Leak testing quick disconnects.
Helium mass spectrometry and vacuum decay are not competing brands of the same test — they resolve different leak rates, take different amounts of time, and prove different things about a coupling. This brief sets the two side by side with the limits we build to (5×10⁻¹² mbar·l/s detector class, ~1.4×10⁻⁶ atm·cc/s release criterion, <1×10⁻⁴ mbar·L/s on 100% of sets), then covers the third test nobody puts on the datasheet: the dry-break spillage check.
DL-004 · Published 2026-09-06 · 9 min read
A dry-break coupling is a pressure vessel that a technician opens on purpose, and it has to hold at two different moments: while it is mated and carrying loop pressure, and in the seconds after it separates. Leak testing is how a supplier demonstrates the first moment. The second one is a spillage test, and it is the test that most often goes unmentioned in a quotation.
Both matter commercially. Water and cooling failures account for a double-digit share of data centre outages, and downtime at the average facility is priced around $9,000 per minute. A coupling that weeps is not a component problem by the time anyone notices — it is an incident. What follows is the protocol arithmetic: what each method can see, how often it should run, and what the paperwork has to say before a batch is worth accepting.
Two methods, two different questions.
Helium tells you how tight a part is and roughly where it leaks. Decay tells you whether it is sealed at all, in the time a production line can afford.
Helium mass spectrometry uses a tracer gas the detector can see at extraordinary dilution. A bench in this class runs a detector sensitivity of 5×10⁻¹² mbar·l/s, and the release criterion applied to sealed sets is in the 1.4×10⁻⁶ atm·cc/s order — a tolerance three to six orders of magnitude tighter than a pressure gauge can resolve. Because helium is introduced on one side of the boundary and sniffed or accumulated on the other, the method also localises a failure: a leak at the threaded end reads differently from a leak past the valve seat.
Vacuum decay is the opposite trade. Evacuate the sealed part, isolate the pump, and watch the pressure rise over a fixed dwell. The resolution is set by gauge accuracy, internal volume and dwell time rather than by a tracer, so practical benches resolve around the 1×10⁻⁴ mbar·L/s class. That is coarse next to helium and entirely sufficient for a go/no-go check — which is exactly why it runs on every set instead of a sample.
A units trap sits in the middle of this. One atm·cc/s equals 1.013 mbar·L/s, so the two common unit families are almost numerically interchangeable — and a leak limit quoted without its unit, or with the wrong one, can be off by a factor that no reader will catch. Every limit in this article carries its unit, and every limit on your RFQ should too.
Method, limit, scope — written as one line each.
An acceptance criterion is only complete when it names the method, the limit and the fraction of the batch it was applied to. All three columns belong on the drawing and in the report.
| Test | Method / limit | Scope |
|---|---|---|
| Vacuum decay | pressure-loss method, <1×10⁻⁴ mbar·L/s | 100% of sets before marking or packing |
| Helium leak test | mass spectrometry, 5×10⁻¹² mbar·l/s detector class | batch verification on leak-critical sets |
| Production release value | ~1.4×10⁻⁶ atm·cc/s order | release criterion for every sealed set |
| Flow verification | class flow at 1.7 GPM on dash-04 — sample recorded within 1.8% of rating | sampled per batch |
| Cycle-then-leak | re-test after 5000 mate cycles (OCP floor) | batch validation sample |
| Dry-break spillage | disconnect at 0 psi, 0.020 – 0.070 ml by size | OCP limit, UQD02 through UQD08 |
| Glycol soak | 72 h at the −40 to +120 °C class endpoints | sampled sets before release |
| Cleanliness on release | ISO 4406 18/16/13 | particle count per batch |
Limits per DryLink outbound sequence and OCP UQD floors; the release value and detector class are bench characteristics, not product marketing figures.
What "100%" has to mean before it means anything.
Every supplier claims testing. The questions that separate the claims are: 100% of what, against which limit, and who signed it.
- 100% of assembled sets — vacuum decay applies to the set as it ships, with end connections fitted. A body tested before the ORB end is assembled has not been tested as a pressure boundary.
- Sample, not 100% — helium verification, flow check, 5000-cycle endurance, 72 h glycol soak and 48 h salt spray belong on batch samples. A supplier who claims 100% helium on every piece for a 50-piece MOQ is either unusually equipped or describing a different test.
- Before and after cycling — the number that matters for service life is the leak rate after 5000 mate cycles, not the rate on a fresh set. Ask for both.
- Per size, not per family — the dash-04 spillage cap is 0.025 ml, the dash-08 cap is 0.070 ml. A record that quotes one number for the whole product family is not a record.
- Per batch, not per year — a test report dated to the year tells you nothing about the carton in your goods-in bay.
There is a cost argument for the split, and it is worth understanding rather than resenting. Helium testing takes minutes per part and consumes tracer gas; vacuum decay takes seconds and consumes nothing. Running decay at 100% and helium on a validation sample buys a tighter release criterion than a datasheet claim and still leaves the set affordable at MOQ 50. What a buyer should not accept is a sample helium test sold as if it covered the batch.
The test the leak bench cannot replace.
Sealed mated and sealed unmated are two separate properties. Spillage is the third, and it lives in the volume trapped between the two valve faces.
When a flat-face pair separates, the fluid held between the poppets has nowhere to go but out. The OCP UQD specification caps that loss per disconnect at 0 psi: 0.020 ml at UQD02, 0.025 ml at UQD04, 0.035 ml at UQD06 and 0.070 ml at UQD08. Those numbers are small enough that a visual check will not catch a failure — 0.025 ml is roughly half a drop of fluid, and a set can exceed the cap by a factor of four while still looking dry.
The bench method is simple enough to run in a service bay. Mate the pair, pressurise to working pressure, hold, vent to 0 psi, then disconnect onto an absorbent pad on a balance readable to 0.1 mg. A 25% propylene glycol mix sits near a specific gravity of 1.02, so the dash-04 cap of 0.025 ml is roughly 26 mg of fluid — a margin a milligram balance resolves by more than two orders of magnitude. Weigh twice, subtract, compare against the cap for the size. Repeat on the same set after cycling, because a valve face that seals at cycle ten can weep at cycle five thousand.
Decide which test you are buying, though. The standard caps spillage at 0 psi, which is the service condition: a technician pulls a tray with the branch isolated, not with the pump running against it. A valve that weeps when disconnected under live system pressure is not compliant with the cap, and may or may not be acceptable in your maintenance procedure. Say which one your site needs and let the supplier test to that.
What a supplier's test record should contain.
If a number cannot travel with the parts, it is a brochure claim. Here is the content a record needs to be usable at goods-in without a phone call.
| Record field | Why it has to be there |
|---|---|
| Batch and serial range, size class, drawing revision | ties the carton in front of you to a specific build, not to a product family |
| Method, tracer or decay, fixture, dwell time | a limit without a method is not reproducible — decay and helium limits differ by orders of magnitude |
| Acceptance value and measured value, with units | shows margin, not just a pass stamp; unit errors are the most common defect in a leak record |
| Test pressure and medium temperature | leak rate and spillage both move with pressure and viscosity |
| Scope per method (100% versus sample size) | lets your incoming inspection decide how much of the batch to re-verify |
| Seal compound, seal lot, body material certificate | the elastomer is the wear part; a leak result without a compound identity cannot be trended |
| Cycle count at test, where cycling was applied | distinguishes a fresh-part result from a post-endurance result |
| ISO 4406 cleanliness code on release | particles hold ions and abrade seat faces; the code is part of the leak argument |
| Spillage result against the size cap | the dry-break claim, measured rather than asserted |
| Tester, date, archived record reference | makes the record auditable a year later, against the original numbers |
Any row can be replaced by your own criterion — send it with the drawing and it becomes the acceptance value in the batch report.
Three artefacts and one real defect.
Not every failed reading is a failed part. Three of the four cases below are test-setup problems; the fourth is the one worth scrapping for.
1. Virtual leaks
A dead-end threaded joint, a blind tapped hole or a trapped volume between two seals holds gas that bleeds back during a vacuum-decay dwell. The pressure rise looks exactly like a leak, and it is a property of the fixture rather than the part. Vent the dead ends, or change the fixture, and the reading disappears.
2. Helium memory in the elastomer
EPDM absorbs helium during a test and releases it afterwards. A set tested twice in quick succession reads worse the second time, which is a false trend rather than a degradation. Purge and allow a settling interval between runs before you conclude anything about a batch.
3. Temperature drift during the dwell
A part still warm from washing, or a bench in a draughty aisle, makes the pressure in a sealed volume move for reasons that have nothing to do with leakage. Stabilise the part and the fixture to the same temperature, then dwell. A short dwell on a cold part is a measurement of the room.
4. A particle on the seat — the real defect
A single hard particle dragged past a valve face holds the seal open just enough to fail a helium test, and it can pass a coarse decay check at low pressure. This is why cleanliness is held to ISO 4406 18/16/13 on release and why sets are wiped and capped before packing. It is also why a failed part should be opened before it is scrapped: a dented or scored seal face is a process escape, whereas a clean face with a stray particle is a handling escape, and the two have different corrective actions.
Three questions this brief answers most.
Is helium testing better than pressure decay?
What spillage is acceptable when a UQD is disconnected?
Does a leak test prove the coupling breaks dry?
Ask for the test record with the quotation.
Sample batch reports ship with any quotation — leak values, flow check, spillage result and cleanliness code. Send your criterion and it becomes the acceptance value in the report.