How to Put a Number on What Leaks Are Costing

Why this matters

Nobody can measure a leak load. There is no meter you can hang on the holes. What you can do is infer it two different ways, and the value of doing it twice is not confidence, it is the disagreement: when the two methods land on different numbers, the gap points at something real that one of them saw and the other did not. A shop that runs both, states the setpoint the number belongs to, and hands over a bound rather than a rate is doing work that stands up when the customer's engineer looks at it.

Before either test

Both tests run with the system fully pressurised and nothing opened. Nobody works on piping while either one is in progress.

To stop the compressor for the decay test, isolate and lock its energy sources under 29 CFR 1910.147 first, because a packaged compressor restarts on its own pressure switch and a machine that looks off is not off. The electrical isolation is its own hazard: de-energize at the disconnect and lock and tag it under 29 CFR 1910.333(b)(2), with 29 CFR 1926.417 as the construction counterpart, and prove dead live-dead-live per NFPA 70E-2021, 120.5, in the edition the employer's electrical safety programme has adopted.

To valve off a drop, close the shutoff and bleed the hose through the tool's own trigger with the tool pointed into a clear area and hands and face out of the path before any coupler is pulled. Never uncouple a pressurised hose, never locate a leak by putting a hand in the jet, and never use a blow gun to clear anything: air driven into a break in the skin can enter the bloodstream, and 29 CFR 1910.242(b) permits compressed air for cleaning only when reduced to less than 30 psi with effective chip guarding and personal protective equipment. Where the survey takes you into an air room or near running equipment at the 85 dBA eight-hour time-weighted average action level, hearing protection and the rest of 29 CFR 1910.95 apply.

Do the ultrasonic walk with every enclosure closed. If a leak is suspected inside an energized enclosure, that is qualified electrical work under the standards above, not a leak survey.

Method one: load and unload duty in a no-production window

With production down and every tool valved off, run the compressor normally for a fixed window and record how long it spends loaded. Leak load is the machine's rated output multiplied by that duty fraction.

What it counts. Everything the compressor had to replace, which includes leaks, any dryer purge that runs on cycle, any timer drain, and anything left switched on that you did not find. That makes it an upper bound on leakage, not a leak rate, and it is quoted as a bound.

Its dominant error. The timing of the load transitions, which is a fixed-count bound and does not shrink with the reading, so it matters more on a short window than a long one.

Method two: pressure decay from a sealed system

Bring the system to its normal top pressure, isolate and lock out the compressor, valve off every drop, and time the fall from one pressure to a lower one.

Leak flow in scfm is the total system volume in cubic feet, times the pressure fall in psi, divided by 14.7 and by the elapsed minutes. This is the same free-air storage relationship used for receiver sizing, run backwards.

Its condition. It assumes the air stays near ambient temperature through the decay, which is true for a fall taking minutes and not true for a fast blowdown taking seconds.

What it gives you is an average, not a setpoint figure. Leak flow is proportional to absolute upstream pressure, because above roughly 13 psig an opening to atmosphere is choked, so the flow at the top of the band is higher than at the bottom. The number the arithmetic returns belongs to the average absolute pressure across the band, and it has to be scaled up to the operating setpoint before it is quoted next to anything else.

Its dominant error. Total system volume, because the piping volume is estimated and the receiver volume is not. That estimate is a bound on the answer in direct proportion.

The plant, run both ways

A 120 scfm rotary screw at a 105 psig cut-out and a 90 psig cut-in. A 400 gallon receiver, which is 53.5 cubic feet, plus a piping volume estimated from the main's length and bore at 25 cubic feet, giving 78.5 cubic feet total.

Decay test. Sunday morning, every drop valved off, compressor isolated and locked out. The system falls from 105 psig to 90 psig in 9.5 minutes.

78.5 x 15 / (14.7 x 9.5) = 1177.5 / 139.65 = 8.43 scfm, and that figure belongs to the average absolute pressure across the band, which is (119.7 + 104.7) / 2 = 112.2 psia. Scaling to the 105 psig setpoint, which is 119.7 psia: 8.43 x 119.7 / 112.2 = 9.0 scfm at 105 psig.

Its bound comes from the volume estimate. Allow 40 percent either way on the 25 cubic feet of pipe and the total moves by 10 cubic feet, which is 12.7 percent of 78.5, so the decay answer is bounded between 7.9 and 10.1 scfm.

Duty test. Same morning, same valve positions. Release the lockout by the procedure that applied it, area checked clear, guards in place, each lock removed only by the person who put it on, then restart and leave the machine to run one hour. It loaded three times for a total of 5.5 minutes out of 60, a duty of 9.17 percent. 120 x 0.0917 = 11.0 scfm at 105 psig.

Its bound comes from timing three pairs of transitions to about 5 seconds each, roughly 15 seconds on 330, which is 4.5 percent, so the air the compressor replaced was 10.5 to 11.5 scfm, which puts leakage at not more than 11.5 scfm and possibly well under it, because the same figure also counts purge, drains and anything left switched on.

The two measurements do not overlap. 7.9 to 10.1 scfm escaping the decay volume, against 10.5 to 11.5 scfm being replaced. Both of those are two-sided measurements of air, and they disagree; reading the duty figure as an upper bound on leakage is a separate step and it does not soften the disagreement. This is the whole reason for running two methods: the duty test is counting something the decay test could not see.

Finding what the gap was

Go back to the boundary each method drew, because that is the only place a difference of this kind can live.

The decay test isolated the compressor package, and the package's discharge check valve held the receiver's pressure back out of it. So anything leaking inside the package, between the airend and that check valve, was outside the decaying volume and invisible to the test. During the duty test the package was running, and any such leak was being replaced by the machine itself and counted in the loaded time.

That put an ultrasonic detector on the package with the machine running, everything closed, standing outside the enclosure. The blowdown vent was discharging continuously rather than only on unload, which is a weeping blowdown valve and a leak inside the compressor's own boundary.

Two scfm, the difference between the methods, is 1.7 percent of the machine's 120 scfm output. Small enough that no one would have gone looking for it, and it is the term that made two honest tests disagree.

Turning the bound into something a customer can act on

The number is not a percentage until you say a percentage of what, and the two available bases give very different answers that are both true.

Against rated output. 11.0 scfm against 120 scfm is 9.2 percent of the machine's capacity. That is the figure that tells you whether the plant is short of air, and 9 percent sounds tolerable.

Against air actually produced. The plant is pressurised all 168 hours of the week, so leaks consume 11.0 x 60 x 168 = 110,880 standard cubic feet. The machine delivers 120 x 60 = 7,200 standard cubic feet per loaded hour, so that is 15.4 loaded hours of full output per week. Measured loaded time for this plant is 40 hours a week, so leaks account for 38.5 percent of the air the compressor actually made.

Both are correct and neither may be quoted without its base in the same sentence. The gap between 9.2 and 38.5 percent is not a discrepancy; it is the fact that leaks run all week and production does not, and it is the single most persuasive number in the whole survey.

The setpoint travels with the number. Quote it as 11.0 scfm at 105 psig, always. Drop the setpoint to 95 psig and the same holes pass 109.7 / 119.7 = 91.6 percent of that, so 10.1 scfm, with no repair performed. That is only half of what the 10 psi buys. The same reduction cuts the compressor's specific power on every unit of air it makes, about 5 percent at the commonly published 1 percent per 2 psi near 100 psig on a lubricated rotary screw. Pressure costs twice, more mass through every fixed hole and more shaft work per unit made, so a setpoint reduction is worth roughly double what the leak arithmetic alone shows and it is worth quoting alongside the repair programme rather than instead of it. A leak figure with no setpoint attached cannot be compared against a later one.

The tag list will not add up to the aggregate, and should not be forced to. Walk the plant, tag each leak, and rank the work by what you can hear and reach. The sum of tagged leaks routinely comes in below the measured aggregate, because small ones are missed, some are behind machines, and deliberate openings such as a manual drain cracked open are excluded from the tag list even though the aggregate counted them. The aggregate is the target; the tag list is the work order. Reporting them as if they were the same quantity is how a repair programme gets judged a failure for closing every ticket.

How to verify you got this right

Re-run the decay test on a different day with the same valve positions. A leak population does not change materially in a week, so a decay time that differs by more than your volume bound explains means the valve positions were not the same, and the shorter time is the one with something still open behind it.

Then re-run both tests after the repairs and compare like with like: same setpoint, same boundary, same method. If the duty test improved and the decay test did not, the repairs were inside the package. If the decay test improved and the duty test did not, something in the plant was switched on during the second duty window.

Then check that your report never quotes a bound as a plus-or-minus. The duty figure is an upper bound that includes unattended demand, and calling it an estimate with an interval around it overstates what the method can deliver.

References

• 29 CFR 1910.147, control of hazardous energy, for isolating and locking out a compressor package before a decay test
• 29 CFR 1910.333(b)(2), with 29 CFR 1926.417 as the construction counterpart, and NFPA 70E-2021, 120.5, in the edition the employer's electrical safety programme has adopted, for electrical isolation and the live-dead-live proving sequence
• 29 CFR 1910.242(b), compressed air used for cleaning, and 29 CFR 1910.95, occupational noise exposure, including the 85 dBA eight-hour time-weighted average action level
• US Department of Energy, Improving Compressed Air System Performance: A Sourcebook for Industry, for leakage as a share of plant output
• See related: What a Compressed Air System Actually Costs to Run; Why a Manual Drain Left Cracked Open Is a Permanent Leak; How to Work Out Whether a Receiver Is Sized for the Demand