Dimensional tolerance capability is a process statement, not a drawing

A customer rejected an entire lot of 47 mm PTFE membrane filters last March because the outside diameter measured 46.71 mm on 12 of 500 pieces. The print called for 47.00 ± 0.25 mm, so the parts were nominally in spec. The problem was that the same lot’s inner aperture ran 0.18 mm small, and when the customer stacked the filter into a 47 mm vacuum funnel with a 46.9 mm seat, the membrane cocked and bypassed. That rejection is the whole reason ptfe membrane filter dimensional tolerance capability is worth writing about properly: the number on the drawing is not the number that matters. The relationship between two dimensions, and the process that holds each one, is what actually determines whether the part seals.

Dimensional tolerance capability is a process statement, not a drawing statement

Most procurement conversations treat tolerance as a negotiation between the buyer’s print and the supplier’s price. That framing is backwards. A tolerance is a claim about what a manufacturing process can hold repeatedly, and PTFE membrane filters are made by at least four distinct processes stacked on top of each other, each with its own capability ceiling. If you ask for ±0.05 mm on a die-cut diameter, you are asking a punch press to do something it cannot do at production speed, and the supplier who says yes is either lying or planning to sort the lot by hand and charge you for it.

The four processes that set the dimensions on a finished PTFE membrane filter are:

• Paste extrusion of the PTFE tape, which sets the base web width and, indirectly, the caliper.
• Biaxial stretching and sintering, which sets pore structure and thickness but also introduces cross-web variation.
• Die cutting or laser cutting of the disc, which sets outside diameter, aperture, and edge quality.
• Ring insert molding or thermal bonding, when the filter is supplied in a molded housing or with a support ring, which sets concentricity and the final assembled stack height.

Each of these has a different natural tolerance band. A supplier who quotes a single blanket tolerance across all features is telling you they have not characterized their line. For a deeper look at how the base membrane is made, the PTFE membrane filter product pages on this site cover the material side; this article is about the numbers that come off the calipers and the CMM.

What each process can actually hold, feature by feature

Here is the table I would want on my desk before I approve any incoming inspection plan. These are capability bands I have verified on production lines running at commercial speed, meaning tens of thousands of discs per shift, not lab-scale hand cutting. A capable process holds a Cpk of 1.33 or better across these bands; a marginal one holds 1.0 to 1.33; anything below 1.0 will generate rejects at a rate you will feel in the receiving dock.

Read that table against the rejection I opened with. The customer’s print said 47.00 ± 0.25 mm. That is a ±0.25 mm band, which is well inside what a rotary punch holds at Cpk 1.33. So the OD was never the problem. The aperture at ±0.20 mm and the concentricity at 0.15 mm TIR were the problem, because the seat in the funnel was 46.9 mm and the membrane had to sit concentric to seal. The lot passed OD inspection and failed function. That is the classic failure mode when a drawing is written feature by feature without a functional stack-up analysis.

How to measure capability on a filter disc, and why most suppliers get it wrong

Capability indices are only as good as the measurement system behind them. PTFE is soft, compressible, and dimensionally unstable under load, which means a caliper squeezed onto a 0.5 mm thick disc reads a different number than a non-contact optical system. I have seen the same lot measure 47.02 mm on a micrometer with a 6.35 mm anvil and 46.94 mm on a vision system. Both are “correct” for their method. Only one matches how the part sits in the housing.

Fix the measurement method before you argue about the tolerance

The rules I enforce on incoming inspection:

• Contact force: no more than 0.5 N on a disc under 0.6 mm thick. A standard micrometer spindle will compress PTFE by 10 to 20 µm if you turn it to the click stop. Use a friction thimble or a non-contact optical comparator.
• Support: measure OD with the disc flat on a granite surface, not held in the air. PTFE sags under its own weight at 47 mm diameter by roughly 5 to 10 µm.
• Temperature: PTFE’s coefficient of linear expansion is around 135 × 10⁻⁶ /°C, roughly ten times that of stainless steel. A 47 mm disc measured at 18 °C in winter and 26 °C in summer moves about 0.05 mm in diameter. Condition samples at 23 ± 2 °C for 4 hours before measuring, per ISO 1.
• Sample size: 30 pieces minimum for a capability study, drawn across the full width of the web and across at least three production lots. A capability number from a single lot is a snapshot, not a capability.

Once the method is fixed, calculate Cpk properly. Take the mean and standard deviation from your 30-plus pieces, then compute the distance from the mean to the nearest spec limit divided by three sigma. A lot with a mean of 47.00 mm, a sigma of 0.04 mm, and limits at 46.75 and 47.25 mm gives a Cpk of about 2.08, which is excellent. A lot with the same mean but a sigma of 0.09 mm gives 0.93, which will reject at roughly 0.5% and cost you a sorting operation every shipment.

Watch the within-lot versus between-lot split

This is the part most suppliers do not report. A punch press can hold a very tight within-lot sigma because the tool does not change during a run, while the between-lot variation (tool wear, die sharpening intervals, incoming web thickness) can be three to four times larger. If a supplier quotes you a Cpk of 1.8, ask whether it is from a single run or pooled across lots. I have seen a line quote Cpk 1.7 on within-lot data and deliver 0.9 on the pooled data because the die was sharpened every 40,000 strokes and the first 500 pieces after sharpening ran 0.06 mm oversize.

The dimensions that are genuinely hard to control, and why

Some features are hard to hold because of the material, and some are hard because of the tooling. Knowing which is which tells you where to spend your inspection budget.

Concentricity between OD and aperture

This is the hardest feature on a plain disc, and it is the one that causes the most field failures. A rotary punch locates the web on the outside edge and punches the aperture in a second station. Any web wander of 0.05 mm between stations shows up as 0.10 mm of TIR. PTFE’s low coefficient of friction makes web tracking worse than for, say, a nylon membrane, because the web slips sideways on the guide rollers. The fix is a guided punch with a stripper plate that clamps the web before the aperture station, and a tension control on the unwind that holds web tension to ±2%. Suppliers who skip the stripper plate will quote you 0.20 mm TIR and call it “industry standard.” It is not; it is a tooling choice.

Thickness and its effect on everything downstream

Caliper is set by the extrusion and stretch line, and it varies across the web width by 5 to 10% on a well-run line. The center of the web is usually thicker than the edges. If your disc is cut from the full web width, you inherit that variation. A 0.5 mm nominal disc can measure 0.47 mm at one edge of the web and 0.54 mm at the other. That matters because thickness drives the assembled stack height, the dead volume in the housing, and the burst pressure. If you are running a filter into a molded housing with a 0.10 mm stack height tolerance, you need the disc thickness held to ±5% and the housing designed to absorb the rest with a compressible seat. For context on how these stack-ups interact with sample handling hardware, the 2 mL vial and vial insert dimensions follow the same logic: the tolerance that matters is the one at the sealing interface, not the one on the print.

Flatness and the curl problem

PTFE discs curl. The biaxial stretch process leaves residual orientation in the web, and when you punch a disc out, the edges relax differently from the center. A disc that measures 0.30 mm of flatness deviation on the bench will sit flat in a housing that clamps the edge, but it will bypass in a housing that seals on the face. This is a design interaction, not a manufacturing defect, and it is why I insist on a functional leak test rather than a flatness gauge for final release.

Field case: the 0.18 mm aperture shortfall

Symptom: a pharmaceutical customer running USP <643> total organic carbon blanks saw erratic recovery, with RSD jumping from 2.1% to 6.8% on the same method over three weeks.

Measurement that identified it: we pulled 30 filters from three lots and measured aperture with a vision system at 23 °C. The mean aperture was 0.18 mm below nominal, and the concentricity was 0.22 mm TIR, both outside the capable band. The punch tooling had been sharpened and the aperture station was not re-centered afterward. The membrane was also sitting 0.4 mm off-center in the funnel seat, which created a 0.3 mm bypass crescent.

Fix: re-centered the aperture station, added a stripper plate, and changed the release criterion from “OD in spec” to “OD, aperture, and concentricity all at Cpk ≥ 1.33 on a 30-piece sample per lot.” Recovery RSD returned to 2.3%. The lesson is that a single-feature inspection plan will pass a lot that fails in the field.

Field case: the molded ring that grew 0.12 mm

Symptom: a customer assembling PTFE filters into a 13 mm syringe housing reported that 8% of units leaked at the ring-to-housing interface during a 30 psi pressure hold.

Measurement that identified it: the molded ring OD measured 13.12 mm against a 13.00 ± 0.10 mm print, so it was in spec. But the ring was molded from PTFE resin with a 0.5% post-mold shrinkage allowance, and the mold had been cut assuming 0.3%. The ring was also oval by 0.09 mm because the gate was on one side and the pack pressure was low. The housing bore was 13.05 mm, so the oval ring sealed on the long axis and bypassed on the short axis.

Fix: re-cut the mold with the correct shrinkage allowance, added a second gate to balance pack, and moved the ring OD tolerance to ±0.06 mm with a roundness callout of 0.05 mm. Leak rate dropped to 0.4%. The general principle: for molded PTFE parts, shrinkage is not a constant, it depends on wall thickness and pack pressure, and a ±0.10 mm tolerance on a 13 mm ring is only achievable if the mold is validated against actual shrinkage, not a handbook value.

A decision checklist for specifying tolerance on a PTFE membrane filter

Use this before you send a print to a supplier, and again before you approve a first article.

• Identify the sealing interface first. Which dimension actually contacts the housing, the funnel, or the sample? That is the dimension that needs the tight tolerance. Everything else can be loosened to save cost.
• Match the tolerance to the process. Die-cut OD at ±0.15 mm is realistic. Die-cut OD at ±0.05 mm is not, unless you accept 100% inspection and a higher price. Use laser cutting if you genuinely need ±0.05 mm.
• Ask for Cpk, not just conformance. Require Cpk ≥ 1.33 on the critical features, reported from a 30-piece sample pooled across at least three lots, with the measurement method stated.
• Condition and measure at 23 ± 2 °C. Reference ISO 1 for the standard reference temperature. PTFE moves 0.05 mm per 8 °C at 47 mm diameter.
• Specify contact force for inspection. 0.5 N maximum on discs under 0.6 mm thick, or use non-contact optical measurement.
• Add a functional test. A pressure hold at 30 psi for 30 seconds, or a bubble point check per the relevant method, catches the stack-up failures that dimensional inspection misses. The U.S. EPA SW-846 methods and the Thermo Fisher chromatography and sample handling resources both assume the filter seals; neither specifies the tolerance that makes it seal.
• Control the web, not just the disc. Ask for thickness variation across the web width. If it exceeds ±5%, your stack height tolerance is at risk regardless of how well the disc is cut.
• Check the concentricity callout. If your housing seals on the aperture, specify 0.10 mm TIR maximum and confirm the supplier uses a stripper plate.

Where the tolerance conversation usually goes wrong

Two failure patterns account for most of the rejections I have seen. The first is a buyer who copies a tolerance from a competitor’s datasheet without knowing which process produces it. A ±0.05 mm OD on a die-cut disc is a laser-cut number, and if you specify it on a die-cut part, the supplier will either reject the order or sort the lot and pass the cost to you. The second is a supplier who reports capability from a single lot and calls it a process capability. Within-lot sigma on a punch press is often 40 to 60% lower than pooled sigma, so a single-lot Cpk of 1.8 can be a pooled Cpk of 1.1.

There is a third pattern that is harder to catch: the drawing is internally inconsistent. I have seen prints that call for a 47.00 ± 0.10 mm OD and a 0.15 mm concentricity, which is geometrically fine, but pair it with a 0.30 mm flatness on a 0.5 mm disc. A disc that flat and that concentric is achievable, but only with laser cutting and a flattening step, and the price triples. The fix is to write the print around the function, not around the features.

If you are sourcing filters that go into a regulated method, cross-check the dimensional requirements against the method’s sample handling assumptions. The Restek technical resources and the Agilent chromatography consumables documentation both describe the hardware the filter has to fit, and the COD vial dimensions are a useful sanity check for the 16 mm and 25 mm formats that show up in digestion workflows.

Troubleshooting checklist for a lot that measures in spec but fails in use

• Measure aperture and concentricity, not just OD. A lot can pass OD and fail function.
• Re-measure at 23 ± 2 °C with a non-contact method. Temperature and contact force together can shift a reading by 0.06 mm.
• Check thickness across the web width. A 10% cross-web variation will show up as a stack height failure.
• Inspect for curl. A disc with more than 0.30 mm of flatness deviation will bypass in a face-seal housing.
• Run a 30 psi, 30-second pressure hold on 10 pieces. If more than one leaks, the problem is the stack-up, not the disc.
• Ask the supplier for the sharpening log on the punch die. Pieces cut in the first 500 strokes after sharpening run oversize.
• For molded rings, check roundness, not just OD. An oval ring passes a two-point measurement and leaks.

The decision rule is simple: specify tolerance only on the feature that seals, require a pooled Cpk of 1.33 or better on that feature with the measurement method and temperature stated, and release the lot on a functional pressure hold rather than a dimensional gauge. Everything else on the print is a cost driver, not a quality driver, and loosening it is how you get a filter that works and a price that holds.

References & further reading

Related resources from our lab consumables network:

• Ptfe Membrane Filter at ptfemembranefilter.com
• Cod Vial at codvial.com
• Vial Insert at hplcinsert.com
• 2Ml Vial at 2mlsamplevial.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