We rejected 180,000 9 mm screw-thread vials in one lot because the neck bore ran 6.38 mm against a 6.20 ± 0.10 mm print. The septa seated, the caps torqued to 1.2 N·m without stripping, and the vials looked fine on the pallet. The problem only surfaced when a customer ran a 96-position autosampler and the needle bent on positions 61 through 72. That is the reality of chromatography supplier dimensional tolerance capability: the failure hides inside a feature nobody inspects until it breaks a run. This piece is about which tolerances a molding process can actually hold on a vial body, how you measure whether a supplier holds them lot to lot, and which dimensions are the ones that quietly drift.
Tolerance and capability are not the same thing. A print that says 6.20 ± 0.10 mm describes the allowed window. Capability describes how much of that window the process consumes with normal variation. If your supplier reports only “within tolerance,” you know nothing about the next lot. If they report a Cpk of 1.33, you know the process mean sits at least four standard deviations from each limit. On a 0.20 mm total window, that means a standard deviation of about 0.025 mm or less.
Why 1.33 and not 1.67? For injection-molded consumables, 1.33 (roughly 32 defects per million on a centered normal process) is a realistic floor for a mature tool. Demanding 1.67 on a glass vial neck is often a way to get a datasheet that lies rather than a part that is better. I would rather see honest 1.33 with a control chart than a claimed 2.0 with no data behind it. Run the numbers yourself: a 0.10 mm half-width divided by three standard deviations gives you the sigma the process must hold. At Cpk 1.33 the process sigma is 0.025 mm; at Cpk 1.67 it is 0.020 mm. That 0.005 mm gap is the difference between a tool that runs three years between rebuilds and one that needs a core pin change every eighteen months.
How to read a capability number without being fooled
Ask three questions. First, is the Cpk calculated from a single cavity or from the whole tool? A 16-cavity mold can show 1.5 overall while cavity 7 sits at 0.9. Second, over how many lots? Thirty consecutive lots is a capability study; three lots is a sales sample. Third, is the dimension measured on the finished part or on the green part before annealing? Glass and molded polymer both move after forming, and a number taken before the stabilizing step is fiction.
There is a fourth trap: the normality assumption. Cpk assumes a roughly normal distribution. Bore diameter on a glass conversion line is often skewed because the fire-polish step pulls one tail in. If the distribution is skewed, a Ppk computed from the actual data is more honest than a Cpk from an assumed normal fit. Ask for both. When they diverge by more than about 0.2, the process is not normal and the Cpk is optimistic.
If you are evaluating a new source, this is the same discipline you apply when you audit a chromatography supplier on anything else. The tolerance capability question is just the one most buyers forget to ask.
What each process can hold on a vial body, in numbers
Different forming routes land in different tolerance bands. Here is what I have seen hold up in production, not in a brochure.
| Process | Typical achievable tolerance (critical features) | Best feature | Weak feature | Realistic Cpk on 0.10 mm window |
|---|---|---|---|---|
| Glass tubing conversion (borosilicate) | ± 0.10 mm ID, ± 0.15 mm OD | Bore concentricity, chemical inertness | Neck roundness after threading, wall thickness | 1.33 to 1.67 |
| Injection molding (PP, PE closures) | ± 0.05 mm on molded features | Thread pitch, sealing land flatness | Post-mold shrinkage drift, warpage | 1.33 to 1.50 |
| Compression molding (PTFE/silicone septa) | ± 0.08 mm thickness | Thickness uniformity, durometer | Flash, edge concentricity | 1.00 to 1.33 |
| Injection blow molding (larger bodies) | ± 0.15 mm | Wall distribution | Neck ID control | 0.9 to 1.2 |
The takeaway: if a print calls for ± 0.03 mm on a molded thread, the process cannot hold it at volume, and the supplier who says yes is quoting a single cavity on a good day. If a print calls for ± 0.03 mm on a glass bore, that is achievable but you will pay for 100% gauging. The cost curve is steep. Moving a molded feature from ± 0.10 to ± 0.05 mm roughly doubles the tooling cost and adds a cooling-fixture step; going to ± 0.03 mm on the same feature typically requires a second operation (machining or post-mold sizing) and pushes unit cost up 40 to 60%.
One more number worth carrying: glass wall thickness on a 1.5 mL vial body typically runs 1.0 to 1.2 mm with a ± 0.15 mm band. If the wall varies more than that, the autosampler gripper can slip on the thinner section. Wall thickness is not usually on the critical-dimension list, but it is the reason vials jam in the carousel.
The three dimensions that actually cause failures, and why they are hard
Most of a vial print is easy. Wall thickness, overall height, thread major diameter: these are forgiving and rarely bite you. The failures cluster on three features.
Neck bore and concentricity
The bore sets how the septum compresses and how the needle enters. On a 9 mm screw vial the bore is typically 6.0 to 6.2 mm. If bore and thread are not concentric within about 0.08 mm, the cap pulls the septum off-center, and the needle meets a compressed edge instead of a flat face. That is where bent needles and coring start. Concentricity is hard because it is a relationship between two features formed at different points in the cycle, and it drifts as tool wear changes the core pin position.
On a glass line, the bore is set by the mandrel that forms the neck, and concentricity depends on the mandrel staying centered as the tube is spun. A mandrel that is out of true by 0.02 mm at the bearing produces a bore that is out by 0.04 mm at the tip. That is half your concentricity budget gone before the tool even warms up. On a molded closure, concentricity depends on the core pin and the cavity gate location; a gate on one side of the cavity pushes the core off-center as the melt fills.
Sealing land flatness
The annular land the septum sits on must be flat within roughly 0.05 mm across its full circumference. A land that is dished by 0.08 mm will not seal even at 1.5 N·m torque, because the contact pressure is uneven. Flatness is hard to hold on glass (it depends on the fire-polish step) and on molded closures (it depends on cooling uniformity across the cavity).
Here is a working number for the septa side: a 1.0 mm PTFE-faced silicone septum compresses about 25 to 30% at 1.2 N·m on a properly flat land. That gives you roughly 0.25 to 0.30 mm of compression to work with. A 0.08 mm land dish eats a third of that budget on one arc. The septum cannot recover the difference, so the seal fails on the thin side first.
Thread lead and pitch diameter
A GPI 9 mm thread has a defined pitch and flank angle. If pitch diameter drifts by 0.1 mm, the cap either binds before it seats or bottoms out before it compresses the septum. Both failures look identical to the operator: the cap “felt tight” but the vial leaked. Pitch diameter is the dimension most sensitive to tool wear, and it is the one least often on a routine inspection plan.
The wear mechanism is simple. Every molding cycle, the thread insert rubs against the cavity steel as the part is ejected. Over 500,000 cycles, a hardened insert loses 0.02 to 0.04 mm off the flank. On a 0.10 mm pitch-diameter tolerance, that is a third of the window gone with no other change. A supplier who does not chart pitch diameter will not notice until the caps start binding, and by then the tool is out of tolerance.
These are the same features that decide whether a closure seals at all, which is why the closure specialists at vialclosures.com treat them as the primary gate rather than a secondary check.
How to measure capability you can trust
A capability number is only as good as the gauge behind it. For a 6.20 mm bore with a 0.10 mm tolerance, your measurement system needs a gauge R&R under 10% of tolerance, meaning the gauge repeatability and reproducibility should be under 0.02 mm. A caliper at the bench will not get you there; it will read 0.02 to 0.05 mm of its own variation and swamp the signal. Use an air gauge, a pin gauge set, or an optical comparator with a fixtured part.
For concentricity, a simple bench method is a plug gauge in the bore and a dial indicator on the thread flank, rotated through one full turn. The total indicated runout is your concentricity number. For land flatness, an optical flat under monochromatic light shows fringes; each fringe is about 0.0003 mm, so a 0.05 mm dish shows as roughly 170 fringes, which is easy to see and easy to reject. For pitch diameter, a thread micrometer or a three-wire method gives you the number without a go/no-go gauge.
A sampling plan that catches drift, not just defects
- 5 parts per cavity at start, middle, and end of run
- Bore, concentricity, land flatness, pitch diameter measured on every part in the sample
- Gauge R&R re-run quarterly or after any gauge change
- Control charts on the two most drift-prone features (bore, pitch diameter) with a Cpk recompute every 20 lots
- Retained samples held for the shelf life so a field failure can be traced to a specific run
I have watched a supplier pass incoming inspection for eight months and then fail on month nine because nobody charted the pitch diameter. The tool had worn 0.06 mm over that period, still inside tolerance, but the Cpk had fallen from 1.4 to 0.8. The parts were “in spec” and unusable.
One detail on the sampling plan that people get wrong: sample across the run, not at the start. Molded parts change as the tool warms up over the first 200 shots. A sample taken at shot 10 is cooler and slightly larger than one at shot 2,000. If you only sample the start of the run, you miss the thermal drift that shows up two hours in. Sample at three points and plot them separately; a trend across the three points is your warning that the tool is heating unevenly.
Two field cases where the dimension, not the part, was the problem
Case one: the autosampler that bent needles on 12 of 96 positions
Symptom: a customer running a 96-well autosampler saw needle deflection on positions 61 through 72, consistently, across three runs. Needles were not worn; the same needles worked on other positions.
Measurement: we pulled the vial from position 61 and measured neck bore concentricity at 0.11 mm against a 0.08 mm limit. The bore itself was in tolerance (6.14 mm), but it was offset from the thread axis. Position 61 through 72 came from cavity 7 of a 16-cavity tool.
Fix: cavity 7’s core pin had shifted 0.04 mm after a maintenance rebuild. We re-centered the pin, re-qualified the cavity with a 30-part capability run (Cpk came back at 1.4), and added a per-cavity concentricity check to the routine plan. The lesson: aggregate Cpk hid a single bad cavity, and only per-cavity data found it. The customer’s needle cost was $340 per incident across the three runs; the re-qualification cost $1,800. The math favors the per-cavity check every time.
Case two: the leak that torque could not fix
Symptom: a pharmaceutical customer reported headspace leaks on 2.1% of vials after 24 hours at 40 °C, using a 1.2 N·m capping torque that had worked for years.
