Ask ten people who have spent time building and operating lab reactors what leak rate is acceptable and most will start with some version of “zero.”
That is the right goal. It is less useful as an acceptance criterion for a real reactor system you have to run on Monday.
For the small lab reactor systems I have worked with, the criterion that has held up is:
No more than 1% of test pressure per hour, temperature-corrected, over a one-hour hold.
At 100 psig, that means no more than a 1 psi pressure loss in an hour.
I did not arrive at that number on my own. Across four sites and two large companies, roughly three quarters of the experienced reactor people I have worked with hold that line. Most of the rest run closer to 2%. Nobody credible runs zero, because no real reactor system holds zero.
What is interesting is that 1% is not written in any code. It has converged across facilities that share almost nothing else with each other, which suggests it is tracking something real. The second half of this article is about what that something is.
Scope and limitations
A few things this article is not, stated once so I do not have to keep qualifying:
This is not a code requirement or a universal standard. It is a practical rule for routine leak checking of assembled laboratory reactor systems. Different facilities, services, reactor volumes, test methods, and hazards may justify different criteria. Where your site has an approved procedure, follow it.
A leak test is not a pressure-rating or proof test. This article assumes the reactor, tubing, valves, fittings, and other pressure-containing components are properly rated for the service, and that any required hydrostatic, pneumatic, proof, or code testing has already been completed. Never exceed the allowable pressure of the lowest-rated component in the system.
A practical leak-test rule does not replace a hazard analysis. For toxic, highly flammable, or otherwise high-consequence service, the allowable release has to come from the actual analysis. More on that below.
What follows is about determining whether an assembled reactor system is leak-tight enough to operate, and more importantly, about what has to be true for a pressure-decay test to tell you anything useful at all.
The Test Conditions Matter More Than the Number
A leak-rate criterion without stated test conditions is not much of a criterion. Several conditions do most of the work.
Test Cold
When you load catalyst, replace a reactor, or break into the process tubing, the system is normally cold. You have already removed insulation or heat tape and cooled the furnace because that is the only practical way to work on it.
Once the system is reassembled there can be a temptation to heat it before leak testing, since that is where it will ultimately operate. For routine turnaround work, test it cold first.
The practical reason is simple. If you heat the entire system and then discover a leak, you now have to cool it, repair it, retest it, and heat it again. A problem that could have taken twenty minutes can consume most of a day.
Cold testing is also a useful baseline for many of the fittings and seals found on lab reactors. PTFE expands considerably more than stainless steel as temperature increases, so in PTFE-packed valves and similar seals the first heat-up can increase sealing compression rather than reduce it.
That does not mean a system that is tight cold is guaranteed to remain tight hot. Thermal gradients, differential expansion between materials, tubing movement, gasket relaxation, valve construction, and repeated thermal cycling can all create leaks that were not present during the initial cold test.
The better way to think about it:
A successful cold test is the first leak-tightness check, not proof that the system will remain leak-tight at temperature.
PTFE also creeps under compression. A joint or packing arrangement that was tight before its first heat cycle may behave differently after repeated cycles. That is one reason historical leak-test data becomes useful. The cold test before run twenty is not necessarily the same test as the cold test before run one.
Test at the Pressure That Matters
Test at the pressure the system will actually experience during operation.
Testing substantially below operating pressure can make a leak appear smaller than it will be in service. For an idealized laminar leak through a small channel, leakage scales approximately with the difference of the squares of the absolute pressures. Testing a 100 psig system at only 50 psig can understate leakage by considerably more than a simple factor of two.
If your available inert-gas supply limits the pressure at which you can test, recognize the limitation rather than treating the lower-pressure test as equivalent. Document the test pressure and interpret the result accordingly.
Watch the relief device. If you are testing at operating pressure and the relief is set only slightly above it, a soft-seated relief valve can begin to simmer well below its set point, sometimes around 90%. That presents as a steady, unfindable pressure decay, because the gas is leaving through the vent line rather than through any joint you can reach with a bubble solution. If a hold test fails and every accessible connection checks clean, the relief path is one of the first things to rule out.
Correct for Ambient Temperature
This is the one that quietly invalidates a lot of informal leak testing.
For a blocked-in gas volume of essentially constant volume:
$$ \frac{P}{T} \approx \text{constant} $$
Pressure and temperature move together. At 100 psig the absolute pressure is approximately 114.7 psia, and near room temperature a change of only 1°C can move the indicated pressure by roughly 0.4 psi. A lab that swings 2°C during an HVAC cycle can therefore move the reading by almost 0.8 psi even if no gas has escaped.
If your acceptance criterion is 1 psi per hour and you are not accounting for temperature, you may not be measuring leakage at all. You may be measuring the building.
Log temperature alongside pressure and normalize using absolute pressure and absolute temperature:
$$ P_{corrected}=P_{measured}\frac{T_{reference}}{T_{measured}} $$
Do not make this correction using psig and °C directly.
Also remember that ambient temperature is only a proxy for gas temperature. If the reactor, tubing, or surrounding equipment is still thermally equilibrating, the gas temperature will lag the room. That is one more reason the shape of the pressure trend matters as much as the endpoints.
Use a Logging Transducer, Not Just a Gauge
For pressure-decay testing, do not choose the transducer based on the headline accuracy class.
You are measuring a change in pressure over time, not an absolute pressure. A fixed zero offset largely cancels in a delta measurement: a transducer reading 2.5 psi high at both 100 psi and 99 psi still reports the 1 psi change correctly. A modest span error has far less effect on a small delta than on the absolute reading.
Three other specifications matter more.
Resolution
Resolution is often as much an acquisition-system problem as a sensor problem, and it is the part people skip.
Take a 0-1000 psi transducer on a 4-20 mA output. Digitized by a 12-bit card, that is roughly 0.24 psi per count, which is marginal when your entire acceptance criterion is 1 psi. The same signal into a 16-bit card gives you around 0.015 psi per count and the measurement becomes genuinely useful.
People spend a great deal of time specifying the pressure sensor and very little looking at what is digitizing the signal.
Short-Term Repeatability
Random scatter is less damaging than it appears, provided you log continuously.
Instead of recording one pressure at the start and one at the end, collect samples throughout the hold and fit the trend. Uncertainty in a fitted slope falls with the square root of the sample count, so a transducer with some noise but little drift still provides an excellent pressure-decay measurement.
Continuous logging also shows you the shape of the curve, which two endpoints cannot. A system still reaching thermal equilibrium produces a pressure change that gradually flattens. A stable leak over a narrow pressure range produces an approximately steady downward trend. A leak that is changing with pressure, temperature, or mechanical movement produces curvature.
Thermal Drift
This is the specification that can quietly dominate the measurement.
A transducer may have an excellent room-temperature accuracy rating and still show a meaningful zero or span shift as its own temperature changes. Unlike a fixed offset, this does not cancel in a delta, because it varies across the measurement window. And unlike random noise, it does not average out under regression, because it is correlated with time rather than random.
Consider a combined thermal effect of 0.02% of full scale per °C:
- A 1000 psi full-scale instrument gives 0.2 psi/°C.
- A 200 psi full-scale instrument gives 0.04 psi/°C.
Over a 2°C change, that is roughly 0.4 psi versus 0.08 psi of apparent pressure movement, against a 1 psi criterion. Both instruments could carry an identical headline accuracy rating.
There is a further subtlety. Sensor thermal drift correlates with the same room-temperature change you are using to correct the gas pressure, so depending on the direction of the sensor’s coefficient, your temperature correction may partially hide or partially exaggerate the instrument effect. You cannot separate the two from the pressure record alone.
The way out is not a more elaborate correction. It is a transducer reasonably ranged for the pressure being measured. For a 100 psig test, a 0-200 or 0-300 psi transducer is far more useful for decay work than a 0-1000 psi unit, assuming it remains appropriately rated for the application.
A conventional Bourdon gauge with coarse graduations is a poor tool here for its own reasons: resolution you cannot read to a tenth of a psi, plus mechanical hysteresis and stiction.
Let the System Reach Thermal Equilibrium
Do not start the one-hour clock immediately after pressurizing.
Rapidly compressing gas into a small volume raises its temperature. As the gas and tubing return toward room temperature, the pressure falls, and that drop looks exactly like a leak if all you are watching is the gauge.
On many small systems something on the order of 15 to 30 minutes is sufficient, but the pressure trace is a better guide than the clock. Start evaluating the leak rate once the rapid equilibration portion of the curve has largely flattened.
Block In the Section You Worked On
One of the highest-value design changes you can make to a lab reactor system is adding an isolation valve shortly upstream or downstream of the point you routinely disconnect.
The reason is sensitivity. Pressure-decay sensitivity is strongly related to the volume being tested. Suppose the full process system holds 1000 cc of gas but the reactor section you actually opened is only 150 cc. For the same physical leak, the smaller blocked-in section shows roughly seven times as much pressure movement.
You are not measuring the leak more carefully. You are testing a smaller box. That makes small leaks much easier to see.
There is a second advantage. If the rest of the system stays isolated under inert pressure while the reactor section is open, you reduce the equipment that has to be repurged and reduce the surface area through which air can enter.
But the isolation valve is now part of your test boundary, so its seat leakage is part of the measurement. Soft-seated valves are often useful here because they can provide very low seat leakage, but do not assume any valve is perfectly tight simply because it has a soft seat. Likewise, do not assume a metal-seated valve leaks significantly. Look at the actual seat-leakage specification.
