Pick the AQL from the defect class, not from a house default

Last March a lot of 2 mL clear glass vials with pre-assembled PTFE/silicone septa arrived at a contract lab in Ohio with a certificate that said “AQL 1.0, passed.” The lab’s incoming check found 4 of 50 vials with septa pushed 0.6 mm below the crimp skirt face, enough to lift the sample needle 0.4 mm off the bottom of the insert. That is a critical defect at roughly 8% of the lot, and the supplier’s outgoing record had sampled 13 pieces. The gap was not dishonesty. It was a sampling plan that classified a dimensional defect as minor and drew a sample size that could not have caught it. What follows is how I build an AQL sampling plan and outgoing QC record for hplc autosampler vial lots, with the levels, sample sizes, and defect classes I actually use.

Most supplier certificates I audit default to AQL 1.0 across the board. That is a lazy number. AQL is a per-defect-class acceptance quality limit under ANSI/ASQ Z1.4 (the U.S. mirror of ISO 2859-1), and it should be set by what the defect does inside the autosampler, not by what is convenient to print. For hplc autosampler vial AQL sampling plan and outgoing QC work, I split defects into three classes and assign different limits:

• Critical (AQL 0.065, c=0): any defect that stops the vial from seating, sealing, or being pierced. Wrong septum thickness outside the 1.0 ± 0.1 mm band for 11 mm crimp, missing septum, cracked neck, dimensional stack that exceeds the 32 × 11.6 mm envelope, visible glass spall inside the cavity. Zero acceptance. One critical defect fails the lot.
• Major (AQL 0.65): defects that degrade performance but do not stop the run. Septum off-center by more than 0.25 mm, crimp skirt not fully rolled (loose cap), liner bonded off-axis, dimensional drift within tolerance but outside the historical mean by more than 2 sigma. These are the ones that produce ghost peaks and needle coring on the third puncture.
• Minor (AQL 1.5 or 2.5): cosmetic and handling issues. Scratches on the exterior, label misregistration, minor color variation in amber glass, cap color mismatch. AQL 2.5 is fine here. AQL 1.5 if the customer is doing automated capping and needs consistent cap geometry.

That split matters because a single AQL number applied across all defect types hides the failure mode that actually costs the lab money. If your supplier is running AQL 1.0 on everything, ask them to break it out. If they cannot, they are not running a real plan, they are running a checkbox.

Sample size follows the lot size, and the table is not negotiable

I have seen suppliers claim “we sample 100% of critical dimensions” and then produce a certificate showing a sample of 13. Both cannot be true. The Z1.4 general inspection level II table is the reference. For a lot of 10,001 to 35,000 vials (a typical single-shift production run), the sample size is n = 315 at general level II. For 3,201 to 10,000, it is n = 200. For 501 to 1,200, it is n = 80. For 151 to 280, it is n = 32. Below 150, you are in the small-lot regime and should be using a tightened plan or 100% inspection of critical dimensions, because the statistics are weak.

Most suppliers sample at general level I, which halves the sample: n = 125 for a 10,001 to 35,000 lot. That is a defensible choice for a mature process with a proven Cpk above 1.33, and I will accept it if the supplier shows the Cpk. Without the Cpk, level I is under-sampling. The math is simple: at AQL 0.65 with n = 125, the acceptance number is c = 2. You can ship a lot with 2 major defects in the sample and still pass, and the actual lot defect rate could be 3% or higher. At n = 315, c = 3, and the discrimination is tighter.

What the switching rules actually do

Z1.4 has normal, tightened, and reduced inspection states with switching rules. Normal to tightened after 2 out of 5 consecutive lots rejected. Tightened to normal after 5 consecutive lots accepted on tightened. Normal to reduced after 10 consecutive lots accepted with total defects at or below the limit number, and production steady. I require suppliers to declare which state they are in on every certificate. A supplier that is permanently in “normal” with no history is either not tracking or not telling you. A supplier in reduced inspection on a new product is a red flag.

Defect classification is where most QC plans fail

The single biggest weakness I see in outgoing inspection records is misclassification. A septum pushed below the skirt face gets coded as “cosmetic” because the inspector is looking at the top of the cap and the defect is inside. A loose crimp gets coded as “minor” because it does not fall off in the hand. Both are critical or major in the autosampler. The classification has to be written from the instrument’s point of view, not the inspector’s.

Here is the classification table I hand to suppliers:

The pull test is the one most suppliers skip. A crimp that releases below 8 N will come off in a robotic gripper or a tray handler. I have seen a lot pass dimensional inspection and fail on the pull test at 6 N average, because the crimping anvil had worn 0.03 mm. That is a tooling maintenance issue, but the outgoing record is where it gets caught.

What a realistic outgoing inspection record contains

A one-page certificate with “AQL 1.0, passed” is not a record. It is a claim. The record I require has six sections, and I have rejected suppliers who could not produce them:

1. Lot identification: lot number, manufacturing date, shift, forming line ID, crimping machine ID, septum lot number, glass lot number. The septum lot matters because a septum change mid-run is a known source of mixed-lot behavior.
2. Sample plan declaration: Z1.4 general level (I or II), inspection state (normal/tightened/reduced), sample size n, acceptance number c, AQL per defect class.
3. Measured data, not just pass/fail: height, OD, ID, septum thickness, crimp diameter, pull force. At least 10 data points per dimension, with mean and standard deviation. A pass/fail line tells you nothing about drift.
4. Defect tally: count per defect class, with the specific defect codes from the classification table. “2 major” is not enough. “2 major, code M-04 (septum off-center 0.31 mm and 0.28 mm)” is a record.
5. Instrument and method: caliper ID and calibration date, pull tester ID, optical comparator ID, environmental conditions (temperature and humidity at time of inspection). Glass dimensions move with temperature, and a 5 °C swing moves a 32 mm height by about 0.002 mm, small but worth recording.
6. Disposition: accept, reject, rework, or concession. If concession, the customer’s written approval reference. If rework, the rework procedure number and the re-inspection sample.

I also want the record to include a retained sample reference: how many vials from the lot are held, and for how long. 20 pieces held for 24 months is standard. That retained sample is what settles a dispute six months later when a customer reports ghost peaks and the supplier claims the lot was fine.

Field case: the certificate that passed but the lot did not

A European supplier shipped 40,000 vials to a pharma QC lab. Certificate showed AQL 1.0, n = 80, all pass. Within two weeks the lab reported needle bending on the third puncture in a Waters autosampler. I pulled the retained sample and ran a puncture test on 30 vials: average 2.4 punctures before needle deflection exceeded 0.1 mm, against a spec of 5 minimum. The septum durometer was 48 Shore A, not the 55 Shore A on the datasheet. The supplier’s outgoing record had no durometer measurement at all, because durometer was classified as an incoming material property, not an outgoing one. The fix was to add durometer to the outgoing plan as a major defect at AQL 0.65, sampled 10 pieces per lot from the assembled vials, not from the septum lot. The supplier’s incoming inspection had passed the septum lot at 55 Shore A because they sampled the top of the drum. The bottom of the drum had aged differently. The outgoing record now catches it because the test is on the finished vial.

The four measurements that belong on every outgoing certificate

You cannot measure everything, and a 40-page record is as useless as a one-line one. Four measurements catch the majority of field failures, in my experience:

• Overall height, 32.0 ± 0.4 mm. This is the dimension that determines whether the vial seats in the tray and whether the needle reaches the bottom. Measure at 3 points 120° apart. A single-point measurement misses tilt.
• Crimp diameter, 11.0 ± 0.2 mm at the skirt. Out-of-round crimps cause gripper misfeeds and incomplete seals. Measure at 2 axes.
• Septum thickness, 1.0 ± 0.1 mm for 11 mm crimp. Thin septa core early. Thick septa lift the needle and cause partial injections. This is the measurement most often missing from certificates.
• Pull force, ≥ 8 N. The single best predictor of cap release in automated handling.

I add durometer (55 ± 5 Shore A for silicone, 60 ± 5 for PTFE-faced) and a puncture count on 10 pieces per lot. Puncture count is destructive and slow, so it goes on the certificate as a lot-release test, not a per-piece inspection. A lot that averages below 5 punctures before coring is rejected regardless of the AQL tally.

Where the AQL plan and the Cpk plan have to meet

An AQL plan is an attribute plan. It tells you whether the lot is acceptable at a stated defect rate. It does not tell you whether the process is stable. That is the Cpk conversation, and it belongs to the supplier’s in-process data, not the outgoing record. But the two have to be consistent. If the supplier reports Cpk 1.67 on height and the outgoing sample shows a standard deviation of 0.18 mm against a 0.4 mm tolerance, those numbers do not agree, and one of them is wrong.

My rule: the outgoing record’s measured standard deviation should be within 20% of the in-process standard deviation for the same dimension. If it is not, the sampling is biased (inspectors picking the good ones) or the measurement method differs. I have caught both. The fix is to require the same gauge and the same fixture on the outgoing line as on the in-process line, and to require the record to state the gauge ID.

What to do when a lot fails outgoing inspection

Rejection is not the end of the story. The disposition decides whether you have a supplier problem or a lot problem. Here is the decision checklist I use:

• Critical defect, any count: reject the lot. Do not accept on concession for critical defects on a validated method. The liability is not worth it.
• Major defect at or below c: accept, but require the supplier to open a corrective action and report the root cause within 10 working days.
• Major defect above c: reject. Offer the supplier the option of 100% screening at their cost, with a re-inspection sample at tightened inspection (n = 315, c = 1).
• Minor defect above c: accept on concession if the defect does not touch the instrument interface. Document the concession reference.
• Any failure on a dimension with Cpk below 1.33: reject and require a process change, not just a re-screen. Screening a process that is not capable just moves the defects to the next lot.

The 100% screening option is the one suppliers resist. I insist on it because it changes the economics. Screening 40,000 vials at 3 seconds each is 33 hours of labor, which is usually more expensive than scrapping the lot, so the supplier suddenly finds the root cause. That is the point.

Field case: the crimp that drifted after a tooling change

A supplier changed crimping anvils mid-run without notifying the customer. The outgoing record for the next three lots showed pull force means of 9.1 N, 8.7 N, and 8.4 N, all passing the 8 N minimum. The fourth lot measured 7.6 N mean with a standard deviation of 0.9 N, and the AQL sample caught 3 majors out of 200, which passed at c = 5 for AQL 0.65 at n = 200. The lot shipped. Two weeks later the customer’s robotic handler dropped caps on 1.2% of vials. The measurement that identified it was the trend on the pull force mean across four lots, not the single-lot AQL result. The fix was to add a trend rule to the outgoing plan: any dimension whose lot mean moves more than 1.5 standard deviations from the previous 10-lot mean triggers a hold and a tooling check. The AQL plan alone would never have caught it, because each lot individually passed. This is why the outgoing record needs the measured means, not just the pass/fail tally.

Torque, crimp force, and the numbers that belong in the purchase order

For screw-thread vials (9 mm and 10-40), the outgoing plan should include a torque measurement. The target is 1.5 to 2.0 N·m for a 9 mm screw cap with a PTFE/silicone septum, measured with a calibrated torque tester on 10 pieces per lot. Below 1.2 N·m the seal leaks on the autosampler tray. Above 2.5 N·m the septum deforms and cores on the first puncture. I have seen both. The purchase order should state the torque window and the test method, or the supplier will use whatever their capping machine happens to produce.

For crimp vials, the equivalent number is the crimp force, typically 180 to 220 N on a pneumatic crimper with a 11 mm anvil. The outgoing record should show the crimper setpoint and a pull-force verification. A crimper that drifts 10 N produces a lot that passes visual and fails pull. The crimp force is not a vial property, it is a process property, but it belongs on the record because it explains the pull-force result.

If you are sourcing vials and want the full dimensional and material context behind these numbers, the Hplc Autosampler Vial reference on this site covers the envelope and septum specs, and the Sample Vial pricing reference is useful when you are comparing suppliers at different AQL levels, because the sampling cost is real and it shows up in the unit price. For the broader instrument-side context, Agilent’s chromatography consumables and columns documentation and Thermo Fisher’s chromatography and sample handling pages both list the tray and gripper tolerances that your outgoing dimensions have to meet. Restek’s technical resources have useful notes on septum durometer and coring, and the EPA SW-846 test methods are worth reading for how method validation treats consumable variability, though SW-846 does not certify vials.

The outgoing plan is a contract, not a certificate

The mistake I see most often is treating the outgoing inspection record as a document the supplier produces to satisfy the customer. It is the other way around. The plan is a contract that specifies what the supplier measures, how they sample, how they classify, and what they do when a lot fails. The certificate is the evidence that the contract was executed. If the contract is vague, the evidence is worthless. Write the AQL levels per defect class into the purchase order. Write the sample size and inspection level into the purchase order. Write the four measurements and the pull-force minimum into the purchase order. Then audit the record against the purchase order, not against the supplier’s template.

The decision rule: critical defects at AQL 0.065 with zero acceptance, major defects at AQL 0.65 with n = 315 at general level II for lots above 10,000 pieces, and a pull-force minimum of 8 N on every lot, with the measured means and standard deviations on the certificate. If a supplier cannot meet that, the unit price they quoted is not the price you are paying. The price you are paying includes the rework, the failed runs, and the investigation time. The sampling plan is how you find out which price you are actually being charged. Suppliers who can produce the full record, with the measured data and the trend rule, are the ones worth the premium. The ones who cannot are selling you a certificate, not a lot. For OEM and private-label programs, the same plan applies to the Chromatography Products catalog and to custom configurations from Aijiren, because the defect classes do not change when the label does.

References & further reading

Related resources from our lab consumables network:

• Hplc Autosampler Vial at hplcautosamplervials.com
• Sample Vial at samplevialprice.com
• Chromatography Products at chromatographyselling.com
• Aijiren at aijirentech.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