How to Baseline a Compressed Air System in One Visit

Why this matters

Compressed air is the most expensive utility in most buildings and the only one nobody meters. There is a meter on the gas, a meter on the water, a meter on the electricity, and on the air there is a pressure gauge that tells you nothing about quantity. So the owner has no number to argue with, and the only proposal anyone ever brings them is a bigger compressor. A baseline visit gives them the first real numbers they have ever had on that utility, and one of those numbers routinely shows that a third of the machine they already own is producing air that leaves through fittings.

The window that has to be agreed before you arrive

This visit is organised around plant states, and one of those states requires the compressor to start without warning while nobody is touching it. Agree in advance that during the no-production window the compressor room is barriered, nobody works on the machine, and everyone within reach knows it will start on its own pressure switch. A compressor that cycles on its own is not an idle machine; the drive coupling and belts are live every time it loads.

Nothing is opened at any point in a baseline. No drain valve, no filter bowl, no joint, no receiver. The system stays charged for the whole visit by design, which means stored pressure energy is present in every component the whole time. If a measurement turns out to require opening something, that becomes a second visit under isolation: close the upstream valve, open the vent, confirm the section gauge reads zero with the vent still open, and lock and tag the compressor disconnect under 29 CFR 1910.147, which covers stored pressure and mechanical energy. Work inside the starter enclosure forks to 29 CFR 1910.333(b)(2) instead, with dead proved by the live, dead, live sequence of NFPA 70E-2021, 120.5 where your employer's electrical safety program has adopted that edition.

Hearing protection goes on at the compressor room door: these rooms commonly sit at or above the 85 dBA eight-hour time-weighted average at which 29 CFR 1910.95 attaches. If you use an ultrasonic leak detector, note that its headset is a listening device and is not hearing protection unless it is separately rated as such, so it does not replace plugs or muffs. Never clear a fitting face or a gauge port with a blow gun; 29 CFR 1910.242(b) limits compressed air used for cleaning to under 30 psi with effective chip guarding and personal protective equipment. Any condensate you encounter is an oil and water mixture: nitrile gloves, sealed container, and confirm the disposal route with the local sewer authority before it goes anywhere, because in most jurisdictions it is a regulated waste.

How the visit is ordered, and by what measure

The measure is how narrow the window is in which the reading exists at all, narrowest first. That is not a preference about importance; it is the only ordering that survives contact with a plant that has work to do. A reading you can take at any hour of any day is a reading you can take while waiting for something else, so it sorts last no matter how much it matters.

There are three windows and one thing that is not a window at all.

Window one: no production, system charged

This is the narrowest window in the plant. It exists at shift start before the floor picks up, at a lunch break if the plant genuinely stops, or after the last shift. It closes without warning and you will not get it twice in one visit, so it is the first thing you do after you have hearing protection on.

What only exists here is the leak load, and it is the number that changes the conversation, because it converts leaks from a nuisance into a stated fraction of a machine the owner has already bought.

The load and unload duty method, which is the one to use whenever the machine is a load and unload compressor. With no production running and the system charged, watch the compressor through several complete cycles and time how long it spends loaded and how long unloaded. The leak load as a fraction of the compressor's capacity is loaded time divided by total time. Multiply that fraction by the machine's rated capacity to get a flow. This is valid for a load and unload machine only: on a modulating or a variable speed machine the relationship between time and delivered flow is not one to one, and the sibling articles on control strategy and variable speed carry why.

The pressure decay method is the fallback where the machine is not load and unload. Charge the system, stop and isolate the compressor, and time the pressure fall between two stated pressures. Free air lost equals system volume in cubic feet times the pressure fall in psi divided by 14.7, over the elapsed time. It is second choice for one reason: it needs the total system volume including every foot of pipe, and that estimate is usually the weakest number in the whole baseline.

Whichever method, write down what was confirmed shut off, by name. This measurement has one large systematic error and it only points one way: anything still consuming during the window inflates the leak figure. A maintenance tech blowing down a machine in the back corner turns a real finding into a wrong one.

Window two: peak production

Wider than window one but still bounded by the shift, and it has to be the peak, because peak is the condition the compressor's discharge setpoint has to cover. Take here:

• Header pressure at the compressor's cut-in, not its cut-out. Cut-in is the worst the plant sees and it is what a starved drop is starving at.
• The pressure drop allocation from receiver to the worst point of use. The method for that is its own article and it produces a per-segment sheet rather than a single number.
• Pressure dew point at the point of use, under load, at line pressure.
• The compressor's loaded fraction under real demand, which combined with window one's leak load tells you how much of the machine production is actually using.

Window three: any time, including while you wait

Everything here is available whenever you are standing in the room, which is exactly why it sorts last in the visit and first in the report.

Compressor nameplate rating and control type. Receiver volume and whether there is a second one downstream of the dryer. Dryer type and its rated pressure dew point at its rated inlet conditions. Filter differentials as they read at rest. Main pipe size and material, and whether the header is a dead end or a loop. Automatic drain function, observed through a cycle rather than assumed. Compressor room intake and discharge ventilation. The date on the last service sticker.

What one visit cannot give you

A baseline is a snapshot and it should be sold as one. The things that need time are the ones that vary: demand across a week, the effect of a shift change, a seasonal ambient swing in the compressor room, and any event that happens less often than your visit lasts. If those matter, the deliverable is a logger left on the discharge and a second appointment, and saying so is what separates a baseline from a guess with a letterhead.

The worked example

A metal fabrication plant, one load and unload rotary screw rated 100 acfm, complaint that the compressor "cannot keep up any more" and a quote already in hand for a second machine.

Window one. Production off, system charged, four complete cycles observed over ten minutes. Loaded time totalled 3 minutes, unloaded 7 minutes. Leak load is 3 divided by 10, which is 30 percent of capacity, so 30 acfm of the 100 acfm machine is producing air that nothing is using. That sits at the top of the 20 to 30 percent band commonly reported for unmanaged industrial plants, against under 10 percent for a plant that manages leaks; those are figures from compressed air system practice rather than a code requirement, and they are useful only as a place to put this plant on a scale.

State the error honestly before the finding leaves your mouth. Four cycles is eight transitions, and calling a transition by ear is good to about a second, so the timing uncertainty is around 8 seconds against 180 seconds of loaded time, roughly 4 percent of the loaded figure. That is a bound, not a spread, and 4 percent of 30 is about 1.3 percentage points, so the finding is 30 percent give or take a point and a bit. Nothing about that moves the conclusion. The error that could move it is the other one: if anything at all was still consuming, 30 percent is too high, which is why the shut-off list is written down and signed rather than remembered.

Window two. At peak production the compressor ran continuously loaded, which is the machine telling you demand is at or above capacity. Header at cut-in read low enough that two drops were starving.

Putting the two together. Total demand at peak is at the machine's 100 acfm. Of that, window one says 30 acfm is leak load, so real production demand is about 70 acfm. The plant does not need a second 100 acfm machine. It needs 30 acfm of fittings repaired, after which the existing machine covers 70 acfm of production demand with 30 percent of itself in reserve.

In hours rather than money. For every 10 hours that compressor runs, 3 hours of its output goes out through joints. That framing is the one an owner acts on, because running hours is a number they already track on the service sticker.

The failure mode of getting this wrong. The second machine gets bought and installed on the same header. Both machines now see the same 30 acfm of leaks plus the 70 acfm of work, both spend most of the day part loaded, and part load is where a fixed speed machine is least efficient. The plant has doubled its compressor room and improved nothing about the fraction of it doing useful work, and the leak load is still there to be found next year by someone else.

How to verify you got this right

Repeat window one on a second date, at the same plant state, with the same shut-off list confirmed item by item. Two leak-load figures within a couple of points of each other is a measurement; one figure is an anecdote, and this is the number the owner is going to spend on.

Then check that your peak reading and your leak reading are on the same basis before you subtract them. Both are flows at the same reference condition, both are against the same machine's rated capacity, and both were taken with the same header band. If the plant raised the setpoint between your two windows, the leak figure from window one no longer applies at window two's pressure, because leak flow rises with upstream absolute pressure. Say so in the report rather than subtracting anyway.

References

• 29 CFR 1910.147, control of hazardous energy, and 29 CFR 1910.333(b)(2) for the electrical fork on starter enclosure work
• 29 CFR 1910.95, occupational noise exposure, and 29 CFR 1910.242(b), compressed air used for cleaning
• NFPA 70E-2021, 120.5, live dead live verification, where the employer's electrical safety program adopts that edition
• Compressed air system practice for leak load as a fraction of compressor capacity in managed and unmanaged plants
• See related: How to Measure Pressure Drop Across a Compressed Air System; What a Compressed Air System Actually Costs to Run; The Compressor That Could Not Keep Up With No New Equipment Added