A shipment of 10,000 HPLC solvent inlet filters left our warehouse on a Tuesday and came back the following Monday. The lab found 14 filters with loose PTFE frits in the first 200 they opened. That is a 7% visible defect rate. The lot had passed a casual “look at a few boxes” inspection. It had never been through an AQL sampling plan. The return freight cost more than the filters, and the lab switched vendors before we could issue a credit. That is the exact failure an outgoing QC sampling plan exists to prevent, and it is the one I will break down here: an hplc filter AQL sampling plan and outgoing QC that catches a 7% defect rate before the truck leaves, not after the customer opens the carton.
AQL stands for Acceptable Quality Limit. It is not a guarantee that zero defects ship. It is the maximum percent defective that you, as the manufacturer, consider acceptable as a process average. For HPLC filters, I use AQL 1.0 for critical defects, AQL 2.5 for major defects, and AQL 4.0 for minor defects. Those numbers are not arbitrary. A critical defect on a solvent inlet filter means the part will not perform its function at all: a cracked stainless steel frit, a PTFE membrane with a pinhole visible under backlight, an ID bore that will not seat on the pump inlet stem. A major defect degrades performance or shortens life: a frit with uneven porosity that causes flow restriction, a housing with a burr that will shed particles, a non-flat sealing surface. A minor defect is cosmetic or non-functional: a light surface scratch on the housing, a slightly off-center laser marking, a label smudge.
The sample size is not a percentage of the lot. It follows ISO 2859-1 (the international version of ANSI/ASQ Z1.4, formerly MIL-STD-105). For a lot of 10,000 filters at General Inspection Level II, the code letter is L. At AQL 1.0, the sample size is 200 pieces with an acceptance number of 5 and a rejection number of 6. That means you pull 200 filters at random, inspect every one for critical defects. If you find 5 or fewer, the lot passes. If you find 6, the lot fails. At AQL 2.5 for major defects, the same 200-piece sample has an acceptance number of 10 and a rejection number of 11. At AQL 4.0 for minor defects, acceptance is 14, rejection is 15.
Here is the part most people miss. If you only inspect for “defects” as one bucket and you set your threshold at AQL 2.5, you will accept lots where 10 out of 200 filters have a major defect. That is 5% of the sample. A lab buying 1,000 filters from that lot will open roughly 50 defective units. That is not acceptable for a consumable that sits in the mobile phase flow path. So I run three separate inspections on the same 200-piece draw: critical at AQL 1.0, major at AQL 2.5, minor at AQL 4.0. The lot must pass all three. A filter with both a critical and a minor defect counts only against the critical tally, but I log the minor defect anyway. The record matters.
Defect classification for HPLC filters has to be written down, not left to inspector judgment
An inspector who has to decide on the spot whether a burr is “bad enough” will make different decisions at 8 a.m. and 4 p.m. The classification sheet removes that drift. I keep a one-page defect library at each inspection station. It lists the defect, the class, the measurement method, and the go/no-go limit. Here is the version I use for solvent inlet filters and inline filters, the two highest-volume HPLC filter types we ship.
- Critical: cracked or chipped frit. Inspect under 10x magnification with a gooseneck light. Any visible crack or chip on the frit face or edge is a reject. No gray area.
- Critical: pinhole in PTFE membrane. Backlight the membrane against a white light panel. A pinhole shows as a bright dot. Reject any dot larger than 0.1 mm. For 0.45 µm and 0.22 µm membranes, a pinhole defeats the entire filter rating.
- Critical: bore ID out of tolerance. The inlet bore on a solvent filter must be 0.125 inch (3.175 mm) nominal to seat on a standard pump inlet stem. Use a go/no-go pin gauge set. Go pin 3.15 mm must enter. No-go pin 3.25 mm must not. If the no-go enters, the filter will leak or fall off under vacuum.
- Major: frit porosity out of spec. Measure with a bubble point tester or a calibrated air permeability rig. For a 10 µm stainless frit, the bubble point in isopropanol should fall between the lot’s validated range, typically 2.5 to 4.5 psi. Outside that range means the frit will restrict flow or pass particles.
- Major: burr on the housing sealing face. Run a clean lab wipe across the face. If the wipe snags, reject. A burr here will score the pump head or the filter holder and generate particles.
- Major: non-flat sealing surface. Place the filter on an optical flat with monochromatic light. More than two interference fringes across the sealing annulus is a reject. That is roughly 0.6 µm of flatness error.
- Minor: cosmetic scratch on the housing. Accept if the scratch does not catch a fingernail and is not visible at arm’s length under normal lab lighting.
- Minor: off-center laser marking. Accept if the marking is fully legible and within the marking area. Reject if any character is clipped or unreadable.
This list changes by product. A PTFE membrane filter for aggressive solvent filtration gets a pinhole test at 100% inspection, not sampling, because a single pinhole in a 0.22 µm membrane shipped to a UPLC lab is a recall-level event. A syringe filter with a Luer lock gets a thread gauge check added to the major defect list. The point is that the classification sheet is product-specific and version-controlled. When a customer returns a filter for a failure mode we did not have on the sheet, the sheet gets revised within one week. That is the only way the sampling plan improves.
Sample size math and what a realistic outgoing inspection record contains
Let me walk through a real lot. Lot 240718-02, 10,000 PEEK inline filters, 0.5 µm, for a contract manufacturer in New Jersey. Lot size 10,000 puts us at code letter L under General Inspection Level II. We pull 200 pieces using a stratified random method: we open cartons from the beginning, middle, and end of the pallet, and we pull from the top, middle, and bottom of each carton. This matters because frit compression can vary by position in the molding tray, and if we only pull from the top carton, we miss the drift.
The outgoing inspection record for that lot has these fields, and I will fill in the actual numbers.
- Lot number and date code. 240718-02, molded 2024-07-18, shift 2.
- Sample size and code letter. n=200, letter L, General II.
- Critical inspection result. 1 defect found (cracked frit at position 37 in carton 8). Acceptance number 5. Lot passes critical at AQL 1.0.
- Major inspection result. 6 defects found (3 burrs on sealing face, 2 bubble point failures, 1 bore ID no-go). Acceptance number 10. Lot passes major at AQL 2.5.
- Minor inspection result. 11 defects found (7 cosmetic scratches, 4 off-center markings). Acceptance number 14. Lot passes minor at AQL 4.0.
- Disposition. Ship. The 18 defective pieces are removed from the sample and scrapped. The remaining 182 inspected pieces are returned to the lot only if the inspection is non-destructive. Bubble point testing is non-destructive. Bore gauging is non-destructive. The cracked frit and the burred housings are destroyed.
- Inspector name and signature. Not “QC” as a department. A person. If the lot fails at the customer, we need to know who inspected it and whether retraining is needed.
- Retained sample. 10 pieces from the same draw, sealed in a labeled bag, held for 24 months or until the lot’s shelf life expires, whichever is longer.
That record takes about 20 minutes to complete for a 200-piece sample. It is not a burden. It is the difference between shipping a 1.5% major defect rate lot with your eyes open and shipping a 7% defect rate lot because nobody looked. The record also gives you the data to see trends. If the burr count creeps from 1 per 200 to 4 per 200 across three lots, the molding tool is wearing. You catch that before the AQL plan rejects a lot, and you fix the tool instead of scrapping 10,000 filters.
Switching rules: when normal inspection is not enough
ISO 2859-1 has switching rules, and most HPLC filter manufacturers ignore them. That is a mistake. The rules exist because a process that has been running clean for 10 lots needs less inspection, and a process that just failed needs more. Here is how I apply them.
Start at normal inspection. If 2 out of 5 consecutive lots fail on the same defect class, switch to tightened inspection for that class. Tightened inspection at the same code letter L and AQL 1.0 uses the same sample size of 200 but the acceptance number drops from 5 to 3. That is a much harder bar. If the next 5 lots pass under tightened inspection, return to normal. If any lot fails under tightened inspection, stop shipment and run a 100% sort on the entire lot, then fix the process before the next lot starts.
If 10 consecutive lots pass under normal inspection with zero critical defects and the major defect count stays below half the acceptance number, switch to reduced inspection. At code letter L, reduced inspection drops the sample size from 200 to 80 pieces. The acceptance number for critical defects at AQL 1.0 drops to 2, and for major defects at AQL 2.5 it drops to 5. If any lot fails under reduced inspection, switch back to normal immediately. I rarely use reduced inspection for critical defects on HPLC filters. The cost of a critical defect reaching a customer is high enough that the labor savings on 120 fewer pieces inspected is not worth it. I use reduced inspection only for minor defects, where the consequence of a miss is a cosmetic complaint, not a failed analysis.
The switching rules are not optional decoration. They are the feedback loop that makes the sampling plan responsive. A process that drifts will fail more lots, which triggers tightened inspection, which forces the process owner to fix the drift or stop shipping. A process that is stable gets rewarded with less inspection, which frees the inspector to spend time on the defect classes that actually need attention.
Field case: the lot that passed AQL but still got returned
A university core facility ordered 500 PEEK solvent inlet filters, 10 µm, for their Agilent 1260 pumps. The lot passed our outgoing AQL inspection at AQL 1.0 critical, AQL 2.5 major, AQL 4.0 minor. Sample size for a 500-piece lot at code letter H is 50 pieces. We found zero critical, one major (a burr), and two minor. Lot shipped. Three weeks later, the facility reported that 12 filters out of the 500 were leaking at the pump inlet. That is a 2.4% leak rate, which our AQL plan should have caught if the leak were a critical defect. It was not on our defect classification sheet as a separate line item. We had a bore ID check and a flatness check, but we did not have a leak test under vacuum.
The measurement that identified it: we pulled the 12 returned filters and tested them on a vacuum rig at 25 inHg with the filter submerged in isopropanol. Ten of the 12 showed a stream of bubbles from the frit-to-housing junction, not from the bore. The frit was not fully pressed into the housing. Our bore gauge and flatness check both passed because they measured the housing, not the frit retention. The fix was two changes. First, we added a frit retention pull test to the critical defect list: a calibrated pull of 2.5 N on the frit edge. Any movement is a critical reject. Second, we added a vacuum leak test on a 10-piece sub-sample from every lot at the same 25 inHg. The AQL plan did not change. The defect classification sheet did. That is the correct order of operations. The sampling plan is the framework. The defect library is the content. If the content misses a failure mode, the framework cannot save you.
Field case: the tightened inspection that caught a molding drift before the customer did
This one is from our own production floor. We mold PEEK filter housings on a 4-cavity tool. Cavity 3 started producing a burr on the sealing face at a rate that our normal AQL inspection did not flag immediately. Over three consecutive lots, the major defect count went from 1 in 200 to 4 in 200 to 7 in 200. The third lot passed at AQL 2.5 because the acceptance number is 10. But the trend was visible in the inspection records, and that visibility is the entire point of keeping the data. The QC supervisor pulled the last 10 inspection records for cavity 3 and saw the burr count climbing in a straight line. That triggered a tool teardown before the next lot. The cavity 3 ejector pin had worn by 0.08 mm, enough to leave a flash line on the sealing face. The fix was a new ejector pin, $40, and two hours of downtime. If we had waited for the AQL plan to reject a lot, we would have scrapped or sorted 10,000 housings and probably shipped some borderline parts in the meantime. The sampling plan is a gate. The inspection record is a trend monitor. You need both.
What the outgoing inspection record must contain for a contract manufacturer or OEM buyer
If you buy HPLC filters from a contract manufacturer and you do not receive an outgoing inspection record with the shipment, you are buying blind. A certificate of conformance that says “this lot meets specifications” is not an inspection record. It is a promise. The record is the evidence. Here is the minimum I require from any supplier, and the minimum I provide to any OEM customer.
- Lot traceability. Lot number, date code, cavity number if molded, raw material lot for the frit and the housing, and the vial insert or filter holder lot if the filter is part of an assembly.
- AQL parameters. Inspection level, code letter, sample size, AQL value for each defect class, acceptance and rejection numbers.
- Defect tallies. The actual count of each defect type found in the sample, not just “pass” or “fail”. A lot that passes with 5 critical defects at the edge of the acceptance number is a different risk than a lot that passes with zero.
- Measurement data. For frit porosity, the bubble point values, not just “in spec”. For bore ID, the go/no-go results. For flatness, the fringe count. Raw numbers let the buyer see the distribution, not just the pass/fail edge.
- Disposition of the sample. Which pieces were destroyed, which were returned to the lot, and the retained sample count and storage location.
- Inspector identity and date. A signature, not a stamp. If the buyer finds a defect that the inspection missed, the supplier needs to know who to retrain.
I have rejected supplier lots that shipped with a clean certificate of conformance but no inspection record. Not because the filters were bad, but because I could not verify the AQL plan was actually run. A supplier that cannot produce the record on request is a supplier that either does not run the plan or does not keep the data. Both are disqualifying for an HPLC filter that sits in the flow path of a chromatography system running a $4,000 column.
Decision checklist: is your outgoing QC plan actually protecting the customer
Run this checklist against your current process. If you answer no to any of the first five, your AQL plan is decorative.
- Is the defect classification sheet written down, version-controlled, and product-specific? A generic “visual defect” category is not a classification. It is a shrug.
- Are critical, major, and minor defects inspected separately with different AQL levels? A single AQL 2.5 bucket will let critical defects through.
- Is the sample drawn randomly from across the lot? Pulling the top carton only is not a random sample. It is a convenience sample, and it will miss positional drift.
- Are the acceptance and rejection numbers written on the inspection record before the inspection starts? If the inspector decides after seeing the defects whether the lot passes, the AQL numbers are theater.
- Does the record include the actual defect counts and measurement values? A pass/fail stamp without the tallies hides the trend data.
- Are switching rules in place and actually used? If you have never switched to tightened inspection, either your process is unusually stable or you are not tracking consecutive lot results.
- Is there a retained sample from every lot? If a customer reports a failure six months after shipment, the retained sample is the only physical evidence you have.
- Does the defect library get revised when a customer reports a new failure mode? The vacuum leak case above is the template. The customer found a mode we missed. We added it within a week. If your defect library has not changed in a year, you are not learning.
The comparison table below shows the sample size and acceptance numbers for the lot sizes most common in HPLC filter manufacturing. Use it as a reference, not a substitute for the ISO 2859-1 tables.
| Lot size | Code letter (General II) | Sample size | Critical accept/reject (AQL 1.0) | Major accept/reject (AQL 2.5) | Minor accept/reject (AQL 4.0) |
|---|---|---|---|---|---|
| 151 to 280 | G | 32 | 1 / 2 | 2 / 3 | 3 / 4 |
| 281 to 500 | H | 50 | 1 / 2 | 3 / 4 | 5 / 6 |
| 501 to 1,200 | J | 80 | 2 / 3 | 5 / 6 | 7 / 8 |
| 1,201 to 3,200 | K | 125 | 3 / 4 | 7 / 8 | 10 / 11 |
| 3,201 to 10,000 | L | 200 | 5 / 6 | 10 / 11 | 14 / 15 |
| 10,001 to 35,000 | M | 315 | 7 / 8 | 14 / 15 | 21 / 22 |
These numbers come from ISO 2859-1 single sampling plans for normal inspection. If you are shipping filters for use with EPA SW-846 methods or other regulated workflows, the buyer may require a specific AQL level in the purchase order. Do not assume your internal standard matches theirs. Ask. A buyer running Restek-validated methods with tight acceptance criteria may demand AQL 0.65 for critical defects, which at code letter L means a sample size of 200 and an acceptance number of 3. That is a much harder bar, and it changes your inspection economics.
One more note on the sample itself. The 200-piece draw for a 10,000-piece lot represents 2% of the lot. That sounds small, but it is statistically sound under the ISO 2859-1 operating characteristic curve. The probability of accepting a lot that is actually 5% defective at AQL 1.0 critical is about 10%. The probability of accepting a lot that is actually 1% defective is about 95%. That is the trade the standard makes, and it is the right trade for a consumable where the cost of a single critical defect reaching a customer is high but the cost of 100% inspection on every lot is higher. If you need a lower miss rate, raise the AQL level or move to 100% inspection on the critical parameter. For PTFE membrane filters, I do 100% pinhole inspection regardless of AQL, because the pinhole test is fast and the consequence of a miss is a failed filtration in a chromatography run that costs more than the filter. For stainless frit porosity, I use the AQL plan because the bubble point test takes 30 seconds per piece and 100% inspection would triple the filter cost.
The decision rule is simple. Run three defect classes at AQL 1.0 critical, AQL 2.5 major, AQL 4.0 minor, on the same random draw, with acceptance and rejection numbers fixed before inspection starts. Keep the defect tallies and measurement values in a record that a buyer can request. Use ISO 2859-1 switching rules to tighten when consecutive lots fail and reduce only when the process has earned it. Revise the defect classification sheet every time a customer reports a failure mode you did not have on the list. If your outgoing QC plan does not include those four elements, it is not a sampling plan. It is a hope.
References & further reading
External references and standards: