Measuring leak rate: eight decades between pressure decay and helium

Hardly any requirement is adopted as unthinkingly as the leak-tightness requirement. "Leak rate below 10⁻³" appears in the functional specification, and nobody asks: at what test pressure, with which gas, over what period, at what temperature stability. Yet it is precisely these four details that decide whether the requirement can be met with a pressure decay test bench costing ten thousand euros or with a helium leak detector costing ten times as much.

This paper ranks the methods by their documented detection limit, explains the conversions between gases and units, and shows with a worked example why the dominant source of error in the most widely used method is not the leak but the temperature.

01 · What a leak rate is

The definition is unspectacular and is nevertheless rarely applied cleanly: a leak rate of 1 mbar·l/s is present when the pressure in a closed, evacuated vessel of one litre volume rises by one millibar per second.^[1] The leak rate is therefore a pV throughput and not a mass — it therefore depends on temperature and gas species, and both belong in the specification.

The standard conditions usually applied are an inlet pressure of 1013.25 hPa against roughly zero hPa outlet pressure at room temperature.^[2] This is where the first pitfall already lies: vacuum technology usually calculates at 20 to 25 °C, while the ASTM practices for mass spectrometers state standard cubic centimetres at 0 °C.^[3] Between the two lie seven to nine per cent. For an acceptance limit that is not an academic point.

How tight is tight enough?

For classification, a table of criteria has become established that has been passed on in the vacuum literature for decades.^[1] It is not a standard, but it places the orders of magnitude correctly:

The last row deserves a note. At the lower end of detectability, at about 10⁻¹² mbar·l/s, that would correspond arithmetically to an opening of one ångström.^[1] At that point no method is measuring a material defect any more, but the passage of gas through the intact wall. A leak-tightness requirement of this order of magnitude is therefore a materials requirement and not a manufacturing one.

02 · The standards situation — and its gaps

Anyone invoking a standard should check whether it still exists. The European family of standards on leak testing has been split since 2017: one part was raised to ISO level, another part still stands unchanged at the 1999 and 2001 editions.

On the American side the practices for mass spectrometers are current and state sensitivity ranges explicitly: ASTM E493/E493M for the inside-out testing mode with 10⁻¹⁰ to 10⁻⁶ Pa·m³/s for helium^[3], ASTM E498/E498M for the tracer probe from 1 · 10⁻⁸ Pa·m³/s^[16] and ASTM E499/E499M for the detector probe from 1 · 10⁻⁷ Pa·m³/s.^[17] The frequently cited guide ASTM E479 on preparing a leak testing specification, by contrast, is inactive.^[18]

03 · The methods and their detection limits

The following overview brings together values from three independent manufacturers' compendia and one technical publication.^{[1] [2] [4] [19]} They scatter in places by an order of magnitude — that is itself a result and not a shortcoming of the presentation. Where a source distinguishes between the theoretical limit and the limit achieved in practice, both are entered.

Three consensus ranges can be read from the figure that can be written into a requirement specification with a clear conscience. The pressure decay method with air reaches around 10⁻⁴ mbar·l/s in theory, but realistically 10⁻³ to 10⁻² mbar·l/s in industrial use.^[4] Helium sniffer testing lies at 10⁻⁷ to 10⁻⁶ mbar·l/s. Helium integral testing in a vacuum chamber reaches around 10⁻¹² mbar·l/s — stated consistently by all three compendia.^{[1] [2] [4]}

The particularities of the individual methods

Pressure decay. Simple, fast, cheap — and temperature-sensitive to the point of being unusable. See section 04. Further documented limitations: water condensation falsifies the measurement, and in vacuum technology the method is practically irrelevant.^[1]

Pressure rise under vacuum. According to the manufacturer it avoids the temperature effects^[4], but takes on wall outgassing in exchange. The distinguishing criterion is elegant and belongs in every test specification: if the time for a given pressure rise stays constant, there is a leak. If it grows longer, it was outgassing.^[1]

Bubble testing. Vivid and covered by standards, but with two catches. The wet test item has to be dried afterwards^[4] — a real time factor in cycle-based production. And with foam spraying the result depends on the attentiveness of the inspector^[4]; that is not a process-capable method.

Acoustic emission. ISO 18081 states a fundamental physical limitation: turbulent flow produces detectable emissions, laminar flow as a rule does not.^[15] Gas leaks give stronger signals than liquid leaks. In two-phase flow, cavitation produces signal peaks that lie at least an order of magnitude above the turbulence signals.^[15]

Forming gas instead of helium. A mixture of 95 % nitrogen and 5 % hydrogen reaches leaks of 5 · 10⁻⁷ mbar·l/s and has three documented advantages over helium: the atmospheric background is 0.5 ppm compared with 5 ppm for helium — a starting point ten times better for the signal-to-noise ratio. The dynamic viscosity is 8.7 µPa·s compared with 19.4 µPa·s for helium. And the gas does not adhere to surfaces and can be purged out considerably faster, which reduces carry-over and purge times. The mixture is classified as non-flammable under ISO 10156 and costs a fraction of helium.^[19]

Dye penetrant testing. Systematically it does not belong here. ISO 3452-1 is a surface crack testing method^[20] and delivers no quantitative leak rate — it locates, it does not measure.

04 · The pressure decay method and temperature

The basic equation is trivial: a volume V loses, at a leak rate q , over the time Δt , a pressure equivalent Δp = q · Δt / V.^[1] The difficulty lies elsewhere — namely in the fact that a temperature change in the closed volume produces the same effect as a leak, only with the opposite sign and usually larger.

From the ideal gas law at constant volume it follows directly that Δp/p = ΔT/T. At room temperature that is 0.34 % of the absolute pressure per kelvin — a value confirmed by two independent sources.^{[21] [22]} At a test pressure of 1 bar absolute, one kelvin therefore corresponds to around 3.4 mbar, at 2 bar absolute to around 6.8 mbar.

A worked example shows how this plays out in a real test: test volume 110 cm³, test pressure 2000 mbar absolute, 23 °C, permissible leakage 0.5 cm³/min, test time 5 s.^[21]

The second limit: the resolution of the sensor

Even without temperature drift the method ends at the resolution of the pressure transducer. An application report puts this unmistakably with an example: a calculated pressure drop of 7.2 · 10⁻⁴ mbar cannot be resolved on a 5 bar scale, because measuring instruments with this resolution are not available.^[23] Around 0.1 Pa is stated as the practical lower limit, that is 0.001 mbar.^[24]

From this follows directly — and this is our derivation, not a formula from a standard — the smallest detectable leak rate as q_min = V · Δp_resolution / Δt. It is proportional to the test volume, inversely proportional to the test time and proportional to the sensor resolution. That is precisely why the method breaks down with large volumes, and precisely why the make-up flow method has the advantage with large test items: there the throttling effect of the laminar flow element governs the test time and no longer the volume.^[22]

05 · Converting between gases, units and media

Units

From gas to gas

Which law applies depends on the flow regime in the leak.^[1] In laminar viscous flow the leak rates are inversely proportional to the dynamic viscosity, in molecular flow inversely proportional to the square root of the molar mass. The transition lies roughly between 10⁻⁶ and 10⁻⁴ mbar·l/s^[1]; a second source gives 10⁻⁴ mbar·l/s as the middle of this range.^[2]

For the most common pair the factors are documented: laminar, q_air = 1,08 · q_helium, molecular q_air = 0,37 · q_helium.^[2] Read differently, and more important in practice: In the molecular regime the measured helium leak rate is around 2.7 times greater than the corresponding air leak rate. Anyone testing an air requirement with a helium leak detector has to know this factor — otherwise they reject parts that meet the requirement.

From gas to liquid — the capillary criterion

A conversion from gas leak rate to water leak rate does not exist in the literature, and that is physically consistent: for liquids neither of the two gas flow laws applies. Instead a capillary criterion governs the largest leak diameter that a liquid will still hold back:^[2]

The documented numerical examples make the difference tangible: at 0.3 MPa pressure difference, water is held back in leaks below about 1 µm in diameter, a urea solution such as AdBlue by contrast only below 0.2 µm — its surface tension is lower than that of water by more than a factor of 50. It therefore creeps through openings that water closes off.^[2]

Refrigerant loss into leak rate

For refrigeration and heat pump applications the requirement is usually formulated as mass loss per year. The conversion is made via q = (m/M) · R · T / t. For R134a one source gives 1 g per year as corresponding to 7.6 · 10⁻⁶ mbar·l/s^[4], another 6.5 · 10⁻⁶ mbar·l/s.^[1] Recalculating with the quantities the second source itself states — molar mass 102.03 g/mol, 298 K — gives 7.7 · 10⁻⁶ mbar·l/s and thus supports the first value.

06 · Calibration: the test leak

Every quantitative leak rate measurement depends on a reference leak. ISO 20486 governs its calibration^[14] and supplies the figures most important for practice:

The difference in the temperature coefficient amounts to an order of magnitude and is thus the most important selection criterion: a permeation leak in a hall without air conditioning already takes on ±17.5 % uncertainty at ±5 K — more than the typical calibration uncertainty. NIST independently confirms the order of magnitude with up to 4 % per degree Celsius for helium permeation leaks.^[25]

ISO 20486 also states handling requirements that are regularly disregarded in practice: temperature equalisation at least twelve hours before the measurement, warm-up time of the leak detector at least two hours, and in vacuum operation at least 30 minutes of pumping down to remove accumulated tracer gas.^[14] The reference conditions are 20 °C and 1000 mbar outlet pressure.

On the uncertainty achievable: for a calibrated helium leak at 23 °C one manufacturer states an expanded measurement uncertainty of ±5.7 % at 95 % coverage probability, with a specification tolerance band of 20 % — giving a test uncertainty ratio of 3.5 to 1. The certificate is valid for at least one year; the leak rate falls by a small percentage each year because the partial pressure difference to the surroundings decreases.^[26]

Such leaks are traceable through the national metrology institutes. NIST calibrates leak standards from 1 · 10⁻⁶ to 1 · 10⁻¹³ mol/s, with temperature-dependent measurements over 0 to 50 °C.^[25]

07 · What is required in our fields

Refrigeration and heat pumps: the F-Gas Regulation

Regulation (EU) 2024/573 entered into force on 2024-03-11.^[27] For operators of stationary equipment it lays down leak checks at graduated intervals:^[28]

From 500 t CO₂ equivalent a leak detection system is mandatory that alerts the operator or a service company; it has to be checked at least once a year.^[28] For test bench design this leads to an observation that is rarely voiced: the entry threshold of 5 t CO₂ equivalent takes effect quite differently with modern refrigerants. With R1234yf, which has a very low global warming potential, it is effectively never reached; with older mixtures it is reached at a few kilograms.

In the series production of refrigeration and air conditioning appliances, a refrigerant loss of 0.5 g per year has become established as the target figure, corresponding to the middle 10⁻⁶ range in standard cubic centimetres per second — vividly, about one bubble three millimetres in diameter under water, every 1.2 hours.^[29]

Liquid-cooled electronics

The Open Compute Project requires for cold plates a pressure decay test to EN 1779 with an acceptance criterion of a pressure drop below 0.5 %, supplemented by an immersion bath bubble test. In addition there is a hydrostatic test at the maximum operating pressure over five minutes and a burst test at three times the operating pressure over two minutes.^[30] The associated requirement specification sets two sets of rules against each other: IEC 62368-1 requires leak tests at three times the normal operating pressure and at twice the pressure in the fault case, ASME B31.3 at 1.5 times the design pressure.^[31]

08 · How to write a leak-tightness requirement

A short list follows from the above. It is our conclusion, not that of a standard.

1. Numerical value with unit and tracer gas. "10⁻³ mbar·l/s helium" and "10⁻³ mbar·l/s air" are requirements that differ by a factor of 2.7 in the molecular regime.^[2]
2. Test pressure and pressure direction. Positive pressure from inside or vacuum from outside — that changes method and result.
3. Test time and settling time. With the pressure decay method, both together with the volume govern the detection limit.
4. Temperature condition. Permissible drift during the test, in kelvin. Without this detail a pressure decay requirement below about 10⁻² mbar·l/s cannot be tested.^[21]
5. Reference temperature of the leak rate figure. 0 °C or 20 °C — the difference is seven to nine per cent.^[3]
6. Calibration evidence. Type of test leak, calibration interval and temperature coefficient. With a permeation leak, the ambient temperature belongs in the test specification.^[14]
7. Order in the test sequence. Leak testing before any wet test.^[2]

And a recommendation on the choice of method: anyone setting a requirement below about 10⁻³ mbar·l/s should no longer consider pressure decay. The gap between the pressure-based and the tracer-gas-based methods is real and amounts to several decades — it cannot be closed by longer test times, only by a different method.

09 · Sources

All references were accessed in August 2026. Where standard texts lie behind paywalls, the catalogue entries of the publishers are given; figures from such standards come from freely available preview versions and are to be checked against the original copy before any binding use.