The lot passed everything on the certificate: 45 Shore A, 0.060 in thick, 9.5 mm OD, siloxane extractables under the USP <381> threshold. Then the customer’s outgoing inspector pulled a 50-piece sample from the 20,000-piece shipment, found three septa with flash bridging the center hole, and rejected the lot on a single attribute. That rejection was correct under the plan, and it was also avoidable, because the flash had been there when the parts left the press. The gap was not quality. It was the sampling plan: nobody had defined what “defective” meant for flash, so the inspector used their own judgment and the shipment died at the dock. A vial septa AQL sampling plan and outgoing QC program exists to make that judgment mechanical, repeatable, and defensible to both sides of the transaction.
Most arguments about outgoing inspection start because someone picked a single AQL for the whole product. That is the wrong unit of analysis. Septa fail in different ways with wildly different consequences, and the plan should price each class separately.
I classify outgoing defects into three tiers. The tiers come from what happens at the customer’s bench, not from what is easy to measure.
- Critical (AQL 0, zero acceptance): any defect that creates a leak path or a chemical interference. Holes through the septum, incomplete pierce, voids at the sealing bead, contamination with mold release, wrong polymer (a silicone part where a butyl was ordered). These do not get an AQL allowance. A single occurrence in the sample rejects the lot. There is no defensible reason to accept a known leak path at any rate above zero.
- Major (AQL 0.65): defects that cause functional trouble without immediate failure. Dimensional drift outside the drawing on the sealing OD or thickness, flash that intrudes into the center hole, hardness outside the 45 ± 5 Shore A window, incomplete cure (tacky surface). A 0.65 AQL on a 20,000-piece lot means you will accept a lot where the true defect rate is around 0.65% or better and reject worse lots most of the time.
- Minor (AQL 1.5 to 2.5): cosmetic and handling defects. Slight discoloration, mold-flow lines that do not cross the sealing surface, minor edge nicks that do not reach the seal. I default to 2.5 for cosmetic and 1.5 where the customer is running an autosampler with a fragile needle guide and a nicked edge can catch.
Why 0.65 for major and not 1.0? Because the sampling tables give you a usable sample size at that level. At AQL 0.65 and inspection level II, a 20,000-piece lot draws a sample of 315 pieces with an acceptance number (Ac) of 5 and a rejection number (Re) of 6. At AQL 1.0 the same lot draws 315 with Ac 7. The difference in protection is real but the sample size is identical, so there is no cost reason to loosen it. Go to 0.65.
The AQL tables in question are ISO 2859-1, the same structure the old MIL-STD-105E used. If your customer’s quality agreement cites ANSI/ASQ Z1.4, the numbers match. Use one table and say which one in the inspection record.
Sample size follows lot size and inspection level, and level II is usually wrong for septa
The default in ISO 2859-1 is general inspection level II. For a lot of 20,000 that is a code letter M and a sample of 315. For a lot of 3,500 it is code letter K, sample 125. For a lot of 500 it is code letter H, sample 50. Those are the numbers most people plug in.
I move septa to general inspection level I for the routine outgoing check and reserve level II for first article, requalification, and any lot where a prior lot was rejected. Level I halves the sample: 20,000 pieces draws 125, not 315. That is a defensible trade because septa are produced on multi-cavity molds with tight process control, and the defect distribution is not random across the lot in the way a level II sample assumes. Cavity-to-cavity variation dominates. If you want to catch a bad cavity, you do not need 315 random pieces. You need to know which cavities fed the lot and sample each one.
That is the practical move most published tables ignore: stratify the sample by cavity. If the mold is 32-cavity and the lot is 20,000, pull roughly four pieces per cavity from the level I sample, not 125 pieces from the top of the tote. A single bad cavity producing 3% of the lot will hide in a random sample of 125 (expected defectives in sample: under 4, often 0 to 2) but show up immediately when you pull 4 from that cavity and 2 are bad.
When to switch to tightened or reduced inspection
ISO 2859-1 has switching rules built in and they are worth using. Start normal. After two consecutive lots accepted on the same part number and process, move to reduced inspection: same sample size, lower Ac. After two lots rejected in five, move to tightened: larger sample, Ac held at the same number, which drops the consumer’s risk. Return to normal after five consecutive lots accepted under tightened. Write the rule into the inspection procedure so it does not depend on who is on shift.
What actually goes in an outgoing inspection record
This is where most suppliers fall short. The record is not a certificate of analysis with a green checkmark. It is a document that lets a customer reconstruct the decision. Here is what mine contain, in order.
- Lot identification: part number, cavity count, press or line, date and shift of molding, cure date, quantity. Traceable to the molding log. If you have a traceability chain, this is the link that ties the shipment to the press.
- Sampling plan statement: standard cited (ISO 2859-1 or ANSI/ASQ Z1.4), inspection level, lot size, code letter, sample size, AQL per defect class, Ac and Re per class. Written out, not implied.
- Defect classification table: the three tiers, with a one-line definition of each defect. “Flash” alone is not a definition. “Flash bridging the center hole or extending more than 0.010 in past the sealing bead” is.
- Measured data, not just pass/fail: thickness at three points per sampled piece (or a subsample of 10), sealing OD, center hole diameter, hardness (Shore A, 5 readings per subsample). Report the mean and the range. A pass/fail hides drift; a range shows it.
- Functional check: a pierce test on a defined subsample, typically 10 pieces, using the customer’s needle gauge (commonly 22 gauge, 0.028 in OD) and counting punctures to failure or to a fixed count. Report the count and the failure mode (coring, tearing, needle drag).
- Attribute results: number inspected per class, number defective per class, accept or reject decision per class, and the overall lot disposition.
- Inspector, date, and the gauge or instrument IDs used for the dimensional and hardness measurements.
If a customer runs USP methods, tie the record to the relevant chapter. USP <381> covers elastomeric closures for containers, and the functional tests there (penetrability, fragmentation, self-sealing) map cleanly onto the pierce and reseal checks above. Cite the chapter in the record so the customer’s auditor can follow the thread. The vial septa you ship should carry a record that survives an audit without a phone call.
Two field cases where the outgoing plan caught what the certificate did not
Case 1: a hardness drift that passed dimensions and failed the needle
A 10 mL screw-thread lot measured fine on thickness and OD. The outgoing hardness subsample read a mean of 52 Shore A against a 45 ± 5 target, so it was technically in spec at the upper edge, but the range across ten pieces ran 49 to 55. The pierce test on the same lot showed needle drag on 4 of 10 pieces and coring on 2. The measurement that identified it was the range, not the mean: a 6-point spread on a 10-piece subsample from a single lot means the cure was uneven, and the hard pieces were under-cured at the surface and over-cured in the body. The fix was a cure-time adjustment on the press and a tightened hardness spec of 45 ± 3 for that part number, with the range reported on every record. The customer’s autosampler vial line stopped seeing bent needles within one lot.
Case 2: flash that only showed up when the sample was stratified by cavity
A 16-cavity mold was producing a 12 mm septum for a crimp vial. Random samples of 125 passed four lots in a row. The fifth lot failed at the customer’s incoming inspection with flash on 2 of 50 pieces. We pulled the retained sample and re-inspected it stratified by cavity: cavity 11 had flash on 3 of 8 pieces, every other cavity had zero. The random sample had happened to draw only two pieces from cavity 11 across the whole 125. The fix was a vent clean on cavity 11 and a change to the outgoing procedure: stratified sampling by cavity from that point on, with a minimum of 4 pieces per cavity regardless of the table’s total. The next 12 lots passed at the customer, including two that would have been rejected under the old plan.
Set the pierce and reseal checks with numbers, not adjectives
Functional checks are where outgoing inspection earns its keep, and where most plans are vague. Here is what I specify.
| Check | Method | Sample | Acceptance criterion |
|---|---|---|---|
| Pierce count to failure | 22 gauge needle, 0.028 in OD, 1 in/min insertion, septum in a crimp or screw cap torqued to spec | 10 pieces | ≥ 50 punctures at the same site without coring or tearing for silicone; ≥ 20 for butyl |
| Reseal (self-sealing) | Pierce once, hold 10 s under 10 psi water or dye, inspect for leak | 10 pieces | Zero leaks |
| Coring | Pierce 10 times, inspect bore with 10x loupe or a 0.5 mm pin gauge | 10 pieces | No visible plug in bore, no pin gauge pass |
| Hardness | Shore A, 5 readings per piece, 1 s dwell | 10 pieces | Mean within spec, range ≤ 4 points |
| Thickness | Micrometer, 3 points per piece | 10 pieces | Within drawing, typically ± 0.003 in |
| Sealing OD | Optical comparator or pin gauge | 10 pieces | Within drawing, typically ± 0.005 in |
Torque matters on the reseal check. A 13 mm crimp cap should be crimped to a cap height of about 5.5 mm and a 20 mm crimp to about 7.5 mm; a screw cap on a 9 mm thread should be hand-tight, roughly 0.4 to 0.6 N·m. If the outgoing reseal test is run on a hand-crimped cap with no height control, you are testing the crimper, not the septum. Use a calibrated crimper and record the setting.
Write the defect definitions before the first lot ships
The most common failure of an outgoing plan is not the AQL level. It is that two inspectors disagree on whether a piece is defective. Fix that with a defect definition sheet that sits next to the inspection station, with photographs and numbers. A few I insist on:
- Flash: any material bridging the center hole, or extending more than 0.010 in past the sealing bead. Flash on the non-sealing face under 0.010 in is minor.
- Voids: any void at the sealing bead, any size, is critical. Voids in the body away from the seal are major if over 0.020 in across.
- Contamination: any visible particle, mold release residue, or discoloration on the sealing face is critical. Discoloration elsewhere is minor.
- Dimensional: out of drawing on thickness, OD, or center hole is major. Out of drawing on the non-sealing flange is minor.
Pin the definitions to the drawing and to the standard. If the customer is running EPA SW-846 methods, the container is part of the method, and the septum’s sealing face contamination is not cosmetic. You can read the method language at the EPA SW-846 page and it will tell you the same thing. That is the reason contamination sits in the critical tier at zero AQL for environmental-grade parts.
Keep the gauge and the plan under control
An outgoing plan is only as good as its instruments. Micrometers and pin gauges need calibration on a 12-month cycle, Shore A durometers on 12 months or sooner if they are used daily, and the pierce fixture’s needle needs a fresh needle every 500 punctures or whenever drag is felt. A dull needle will manufacture coring defects that are not in the part.
Record the gauge IDs on the inspection record. When a customer disputes a rejection, the first question is which instrument measured it and when it was last calibrated. If you cannot answer in one line, the record is incomplete.
What the plan cannot do, and where to put the effort instead
Outgoing sampling catches bad lots. It does not make good ones. A 125-piece sample from a 20,000-piece lot will miss a 1% defect rate most of the time; that is arithmetic, not a flaw in the plan. The plan’s job is to protect the customer from a bad lot and to give both sides a shared, documented basis for the accept/reject decision. The work that actually reduces defects happens upstream in the press, the cure, and the cavity, and the outgoing data should feed back there. If cavity 11 keeps showing flash in the stratified sample, the answer is a vent clean, not a tighter AQL.
If you are sourcing septa and want to see how the parts and the documentation fit together, the wider catalog at vials and filters and the application notes at HPLC and GC consumables are useful references. For instrument-side specifics, the Agilent GC columns and accessories pages and Thermo Fisher’s chromatography and sample handling documentation cover the needle and vial interfaces that drive the pierce and reseal specs. Restek’s technical resources are worth reading for septa bleed and background data that inform the contamination tier.
A working rule: critical defects at AQL 0, major at 0.65, minor at 2.5, level I inspection stratified by cavity with a minimum of 4 pieces per cavity, functional pierce and reseal checks on 10 pieces per lot, and a written defect definition sheet with numbers on it. If a supplier cannot produce that record on request, the certificate is decoration and the shipment is a coin flip.
References & further reading
Related resources from our lab consumables network:
- Vial Septa at vialsepta.com
- Hplc Vial at hplcgc.com
- Autosampler Vial at autosamplerhplc.com
- Vial at vials-filters.com
External references and standards:
