Why a Leak Survey Has to Be Repeated to Be Worth Anything

Why this matters

A one-time leak survey buys a benefit that starts decaying the day it finishes. Fittings loosen, hoses chafe, seals age, drains get cracked open and left, and the point-of-use connections that carried the largest share of the original loss are the ones handled by hand every shift. A shop that sells a single survey and moves on has sold the customer a number that will be wrong within a year. A shop that sells a survey plus a measured cadence has sold a program, and the cadence is not a calendar convention you pick, it is a rate you measure on that specific system.

Leak load regenerates, and the rate is a property of the plant

Three mechanisms put leaks back, and they act at different speeds on different plants:

• Handling. Couplers, hose whips, blow guns and tool inlets are connected and disconnected many times a day. Wear on a coupler seal is a function of cycles, so a plant with many hand-connected points regenerates fast regardless of how well the pipe was installed.
• Movement and thermal cycling. Threaded joints on hard pipe loosen where the pipe expands and contracts or where a hanger lets it move. Slow, and it concentrates on long straight runs and at equipment connections that get bumped.
• Repairs that were never verified. A joint remade under time pressure and never re-tested is a leak that was counted as fixed. This one shows up as regeneration but is really an unfinished repair, and it is the reason the verification sample in a survey matters.

Because the mix of those three differs between plants, so does the regeneration rate. A single-shift plant of hard-piped drops with a handful of couplers behaves nothing like a three-shift assembly area where every station has a hose and a quick connect. Assuming an annual cadence for both means one plant is surveyed too often and the other has been leaking for eight months before anyone looks.

Measure the rate with the cheap test, not the expensive one

The survey is expensive: a technician for most of a day, plus the repair work. The pressure decay test is cheap: one no-production window, one technician, under an hour. That asymmetry is what makes a measured cadence practical. Run the decay test quarterly to track the regeneration, and run the full traverse only when the tracking says it is time.

Run the decay test with the compressor stopped and its disconnect locked open under 29 CFR 1910.147, and leave the receiver pressurised and untouched for the duration, because that stored energy is the test. Nobody opens a drain, a filter bowl or a joint while the test is running, and the drain and the indicating gauge the test is read against are the ones 29 CFR 1910.169 requires the receiver to carry in the first place. If the test has to be aborted and something opened, the leg is isolated and bled and the local gauge confirmed at zero before any joint is broken.

Express the result the same way every time so the series is comparable: total leak load as a percent of the compressor's rated delivery at its rated discharge pressure, measured in a genuine no-production condition. The decay method itself is covered under leak costing in this library; what this card owns is what to do with the series it produces.

Why the difference between two tests is stronger evidence than either test alone

Before you compare two decay figures, you need to know what kind of error each one carries, because that decides whether the errors cancel or accumulate.

Basis. A decay test's uncertainty is dominated by the operator's start and stop criteria and by the estimate of system volume, and both of those are percent of reading rather than percent of full scale. That means the error shrinks as the leak load shrinks, so a small post-repair figure is not swamped by an error sized for the original figure.

Character. The volume estimate is a fixed systematic offset. If the technician's assumed system volume is 8 percent high, every test that technician runs with that assumption reads 8 percent high. A fixed offset from one source cancels in a difference, leaving 8 percent of the difference rather than 8 percent of either reading. So a series run by the same person, with the same volume figure, the same gauge and the same pressure marks, tracks change far more reliably than any single number in it tracks truth.

That cancellation stops the moment anything changes. A different technician with a different volume assumption introduces an independent offset, and the two no longer subtract out. If the technicians' spreads are independent and roughly equal, they combine in quadrature, so the uncertainty on the difference is about 1.4 times one of them rather than the same as one of them. If either figure is stated as a worst-case bound instead of a spread, the two bounds add linearly and the result must be reported as a bound, never as a plus-or-minus interval.

The operational consequence is short: the same person, the same assumed volume, the same marks, the same no-production condition, every time. Write those four on the test sheet.

The gate

Set the next full survey by this rule, and state both arms of it out loud because they are joined by OR, not AND:

Repeat the full traverse when the measured leak load has regenerated back to half the reduction the last campaign bought, OR at 24 months, whichever comes first.

Unit of analysis: system-level leak load as a percent of the compressor's rated delivery, from the quarterly decay series, not per-zone and not per-leak.

Why half the reduction rather than a fixed percentage of rated delivery: a fixed threshold punishes the plant that started clean and lets the dirty plant coast. Half the reduction scales the trigger to what the last campaign actually achieved on that system, which is the only measure of what a traverse is worth there.

Why the 24-month arm exists at all: a plant can regenerate so slowly that the half-the-reduction arm never trips, and at some point the survey stops being about regeneration and starts being about the joints that were never found the first time. Twenty-four months is a common starting point; move it if your own verification sampling says the traverses are finding a lot of long-standing leaks.

One gate, two plants, opposite answers

Plant A: three shifts, assembly work, many hand-connected points. First survey ran 22 percent of rated delivery down to 12 percent, a reduction of 10 points. Half the reduction is 5 points, so the trigger sits at 17 percent.

Quarterly decay series, same technician, same assumed volume: 12.0, then 13.5 at month 3, 15.2 at month 6, 17.4 at month 9.

The trigger was crossed between month 6 and month 9. The regeneration is running roughly 1.7 points per quarter and shows no sign of flattening, which fits a plant whose largest leak population is coupler and hose seals turning over on a cycle count. Their answer is a traverse at month 9, and going forward a survey cadence near 9 months rather than 12, aligned to the nearest planned shutdown so the branch repairs can be worked at the same time. Theirs happens to fall at month 8, so that is the date.

Plant B: one shift, hard-piped drops, few couplers. First survey ran 9 percent down to 5 percent, a reduction of 4 points. Half the reduction is 2 points, so the trigger sits at 7 percent.

Quarterly series: 5.0, then 5.3, 5.6, 6.0, 6.3 at month 12.

Regeneration is about 0.3 points per quarter and steady. Extrapolating at that rate, 7 percent arrives around quarter 7, roughly month 21. That is inside the 24-month arm, so the regeneration arm governs and the next traverse goes on the schedule at month 21.

Two plants, the same rule, cadences differing by more than a factor of two, and neither one is the annual interval both would have been given by default. That is the whole argument for measuring instead of assuming.

Reading the series when it does something other than climb steadily

A decay series is a diagnostic in its own right, and three shapes are worth naming.

A step rather than a slope. Load jumps between two quarterly tests and then holds flat. That is not regeneration, it is one event: a drain valve left cracked open, a hose burst and capped badly, a machine removed and its drop left open. Look at the maintenance log for that quarter before scheduling a traverse, because one visit may fix all of it.

A rise that flattens. Common on plants dominated by handling wear, where the population reaches a steady state between new leaks and incidental repairs. If it flattens below your trigger, the trigger never trips and the 24-month arm is doing the work, which is exactly what that arm is for.

A fall with no repairs. Suspect the test, not the system. The usual cause is a different starting pressure or an isolation valve closed during one test that was open during the other, which shrinks the volume the test sees. This is the failure mode the same-person, same-marks discipline exists to prevent, and it is why the test sheet records the valve positions.

Verifying that the cadence is right rather than just followed

Two checks, run at each traverse, and both of them are about the cadence rather than about any leak.

First, sort the newly found leaks by whether the fitting was on the previous survey's log. A traverse where most finds are at fittings that were clean last time confirms genuine regeneration and confirms the interval. A traverse where most finds are at fittings never logged before means the previous traverse missed them, and the fix is traverse quality, not a shorter interval. Shortening the cadence to compensate for a bad traverse is the most expensive way to solve that problem.

Second, compare the reduction this campaign bought against the last one. A campaign that buys much less than the previous one, on a system with the same trigger, usually means the remaining population has shifted toward leaks that need an isolation and are being deferred. That is a scheduling problem, and it argues for aligning the traverse to a shutdown rather than for surveying more often.

References

• Manufacturer documentation for coupler and hose assembly service life and replacement intervals
• 29 CFR 1910.169, requiring every air receiver to carry a drain at its lowest point and an indicating pressure gauge with one or more spring-loaded safety valves; 29 CFR 1910.147 for locking the compressor out before the decay test is left standing
• See related: How to Find Leaks in a System Nobody Has Surveyed; How to Put a Number on What Leaks Are Costing; Why a Manual Drain Left Cracked Open Is a Permanent Leak