Why a drawing tolerance and a process capability are not the same number

A customer sent back a lot of 500 vials last March because their autosampler kept throwing “vial not found” errors on roughly 3% of the tray. The vials met every dimension on the drawing. What they did not meet was the dimensional tolerance capability the process needed: the crimp shoulder sat at 12.41 mm instead of the nominal 12.35 mm, and the 0.06 mm drift was enough to lift the vial out of the gripper’s reach on one out of every thirty-three positions. The drawing said ±0.10 mm. The process could only hold ±0.06 mm on that feature with any reliability. Nobody had calculated the difference between the two until the complaint arrived.

This is the gap I want to write about. Not whether a vial is “in spec,” because in-spec is a binary that hides almost everything useful. The question is whether the process that made it can hold a tolerance repeatably, how you measure that capability, and which dimensions on an HPLC vial are genuinely hard to control versus which ones are easy to hit and easy to inspect. The primary intent here is hplc vial dimensional tolerance capability: what a given molding or forming process can actually achieve, and how a buyer should read a capability number rather than a nominal value.

Every vial drawing I have seen from a reputable supplier lists nominal dimensions with symmetric tolerances. The 12 x 32 mm format, defined in the industry by the Agilent and Thermo Fisher autosampler ecosystems, typically specifies:

• Body outside diameter: 12.00 mm ± 0.20 mm (loose, because the gripper has clearance)
• Body height: 32.00 mm ± 0.30 mm (loose, because the tray pocket is deep)
• Crimp or screw shoulder height: ± 0.10 mm (tight, because the gripper and the cap engage here)
• Neck inner diameter: 6.00 mm ± 0.10 mm (tight, because the septum and the needle path depend on it)
• Bottom thickness: 1.00 mm ± 0.10 mm (moderate, because the needle must not bottom out but must also reach the sample)
• Concentricity of the neck to the body: ≤ 0.15 mm TIR (tight, because a cocked vial in the gripper is a coring risk)

Those tolerances are what the drawing permits. Capability is what the process delivers. A glass forming line or a plastic injection molding cell might hit the ±0.30 mm on height with a Cpk of 2.0 and struggle to hold ±0.10 mm on the shoulder at a Cpk of 0.9. Same drawing, same supplier, two very different realities. If you buy on the drawing alone, you are buying the tolerance, not the capability, and you will find out which one you got on the autosampler.

How capability is actually measured on a vial lot

Capability is a statistical statement about a process, not a statement about a single part. The standard framework is Cp and Cpk, and the math is straightforward if you have the data. Take 30 to 50 vials from a production run, measure the critical dimension with a tool that resolves to at least one-tenth of the tolerance band, and compute:

• Cp = (USL − LSL) / (6σ), where USL and LSL are the upper and lower specification limits and σ is the process standard deviation
• Cpk = min[(USL − x̄) / 3σ, (x̄ − LSL) / 3σ], which penalizes a process that is centered off the nominal

A Cpk of 1.33 means the process is 4σ from the nearest spec limit and will produce roughly 63 defects per million if it stays centered. A Cpk of 1.67 is 5σ, about 0.6 defects per million. Below 1.0, the process is producing out-of-spec parts on a routine basis and you are relying on inspection to catch them, which is a losing strategy at vial volumes.

Here is the part most buyers miss. The σ that matters is not the within-lot σ. It is the between-lot σ. A molder can hold ±0.02 mm inside a single 10,000-piece run and drift ±0.08 mm from lot to lot as the mold temperature controller cycles, the glass gob weight shifts, or the annealing oven profile changes. A within-lot Cpk of 1.8 with a between-lot Cpk of 0.7 is a supplier you will fight with every quarter. Ask for the control chart, not the histogram. The histogram tells you one lot. The control chart tells you whether the process is in statistical control across lots, which is the only thing that predicts what arrives next month.

The measurement system is half the capability number

I have watched a supplier report a Cpk of 1.5 on shoulder height using a caliper with 0.02 mm resolution and a 0.05 mm repeatability. That measurement system contributes more variance than the process it is trying to characterize. Before you accept any capability number, ask what gauge was used and whether a Gage R&R was run. For a ±0.10 mm tolerance, the gauge needs to resolve to 0.01 mm (a digital micrometer or an optical comparator, not a caliper) and the R&R should be under 10% of the tolerance band. A capability study built on a bad gauge is a number with a decimal point and no meaning.

Which dimensions are hard to control, and why

Not all features on a vial are equally difficult. The easy ones are the ones the forming process naturally controls: overall height on a glass vial is set by the mold cavity and the gob weight, and a well-run line holds ±0.15 mm without breaking a sweat. The hard ones are the ones where two process variables interact, or where the feature is a secondary operation rather than a primary form.

Shoulder height and concentricity are the two that bite hardest. Shoulder height because it is a secondary forming step: the neck is drawn after the body, and the draw length depends on glass temperature, which depends on the annealing oven, which drifts. Concentricity because it requires the neck axis and the body axis to align within 0.15 mm total indicated runout, and any misalignment in the mold halves or the blow station shows up here. A vial with 0.25 mm TIR will sit cocked in the gripper, and a cocked vial is a coring event waiting to happen.

What injection molding can hold versus what glass forming can hold

The process determines the capability ceiling. This is the single most useful thing to know before you write a spec, because it tells you what to ask for and what to reject as physically impossible.

Injection-molded plastic vials (polypropylene, COC, COP) can hold tighter tolerances than glass on most features because the mold cavity is machined to a known dimension and the cycle is repeatable. A well-run PP vial mold holds ±0.05 mm on neck ID and ±0.05 mm on shoulder height at Cpk 1.5 or better, provided the mold temperature is controlled to ±2 °C and the resin lot is consistent. The failure mode is shrinkage variation: PP shrinks 1.5 to 2.5% from mold to room temperature, and the exact number depends on mold temperature, packing pressure, and cooling time. Change any of those and the part moves. A molder who does not control resin moisture and mold temperature will show you a beautiful first article and a drifting third lot.

Glass forming (tubing conversion or molded glass) holds looser tolerances on the secondary features because the material is formed hot and shrinks on cooling in a way that depends on the thermal profile. Tubing-converted vials, where a glass tube is cut, heated, and formed into a neck and bottom, hold ±0.10 mm on neck ID at Cpk 1.0 to 1.4 on a good line. Molded glass holds similar numbers on the body but struggles more on concentricity because the parison has to be centered in the blow mold. Neither process matches injection molding on the tight features, and a supplier who claims ±0.03 mm on a glass neck ID at Cpk 1.5 is either measuring wrong or talking about a different feature.

This matters for procurement because the spec you write should match the process you are buying. If you need ±0.05 mm on neck ID, you are buying an injection-molded vial, full stop. If you write ±0.05 mm on a glass vial drawing, you will get a supplier who signs the drawing and ships at ±0.10 mm, and you will spend the next two years arguing about it. The HPLC vial catalog should tell you which process made the part; if it does not, ask.

Two field cases where capability, not conformance, was the problem

Case 1: The shoulder that passed inspection and failed the gripper

A pharmaceutical QC lab running a Waters autosampler reported intermittent “vial present” sensor failures on a 12 x 32 mm screw vial lot. Incoming inspection passed: 100% of sampled vials were within ±0.10 mm on shoulder height. The lab pulled 50 vials from the failing tray and measured shoulder height with a 0.001 mm micrometer. The mean was 12.38 mm against a nominal of 12.35 mm, σ was 0.045 mm, and the upper spec was 12.45 mm. Cpk was 0.52. The process was centered 0.03 mm high and the σ was wide enough that roughly 6% of parts sat above 12.44 mm, close enough to the gripper’s engagement window to cause the sensor to miss. The fix was not a tighter drawing. It was a supplier conversation about centering the shoulder form and reducing the between-lot drift, plus a switch to a vial from the Aijiren line that published a Cpk of 1.4 on that feature. The lab now measures 30 vials per lot and plots the mean on a control chart.

Case 2: The neck ID that drifted with the annealing oven

A contract manufacturer of cannabinoid standards was seeing needle bending on about 1 in 200 injections. The needle was a 22-gauge (0.72 mm OD) autosampler needle, and the neck ID was specified at 6.00 mm ± 0.10 mm, so the needle had 5 mm of clearance. The bending made no sense until we measured neck ID on 40 vials from the suspect lot with a pin gauge set. The mean was 5.91 mm and the σ was 0.06 mm, putting the lower tail below 5.85 mm. Cpk was 0.7. The root cause was an annealing oven that had drifted 8 °C low, which changed the glass shrinkage and pulled the neck in. The vials were in spec on the drawing (5.90 mm is inside 6.00 ± 0.10), but the process was not capable, and the tail of the distribution was catching the needle. The fix was a re-set of the oven profile and a supplier agreement to report Cpk on neck ID per lot, with a minimum of 1.33.

A decision checklist for reading a capability claim

When a supplier sends you a capability report, run it through this list before you accept it. I have used this exact sequence for years and it catches most of the inflated numbers.

• Is the sample size at least 30 per lot and 5 lots? A 10-piece study is a snapshot, not a capability.
• Is the gauge resolution at least 10x finer than the tolerance? For ±0.10 mm, that means 0.01 mm resolution.
• Was a Gage R&R run? Under 10% of tolerance is acceptable; over 30% invalidates the study.
• Is the Cpk reported per feature, or as a single number? A single number is marketing. Per-feature is engineering.
• Is the process in statistical control? Ask for the control chart. Points outside the limits or a trend means the Cpk is not predictive.
• Is the between-lot σ reported separately from the within-lot σ? If not, the Cpk is optimistic.
• Does the report name the process? Injection molded and glass formed have different ceilings; the Cpk should be plausible for the process.
• Is the nominal centered in the tolerance band? A Cpk of 1.5 with the mean 0.04 mm off nominal on a ±0.10 mm tolerance is a process one drift event away from failure.

If a supplier cannot answer three of these, treat the capability claim as a nominal value with a decimal point. The chromatography consumables distributors who publish real capability data are the ones worth a long-term agreement; the ones who send a one-page certificate with a single Cpk and no gauge information are the ones you will be re-qualifying every year.

What to write into a purchase specification

Stop writing drawings that specify only nominal plus tolerance. Write a spec that specifies the feature, the tolerance, the required Cpk, the gauge, and the sampling plan. A workable clause looks like this:

“Neck inner diameter: 6.00 mm ± 0.10 mm. Process capability Cpk ≥ 1.33 on within-lot data and ≥ 1.00 on between-lot data, demonstrated on 5 consecutive lots of ≥ 30 pieces each, measured with a pin gauge or optical comparator resolving to 0.01 mm, with Gage R&R ≤ 10% of tolerance. Supplier to provide control chart and per-lot Cpk with each shipment.”

That clause does three things. It tells the supplier what process performance you expect, not just what part you will accept. It gives you a rejection basis that is statistical rather than a single bad part. And it forces the supplier to measure, which is the only way the capability number stays honest over time. Pair it with the format and closure requirements you already have (9 mm screw, 12 x 32 mm, septa rated to the puncture count your method needs) and you have a spec that survives contact with a real autosampler.

For reference standards, the relevant documents are U.S. EPA SW-846 for method-level sample handling requirements and the dimensional conventions published by the major autosampler vendors; Restek also publishes useful dimensional guidance for vial and autosampler compatibility. None of these documents specifies a Cpk, which is exactly why the buyer has to write it into the purchase order.

The decision rule

Buy on capability, not on conformance. A vial lot that passes a go/no-go gauge tells you nothing about the next lot; a supplier who can show a control chart with Cpk ≥ 1.33 on shoulder height, neck ID, and concentricity, measured with a gauge that resolves to 0.01 mm and validated by Gage R&R, is telling you something the drawing cannot. If the supplier cannot produce that chart, the tolerance on the drawing is a wish, and the autosampler will eventually tell you which one you actually bought. The lab vials you specify should carry a capability number in the catalog, not just a nominal dimension, and if they do not, that absence is the first data point in your evaluation.

References & further reading

Related resources from our lab consumables network:

• Hplc Vial at vialforhplc.com
• Chromatography Consumables at chromatographyconsumables.com
• Lab Vials at labvials.com
• Aijiren Brand at aijirenvials.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