Measurement: we measured sealing land flatness on the failing vials and found a dish of 0.09 mm on the closure land. The cap was applying 1.2 N·m, but the contact pressure was concentrated on one arc. A leak test at the point of failure confirmed it.
Fix: the closure tool had a cooling imbalance on one side of the cavity, warping the land. We adjusted the cooling circuit, brought flatness back to 0.04 mm, and the leak rate dropped to 0.1%. No change to torque was needed; the torque was never the problem. The customer had already spent two weeks chasing capping equipment and had replaced a torque tester before calling us, which is the usual order of investigation and the usual waste.
Both cases share a pattern: the failure was a relationship between features, not a single dimension out of tolerance. That is the class of defect a capability study catches and a go/no-go gauge misses.
The failure modes that live in the tolerance stack
When bore, land, and thread each sit at the edge of their individual tolerances, the stack can produce a vial that passes every single-feature check and still leaks. A 0.10 mm bore offset plus a 0.05 mm land dish plus a 0.08 mm pitch diameter drift can put the effective seal contact outside the septum’s compression range. This is why suppliers who only inspect features in isolation ship defects that only appear in the field.
Run the stack arithmetic to see how bad it gets. Worst case, a 0.10 mm bore offset moves the septum contact 0.10 mm off the land centerline. A 0.05 mm land dish removes 0.05 mm of compression on the thin side. A 0.08 mm pitch-diameter drift tilts the cap so the compression is uneven by another 0.04 mm. Add them and the thin side has lost 0.19 mm of its 0.25 to 0.30 mm compression budget. The septum is now barely touching. Any temperature swing or solvent swell finishes the job.
The fix is a functional check on the assembled vial: a leak test or a septum-compression measurement on a sample from every lot. A helium leak test at 1 × 10⁻⁶ mbar·L/s on a 30-vial sample catches stack failures that dimensional inspection cannot. It costs more per lot, but it is cheaper than a recall. For suppliers who cannot run it in-house, a third-party analytical testing lab can run the leak and compression checks on retained samples.
There is a cheaper functional gate if helium is out of reach: a dye ingress test at 1 bar for 10 minutes on a 30-vial sample. It catches gross stack failures, not the marginal ones, but it runs for a few dollars a lot and it will flag the 2% leak case above before the customer does.
What to put in a purchase specification so tolerance capability is enforceable
Most vial purchase specs list dimensions and stop. That leaves capability unenforceable. Here is what I put in ours.
- Critical dimensions with tolerance AND minimum Cpk: bore 6.20 ± 0.10 mm at Cpk ≥ 1.33; concentricity ≤ 0.08 mm at Cpk ≥ 1.33; land flatness ≤ 0.05 mm at Cpk ≥ 1.33
- Per-cavity reporting: capability data broken out by cavity, not aggregated
- Gauge R&R requirement: under 10% of tolerance, method stated
- Functional gate: leak test on every lot, acceptance at ≤ 0.5% failures on a 30-vial sample
- Change control: any tool repair, cavity re-qualification, or material lot change triggers a fresh capability run before shipment
- Retained samples: held for shelf life, traceable to lot and cavity
If you are sourcing molded components alongside glass, the OEM side of the network at oemsamplevials.com carries the same capability language, because the tolerance logic does not change with the material.
Where the standards leave tolerance to you
No chromatography standard sets a bore tolerance or a Cpk floor. U.S. EPA SW-846 treats the container as a method parameter, so it tells you the vial must not contaminate the sample but not how round the bore must be. USP <1058> names the container loosely. That means the tolerance capability is entirely a commercial specification you write, and if you do not write it, the supplier sets it for you at whatever their tool happens to hold.
The instrument makers publish their own consumable specs, and those are a useful reference floor. Agilent’s GC columns and accessories documentation and Thermo Fisher’s chromatography and sample handling pages both list dimensional requirements for the vials their autosamplers accept. Restek’s technical resources go further and publish septa compression and bleed data that let you back-calculate the land flatness your method can tolerate. Use those as the starting point, then add the Cpk floor and the functional leak gate on top.
A practical note on the GPI reference: the 9 mm thread standard defines pitch and flank angle but not the bore or the land. That is why two suppliers can both claim “GPI 9 mm compliant” and ship vials that fail on the same autosampler. The standard is a compatibility floor, not a quality ceiling. Your spec has to add the features the standard leaves out.
The decision rule for tolerance capability
Do not buy on a tolerance print. Buy on a capability report with per-cavity data, a gauge R&R under 10% of tolerance, and a functional leak gate on every lot. If a supplier cannot show you Cpk on the bore, concentricity, and land flatness broken out by cavity, their “within tolerance” claim is a promise about a single part, not about the lot you are about to run. The dimensions that matter are the ones that define relationships between features, and those are the ones a go/no-go gauge will never catch.
References & further reading
Related resources from our lab consumables network:
- Chromatography Supplier at chromatographysupplier.com
- Oem Vial at oemsamplevials.com
- Vial Closures at vialclosures.com
- Analytical Testing at labanalyticaltesting.com
External references and standards: