What “tolerance capability” means when the drawing says ±0.10 mm

A customer sent back 12,000 syringe filters last March because their autosampler kept throwing “vial not found” errors on a batch of 2 mL vials. The filter bodies were fine. The problem was the outlet spigot: we had quoted 4.05 ± 0.10 mm and shipped 4.02 to 4.18 mm off a mold that had been running for 340,000 shots. The spigot still fit the vial, but the extra 80 µm of interference changed the insertion force enough that the autosampler’s gripper slipped. That is what dimensional tolerance capability actually means in HPLC filter production. It is not a number on a drawing. It is a measured, ongoing property of a specific mold, resin, and process window, and it drifts in ways that a first-article inspection will never catch.

Most HPLC filter drawings carry tolerances that look generous until you map them against what the process can actually hold. A ±0.10 mm callout on a 4 mm spigot is 2.5% of nominal. On a 13 mm housing diameter it is 0.77%. Same number, very different difficulty. The question a procurement manager should ask is not “can you hold ±0.10 mm” but “what is your Cpk on that dimension, measured on production parts, over the last 90 days.”

Tolerance capability is the ratio of the specification width to six times the process standard deviation. A process running at Cpk 1.33 has about 63 parts per million outside the spec limits on a normal distribution. At Cpk 1.67 it is about 0.6 ppm. At Cpk 1.0 you are shipping roughly 2,700 ppm, which on a 500,000-unit lot is 1,350 nonconforming filters. For a consumable that goes into a validated method, that is a lot of 483 exposure.

I have seen suppliers quote ±0.10 mm and deliver Cpk 0.9 because they measured 30 parts from one cavity on one shift. The capability number you want is per-cavity, over at least 30 consecutive shots, with the mold at thermal steady state. A 16-cavity tool has 16 separate capability distributions. The worst cavity sets the lot.

There is a second trap that catches even careful buyers: the difference between capability and performance. Cpk describes a process in statistical control. Ppk describes what actually shipped, including the drift between studies. A supplier can show you a beautiful Cpk of 1.5 from a study run in February and ship you a lot in August with a Ppk of 0.8, because the tool wore and nobody re-ran the study. Ask for both, and ask for the date on each. A capability report older than six months on a tool past 100,000 shots is a historical document, not a control plan.

How to read a capability report

• Sample size: 30 parts minimum per cavity, 50 preferred. Anything under 30 is a guess.
• Conditions: steady-state mold temperature, documented. Cold-mold data is fiction.
• Gage R&R: the measurement system must contribute under 10% of the tolerance. A 0.10 mm tolerance measured with calipers that repeat to ±0.03 mm is not a valid capability study.
• Time span: capability drifts with tool wear. A study from shot 1,000 does not describe shot 300,000.
• Reported statistic: Cpk, not Cp. Cp ignores centering, and centering is exactly what drifts.
• Date and shot count: a report without the tool’s shot count at the time of study is not auditable.

Which process holds which tolerance, and where the line actually sits

Not every dimension on an HPLC filter comes off the same process. A typical two-piece housing is injection molded, the membrane is die-cut or laser-cut, and the assembly is either ultrasonic welded or press-fit. Each step has its own capability floor, and the stacked tolerance is what the customer feels.

Those numbers are what I would sign off on after watching the process. A supplier claiming ±0.03 mm on a 16-cavity housing OD is either running a single-cavity tool, sorting 100%, or measuring the wrong thing. Ask which.

The membrane seat depth is the one that catches people. It is a mold feature, so it looks easy, but it sits between two cooling surfaces of different mass. Uneven cooling warps the seat, and the warp does not show up in a go/no-go check. It shows up as a leak at 4 bar. A PTFE membrane filter with a 0.2 µm pore and a seat that is 40 µm too deep will bypass at the edge before the membrane ever sees the pressure. If you are specifying a PTFE membrane filter, get the seat depth capability number in writing, not just the housing OD.

One more process note that matters for anyone comparing quotes. Die-cut membranes and laser-cut membranes are not interchangeable on tolerance, and the gap is bigger than the table suggests once you account for burr. A die-cut disc at ±0.15 mm typically carries a burr of 20 to 50 µm on the cut edge. That burr sits against the housing wall and creates a path for edge bypass. A laser-cut disc at ±0.05 mm has a heat-affected zone of 10 to 30 µm, which is cleaner but can char the polymer edge and shed particles. Neither is free. The right choice depends on whether your failure mode is bypass or particulate. For most 0.45 µm applications, die-cut with a deburr step wins on cost; for 0.2 µm sterile filtration, laser-cut wins on edge integrity.

The dimensions that fight back: spigot ID, seat depth, and concentricity

Three dimensions cause most of the field failures I have traced back to molding, and none of them are the ones customers ask about.

Spigot inner diameter

The spigot ID sets the dead volume and the back pressure. It is formed by a core pin, and core pins wear. A 1.50 mm core pin running PBT at 260 °C will lose 5 to 10 µm over 200,000 shots from abrasive filler and thermal cycling. The ID grows, dead volume grows, and peak shape drifts. Nobody notices until a method transfer fails on resolution. The capability number on a new tool is meaningless without a core pin replacement schedule. I specify replacement at 150,000 shots or when the ID drifts 15 µm from nominal, whichever comes first.

Do the arithmetic on what 10 µm of ID growth does. On a 1.50 mm bore, a 10 µm increase is a 1.3% increase in cross-sectional area, which sounds trivial. But dead volume scales with that area times the spigot length. On a 6 mm spigot, the added volume is about 0.07 µL. That is small, and on a 10 µL injection it is noise. On a 1 µL injection into a 2.1 mm column, it is 7% of the injection volume sitting in the wrong place, and that shows up as peak tailing on early-eluting compounds. The dimension is not a cosmetic callout. It is a chromatographic variable.

Membrane seat depth

Covered above, but the practical control is to measure it with a non-contact optical comparator, not a pin gauge. A pin gauge reads the high spot. The seat is a recessed annular surface, and the relevant number is the depth at four points 90° apart. If the four-point spread exceeds 20 µm, the part will leak regardless of the mean.

I also want to see the seat flatness, not just the depth. A seat that is 0.500 mm deep on average but dished by 30 µm across the annulus will hold the membrane on the rim and leave a gap in the center. The membrane deflects under pressure and the center gap becomes a leak path. Flatness on a 9 mm seat annulus should be under 15 µm. That is a lapping or a well-cooled insert job, and it is the feature most likely to be missing from a supplier’s capability report because most suppliers never measure it.

Concentricity of spigot to housing

This is the dimension that kills autosampler compatibility. The spigot must be concentric to the housing OD within about 0.05 mm total indicator reading, or the filter cocks in the vial and the needle path shifts. On a two-plate mold, concentricity depends on core-to-cavity alignment, which depends on guide pin wear. A tool at 400,000 shots with worn guide pins will hold diameter but lose concentricity. Diameter passes, function fails.

Concentricity is also the dimension most likely to be measured wrong. A CMM that touches four points and fits a circle will report a number that looks fine while missing a bow. The bow is what causes the cocking. Use a vision system that captures the full profile at two heights, or a dedicated concentricity fixture with a 0.001 mm indicator. If the supplier reports concentricity from a caliper, they are not measuring concentricity.

Two field cases where the measurement found the drift

Case 1: the lot that passed AQL and failed the autosampler

A pharma customer running a 96-well autosampler reported intermittent needle crashes on one lot of 0.45 µm PVDF filters. AQL sampling passed. We pulled 200 retained parts and measured spigot-to-housing concentricity on a vision system. The mean was 0.03 mm TIR, well inside spec. But the distribution was bimodal: one cavity was running 0.08 mm TIR. The mold had a bent ejector pin on cavity 11 that was kicking the part sideways at ejection, and the resulting stress set a slight bow. The fix was a $340 ejector pin and a re-qualification of cavity 11. The lesson: measure concentricity per cavity, and look at the distribution, not the mean.

The measurement that found it was a histogram, not a pass/fail count. The AQL plan sampled 80 pieces across the lot and found zero defects because the failing cavity was 1 of 16 and the sampler hit it fewer than five times. A per-cavity histogram of 200 retained parts put 12 pieces from cavity 11 outside 0.06 mm TIR. That is a 6% defect rate on one cavity and a 0.4% defect rate on the lot, which is exactly the rate that slips past a general inspection level II plan. If you are not stratifying by cavity, you are not measuring capability.

Case 2: the seat depth that drifted with ambient humidity

A supplier running nylon 6/6 housings shipped a lot in July that leaked at 3 bar on a bubble point test. The seat depth measured 0.02 mm shallow on incoming inspection, inside the ±0.05 mm callout. But nylon absorbs moisture, and the parts had been molded in a dry spell and shipped in a humid week. The dimensional change from moisture absorption was enough to change the weld collapse. The fix was twofold: switch to a resin with lower moisture uptake (PBT or PP depending on the solvent), and add a 24-hour conditioning step before final dimensional inspection. If your filter sees aqueous mobile phases, moisture-conditioned dimensions are the only ones that matter. This is also why the vial you pair it with matters: a vial insert with its own tolerance stack can compound the problem.

The numbers on that case are worth recording. Nylon 6/6 at 50% relative humidity holds about 2.5% moisture by weight. At 80% RH it holds about 4%. The linear dimensional change between those two states is roughly 0.3%, which on a 13 mm housing is 39 µm. That is most of a ±0.05 mm tolerance band consumed by the weather. A supplier who inspects dry and ships humid is shipping out of tolerance without ever changing the process.

How to audit a supplier’s tolerance claim without a plant visit

You can get most of the way there with documents and a small sample. Here is the checklist I use.

• Request the per-cavity capability report for the three critical dimensions: spigot ID, seat depth, concentricity. Reject any report with fewer than 30 parts per cavity.
• Ask for the gage R&R on each dimension. If the measurement system consumes more than 10% of the tolerance, the capability number is noise.
• Get the core pin and guide pin replacement schedule in writing. A supplier with no wear schedule has no capability control, only incoming luck.
• Request 50 production parts from a lot at least 100,000 shots into the tool. Measure them yourself on a vision system. Compare the distribution to the report.
• Ask what happens at the tolerance limit. Is the part sorted, scrapped, or shipped with a deviation? The answer tells you how the supplier thinks.
• Confirm the resin grade and lot on the certificate of analysis. A resin change without a re-qualification is a capability reset.
• Ask for the last three deviation reports. A supplier with zero deviations in three years is either perfect or not documenting. The second is more likely.

The 50-part check is the one that catches the most problems. I have run it on a dozen suppliers and found distribution shifts on four of them. None were lying. Their tools had simply worn since the last study. The cost of the check is a few hundred dollars in measurement time. The cost of skipping it is a recall.

One practical note on the 50-part sample: ask the supplier to pull from the end of a production run, not from retained first-article stock. First-article parts are the best parts the tool ever made. End-of-run parts are the ones that ship. If the supplier cannot tell you which shot number the sample came from, the sample is not useful.

For reference on how the major instrument vendors specify their own consumables, the Agilent chromatography consumables documentation and the Thermo Fisher sample handling specifications are worth reading as benchmarks for what a defensible tolerance callout looks like. The Restek technical resources also publish dimensional data on their filter and vial lines that you can compare against.

The tolerance stack that matters is the one the autosampler sees

Individual dimensions can all be in spec while the assembly fails. The autosampler cares about three things: the filter’s overall length, the spigot OD at the point of insertion, and the concentricity of the two. If each is at its limit in the wrong direction, the stack exceeds the gripper’s compliance and the needle misses.

Run the stack on a typical 13 mm filter. Overall length 32.0 ± 0.20 mm. Spigot OD 5.85 ± 0.05 mm. Concentricity 0.05 mm TIR. The vial neck ID is 6.00 ± 0.10 mm and the gripper allows 0.15 mm lateral. Worst case: filter length at +0.20, spigot at +0.05, concentricity at 0.05, vial neck at -0.10. The filter sits 0.20 mm too high, the spigot is 0.05 mm too wide on one side, and the eccentricity adds another 0.05. The total lateral misalignment at the needle entry is 0.10 mm plus the length error. If the gripper compliance is 0.15 mm, you are at 67% of the budget before any other variation. That is why the stack matters more than any single callout.

I recommend a functional gauge rather than a dimensional inspection for final release. A go/no-go gauge that mimics the autosampler’s gripper and vial seat, built to the vial drawing’s tolerance, will catch stack-up failures that a CMM will miss. Build the gauge to the SW-846 method’s vial specification if you are running environmental methods, or to the instrument vendor’s drawing if you are running pharma. The gauge is cheap. The recall is not.

For the housing itself, the HPLC filter product line should be specified with a functional gauge dimension, not just a drawing dimension. And if you are pairing the filter with a crimp or screw-top vial, check the COD vial neck tolerances against the filter spigot before you commit to a lot.

A decision rule for specifying tolerance on a new filter

Start from the function, not the drawing. The function is: the filter seats in the vial, the needle passes through the septum into the filter inlet, and the outlet seals against the vial bottom without dead volume. Work backward from the instrument’s gripper compliance and the vial’s neck ID tolerance.

If the vial neck ID is 6.00 ± 0.10 mm and the gripper allows 0.15 mm of lateral misalignment, the filter spigot OD should be 5.85 ± 0.05 mm to guarantee insertion. That is a tighter callout than most drawings carry, and it is the one that matters. Specify it, then require Cpk 1.33 per cavity on that dimension, then audit with the 50-part check. Everything else on the drawing is secondary.

Two more numbers to put in the purchase specification. First, a torque or weld energy window for the assembly step, because the assembled length depends on it: ultrasonic weld energy at 40 to 60 J with a hold time of 0.2 to 0.4 s, or a press-fit force of 80 to 120 N. Second, a particle count limit on the finished part, because a dimension that is in tolerance but sheds is still a failure: under 100 particles ≥ 10 µm per filter in a 10 mL rinse, measured by light obscuration. The dimension and the cleanliness are both part of the specification, and a supplier who only quotes one is quoting half a product.

The decision rule: specify tolerance on the two or three dimensions that touch the instrument, require per-cavity Cpk 1.33 with a documented gage R&R under 10%, and release on a functional gauge rather than a CMM. If a supplier cannot produce the per-cavity capability report and the wear schedule, the tolerance number on their datasheet is a hope, not a capability. Buy from someone who can show you the distribution.

References & further reading

Related resources from our lab consumables network:

• Hplc Filter at hplcfilters.com
• Ptfe Membrane Filter at ptfemembranefilter.com
• Cod Vial at codvial.com
• Vial Insert at hplcinsert.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