How to Find Leaks in a System Nobody Has Surveyed

Why this matters

What you hand back is not a leak count. It is a located, tagged, repair-ready list that somebody else can work without re-finding anything, and that is the part most surveys get wrong. A number on its own changes nothing, because the person holding the wrench two weeks later cannot act on it. The first survey is also the easy one, since nothing has been picked over yet, which makes it the survey most worth doing properly. What it costs is a quiet window and a disciplined traverse; what a shop usually brings back instead is a count.

The hazards you are walking into before you find anything

Lead with these, because a survey puts you next to running machinery and a live pressurised header for several hours.

• Noise. A compressor room commonly sits at or above the eight-hour time-weighted average of 85 dBA where 29 CFR 1910.95 attaches its hearing conservation requirements, so wear hearing protection from the door and keep it on for the whole room, not just while you are beside the machine.
• Compressed air on skin. Never put a hand, a probe tip or a face into a leak stream to feel for it. Air driven under the skin can enter the bloodstream, so locate leaks with the detector at a distance and confirm with soap solution applied from the side, never by feel.
• Whipping lines. A hose or a drop that lets go under pressure is a struck-by hazard. Stand out of the arc of any hose you are testing, and where a connection is above head height, treat the whip restraint as required equipment rather than optional.
• Opening anything. No repair in this procedure starts until the drop is isolated at its valve, the leg is bled through a downstream opening, and the local gauge is confirmed reading zero. For work on the compressor itself, that means lockout under 29 CFR 1910.147 with the stored air in the receiver relieved and verified at the receiver gauge, because receiver energy does not dissipate when the motor stops. For work inside the starter or the drive enclosure, the fork is 29 CFR 1910.333(b)(2) rather than 1910.147, with dead confirmed by the live-dead-live sequence in NFPA 70E-2021, 120.5 as adopted by your employer's electrical safety program.
• Airborne material at blow-off stations. If your traverse takes you past an operator blowing parts down, do not walk through the plume. That operation puts whatever was on the part into the breathing zone, and cleaning with compressed air is limited by 29 CFR 1910.242(b) to reduced pressure with effective chip guarding and personal protective equipment. If you must stand in that area, respiratory protection under a 29 CFR 1910.134 program is the control, not safety glasses.

Step 1: pick the condition before you pick the day

Survey during a genuine no-production window if you can get one: a weekend, a shift break, a shutdown. Two reasons, and the second one matters more.

The first is that ambient noise falls, so the ultrasonic detector has a cleaner floor to work against. The second is that with production stopped, every remaining flow is loss. A leak you find during a running shift has to be argued for; the same leak found in a silent plant is unarguable, because nothing else could be consuming.

If you cannot get a quiet window, survey during the shift and accept that the point-of-use zone will be the hardest part of the traverse. Schedule that zone for a break.

Step 2: get the total before you go looking for the parts

Ask the plant for, or run, a pressure decay test so you know the size of what you are hunting. This library covers the decay method and the arithmetic that converts it into a load figure under leak costing, so use it rather than re-deriving it here. What you need out of it is one number: total leak load expressed as a percent of the compressor's rated delivery at its rated discharge pressure.

That number is your completeness check. Without it you have no way to know whether your 40 tagged leaks were most of the problem or a quarter of it, and a survey that cannot say which is a list, not a finding.

Step 3: traverse by fitting density, working from the most joints per foot to the fewest

Leaks happen at joints, so walk the system in the order of how many joints there are per unit of pipe, not in the order the air flows. That ordering puts the point of use first and the compressor room last, which is the reverse of how most people walk a plant.

1. Point-of-use connections. Quick couplers, hose whips, tool inlets, filter-regulator-lubricator assemblies, blow guns, cylinders and their fittings. The highest joint count in the system and the only joints that get handled by hand every day.
2. Drops and their drain legs. The drop valve, the takeoff fitting, the drain leg and its manual drain.
3. Branch takeoffs. Where a branch leaves the header, plus the isolation valve on each branch.
4. Header runs. Long stretches with few joints. Fast to walk, low yield per foot, but check every union, elbow and hanger where the pipe can move.
5. Compressor room. Aftercooler, receiver, dryer, filters, drains, relief valve seat, unloader and its piping.

Zone the plant on the drawing before you start and clear one zone completely before moving. A traverse that jumps around leaves gaps you cannot see, because the thing you missed makes no mark.

Step 4: what the detector is hearing, and how to close on a leak

An ultrasonic detector is listening to the broadband ultrasound that turbulent flow generates at the orifice, then shifting it down into the audible range. Two properties of ultrasound drive the whole technique: it attenuates quickly in air, and it reflects off hard surfaces.

Attenuation is what makes the instrument directional. Start at high gain and sweep slowly across the zone at a distance. When you get a rise, reduce the gain and walk in. Each gain reduction narrows the field, and by the time you are within arm's length the peak should be sharp enough to point at a single fitting.

Reflection is the failure mode at the far end of that same behaviour. A strong signal coming off a flat panel, a machine guard or a concrete wall reads as a leak in the wall. Confirm direction by breaking the path: move your body or a hand between the suspected source and the sensor. A real source drops out. A reflection changes character but keeps coming from somewhere else, and moving to a second angle resolves it.

Confirm the last few inches with soap solution applied with a brush from the side of the joint, and mark the exact fitting, not the assembly.

Step 5: tag at the leak, log away from it

Tag where you find it, immediately, with a tag that carries a unique number. Then record against that number: zone, equipment or drop identifier, the specific fitting, whether the repair needs an isolation, and a relative size ranking taken at a fixed gain and a fixed standoff distance.

That size ranking is ordinal, not quantitative. It sorts your repair list; it does not convert to flow. Anyone who wants a per-leak flow figure needs an orifice estimate, which belongs to the costing article and needs a measured or estimated hole size you almost certainly do not have.

The log is what makes the survey repeatable. A tag with no record is a leak somebody will find again next year and count as new.

Step 6: split the list by what the repair requires

Sort the tagged list into repairs that need no isolation beyond a drop valve, repairs that need a branch or header isolation, and repairs that need the compressor down. This is the sort that decides whether anything actually gets fixed, because the first group can be worked the same week and the third group waits for a shutdown that may be months out.

Every repair in the first group still carries its own gate: close the drop valve, bleed the leg through the coupler or a downstream opening, and confirm the local gauge reads zero before breaking the joint.

Worked example: a first survey on a single-shift plant

The plant runs one shift, one rotary screw compressor, and has never been surveyed. A decay test over a Saturday returns a total leak load of 22 percent of the compressor's rated delivery.

The traverse takes one technician most of a Saturday and tags 47 leaks:

• 31 at point-of-use connections, mostly couplers and hose whips
• 9 at branch takeoffs and header joints, all needing a branch isolation
• 7 in the compressor room, at drains, the dryer inlet union and one filter bowl

The 31 point-of-use repairs and the 7 compressor-room repairs get worked over the following two weeks: 38 of the 47. The 9 branch repairs wait for the quarterly shutdown.

A repeat decay test two Saturdays later, once the repairs are actually complete, run by the same technician with the same method and the same starting and ending pressure marks, returns 12 percent of rated delivery. So the 38 repairs captured 10 points of the original 22, which is 45 percent of the leak load, on a base of the compressor's rated delivery.

Read the leftover honestly. Twelve points remain against 9 known unrepaired leaks, and 9 branch leaks are unlikely to account for the whole 12 given that they ranked mid-size against 31 point-of-use leaks that together carried the largest share. That gap is the part of the survey that was not found, and the right conclusion is a second traverse of the point-of-use zone during a break rather than a claim that the system is now clean.

The failure mode here is the survey that stops at "47 leaks found and 38 repaired" and never runs the second decay test. That report cannot distinguish a 45 percent reduction from a 90 percent one, and the customer's next question is exactly that.

Confirming the survey rather than the repairs

Three checks, all of them on the survey rather than on any single fitting.

Completeness against the total. Compare the post-repair decay figure against the pre-repair figure, as above. Anything under about half the original load captured means the traverse missed a zone, and the zone it missed is almost always the one you walked during production noise.

Reproducibility of a sample. Have a second person re-find five tagged leaks from the log alone, without the tags visible. If they cannot, the log is under-specified and the repair crew will not find them either.

No false positives. Re-check every leak you tagged from more than arm's length without confirming with soap solution. Reflections concentrate in exactly those, and a repair crew sent to a wall loses faith in the whole list.

References

• 29 CFR 1910.95, hearing conservation requirements attaching at an eight-hour time-weighted average of 85 dBA
• 29 CFR 1910.147, lockout and control of stored mechanical energy for work on the compressor and receiver; 29 CFR 1910.333(b)(2) for work inside the starter or drive enclosure, with NFPA 70E-2021, 120.5 for the live-dead-live sequence where your employer's electrical safety program adopts it
• 29 CFR 1910.242(b), the reduced-pressure, chip-guarding and personal protective equipment limits on cleaning with compressed air
• Manufacturer documentation for the ultrasonic detector on gain settings, standoff distance and reflection behaviour
• See related: How to Put a Number on What Leaks Are Costing; Why a Leak Survey Has to Be Repeated to Be Worth Anything; What Pressure Drop Through Distribution Actually Costs