Control loops close on glass dimensions, not on chemical durability

A lot of 2 mL clear glass vials left the annealing lehr at 11 ppm residual strain against a 40 ppm reject limit, passed dimensional go/no-go, and still failed the customer’s USP <661> hydrolytic resistance screening six weeks later. The strain was fine. The problem was upstream: the forming machine’s gob weight had drifted 0.4 g high for ninety minutes, the wall thinned at the heel, and the lehr pulled the stress out of the body but not the base. Nobody on the line controlled wall thickness. It was only checked at the end, on a sample of thirty.

That is the whole argument of this piece. When people ask how analytical testing consumables are manufactured and what the factory actually controls, the useful answer is not a process flow diagram. It is a map of which parameters sit inside a closed control loop and which parameters are merely inspected after the fact. Everything I write below comes from running lines that make vials, caps, septa, and filter hardware, and from the arguments I have had with quality managers about where the loop should close.

If you want the catalog view of the product family, analytical testing covers it. This article is about the line behind it.

On a vial forming line, the parameters inside a closed loop are the ones a sensor can read in under a second and an actuator can correct without stopping the machine. That list is short and specific:

  • Gob weight, closed loop via shear timing and feeder speed, held to ±0.15 g on a 2 mL vial.
  • Mold temperature, closed loop via thermocouple and cooling air, held to ±5 °C.
  • Blow pressure and timing, closed loop on the servo valve, held to ±0.02 bar.
  • Lehr belt speed and zone temperatures, closed loop, with the annealing zone held to ±3 °C.
  • Inside diameter at the neck, closed loop on the finish mold, checked by inline laser every 90 seconds.

Now the parameters that are not in a loop, and this is the part that matters to a procurement manager:

  • Wall thickness distribution. Measured offline on a sectioned sample, typically 10 to 30 pieces per shift. It is a check, not a control.
  • Residual strain. Measured by polariscope on a sample, not on every unit. The lehr is controlled; the strain result is inspected.
  • Hydrolytic resistance. USP <661> and the ISO 719 / ISO 720 powder and grain tests run on a lot sample, days after the glass was made. You cannot correct a lot that fails. You can only reject it.
  • Extractables and leachables. A qualification activity, not a line control. It runs once per glass type and formulation change.

So the honest statement is this: the factory controls the shape of the glass in real time and controls the chemistry of the glass only by controlling the batch recipe and the furnace, then verifying it later. If your supplier tells you they “control” hydrolytic resistance on every lot, ask what the control variable is. The answer is usually the batch recipe and the cullet ratio, verified by sampling. That is legitimate. It is just not a closed loop.

What the batch house actually locks down

The batch recipe on a Type I borosilicate line is where the chemistry gets fixed. The parameters a serious factory holds are the silica, boron, and aluminum oxide ratios to within about ±0.3 wt%, the cullet fraction (typically 20 to 40%, and it must be the same cullet, not mixed), and the furnace pull rate. Change the cullet source and you change the surface chemistry, even if the certificate of analysis looks identical. I have seen a lot fail USP <661> powder test after a cullet supplier switch that nobody flagged because the batch sheet was “the same.”

Septa are where the line controls the least and the customer feels it the most

A 11 mm silicone/PTFE septum for a crimp vial is a laminated part. The silicone provides the reseal, the PTFE provides the chemical barrier. What the factory controls on the line is the lamination, the die cut, and the PTFE film thickness. What it checks at the end is puncture performance and extractables, and those two things are exactly what an autosampler punishes you for.

Here is the comparison that matters when you are specifying septa, and it is a decision checklist rather than a spec sheet:

Parameter Typical controlled value How it is verified Failure mode if it drifts
PTFE film thickness 0.10 mm ±0.01 mm (premium), 0.05 mm ±0.01 mm (standard) Inline beta gauge, closed loop Thin film tears at puncture, silicone exposed to sample
Silicone durometer 45 ±3 Shore A Offline, per batch Too soft: coring and reseal loss. Too hard: needle bend
Puncture count to reseal failure >100 punctures at 22 G, <5 µL leak Offline, lot sample of 12 Leak at injection port, ghost peaks, area RSD climb
Die cut concentricity ±0.15 mm Vision system, sampled Off-center septum, crimp seal leak
Extractables (total) <2 mg per septum after 24 h at 40 °C in hexane Offline, qualification only Baseline rise, interfering peaks in early elution window

Notice the pattern again. Film thickness is closed loop. Puncture count and extractables are end-of-line checks. The line can make a million septa a day and never puncture one. It punctures twelve per lot and calls that evidence. If you are qualifying a septum for a method with a 40 °C autosampler tray and 200 injections a week, that sample of twelve is the only thing standing between you and a reseal failure at injection 70. Ask for the puncture data, not the certificate. The chromatography vial supplier who can hand you a puncture-count distribution is the one who actually runs the test.

Caps and closures: the torque window is the control, and most factories do not close it

Crimp caps and screw caps are where dimensional control and assembly control meet. For a 9 mm screw cap on a 2 mL vial, the useful number is the torque window. A typical target is 1.2 to 1.5 N·m for a 9 mm cap, and the window is narrow because below it you get evaporation and above it you get septum distortion and, on some caps, cracking of the plastic at the thread root.

What the factory controls on the capping line is the torque, and it does so with a servo capper that reads torque on every cap and rejects out-of-window caps. That is a genuine closed loop and it is the one place where the line is better than the lab. In the lab, a human hand-tightens and the torque is whatever the analyst’s wrist says. If you want to know why two analysts get different results on the same lot, the cap torque is a good place to look. I have measured 0.6 N·m on one analyst’s caps and 2.1 N·m on another’s, on the same vial lot, and the 2.1 N·m caps showed septum doming.

The parameters that are checked at the end on closures are the ones you would expect: pull-off force for crimp caps (target typically 25 to 40 N), penetration force through the cap, and seal integrity by dye penetration or vacuum. Those are lot-sample checks. The line controls torque and cap dimensions; it inspects the seal.

Where GPI and ISO threads bite

Screw thread dimensions follow GPI 9-425 for the 9 mm finish and the ISO 4796 family for the vial bodies. The line controls the thread pitch and the major diameter to within about ±0.05 mm. What it does not control, and cannot, is the mismatch between a cap from one supplier and a vial from another. If you are mixing a HPLC vial from one factory with caps from another, the torque window you measured on the cap supplier’s reference vial may not hold on yours. Buy the pair together, or verify the pair yourself.

Filter hardware: the membrane is inspected, the housing is controlled

Syringe filters and inline filter housings follow the same split. The housing dimensions (Luer lock taper per ISO 594, inlet and outlet bore, housing wall) are closed-loop controlled on the injection molding machine: cavity pressure, melt temperature, and cooling time all sit in loops, and the parts come off within ±0.03 mm. The membrane inside is a purchased component, and what the factory does to it is inspect: bubble point, flow rate, and extractables are checked on a lot sample, not controlled per unit.

The number that matters for a 0.22 µm PVDF filter is the bubble point, typically >3.5 bar with water or the manufacturer’s specified wetting fluid. A low bubble point means a pinhole or a wetting failure, and it will show up as particulate in your sample, not as a failed filter test. The line cannot close a loop on bubble point. It can only test a sample and reject the lot.

Two field cases where the end-of-line check was the only thing that caught it

Case 1: the wall-thickness drift that passed strain

Symptom: a customer running a stability study saw vial breakage in the centrifuge at 3,000 × g, about 1 in 400 vials, on a lot that had passed all incoming checks.

Measurement that identified it: we sectioned 40 vials from the lot and measured wall thickness at the heel and the body with a micrometer. Body wall was 1.05 mm ±0.04 mm, fine. Heel wall was 0.72 mm on the failing vials against 0.95 mm nominal. The gob weight log showed a 0.4 g high excursion for 90 minutes, four days before the lot shipped. The lehr had annealed the body; the heel was thin and under-annealed.

Fix: we added heel-wall measurement to the offline check at 20 pieces per shift and put an alarm on gob weight drift above 0.2 g. The loop still does not close on heel wall, but the alarm catches the cause within the shift instead of four days later. That is the realistic improvement: shorten the detection lag, not pretend you can control it.

Case 2: the septum that passed puncture count and failed at injection 70

Symptom: a lab running 180 injections a day on a 40 °C tray saw area RSD climb from 0.8% to 4.5% over a sequence, with the drift starting around injection 60 to 70 and getting worse.

Measurement that identified it: we ran a reseal test on the lot: 22 G needle, 100 punctures in the same spot, pressure decay measured every 10 punctures. The lot passed the supplier’s 100-puncture criterion at the 5 µL leak limit, but the leak at puncture 70 was 3.8 µL and climbing. The PTFE film was 0.07 mm, inside the 0.05 to 0.10 mm band but at the thin end, and the silicone under it had a durometer of 41 Shore A, below the 45 ±3 target.

Fix: we tightened the incoming spec to PTFE ≥0.09 mm and durometer 44 to 48 Shore A for tray applications, and the drift moved out past injection 150. The lesson is that a pass/fail puncture count hides the shape of the reseal curve. Ask for the curve.

What to ask a factory before you trust its line

This is the checklist I would use on a supplier audit, in order, because it separates the parameters that are actually controlled from the ones that are inspected and called controlled:

  • Show me the closed-loop list. Which parameters have a sensor, an actuator, and an automatic correction? If the answer is “all of them,” the supplier does not understand the question.
  • Show me the sample size and frequency for the end-of-line checks. Wall thickness, residual strain, puncture count, bubble point. Thirty pieces per shift is a check. One piece per shift is a gesture.
  • Show me the drift alarm setpoints. Gob weight, mold temperature, lehr zone temperature, capping torque. A factory that cannot name its alarm limits is running open loop and calling it control.
  • Show me the batch recipe control. Cullet fraction, furnace pull, and the last recipe change. This is the only real control on glass chemistry.
  • Show me the retention samples. If a lot fails six weeks later, what do you have to test? A factory with no retained samples cannot investigate its own failure.

For reference material on what the end tests are supposed to look like, the U.S. EPA SW-846 test methods are a useful benchmark for how a method defines its own acceptance criteria, and the technical notes from Restek and the consumables documentation from Agilent and Thermo Fisher show how the vendors themselves characterize septa, liners, and vials. If your supplier’s data sheet has fewer parameters than the vendor’s application note, that tells you something. The broader product range is catalogued at chromatography products if you need to compare across formats.

The decision rule

Treat any parameter a factory claims to “control” as controlled only if it has a sensor, an automatic correction, and a logged alarm limit; treat everything else as an inspection, and size your incoming verification to the inspection frequency the factory actually runs, not the frequency on the certificate. If a critical attribute for your method (heel wall, PTFE thickness, durometer, bubble point, cap torque) is on the inspection side of that line, buy the data, not the promise.

References & further reading

Related resources from our lab consumables network:

External references and standards: