Pick the AQL by defect class, not by product family

A lot of 47 mm, 0.45 µm PTFE membrane filters failed outgoing inspection last March at a point most suppliers never check: the bubble point on the 60th disc pulled from carton 12 read 18.2 psi against a 22.0 psi minimum, and the lot was already on a pallet with a released stamp. The AQL sampling plan had passed it. That is the whole problem with outgoing QC on PTFE membranes. A single attribute AQL plan on a variable that drifts within a lot will pass a bad lot roughly half the time at the boundary. What actually catches drift is a layered plan: an AQL attribute gate for gross defects, a variables check on the specifications that matter, and a record that shows a customer’s auditor exactly what was measured and where.

This is a working account of how I build and run a PTFE membrane filter AQL sampling plan and outgoing QC program. It assumes you already know what the product is. It covers the acceptance levels, the sample sizes, how to classify defects so the plan means something, and what an outgoing inspection record has to contain to survive a supplier audit.

The mistake I see most often is a supplier applying one AQL, usually AQL 1.0, to everything on the inspection sheet. That is too loose for a pinhole and too tight for a label smudge. Split the sheet into three defect classes and set a separate acceptance number for each.

Critical defects: AQL 0, always

A critical defect is one that creates a safety, method-validity, or regulatory failure downstream. On a PTFE membrane filter that means: a pinhole or rupture visible under 10x magnification, a bubble point below the released minimum, a pore size mislabel on a lot that was actually extruded to a different spec, a membrane that is not PTFE (a PTFE-laminate substitute slipped in), or a sterility breach on a sterile-packed SKU. Critical defects get AQL 0 and a c=0 acceptance number. Any critical defect fails the lot. No exceptions, no rework-and-retest on the same lot number.

For sterile SKUs I would also apply the sampling from ANSI/AAMI/ISO 11737-2 for bioburden and treat a positive as critical. That is not an AQL plan, but it belongs on the same outgoing record.

Major defects: AQL 0.65 to 1.0

A major defect degrades function or forces a customer to reject the part but does not create a safety or method-validity failure. On PTFE membranes: dimensional out-of-tolerance on diameter or thickness, an out-of-spec bubble point within the marginal band (say 22.0 to 23.5 psi when the release floor is 22.0), a wrinkle or crease in the membrane, a support ring that is not bonded, an extractables value above the release limit, or a lot that fails the visual for particulate contamination above the acceptance limit.

I run majors at AQL 0.65 for critical-use SKUs (HPLC mobile phase filtration, USP <797> compounding, EPA SW-846 sample prep) and AQL 1.0 for general filtration. The difference is real. At a lot size of 5,000 and AQL 0.65, general inspection level II gives a sample size of 200 with Ac=3, Re=4. At AQL 1.0 the same sample size gives Ac=5, Re=6. That is two extra defect allowances across 200 units, which sounds trivial until you are the lab that received the two extra.

Minor defects: AQL 2.5 to 4.0

Cosmetic and packaging issues. Label misalignment, a scuffed outer carton, a lot code printed at the edge of legibility, a bag seal that is intact but not clean. AQL 2.5 is my default. I will go to 4.0 on a high-volume commodity SKU where the customer has told me the carton goes straight into a dispenser and the outer box is discarded.

One caution: do not let “minor” absorb packaging defects that compromise the product. A bag seal that is intact but not clean is minor. A bag seal with a channel that lets ambient air in is major, because moisture uptake changes the membrane’s wetting behavior and therefore its bubble point.

Sample size follows ISO 2859-1, but the inspection level does not

The sample size comes from ISO 2859-1 (the same tables as ANSI/ASQ Z1.4). Lot size and inspection level set the code letter, the code letter sets n, and the AQL sets Ac/Re. The part most suppliers get wrong is the inspection level.

General inspection level II is the default and is what most customers expect to see on the record. For PTFE membranes I use level II for majors and minors and level III (or 100% for critical characteristics) on the characteristics that are hard to detect downstream. Bubble point and pore size are the two. A customer cannot measure bubble point on a filter after it is in a filter holder mid-method. If it is bad, they find out when the method fails and they call you. So I take bubble point out of the AQL plan entirely and run it as a variables sample.

Why I run bubble point as a variables sample, not an attribute

An attribute plan on bubble point turns a continuous measurement into pass/fail at the release floor. That throws away the information that predicts drift. Instead I pull n=13 discs per lot (the variables sample size for a 95%/99% one-sided tolerance with roughly 90% confidence at normal inspection), measure bubble point on each to 0.1 psi, and compute the mean and standard deviation. The lot releases if the lower tolerance limit, mean minus 2.5 × SD, is at or above the release floor. That is a stricter gate than an attribute plan at the same n, and it gives me a trend line. The 18.2 psi disc in the opening paragraph would have shown up as an outlier in the variables sample two lots earlier, when the mean was still passing but the SD was climbing.

The same variables approach applies to thickness (target ±0.02 mm on a 0.10 mm nominal) and diameter (47.0 ±0.25 mm, or whatever the drawing says). Attribute plans on dimensions are fine for gross nonconformance but will not catch a slow drift toward the tolerance edge.

Classify defects against the customer’s method, not your drawing

Defect classification is where the outgoing plan either earns its keep or becomes paperwork. A defect is only major if it would make the customer’s method fail. That means the classification has to be written against the use case, not against the internal drawing.

The classification table I actually use

Note the n=200 on the visual characteristics and n=13 on the dimensional ones. The visual plan is a level II attribute plan; the dimensional plan is a variables plan at the same code letter. Both are defensible. Mixing them on one sheet is what makes the record readable.

What a realistic outgoing inspection record contains

I have sat on both sides of supplier audits. The record that passes is not the one with the most fields. It is the one where every field can be traced to a measurement and every measurement can be traced to a lot. Here is what I put on a PTFE membrane outgoing inspection record, in order.

• Lot number and date of manufacture. The lot number must match the carton label exactly. I have seen a supplier’s carton label and inspection record disagree on the last digit, and the whole shipment was held for three weeks.
• Product code, pore size, diameter, and packaging configuration. “PTFE, 0.45 µm, 47 mm, 100/pk, non-sterile” is a record. “PTFE filter” is not.
• Sampling plan reference. ISO 2859-1, general inspection level II, AQL 0.65 (major) / 2.5 (minor) / 0 (critical). State the standard, the level, and the AQLs. Do not just write “AQL sample.”
• Sample size and code letter. n=200, code letter K for a 5,000-piece lot at level II. The auditor will check this against the table.
• Individual measurements for the variables characteristics. All 13 bubble point readings, not just the mean. All 13 diameters. The auditor wants to see the spread, because the spread is what predicts the next lot.
• Computed statistics. Mean, SD, and lower tolerance limit for each variables characteristic. State the multiplier used (2.5 for the one-sided 95/99 at n=13).
• Defect tally by class. Critical 0, major 1 (wrinkle, carton 7), minor 4 (labels, cartons 2, 5, 9, 14). The location within the sample matters. A defect cluster in the last 20 units pulled suggests a packaging-line problem, not a random defect.
• Disposition. Accept, reject, or accept-with-deviation. If accept-with-deviation, name the deviation and the customer approval reference. Never accept-with-deviation on a critical.
• Inspector ID and date. Initials and a date. Not a signature stamp applied later.
• Retained sample location. Where the 13 bubble point discs and the 200 visual units are stored, and for how long. I keep retained samples for 24 months minimum, longer if the customer’s method is regulated.
• Reference to the incoming and in-process record for the lot. One line, one lot number. The outgoing record does not need to reproduce the in-process data, but it must point to it.

That is 11 fields. A one-page form fits them. If your form is three pages, you are probably recording things that do not change a release decision.

Two field cases where the AQL plan missed and the variables plan caught it

Case 1: bubble point drift on a 0.2 µm lot

A customer running USP <797> compounding flagged intermittent failures on a 0.2 µm PTFE SKU. Their bubble point test on the filter holder read 28 to 30 psi on most discs and 24 psi on roughly one in eight. Our outgoing attribute plan at AQL 0.65 had passed every lot for six months. When I pulled the variables data, the mean bubble point had drifted from 31.4 psi to 28.9 psi over four lots, and the SD had grown from 0.6 to 1.9 psi. The lower tolerance limit had crossed the 26.0 psi release floor two lots before the customer’s first complaint. The cause was a stretch-ratio change on the paste extrusion line after a die swap. The fix was a die re-qualification and a re-baselined LTL on the outgoing record. The lesson: the attribute plan was never going to catch it because no single disc was out of spec until the lot was already marginal.

Case 2: a wrinkle cluster that was a packaging problem

A lot of 100 mm discs failed outgoing at 6 major wrinkles in a 200-piece sample against Ac=3. The wrinkle was in the membrane, but the location in the sample was the tell: 5 of the 6 were in the last 40 units pulled, which came from the bottom of the shipping carton. The cause was not the membrane. It was the bag seal on the inner pouch, which had been set 2 mm off-center and was compressing the stack at the bottom of the carton. The fix was a bag-seal die adjustment and a torque spec on the cartoner’s sealing jaw. The wrinkle classification stayed major. The corrective action was on the packaging line, not the membrane line. Without the location tally, we would have quarantined a perfectly good extrusion lot.

A pre-shipment checklist that catches what AQL misses

The AQL plan is the gate. The checklist is what runs before the gate opens. I use this on every PTFE membrane lot before the release stamp goes on.

• Lot traceability. Lot number on the carton matches the inspection record and the extrusion batch record. One digit off is a hold.
• Retained sample pulled and labeled. 13 bubble point discs, 5 pore size discs, 200 visual units, stored at ambient in a sealed bag with a desiccant.
• Bubble point variables run. n=13, IPA wet, ASTM F316, mean and SD computed, LTL compared to the release floor.
• Pore size correlation check. The bubble point to pore size correlation curve is current for this lot’s extrusion run. A curve older than the last die change is not valid.
• Dimensional variables run. Diameter and thickness, n=13 each, LTL compared to the drawing.
• Extractables. Gravimetric on n=3, compared to the release limit. For HPLC-grade SKUs I also run a blank injection on an Agilent or Thermo Fisher system at the customer’s method wavelength to confirm no ghost peaks, using a COD vial for the digestion blank if the method calls for one.
• Visual attribute run. n=200, 10x magnification, defect tally by class and by sample position.
• Packaging check. Bag seal integrity, carton count, label placement, lot code legibility. Include a vial insert and 2 mL vial compatibility note if the SKU ships as a kit.
• Disposition and stamp. Accept, reject, or accept-with-deviation. Inspector initials and date.
• Retained sample location logged. Shelf, bin, date in, date out (24 months).

If the SKU is used in EPA SW-846 methods, I also confirm the lot’s extractables data against the method’s blank acceptance criteria. The SW-846 methods leave the filter blank to the lab, so the supplier’s extractables number is often the only data the lab has. I would rather it be on the outgoing record than discovered in the customer’s method blank.

Common failure modes in outgoing PTFE membrane QC, and how to spot them

These are the ones I have dealt with personally. Each has a symptom you can see on the record before it becomes a customer complaint.

• Attribute plan on a drifting variable. Symptom: every lot passes, then a customer rejects three in a row. Check the SD trend on the variables characteristics. If SD is climbing, the plan is blind.
• Wrong inspection level. Symptom: the sample size does not match the lot size on the ISO 2859-1 table. A 5,000-piece lot at level II is n=200, code letter K. If your record says n=80, you are at level I.
• Defect classification that has not been reviewed against the customer’s method. Symptom: a “minor” defect that the customer rejects. Re-classify against the method, not the drawing.
• No location tally. Symptom: you cannot tell whether a defect cluster is random or positional. Add the sample position to the tally.
• Bubble point measured on the wrong wetting fluid. Symptom: bubble point values that are 15 to 20% higher than expected. IPA wet and water wet give different numbers on the same membrane. The record must name the fluid.
• Retained sample not kept. Symptom: a customer complaint arrives three months after shipment and you have nothing to retest. Keep 24 months minimum.
• Release stamp applied before the variables data is computed. Symptom: the record shows the stamp date before the measurement date. That is a documentation failure even if the lot is good.

For the analytical side of the extractables check, Restek publishes blank and ghost-peak guidance that is worth reading against your own release limits. The point is not to copy their numbers. It is to make sure your outgoing extractables limit is at least as tight as the blank acceptance the customer’s method will apply.

The decision rule

Run outgoing QC as a layered plan, not a single AQL. Critical defects get AQL 0 and c=0 with 100% or level III inspection. Majors get AQL 0.65 for regulated-use SKUs and AQL 1.0 for general filtration at level II. Minors get AQL 2.5. Pull bubble point, diameter, and thickness as variables samples at n=13 and release on the lower tolerance limit, not on an attribute pass. Classify every defect against the customer’s method, not the internal drawing. Put all 13 individual readings, the mean, the SD, the LTL, the defect tally with sample position, and the retained sample location on a one-page record. If any of that is missing, the lot does not ship. That is the rule I would defend in an audit, and it is the rule that would have caught the 18.2 psi disc before the pallet left the dock.

References & further reading

Related resources from our lab consumables network:

• Ptfe Membrane Filter at ptfemembranefilter.com
• Cod Vial at codvial.com
• Vial Insert at hplcinsert.com
• 2Ml Vial at 2mlsamplevial.com

External references and standards:

• Agilent chromatography consumables and columns
• Thermo Fisher chromatography and sample handling
• U.S. EPA SW-846 test methods for waste and materials
• Restek technical resources for chromatography