than negotiated on the final morning.

A worked example: 5,000 pieces at AQL 2.5

Take a real-scale purchase. The order is 5,000 roll-top bags across four colourways, produced over 22 days, and inspection is booked for the fourth day after packing begins. The lot at inspection is 5,000 units, which sits in the 3,201 to 10,000 band, giving code letter L and a sample of 200 units drawn randomly across cartons rather than picked from the accessible top layer.

Read the last row carefully: twelve non-critical defects exceeds the acceptance number of ten, so the lot fails even before the three critical units are considered. In practice the critical findings would have stopped the inspection anyway, because most programmes treat them as immediate cause for full screening. The worked example makes the arithmetic visible, and it is the kind of exercise worth running with your own numbers before you agree a plan.

What AQL inspection does not cover

Sampling plans say nothing about the mean quality of a lot, only about its probability of acceptance. They also do not protect against a systematic error that affects every unit identically, which is the single most dangerous category in this product group. If the wrong catalyst ratio was used in a batch of film, or if a machine die has drifted 1 mm, every unit may be affected and no amount of random sampling from that lot will reveal a trend.

This is why the plan must sit alongside first-article approval, in-line checks during the first two hundred pieces, and periodic destructive testing sent to an independent laboratory. The comparison between those two approaches is set out in our field and laboratory testing comparison, and it is worth reading before you write a single acceptance criterion. Systematic problems need process controls, not better statistics applied at the very end.

Sampling also ages badly if the plan never changes. A programme that ran happily for three years should still be reviewed after any material change, any new colourway, or any move to a second line, because the defect population changes with every one of those events and historical acceptance numbers stop being meaningful. The same logic underpins the periodic re-qualification described in our warranty testing standards.

Building a defect catalogue before bulk starts

1. List every plausible fault with a unique code, so findings are recorded consistently across inspectors.
2. Photograph a physical reference for each code with lighting and distance stated.
3. Assign a class to each code and get that classification signed by both parties.
4. Set threshold quantities for automatic 100% screening, independently of the AQL outcome.
5. Agree who pays for sorting and repacking when a lot fails and is later re-presented.
6. Decide the switching rules in advance rather than renegotiating them after a rejection.

Two of those lines deserve extra emphasis. The threshold quantities are what turn a statistical plan into a practical one: even at accept counts, most brands will demand full screening once any critical defect appears, because the reputational asymmetry between shipping one leaking bag and inspecting five thousand units is enormous. Write the threshold into the document rather than relying on goodwill on the day.

The cost clause is equally important and frequently skipped. A failed lot typically costs five to eight working days to sort, screen and re-present. Knowing in advance whether that expense falls on the production partner concentrates attention on avoiding it, which is precisely the behavioural effect you want from the clause. An unallocated cost is a cost that will eventually be argued about at the worst possible moment.

Fitting inspection into the production calendar

Our team has produced waterproof bags since 2014 in an SGS-verified 4,950 m2 facility with 7 production lines and 149 machines, running monthly output near 200,000 pieces, and inspection booking is the most common cause of avoidable delay we see. A final random inspection typically takes one to two working days on site, so it must be scheduled before the container cut-off, not on the same day.

Standard timing works like this: sampling in 6 to 10 working days once materials are confirmed, bulk in 35 to 50 days, then final inspection with results issued within 24 hours of completion. A failed lot adds five to eight days for sorting. Building a buffer of ten days between packed stock and vessel cut-off costs nothing if unused and saves the entire shipment if it is not.

Every programme we run ships against AQL 2.5 with FOB Xiamen terms and T/T 30/70 payment, and the inspection clause is part of the same purchase document rather than a separate email thread. Written that way, there is one place to look when someone asks what happens next, which is most of what makes a dispute resolvable quickly on our roll-top platform and across the wider product range.

Writing the inspection clause that protects you

A usable clause runs to about ten lines and answers six questions: which standard and revision, which inspection level, which AQL values per defect class, whether lot size equals total presented quantity, who pays for sorting after rejection, and what happens on re-inspection. Anything longer usually indicates that some question remains unanswered and someone is hoping volume will hide it.

Add two procedural sentences that save hours later. State that random sampling follows a documented random-number map across cartons, so nobody selects units from the top of the nearest pallet. State that the inspector may open and reseal retail packaging for destructive tests, with reasonable replacement, so access disputes do not consume the morning. Both sentences take half a minute to agree and remove two of the three arguments that routinely delay despatch.

Finally, tie the clause to a measurable product claim. If your warranty rests on the bag surviving immersion, then the warrantied requirement and the inspection criterion must describe the same event, otherwise you will be arguing about whether a bag that passed inspection can be rejected by a user. Whatever terms you write should be validated against realistic use rather than the other way round.