Put the isolation point reasonably close to the connection that is routinely disturbed. There is little benefit in testing several feet of untouched tubing every time you replace a reactor.
Internal Leakage Reads the Same as External Leakage
This is worth stating generally, because it is where failed hold tests go to die.
Your transducer knows only that gas left the box. It cannot tell you whether it left through a fitting or through a valve seat into a lower-pressure or vented section. Needle valve seats, soft-seated check valves, and three-way valves passing to a vent line all produce the same trace as a leaking union.
So when a section fails the hold and a careful joint-by-joint bubble check comes up clean, stop re-checking the same fittings. Start subdividing the volume with block valves and re-testing the pieces. The answer is usually a seat, not a connection.
The Gas You Test With Is Not Necessarily the Gas You Run
Most reactor systems are leak tested with an inert gas such as nitrogen. The process gas may behave differently, and hydrogen is the important example. There are two separate effects.
Leakage Through an Actual Leak Path
For an idealized leak path in the viscous-flow regime, leakage is influenced by gas viscosity. At approximately room temperature, nitrogen is around 18 µPa·s and hydrogen around 9 µPa·s, so for the same geometry and pressure conditions hydrogen could flow through a viscous leak path at roughly twice the nitrogen rate.
In the molecular-flow regime the relationship depends on molecular mass instead:
$$ \sqrt{\frac{28}{2}}\approx3.7 $$
Real fitting leaks do not behave as perfect capillaries or pure molecular-flow orifices. They may fall between regimes, or change behavior as the joint moves. So the useful conclusion is not that 1 psi/hr on nitrogen becomes a specific number on hydrogen. It is:
A nitrogen leak test can be optimistic for hydrogen service, plausibly by a factor of two to four depending on the flow regime.
That is one more reason I prefer 1%/hr to a looser 2% criterion. The margin is not really there once you convert it.
Permeation Through Seal Materials
Permeation is a different phenomenon. Gas does not have to travel through a mechanical gap; it can dissolve into a polymer and diffuse through the material itself.
That process does not follow the viscous- or molecular-flow relationships at all. It depends on the specific gas, polymer formulation, temperature, pressure, thickness, and geometry, and hydrogen can permeate PTFE and elastomers substantially differently from nitrogen.
The consequence is that a nitrogen pressure-decay test is structurally blind to the permeation behavior of your polymeric sealing elements in hydrogen service. There is no correction factor that fixes this.
The solution is material selection and an appropriate commissioning procedure for the actual process gas: selecting suitable valves and seals, minimizing unnecessary polymeric sealing surfaces where practical, using appropriate hydrogen detection, and following the site’s approved procedure when introducing hydrogen into a newly assembled or disturbed system.
Helium Is Useful for a Different Reason
Helium is sometimes described as a good hydrogen substitute because it is a small molecule. That is only partly true. In molecular flow it is closer to hydrogen than nitrogen is, but in viscous flow helium’s viscosity is actually slightly higher than nitrogen’s, which makes it a worse stand-in, not a better one.
The real advantage of helium is detectability. A helium leak detector can identify extraordinarily small leaks and, more importantly, tell you where they are. That turns “the pressure transducer says something is leaking” into “it is this union,” and those are very different troubleshooting problems.
For systems where extremely low leakage matters, helium testing is valuable even though helium is not a perfect physical stand-in for the process gas.
Why 1% and Not 0.1% or 5%?
Here is the part that usually gets skipped, and it is why the number is stable across facilities that share nothing else.
For a routine build-quality leak check, the criterion is set by findability, not by the hazard limit.
Work it out. A 150 cc blocked-in section losing 1 psi in one hour corresponds to about 2.8 × 10⁻³ atm·cc/s, or roughly 245 cc/day at atmospheric conditions. Bubble leak-detection solutions are generally capable somewhere in the range of 10⁻³ to 10⁻⁴ atm·cc/s under good conditions.
Those two numbers sit right on top of each other. That is not a coincidence, and it is the whole argument: the acceptance criterion has converged on the detection floor of the method people actually use to find the leak.
The consequences of getting that relationship wrong run in both directions. Set the criterion well below your leak-location capability and you create an ugly situation: the system fails, you know something is leaking, and you cannot find it. Someone spends two hours checking every fitting, finds nothing, and eventually runs the unit anyway because it always does this. A criterion that is routinely ignored is not a criterion. Set it well above your capability and you are accepting defects you could have found and repaired in twenty minutes.
So the design principle behind the 1% rule is:
Set the routine build-quality criterion at a level that reliably identifies meaningful assembly problems and still gives the person troubleshooting a reasonable chance of finding them.
This also explains the 2% camp, and they deserve more credit than they usually get, because percentage pressure decay is volume blind. A 1% loss from 1000 cc is far more gas than a 1% loss from 150 cc. Someone testing a whole unit at 2%/hr may be holding a physically tighter standard than someone testing an isolated section at 1%/hr. They are not necessarily looser, they are expressing the same thing in a unit that hides the difference.
Which is the practical fix: specify both the allowable pressure decay and the approximate isolated volume. The criterion becomes portable between people and between systems.
What the Criterion Actually Protects
For the small blocked-in reactor volumes discussed here, 1%/hr will often sit well inside the leak rate a properly designed ventilated enclosure can accommodate. That does not make it a safety limit.
The acceptable release from any particular reactor system has to come from the hazard analysis: gas composition, toxicity, flammability, operating pressure, inventory, enclosure volume, ventilation rate and reliability, detection, ignition sources, and failure scenarios. A practical leak-test rule cannot replace that.
What the 1% criterion is good at is something else entirely. It catches defects long before they become obvious operating problems.
Data Quality
On a continuous reactor with mass flow controllers setting feed and a GC closing the material balance, a leak creates an unmetered flow path. The result is often not an alarm. It is a mass balance that never quite closes, or a conversion calculation that is consistently wrong.
Those are dangerous in a different sense. You can generate weeks of apparently good experimental data before realizing the reactor was never operating under the conditions you thought it was.
Bed Integrity
A path that lets gas out under positive pressure can let air in during a cooldown or any subatmospheric excursion. For an air-sensitive or reduced catalyst, a small leak can destroy a run without ever becoming a personnel hazard.
Build Quality
Most leaks are not mysterious porosity. They are something specific: a fitting that was not assembled correctly, a ferrule reused one time too many, a valve packing arrangement that needs attention, a damaged gasket surface, a connection disturbed during reactor replacement.
The leak test forces you to inspect your own workmanship before starting the experiment. On a system where most of the joints disappear behind insulation and enclosure panels the moment you close it up, nothing else forces that check.
That is the strongest argument for the criterion. It is a build-quality gate, and the fact that it sits well inside the hazard limit is not a weakness. A criterion set at the enclosure’s actual tolerance would pass a unit that is coming apart.
Two Refinements Worth Adopting
Distinguish a Step Change From a Slope
Every reactor develops a history, and if you record the decay result each time the system is opened, that history becomes useful.
Most drift is benign and not worth chasing. Packing creep, valve wear, and normal aging produce slow, monotonic changes over many cycles, and the right response is to keep running until the unit stops holding at an acceptable level.
A sudden change is a different animal. If a section has repeatedly tested around 0.3%/hr for six catalyst changes and then tests at 0.9%/hr, it still passes a 1% criterion, but creep does not triple in one cycle. Something happened, and right now you know exactly which connections you touched. The repair probably takes twenty minutes today. If you wait until the system finally exceeds the acceptance limit several months later, you will no longer know where to look.
The distinction worth acting on is not upward drift. It is a step change that the slope does not explain.
That is the real argument for logging the number rather than writing PASS on a checklist.
Derive a Tighter Criterion for Higher-Consequence Service
Do not automatically apply the general 1% rule to every gas.
For gases with low occupational exposure limits, unusual toxicity, or other high-consequence properties, determine the allowable leak rate from the hazard analysis: release rate, operating pressure, inventory, enclosure, ventilation, detection, and the applicable exposure or safety criteria. Then verify your leak-test method is sensitive enough to detect leakage comfortably below that value.
The distinction matters:
The routine 1% criterion is a build-quality rule. It is not a substitute for a consequence-based leak limit where one is required.
The Short Version
For routine pressure-decay leak testing of a small lab reactor system:
- Test the system cold, before heat-up.
- Use an inert test gas appropriate for the equipment and procedure.
- Test at the relevant operating pressure, and never above the rating of the lowest-rated component.
- Isolate the smallest practical section containing the connections that were disturbed.
- Let the gas and equipment reach thermal equilibrium before starting the clock.
- Use a properly ranged logging transducer, and check what is digitizing the signal.
- Record temperature alongside pressure, and correct using absolute pressure and absolute temperature.
- Hold for one hour.
- Accept 1% of test pressure per hour, and record the isolated volume alongside it so the number means something to the next person.
- Compare the result against that section’s historical baseline.
- Derive a tighter criterion where the process hazard requires one.
And remember what the number is telling you.
It is not proof that the reactor is safe. Safety comes from proper design, pressure ratings, materials, relief protection, ventilation, detection, operating procedures, and hazard analysis.
The leak test answers a narrower and extremely useful question:
Did I put this reactor system back together well enough that I should trust it?
That is the question worth answering before you start the run.