A lot of 500 autosampler vials came back from a customer with a 0.4 mm spread in overall height, and the symptom was not a leak. It was a rack of bent needles. The gripper on their CTC PAL had been set to a fixed push depth of 24.0 mm from the rack datum, and the tallest vials in the lot put the septum 0.35 mm higher than the setpoint, which drove the needle shoulder into the crimp ring on every tenth injection. The dimensional tolerance capability of that lot was never specified on the drawing, so nobody had a number to reject against. That is the gap I want to talk about here: what tolerance a molding or machining process can actually hold on chromatography parts, how you measure whether it is holding it, and which dimensions are the ones that drift first.
If you buy septa, caps, ferrules, vial bodies, or column hardware in volume, you are already doing dimensional tolerance capability work whether you call it that or not. The spec sheet is the easy part. The hard part is knowing which 0.05 mm on a drawing is a real control point and which 0.05 mm is decoration that will never be inspected. Most of the field failures I have chased came from a dimension that was never on the drawing at all.
Process capability is not a single number. It is a number per feature, per material, per tool, and it moves after the first few thousand shots. Here is what I have seen hold up in production, not in a sales brochure, across the parts we make and buy.
| Process | Typical achievable tolerance | Where it is used in chromatography | What breaks first |
|---|---|---|---|
| Injection molding, PP vial body | ±0.10 mm on OD, ±0.15 mm on height | 9 mm and 12 x 32 mm autosampler vials | Height drift from mold wear and cycle-time creep |
| Injection molding, PE cap | ±0.05 mm on thread pitch diameter | Screw caps, snap caps | Thread core shift, gate blush |
| Compression molding, silicone/PTFE septum | ±0.08 mm on thickness, ±0.10 mm on diameter | 11 mm and 20 mm septa | Thickness taper across the disc |
| Machining, PTFE ferrule | ±0.025 mm on cone angle, ±0.05 mm on bore | 1/16 in and 1/8 in fittings | Cone angle drift from tool wear |
| Machining, stainless column end fitting | ±0.013 mm on thread, ±0.02 mm on seat face | HPLC column hardware | Seat face flatness after repeated cuts |
| Extrusion, PEEK tubing | ±0.025 mm on OD, ±0.013 mm on ID | 0.005 in to 0.030 in ID tubing | ID ovality at the cut end |
Those numbers are the ones a well-run tool holds at Cpk 1.33 or better. A new tool with a fresh core can hold half of that for the first 5,000 shots. The question is not what the tool can do on day one. It is what the tool does at shot 250,000, after the core has worn 0.02 mm and the cooling lines have scaled.
For septa specifically, thickness is the dimension that controls everything downstream. A 11 mm septum specified at 1.00 mm ± 0.05 mm that runs at 1.08 mm on one edge and 0.95 mm on the other will still pass a go/no-go check at the center, and it will still fail on a 30 psi headspace run because the thick side does not seat. I have pulled septa from three lots of the same part number and measured a 0.13 mm taper across the disc with a bench micrometer at four clock positions. That is not a molding defect you catch with a single-point gauge.
How to measure capability on a part that is mostly a hole
Capability on a chromatography consumable is awkward because the functional dimension is often an internal feature. The ID of a vial, the bore of a ferrule, the cone angle of a fitting. You cannot measure those with a pair of calipers and expect a useful Cpk.
Pick the functional dimension, not the drawing dimension
For a 12 x 32 mm vial, the drawing usually calls out OD, height, and neck finish. The functional dimensions are different. They are:
- Height from the base to the crimp ledge, because that sets where the septum sits relative to the gripper datum.
- Neck ID at the constriction, because that sets the needle guide and the seal area.
- Base concentricity to the neck, because a vial that leans 0.2 mm in the rack changes the needle path on the far side of the tray.
If your supplier only reports OD and height, you are not measuring the dimensions that fail. I ask for a first-article inspection that includes the neck constriction ID and base-to-neck runout, and I ask for it on 30 parts from a single shot, not 5 parts from 5 shots. The 30-part sample from one shot is what tells you the within-cavity spread. Five parts from five shots tells you nothing about the tool.
Run a 30-piece capability study before you approve the tool
The study I want on a new mold or a new machining program is 30 consecutive parts from one cavity, measured on the functional dimensions, with the mean and standard deviation recorded. Then the same on the second cavity, and so on. For a 32-cavity vial mold, that is 960 parts, which is a day of production and a reasonable ask for a tool that will run for a decade.
Compute Cpk as (USL minus mean) divided by 3 sigma, or (mean minus LSL) divided by 3 sigma, whichever is smaller. On a 0.30 mm total tolerance band for vial height, a Cpk of 1.33 means sigma is about 0.038 mm. That is a tool that will still be in tolerance at shot 300,000 if wear is linear. A Cpk of 0.9 on the same band means sigma is 0.056 mm and the tool is already spending time outside the band on a normal day.
One more number that matters: the drift rate. Measure the same 30 parts at shot 10,000 and again at shot 100,000 and plot the mean. If the mean moves 0.01 mm per 100,000 shots, you have a predictable tool life. If it moves 0.04 mm, you have a tool that will surprise you on a Friday afternoon.
The dimensions that are actually hard to control
Some features are easy. A vial OD on a round core is about as forgiving as molding gets. Others are the ones that eat your yield. These are the ones I watch.
Septum thickness taper
A septum is a disc of soft material compressed in a mold. The material flows from the gate outward, and the far side from the gate is always slightly thinner because the melt has cooled by the time it arrives. On a 11 mm silicone/PTFE septum, a 0.06 mm taper from gate side to far side is normal. A 0.12 mm taper is a tool that needs a gate redesign or a hotter barrel. The functional consequence is that the septum seals on one side and leaks on the other at the same compression, and on a headspace run at 80 °C the leak shows up as a drifting internal standard, not a visible drip. I have chased that one for two weeks on a GC-FID method before pulling the septa and measuring them.
Ferrule cone angle
A 1/16 in graphite or PTFE ferrule has a cone that has to match the receiving port to within about 0.5 degrees. The angle is cut on a lathe, and the tool wears on the flank, so the angle drifts slowly. A ferrule that is 1 degree off will still seal at low pressure and will weep at 400 bar. The symptom is a baseline that wanders on a gradient method and a puddle under the column fitting after a long run. Measure cone angle with an optical comparator on a sample from each lot, not a go/no-go ring gauge, because the ring gauge will pass a ferrule that is off-angle but on-diameter.
Vial neck constriction and thread start
The neck constriction on a 9 mm vial is a small feature on a small core, and it is the last place the plastic fills. Short shots show up here first. The thread start position is also hard to hold because it depends on the core rotation during ejection, and a 0.3 mm shift in thread start will make a cap cross-thread on the capper and still pass a visual check. I have seen a lot where 2% of caps cross-threaded on an automated capper and 0% failed a hand-check, because the hand can feel the start and the machine cannot.
Column end fitting seat flatness
On HPLC hardware, the seat face where the ferrule compresses has to be flat to about 0.005 mm and square to the thread axis. A seat that is 0.01 mm out of square will seal at 200 bar and leak at 600 bar, and the leak shows up as a pressure ripple, not a drip. This is a machining-center problem, not a molding problem, and it is the one I see most often on aftermarket fittings that were reverse-engineered from a sample rather than from a drawing.
If you are sourcing any of these parts, the chromatography products catalog is only useful if the drawing behind each part number names these features with a tolerance and a measurement method. A part number without a control plan is a part number you cannot audit.
Two field cases where the tolerance number was the diagnosis
Case 1: the headspace vial that sealed at 60 °C and leaked at 90 °C
A customer running EPA method 8260 on a headspace GC-MS started seeing their internal standard response drop 15% to 20% on the last third of a 40-vial tray. The first ten vials were fine. The instrument was fine. The method had not changed. What had changed was a new lot of 20 mm crimp caps with septa.
We pulled 30 caps from the lot and measured septum thickness at four clock positions with a bench micrometer. The mean was 3.02 mm against a 3.00 mm nominal, which passed. The spread was 0.18 mm across the lot, and the thin side was 2.91 mm. The crimper was set to a fixed crimp height, so the thin side of the septum got less compression and the thick side got more. At 60 °C the thin side still sealed. At 90 °C the thin side relaxed and the seal opened. The fix was not a new crimper setting. It was a tighter thickness tolerance on the septum (3.00 mm ± 0.05 mm) plus a compression check on the crimper every shift. The lot that failed had a Cpk of 0.71 on thickness. The replacement lot ran at 1.4.
That is the pattern I see most: the average is fine, the spread is not, and the failure only shows up at the temperature or pressure extreme. A supplier who reports only the mean is telling you half the story.
Case 2: the ferrule that passed the gauge and failed the pump
A method development lab was running a 600 bar UHPLC method and getting a pressure ripple of about 8 bar that appeared after 40 minutes and got worse. They had swapped the column, the mobile phase, and the detector cell. The leak was at the column inlet fitting. The ferrule was a PEEK part from a lot that had passed the incoming go/no-go ring gauge.
We cut three ferrules from the lot in half and measured the cone angle on an optical comparator. Nominal was 60 degrees. The lot ran 59.1 to 59.4 degrees. The ring gauge checks the diameter at the cone, not the angle, so it passed. At 600 bar the 0.6 degree mismatch opened a leak path at the top of the cone that was invisible at 200 bar. The fix was a drawing change: cone angle 60° ± 0.25°, measured on an optical comparator, with a first-article on every lot and a 5-piece audit at incoming. The pressure ripple went to under 1 bar.
Both cases have the same lesson. The dimension that failed was not the dimension on the certificate. If you want the full background on how the method drives the part choice, the resources at Restek and the application notes at Agilent are worth reading alongside your own drawing review. The EPA SW-846 methods also set the temperature and pressure envelope your part has to survive, and that envelope is what defines which dimensions are critical.
A decision checklist for buying to a tolerance, not to a catalog
When you write the next purchase order for a molded or machined chromatography part, work through this list before you sign. It takes an hour and it saves the two-week investigation.
- Name the functional dimensions on the drawing, not just the outline dimensions. For a vial: base-to-ledge height, neck constriction ID, base-to-neck runout. For a ferrule: cone angle, bore, face flatness. For a septum: thickness at four clock positions, diameter, durometer.
- Set a tolerance band you can measure. A 0.30 mm band on vial height is realistic for injection molding. A 0.05 mm band on the same feature is a machining number and will price you out of the category.
- Require a Cpk of 1.33 or better on the functional dimensions, computed from a 30-piece single-shot sample, and require it again at first production lot.
- Ask for the measurement method. A micrometer at four clock positions is not the same as a single-point gauge, and the certificate has to say which one was used.
- Set an incoming audit sample size. I use 30 pieces per lot on the functional dimensions and a 5-piece optical check on any angled feature. That is enough to catch a 0.1 mm mean shift on most features.
- Watch the drift, not just the pass. Ask for the mean at shot 10,000 and shot 100,000 on the tool. A supplier who tracks tool wear will have the number.
- Match the material to the temperature. Silicone septa rated to 200 °C, PTFE-faced to 250 °C, and a crimp cap that holds compression at the method’s top temperature. A tolerance that holds at 40 °C and fails at 90 °C is not a tolerance, it is a hope.
For the parts themselves, the suppliers in the vials with caps range and the OEM work at Aijiren publish first-article data on request, and that data is the thing to read before you approve a tool. If you are sourcing at volume and want a supplier who will quote a Cpk alongside a price, vial supplier is a reasonable starting point for the vial and cap side of the bill of materials. The Thermo Fisher consumables catalog is also useful for cross-checking what tolerance band the major instrument makers assume on their autosamplers, because that band is the real upper limit on what you can buy.
What to do when a lot fails on a dimension nobody specified
This happens more than it should. A lot fails in the field, you pull samples, and the failing dimension is not on the drawing. The temptation is to blame the supplier and move on. The better move is to add the dimension to the drawing with a tolerance and a measurement method, and to require a capability study on the next tool.
The order of operations I use is: identify the functional dimension from the failure mode, set a band that the process can hold at Cpk 1.33, specify the gauge and the sample size, then re-qualify the tool. If the supplier cannot hold the band, the part is the wrong process for the application, and no amount of incoming inspection will fix it. A 0.02 mm band on a molded vial height is not a quality problem, it is a process-selection problem. Move to machining or change the design.
The one number I would put on any purchase order for a molded chromatography part is the Cpk on the functional dimension, not the pass/fail on the certificate. A certificate says the parts were in tolerance on the day they were measured. A Cpk of 1.33 says the tool will still be in tolerance at shot 200,000, which is the only thing that matters if you are going to run that lot through an autosampler for the next six months. Buy the capability, not the certificate.
References & further reading
Related resources from our lab consumables network:
- Chromatography Products at chromatographyselling.com
- Aijiren at aijirentech.com
- Vials With Caps at vialswithcaps.com
- Vial Supplier at vialsupplier.com
External references and standards: