A 40,000-piece lot of 13 mm butyl stoppers failed incoming inspection because the flange thickness ran 0.062 in. at the mean with a standard deviation of 0.0031 in. against a 0.060 ± 0.005 in. print. The mean was in tolerance. The capability index was Cpk = 0.43. Two out of every hundred stoppers would have sat proud in the crimp and lost seal on the capper. Nobody on the buyer’s side had asked for a capability number, only a conformance certificate, and the molder had never calculated one. This is the gap I run into most often when a purchasing team starts asking about vial closures dimensional tolerance capability: they get a pass/fail sheet when what they need is a distribution.
Tolerance is a limit. Capability is what the process does relative to that limit, run after run, cavity after cavity, lot after lot. The two are not the same conversation, and the closure industry is very good at having the first one while quietly avoiding the second. Below I’ll lay out what each molding process can actually hold, how to measure it without fooling yourself, and which dimensions on a vial closure are the ones that will bite you.
If you are specifying a closure and you have not matched the tolerance callout to the process, you have already lost. A ±0.002 in. callout on a compression-molded stopper is a request the molder cannot meet at volume, and a ±0.010 in. callout on an injection-molded cap is money you are leaving on the table. Here is the working envelope I use when I quote or audit a job.
| Process | Typical achievable tolerance (critical dim.) | Best-in-class with SPC | Where it struggles |
|---|---|---|---|
| Compression molding (elastomer stoppers) | ±0.005 in. (±0.13 mm) | ±0.003 in. (±0.08 mm) | Flange thickness, crown radius, flash at parting line |
| Transfer molding (elastomer) | ±0.004 in. (±0.10 mm) | ±0.0025 in. (±0.06 mm) | Coring resistance (compound-driven, not dimensional), gate witness |
| Injection molding (thermoplastic cap) | ±0.003 in. (±0.08 mm) | ±0.0015 in. (±0.04 mm) | Concentricity of thread to bore, warp on thin skirts |
| Injection molding (TPE stopper) | ±0.004 in. (±0.10 mm) | ±0.002 in. (±0.05 mm) | Shrinkage variation across cavities, hardness drift |
Two things to read from that table. First, the elastomer processes are looser than thermoplastic by roughly a factor of two, and that is physics, not laziness. Elastomer shrinks 1.5 to 3.5% depending on compound and cure, and the shrink is not uniform across a part with a thick flange and a thin diaphragm. Second, the “best-in-class” column only exists when the molder is running statistical process control on the dimension, not just final inspection. A molder who measures 32 pieces at end of run and calls it good will never hit that column.
For context on how tight the downstream hardware expects things to be, look at what the instrument makers publish for their own consumables. Agilent’s chromatography consumables and Thermo Fisher’s sample handling line both assume a crimp seal that seats within a few thousandths. The closure does not get to be the loose link.
Capability is Cpk on the right dimension, not a COA with a checkmark
Here is the trap. A certificate of analysis that says “dimensions conform” tells you the sample passed. It tells you nothing about the population. To get vial closures dimensional tolerance capability you need three numbers per critical dimension: the process mean, the standard deviation, and the specification width. From those you compute Cpk:
Cpk = min[(USL − mean) / 3σ, (mean − LSL) / 3σ]
A Cpk of 1.33 means the process is centered and the nearest spec limit sits four sigma away. That is the floor I will accept on a critical closure dimension. A Cpk of 1.67 (five sigma) is what I want on a flange thickness or a bore diameter that controls seal. Anything below 1.0 means the process is generating defects by design and you are sorting good parts from bad at your own expense.
Why the sample size in most COAs cannot support a Cpk claim
If a molder measures 10 parts and reports a standard deviation, that σ has a confidence interval so wide it is worthless for capability. To estimate σ within ±20% at 95% confidence you need roughly 100 pieces, and they must be pulled across cavities and across the run, not from one cavity at the end. A 16-cavity tool sampled at 8 shots gives you 128 pieces and a defensible σ. A 10-piece grab sample gives you a number that will embarrass everyone when the next lot drifts.
Measure capability per cavity, then pool
This is where a lot of molders get caught. They pool all cavities into one distribution, and the pooled σ looks acceptable. But if cavity 7 runs 0.004 in. high on flange thickness and cavity 12 runs 0.003 in. low, the pooled mean hides two cavities that are individually out of capability. I ask for a per-cavity X-bar and R chart over at least 25 subgroups. When I get it, cavity-to-cavity spread on a 16-cavity compression tool routinely shows a 0.006 in. range on flange thickness even when the pooled Cpk reads 1.4. That spread is a tooling problem (uneven cavity fill, worn cavity inserts, thermal gradient across the platen), and it will not show up in a pooled capability number.
The three dimensions that refuse to stay in tolerance
Not all dimensions on a vial closure are equally hard. Three of them account for most of the capability failures I have investigated, and each fails for a different reason.
Flange thickness on an elastomer stopper
The flange is the sealing surface and the crimp seat at the same time. It is also the thickest section on the part, which means it cures last and shrinks most. On a 13 mm butyl stopper with a nominal 0.060 in. flange, I have seen within-lot σ of 0.003 in. when the press temperature drifted 4 °F across the platen. The fix is not a tighter callout; it is platen temperature uniformity to ±2 °F and a cure-time window held to ±3%. When a molder tells me they cannot hold ±0.005 in. on flange thickness, I ask for their platen map. Nine times out of ten there is no map.
Bore diameter and concentricity on a cap
On an injection-molded cap, the bore that receives the vial neck and the thread that engages it must be concentric within about 0.003 in. TIR, or the cap cocks on the neck and the seal loads unevenly. Concentricity is the hard one because it depends on core-to-cavity alignment, and alignment walks as the tool heats. A cap that gauges 0.002 in. TIR cold can run 0.005 in. TIR after 500 shots. If your supplier inspects first-shot parts, they are inspecting the best parts the tool will ever make.
Diaphragm thickness on a septum
A 0.010 in. PTFE-faced silicone septum with a ±0.001 in. callout on the silicone layer is a genuine capability challenge, because the PTFE lamination and the silicone cure are two processes that have to agree. The failure mode is not a leak at time zero; it is a puncture-count shortfall. A septum that should survive 100 injections on a chromatography vial starts coring at 60 because the silicone is 0.0015 in. thin in the puncture zone. That is a dimensional capability problem wearing a performance costume.
How to audit a molder’s capability without a two-week site visit
You do not need to live in the plant. You need four documents and one phone call. Here is the checklist I send to any new closure supplier before I release a purchase order.
- Per-cavity capability study on the three critical dimensions (flange thickness or bore, concentricity, diaphragm thickness), minimum 25 subgroups of 5, Cpk reported per cavity. Reject pooled-only reports.
- Gauge R&R under 10% of tolerance on each dimension. If the measurement system eats more than a tenth of the tolerance, the capability number is measuring the gauge, not the part.
- Platen or mold temperature map with the uniformity figure stated in °F. Ask for ±2 °F on elastomer, ±5 °F on thermoplastic.
- Resin or compound lot traceability tied to the capability study. A study run on one compound lot does not describe the next lot.
- Change-control notification clause covering tool refurbishment, cavity insert replacement, and compound source. A silent tool rebuild resets your capability to zero.
That last item is the one buyers forget. I have watched a supplier swap a worn cavity insert, hold the nominal dimension, and still blow the capability because the new insert changed the shrink behavior by 0.002 in. The dimension was in tolerance; the distribution moved. Without a notification clause you find out at incoming inspection, which is the most expensive place to find out.
Two field cases where the number told the story
Case 1: the stopper that passed the print and failed the capper
Symptom: a fill line running 11 mm lyophilization stoppers reported intermittent seal failures at the crimp station, roughly 1 in 300 vials, always on the same capper head.
Measurement: we pulled 200 stoppers from three lots and measured flange thickness on a contact gauge with 0.0001 in. resolution. The mean was 0.058 in., dead center of a 0.055 to 0.061 in. print. But the distribution was bimodal: one mode at 0.056, one at 0.060, split roughly 70/30. The 0.060 mode came from two cavities on the 16-cavity tool. Per-cavity Cpk on those two cavities was 0.71.
Fix: the molder re-shimmed the two cavity inserts and re-ran the capability study. New per-cavity Cpk on the worst cavity was 1.41. The capper head was fine the whole time; the stopper was presenting a taller crimp seat on 30% of parts, and the crimp die could not take up the difference. The lesson is that a centered mean with a bimodal distribution is the most dangerous capability report there is, because the pooled number looks beautiful.
Case 2: the septum that cored early
Symptom: a contract lab running EPA Method 8260 on HPLC and GC vials saw coring debris in the injector at 60 to 70 punctures, against a supplier claim of 100. The septa were 0.010 in. PTFE-faced silicone, 11 mm.
Measurement: we sectioned 30 septa and measured silicone thickness at the puncture zone with a digital micrometer (0.0001 in.). Nominal was 0.008 in.; the lot ran 0.0065 in. mean with a 0.0008 in. σ. Cpk against a 0.007 in. lower limit was 0.21. The PTFE facing was also 0.0005 in. thinner than the prior lot, which reduced the reinforcement that spreads puncture load.
Fix: the molder’s lamination line had drifted on calender gap. They brought silicone thickness back to 0.008 in. mean, Cpk 1.38, and puncture count returned to 100+ on the same needle and the same 10 µL injection volume. If you are running volatile methods and your coring starts early, measure the septum thickness before you blame the needle. The Restek technical library has good background on how septum quality interacts with inlet maintenance, but the root cause here was dimensional.
The spec sheet that actually controls capability
Most closure drawings I see specify a nominal and a symmetric tolerance and stop there. That drawing does not control capability, because it does not tell the molder which dimensions are critical, how they will be measured, or what distribution is acceptable. Here is the structure I put on our own drawings and ask for from suppliers.
- Critical dimensions flagged with a capability requirement, not just a tolerance. Example: “Flange thickness 0.060 ± 0.005 in., Cpk ≥ 1.33 per cavity, gauge R&R ≤ 10%.”
- Measurement method named. Contact gauge vs. optical comparator vs. CMM gives different σ. The method is part of the spec.
- Sampling plan tied to the capability study, not to a generic AQL. If the process is at Cpk 1.67 you can sample lightly; at Cpk 1.0 you cannot.
- Dimensional stability over time. Elastomer closures relax after cure. A flange measured at 0 hours can move 0.001 to 0.002 in. by 72 hours. Specify the measurement window.
- GPI and ISO references where they apply. Thread finish on caps should reference the GPI 400 series for the neck finish it engages, and elastomer closures should be dimensioned against the ISO 8362-2 vial finish so the closure and the vial are speaking the same language.
That ISO reference matters more than people think. When a stopper is dimensioned against a drawing that was never tied to the vial finish standard, you get a closure that is perfectly capable against its own print and perfectly wrong against the vial it is supposed to seal. I have seen a 0.004 in. plug diameter shortfall that was invisible on the stopper drawing and fatal on the fill line, because the vial bore sat at the top of its own tolerance band. Capability against the wrong datum is not capability at all.
For buyers running regulated methods, the container closure system also has to survive the way the method treats it. EPA SW-846 and the associated preparation and analysis protocols put the closure through thermal and chemical cycles that will expose a marginal dimension faster than any bench gauge. If you want to see how a closure behaves under that kind of stress before you commit to a lot, the analytical testing resources at labanalyticaltesting.com are a reasonable starting point for the test matrix. And when you are ready to specify the closure itself against a real capability target rather than a generic conformance sheet, the product and specification pages at vialclosures.com lay out the dimensions and the process envelope together.
Decision rule
Do not accept a closure lot on a conformance certificate. Accept it on a per-cavity capability study with Cpk ≥ 1.33 on the flange or bore, Cpk ≥ 1.67 on the diaphragm thickness, gauge R&R under 10%, and a change-control clause that forces notification on any tool or compound change. If the supplier cannot produce those four things, the tolerance on the drawing is decoration and the capability is whatever the last lot happened to do.
References & further reading
Related resources from our lab consumables network:
- Vial Closures at vialclosures.com
- Analytical Testing at labanalyticaltesting.com
- Chromatography Vial at chromvials.com
- Hplc Vial at hplcvialfactory.com
External references and standards: