An AQL plan that only counts visual defects is a shipping form, not a QC gate

A 500-lot shipment of 9 mm moulded septa left our plant with a certificate that said “AQL 1.0, passed.” Four months later a customer running USP <621> method suitability on a 1260 Infinity II hit an RSD of 4.8% on six replicate injections of a 0.1 mg/mL standard. The septa were not the problem. The sampling plan was. We had drawn 13 pieces from a 5,000-piece lot, inspected them for visible flash and thickness, and accepted on attributes. Nobody had measured compression set after three autoclave cycles, because compression set was not a defect class in our plan. That is the gap this article is about: the hplc septa AQL sampling plan and outgoing QC record that actually catches the failures a customer will see, rather than the ones that are easy to see on a bench under a lamp.

Most septa suppliers publish an AQL of 1.0 or 0.65 and stop there. The number means nothing until you say what is being counted. ANSI/ASQ Z1.4 (the US adoption of ISO 2859-1) gives you the sample size from the lot size and inspection level, and the accept/reject number from the AQL and the sample size. For a 5,000-piece lot at general inspection level II, code letter L gives n = 200. At AQL 1.0, Ac = 5, Re = 6. At AQL 0.65, Ac = 3, Re = 4. At AQL 0.25 with n = 200, Ac = 1, Re = 2. Those are the numbers I keep on the wall.

The trap is using one AQL for everything. A flash tab on the outside diameter of a moulded septum is cosmetic. A 0.15 mm shortfall in the sealing land thickness is a leak. A durometer drift of 3 Shore A on a silicone/PTFE laminate is a coring problem at injection 40. If all three sit in the same “major” bucket at AQL 1.0, you are accepting roughly 1% of a defect class that will stop a run. Split the classes and set different AQLs. That single change did more for our escape rate than any equipment purchase.

Defect classification: what a critical, major, and minor septum defect actually is

I use four classes, not three, because the fourth (dimensional) needs its own sampling because it is measured, not looked at. Here is the classification we run on a 9 mm septa line, and the reasoning behind each threshold.

Class Defect examples AQL Inspection method Why this level
Critical PTFE facing delamination, pinhole through the barrier film, wrong polymer grade (e.g. EPDM substituted for silicone) 0 (zero acceptance) 100% visual on facing; FTIR spot check on 5 pieces per lot Any single occurrence causes sample contact with the septum body or extractables outside the validated profile
Major A (functional) Coring debris on the piercing face, sealing land thickness outside 1.50 ± 0.10 mm, durometer outside ±3 Shore A, compression set > 25% after 3 cycles at 121 °C 0.65 n = 200 at level II; thickness by micrometer, durometer by ASTM D2240 Type A Directly linked to leak and coring at the injection count the customer validated
Major B (fit) OD outside 9.00 ± 0.10 mm, thickness outside tolerance, cap interference on crimp 1.0 n = 200; go/no-go ring gauge plus micrometer Fails autosampler seating or causes cap tilt, but usually caught before injection
Minor Flash on non-sealing surfaces, mould parting-line marks, colour variation within batch 2.5 n = 200 visual Cosmetic; does not reach the fluid path or the seal

Note the critical class: zero acceptance. ANSI/ASQ Z1.4 has no AQL of zero, so you cannot express it in the tables. The practical route is 100% inspection of the attribute that matters (the PTFE facing) and a documented destructive check (FTIR, or a dye penetration test on a 10-piece subset) per lot. If a supplier tells you their plan is “AQL 0,” ask which attribute is 100% inspected. If they cannot name it, the zero is a slogan.

Sample size is a function of lot size and inspection level, not of how many you feel like pulling

The number of pieces you pull is fixed by the standard, not by convenience. A few code-letter values that cover most septa lots:

  • Lot 501 to 1,200: code J, n = 80 at level II
  • Lot 1,201 to 3,200: code K, n = 125 at level II
  • Lot 3,201 to 10,000: code L, n = 200 at level II
  • Lot 10,001 to 35,000: code M, n = 315 at level II

Two things I insist on. First, use normal inspection as the default, and switch to tightened (level III, or the tightened Ac/Re column) after two consecutive lots with any Major A reject. That is the switching rule in Z1.4 and it exists for a reason: a supplier drifting on durometer will show it in tightened inspection long before it shows in a customer’s RSD. Second, stratify the sample across cavities. A 16-cavity mould sampled only from the first 200 pieces off the take-off conveyor is a sample of one cavity. We pull in proportion to cavity output and record the cavity number on every measured piece.

For dimensional characteristics, I run a separate variables plan (ISO 3951-1, or the variables tables in Z1.4’s companion) on thickness and durometer, because attributes sampling throws away the magnitude. A lot with a mean thickness of 1.48 mm and σ = 0.02 mm will pass attributes at AQL 0.65 and still fail a customer’s compression test. Variables sampling with a specified AQL of 0.65 and a target Cpk of 1.33 catches the drift. That is the same Cpk number the customer’s incoming inspection will eventually apply, so measuring it at outgoing keeps the two ends of the chain honest. If you want the deeper argument on capability, the sibling piece on Cpk covers it; here I only care that the outgoing plan reports the same statistic the customer will.

What the outgoing inspection record actually contains, line by line

A certificate of analysis is a summary. The record behind it is what an auditor or a customer’s supplier-quality engineer will ask for, and most septa suppliers cannot produce it. Here is the record set we ship (or hold for 24 months) with each lot. Every field is populated, not boilerplate.

  • Lot and sub-lot ID with cavity map and mould number
  • Sampling plan reference: standard (ANSI/ASQ Z1.4), inspection level (II), sample size (n = 200), AQL per defect class
  • Measured data, not pass/fail: thickness (mean, σ, min, max, n), durometer (mean, σ), OD (min/max), for every characteristic
  • Compression set: percent set after 3 cycles at 121 °C, 30 min per cycle, per ASTM D395 Method B, n = 10
  • Extractables screen: gravimetric residue after 24 h in HPLC-grade acetonitrile at 40 °C, reported in mg per septum, n = 5
  • Piercing / coring count: number of injections before a 0.5 mm particle is detected in the eluate, n = 10, reported as median and minimum
  • FTIR identity of facing film and body polymer, with reference spectrum ID
  • Accept/reject decision per class, with the actual defect counts and the Ac/Re numbers used
  • Inspector ID, date, and instrument calibration due dates for micrometer, durometer, and balance

That is roughly a page and a half. It is not burdensome. What is burdensome is the alternative: a customer returns a lot, you cannot reconstruct which cavity it came from, and you scrap 5,000 pieces to protect the relationship. I have done both. The record is cheaper.

Two field cases where the sampling plan, not the product, was the failure

Case 1: the compression-set escape

Symptom: a pharma customer reported peak-area RSD climbing from 0.4% to 3.1% across a 96-injection sequence on a stability batch, with the drift starting around injection 60. Their method used a 40 °C autosampler tray and 12-hour runs.

What identified it: we pulled the retained lot and ran compression set at 121 °C, three cycles. Mean was 31%, against a spec of ≤ 25%. Our outgoing record showed only thickness and durometer, both in spec. The plan had no compression-set characteristic at all, so the defect class did not exist and the AQL never applied to it. The durometer was fine at time zero and the seal force decayed under the customer’s thermal cycling.

Fix: added compression set as a Major A characteristic at AQL 0.65, n = 10 destructive per lot, and changed the moulding cure schedule (post-cure 4 h at 200 °C) for the affected silicone grade. Escape rate on that characteristic went to zero over the next 40 lots. The lesson: a defect class you have not defined cannot be sampled, and an AQL table will happily accept a lot on the characteristics you did measure.

Case 2: the cavity that ran hot

Symptom: a distributor flagged that septa from one shipment sat 0.08 mm proud in a crimp cap and caused cap-seating faults on their customer’s autosampler. Our dimensional data said mean thickness 1.51 mm, well within 1.50 ± 0.10 mm.

What identified it: we broke the data down by cavity. The mean was fine because 15 cavities ran at 1.49 mm and one cavity (cavity 7, near the hot runner) ran at 1.62 mm. Our sample of 200 had been drawn from the take-off end of the conveyor, which was dominated by the first cavities to fill. Cavity 7 was under-represented. A stratified re-sample of 12 pieces per cavity found the outlier immediately.

Fix: stratified sampling by cavity with cavity ID recorded, plus a per-cavity thickness chart reviewed at lot release. The hot-runner zone got a 6 °C setpoint correction. The point is not that the mould was wrong. The point is that an unstratified sample of 200 gave us false confidence that a stratified sample of 192 would not have.

Where the plan meets the customer: linking outgoing AQL to incoming acceptance

The outgoing plan and the customer’s incoming plan should agree on the defect classes and the measurement methods, even if they use different AQLs. If your outgoing durometer is ASTM D2240 Type A on a 6 mm stack and theirs is Type A on a single piece, you will disagree about the same lot. Method drift between the two ends is the most common reason a “passed” lot gets rejected on receipt.

I send the customer the defect classification table and the measurement methods with the first shipment, and ask them to run their incoming plan against it. Most labs will not, but the ones that do stop arguing about lots. If the customer’s incoming inspection is run by a third-party lab, the method statement matters even more; the analytical testing lab will apply its own methods unless you hand them yours. For the closure side, the fit characteristics (OD, thickness, cap interference) should reference the same dimensions the vial closures supplier uses, or you will chase a 0.05 mm disagreement for a month.

A release checklist you can run at the bench in under an hour

This is the sequence we use at final release for a 5,000-piece lot. It assumes the in-process data is already in hand (that is a different gate and a different article).

  1. Confirm lot identity and cavity map. No map, no release.
  2. Draw n = 200 stratified by cavity at level II. Record cavity on each piece.
  3. Critical: 100% visual on PTFE facing for delamination and pinholes. Any reject holds the lot.
  4. Major A: thickness (micrometer, 0.001 mm resolution), durometer (ASTM D2240 Type A, 1 s reading), compression set (n = 10, 3 cycles at 121 °C). Compare to Ac/Re at AQL 0.65.
  5. Major B: OD by go/no-go ring gauge, thickness against 1.50 ± 0.10 mm. AQL 1.0.
  6. Minor: flash and parting-line visual. AQL 2.5.
  7. Destructive subset: FTIR identity (n = 5), extractables gravimetric (n = 5), coring count to first 0.5 mm particle (n = 10).
  8. Variables check: thickness and durometer Cpk against 1.33 target. If Cpk < 1.00, hold and investigate even if attributes pass.
  9. Record and sign. Inspector ID, instrument calibration dates, accept/reject per class.
  10. Switching rule: two consecutive lots with any Major A reject moves the next lot to tightened inspection.

Step 8 is the one people skip. Attributes sampling at AQL 0.65 can accept a lot whose mean is drifting toward the limit; the Cpk check is the only thing standing between that lot and a customer’s failed system suitability run. If you only do one thing differently after reading this, add the variables check to the attributes plan.

The recommendation: AQL 0.65 on functional, zero on critical, and a record that carries the data

For a 9 mm moulded septum shipped to a regulated lab, run AQL 0.65 on Major A (functional) characteristics, AQL 1.0 on Major B (fit), AQL 2.5 on minor, and zero acceptance with 100% inspection on the PTFE facing and polymer identity. Sample at general inspection level II, stratify by cavity, and report measured values rather than pass/fail. Hold the record for 24 months. Switch to tightened inspection after two consecutive Major A rejects. That plan will cost you about 40 minutes of bench time per lot and will catch the compression-set and cavity-drift escapes that a visual-only plan at AQL 1.0 will wave through. The septa themselves are usually fine; the plan is what decides whether you know it. If you are specifying septa and want the dimensional and material baseline to write the plan against, start from the HPLC septa product data and cross-check the fit dimensions against your HPLC vial drawing before you set a single AQL. The published methods and consumable references from Agilent, Thermo Fisher, and Restek are useful for the measurement methods, and EPA SW-846 is worth a look for how a regulator frames acceptance criteria on a consumable that reaches the sample path.

References & further reading

Related resources from our lab consumables network:

External references and standards: