Ask any plant where their leak limit came from and you will get one of three answers: the drawing says so, the customer specified it, or it has always been that number. Almost nobody can trace the limit back to a functional requirement — the quantity of fluid the assembly must retain, or exclude, over its service life. That gap is the root of most leak-testing misery I have seen across two decades of launch programmes in automotive and aerospace.
A limit divorced from function creates two failure modes at once. It is either too loose, letting parts pass that will weep coolant or breathe moisture in the field, or too tight, scrapping good parts and burning hours on retest and containment. It also makes method selection arbitrary: if nobody knows what leakage rate actually matters, nobody can say whether pressure decay at end of line is adequate or whether helium mass spectrometry is required. The instrument gets chosen by capital budget, and the limit gets habit.
I have sat in rooms where engineers argued passionately about the fourth decimal place of a rejection threshold in sccm, while admitting, when pressed, that no one had ever measured how much leakage the sealed assembly tolerates in service. The argument about the limit is a proxy war. The real fight — method selection and correlation to function — never happens because it is harder and requires data.
Pressure Decay: Honest About Its Strengths
Pressure decay testing is cheap, fast, dry and non-destructive. You pressurise the part, isolate it, wait, and watch the transducer. For machined castings, welded housings, cooling circuits and most automotive fluid components, it is the right first choice. A good system with temperature-compensated transducers and a properly sized fill-and-stabilise cycle will detect leaks in the range that matters for liquid-retaining assemblies, in a takt-friendly time.
Its physics imposes limits you must respect rather than argue with. The measurable leak is a function of test volume divided by pressure change: a large-volume part with a small leak produces a pressure slope buried in noise. Temperature is the killer. A few tenths of a degree of air warming after pressurisation produces a pressure rise that swamps a genuine decay signal — which is why stabilisation time exists, and why shortening it to hit takt quietly destroys your capability to find anything.
The failure modes are well known and worth listing on the wall next to the tester: seal bypass at the fixture leaking more than the part ever could, adapter o-rings hardened from thousands of cycles, and thermal drift from parts arriving warm off a washing machine. Every one of these shows up as a tester that rejects everything on afternoon shift and passes everything at 6 a.m. When you see that pattern, look at temperature before you look at the parts.
Helium Mass Spectrometry: Buy It for the Right Reason
Helium testing exists because pressure decay cannot go below a certain leak rate without absurd cycle times, and because tracer-gas methods localise as well as detect. In sniffing mode you find where the leak is; in hard-vacuum bomb-and-scan mode you measure total leakage at sensitivities pressure decay will never reach. Hermetic assemblies, refrigerant circuits, fuel systems with evaporation requirements and aerospace components genuinely need it. Pretending a pressure decay test is equivalent is self-deception.
What helium is not, is a more moral version of pressure decay. It is a different instrument measuring a different thing under different conditions, and the numbers do not translate one-to-one. Helium finds molecular flow through porosity that liquid may never pass through; pressure decay with a wetting fluid behaves differently again. When a customer demands helium-equivalent results from a pressure decay tester, or vice versa, someone must build an empirical correlation part by part, leak path by leak path.
The hidden costs catch plants out: helium supply, recovery systems, background helium contamination in a factory that has run helium for years, calibration leaks whose certified values drift, and the skill needed to maintain vacuum integrity on a production line. I have seen helium testers converted into very expensive screening gates whose results were then re-verified by pressure decay anyway, because that was what the process documentation allowed. Buy helium when the function demands that sensitivity, not when a specification writer wanted a smaller number.
Temperature: The Variable That Decides Your Fate
Every leak test is secretly a thermometry exercise. Air in a sealed volume follows the gas law; if temperature changes during the measurement phase, pressure changes with it, and the tester reads a leak that does not exist — or masks one that does. Parts straight from machining carry cutting-fluid heat. Parts from a washer carry hot water. Sunlight through a skylight over the accumulation queue warms the top layer of totes differently from the bottom. All of it lands in your measurement.
Mitigation is unglamorous but effective. Stabilisation dwell before measurement, sized to the part's thermal mass and verified empirically rather than copied from a similar programme. Reference volumes — an unpressurised or matched-pressure reference on the far side of a differential transducer — cancel ambient thermal shifts that hit both sides equally. Fixturing in consistent contact patterns so conduction paths repeat. And test in the same orientation and location, because moving a tester three metres closer to a compressor line once changed readings on a programme I ran, purely through vibration and exhaust heat.
Characterise the tester with a non-leaking master part: run it fifty times across a shift and plot the pressure-decay readings. If that plot has a slope or a shift across the day, your leak signal contains temperature, and every capability study you have done on the tester is measuring the factory's HVAC as much as the parts. Fix that first. It is the cheapest improvement available in leak testing.
Correlating the Test to Actual Leakage
The test exists to predict function: will this assembly hold coolant for its life, keep water out of the electronics bay, retain refrigerant for a decade? The only honest way to answer that is a correlation study, and it is simpler than people fear. Take deliberately leaked parts — cracked castings, thinned welds, laser-drilled orifices, seals with known defects — and run them through the production test. Then run them through a functional proxy: submerged pressurisation watching for bubbles, an extended hold measuring actual fluid loss, or a tracer-gas measurement at a known sensitivity.
Building a defensible leak limit
- 01Define functionState in engineering units what the assembly must retain or exclude over its service life.
- 02Create known leakersCracked castings, thinned welds, laser-drilled orifices, defective seals.
- 03Run both testsProduction leak test plus a functional proxy: bubble test, extended hold or tracer gas.
- 04Find the boundaryPlot reading against outcome; identify where function actually degrades.
- 05Set the limitBoundary plus margin for measurement noise and field severity.
Plot production test reading against functional outcome and find the boundary — the leakage level at which function actually degrades. That boundary, with margin for measurement noise and field severity, is your defensible leak limit. Now the limit is traceable, auditable and arguable on evidence. When the customer pushes for a tighter number, show the correlation data and ask what functional failure mode the tighter limit prevents. Sometimes they have an answer. Often they do not, and the conversation improves considerably.
No one may argue about the number until they can state what leaks in service, how much, and what fails as a result.
One caution from hard experience: leak paths are not equivalent. A long, tortuous casting porosity path may pass gas slowly but pass nothing at all once wetted by glycol; a single sharp crack of the same gas rate may pass liquid freely. Correlate against the fluid and the failure mode, not against air in the abstract. Where the distinction matters, consider a two-stage test: gas screening followed by a wet test or pressure-hold on the boundary population.
Choosing the Method: A Practical Sequence
Start from the function, then the fluid, then the geometry, then the takt, and only then the budget. Write down what the assembly must retain or exclude over life in engineering units — that single sentence determines the target sensitivity. Convert the lifetime requirement into a per-second leak rate at test pressure; this arithmetic, rough as it is, immediately tells you whether pressure decay is even in the running. Large sealed volumes with tight requirements push you towards tracer gas whether you like it or not.
Second decision: whole-part go/no-go or localisation? If your process can produce a leak anywhere on a welded seam, and repair is possible and economic, a sniffing method or a multi-channel pressure test that brackets the failure zone saves rework. If the part is sealed, non-repairable and tested once at end of line, total-leak methods are correct and localisation belongs to the failure analysis lab, not production. I have watched plants buy localisation capability they never used because their repair strategy was scrap-and-replace anyway.
Third decision: what happens at the boundary? Every method has a grey zone where discrimination is poor. Decide in advance — retest rules, wet-test arbitration, quarantine — rather than letting operators invent the policy on nights. The worst leak tests I have inherited were not the insensitive ones; they were the sensitive ones with no defined boundary protocol, where the line retested until a marginal part eventually passed and everyone felt good about it.
How the method decision goes wrong
Chosen by default
- Limit inherited from the drawing or the customer
- Instrument selected by capital budget cycle
- Stabilisation time cut to protect takt
- Boundary policy improvised by operators
Chosen by function
- Limit traced to a measured functional boundary
- Sensitivity from lifetime leak-rate arithmetic
- Thermal dwell sized to part mass and verified
- Retest and arbitration rules fixed before launch
Governance: Keeping the Argument Honest
Once the correlation exists, protect it. Master parts — golden samples with certified leak rates, or at minimum a non-leaking reference and a known leaker near the limit — must be run at shift start, after any fixture change and after calibration. Log those results as a chart. When the tester drifts, you want to see it in the master data that morning, not in a field return eighteen months later. Fixture seals, the highest-wear item on any leak tester, deserve a scheduled replacement interval based on cycle count, not visible deterioration.
Review the limit annually against warranty and field data. If no field leaks have ever occurred on parts rejected well above the limit, the margin conversation is legitimate; if field leaks occur on parts that passed, your method or your limit is wrong and the correlation must be revisited. The limit is not scripture. It is an engineering estimate attached to a specific method, fixture and thermal environment, and it inherits all their weaknesses.
The disputes that survive the standing rule are worth having, because they are now about something real: the relationship between a pressure transducer's reading on a Tuesday afternoon and a customer's coolant puddle five years later. That relationship is the whole job of leak testing. Everything else — the instruments, the arguments, the decimal places — is instrumentation around it.