What to Record About a Compressed Air System So It Can Be Compared Later

Why this matters

Most compressed air records are a list of pressures with a date on them, and a list of pressures cannot be compared to anything. A year later somebody puts this year's number next to last year's, announces that demand grew, and is wrong, because in between the plant raised its setpoint and the compressor room got hotter and nobody wrote either of those down. The record failed not because a reading was inaccurate but because it had no denominators. What you record has to make a future reading correctable onto the same basis as this one, and that is a different design brief from writing down what the gauges said.

Everything in this record is taken with the system charged and nothing opened. Hearing protection goes on at the compressor room door, since these rooms commonly sit at or above the 85 dBA eight-hour time-weighted average at which 29 CFR 1910.95 attaches. If a field turns out to need a component opened, that field moves to a separate 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 open under 29 CFR 1910.147, which covers stored pressure energy. Never clear 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.

What you are actually going to compare

Not readings. Ratios. The four that carry a compressed air system's condition over time are:

• Loaded fraction: loaded hours divided by running hours, from the compressor's own controller.
• Leak load: leak flow as a fraction of the machine's rated capacity, with the method named.
• Loss from receiver to the worst point of use, in psi, at a stated flow state.
• Dew point margin: the coldest temperature the air sees downstream, minus the measured pressure dew point at line pressure.

Each of those is a quotient or a difference, which means each one needs its bottom half recorded as carefully as its top half. A pressure written alone has no bottom half at all.

Group A: the identity fields

These change rarely, and when one of them changes, every comparison across that change is void rather than interesting. Record them once and re-confirm them each visit rather than copying them forward.

Compressor rated capacity, control type (start and stop, load and unload, modulating, variable speed) and drive. Receiver volume, and whether there is a second receiver downstream of the dryer as well as one upstream. Dryer type and its rated pressure dew point at its rated inlet conditions, transcribed from the plate including those conditions, because the rating without them is not a specification. Filter stages in order, with each element's rated clean differential from the manufacturer's data. Header size, material, and whether it is a loop or a dead end.

Group B: the denominators

This is the group that separates a record from a note, and it is the group that gets skipped.

Ambient in the compressor room, at the reading. Inlet air density falls as it warms, so a machine drawing warmer air delivers less mass per revolution and has to load longer to meet the same demand. The direction is fixed and worth knowing by heart; the magnitude belongs to the machine's own inlet condition curve rather than to a coefficient you carry around.

Site elevation, once, because it is fixed for that plant and because it is what stops someone comparing this plant's numbers to another one naively.

Compressor running hours and loaded hours off the controller, as absolute counter values rather than as a fraction. Two absolute counters at two dates give you the fraction for the interval between them; a fraction alone gives you nothing to subtract.

Header band as found, cut-in and cut-out, both. Not "about 100". The band is the single field most likely to have been changed by somebody between visits and least likely to have been written down when they did.

What was running, named machine by machine. "Normal production" is not a state, it is a hope. If you took the reading during a shift change, write shift change.

Date and clock time, because a reading at 06:40 and a reading at 14:00 in the same plant are two different plants.

Group C: the readings

Header pressure at the compressor's cut-in, not its cut-out, because cut-in is the worst the plant sees and is what a starved drop starves at. Pressure at the worst point of use under load, at a named drop. Pressure dew point at a named point, under load, at line pressure. Each filter's differential, with the flow state it was read at. Leak load with the method named and the shut-off list attached item by item, because that measurement's largest error is systematic, points one way only, and comes entirely from something still consuming.

Every one of those carries its location and its state in the same line as its value. A dew point without a location and a load state is a number somebody will later assume was taken at the dryer.

Group D: the instrument row, which is not optional

Record which instrument took which reading, its identity, its accuracy specification with the basis, and when it was last calibrated. Three reasons, all of which decide whether a future comparison means anything.

Basis decides whether the error shrinks with the reading. An accuracy quoted as a percent of full scale does not shrink as the reading falls. A gauge specified at 1 percent of full scale on a 0 to 300 psig range carries 3 psi at any reading on it, which is 3 percent of a 100 psig reading; the same class on a 0 to 160 psig range carries 1.6 psi, which is 1.6 percent of the same reading. So the record has to state the gauge's full scale, not just its accuracy class. A dew point transmitter is different again: usually a fixed accuracy in dew point degrees over a stated range, which is a bound rather than a spread and does not shrink at all.

Character decides what happens when two readings combine. A fixed offset in one instrument cancels when you subtract two readings taken with that same instrument, leaving that bound applied only to the difference. Two different instruments each bring their own bound and those bounds add. That is why "same gauge, moved" is worth writing down, and why a difference computed from two permanently installed gauges is a weaker number than the same difference read with one.

Identity decides whether next year's reading is comparable at all. If the instrument changed between visits, the change in the reading may be the instrument.

The filled record

Site and system. One rotary screw, rated 100 acfm, load and unload control. One 400 gallon receiver upstream of the dryer, none downstream. Refrigerated dryer, plate rating 38 F pressure dew point at 100 psig, 100 F inlet, 100 F ambient. Two filter stages: coalescer then particulate, rated clean differentials taken from the element data sheets. Header 2 inch steel, dead end. Site elevation recorded once, at the top of the sheet.

Year one, taken 06:40, before the floor picked up and then again at 13:10 at peak. Compressor room ambient 88 F. Controller running hours and loaded hours recorded as counter values; over the trailing period they give a loaded fraction of 0.62. Header band as found, 100 psig cut-in to 125 psig cut-out. Leak load 22 percent of rated capacity by the load and unload duty method, shut-off list attached, seven items, each initialled. Same gauge used at both pressure points, its full scale and accuracy class written on the sheet.

Year two, same two clock times, same shut-off list. Compressor room ambient 104 F. Loaded fraction over the trailing period 0.71. Header band as found, 105 psig cut-in to 130 psig cut-out, and nobody at the plant could say when it moved.

The comparison, done three ways

The way it is usually done. Loaded fraction went from 0.62 to 0.71, which is a rise of 14.5 percent relative. Report: demand grew about 15 percent. This treats loaded fraction as if it were production demand, which it is not, because leaks are inside it.

Splitting leaks out, but leaving the comparator uncorrected. Year one production demand is loaded fraction minus leak load, 0.62 minus 0.22, which is 0.40 of capacity. Apply year one's 22 percent leak load to year two: 0.71 minus 0.22 gives 0.49, and 0.49 against 0.40 is a rise of 22.5 percent. Better, and still wrong, because the leak figure being subtracted from year two is a year one figure taken at a year one pressure.

Correcting the comparator, which is the point of having recorded the band. Leak flow through a fixed opening rises with upstream absolute pressure. Cut-in moved from 100 psig to 105 psig, so absolute moved from 114.7 to 119.7, a ratio of 1.044. The same physical holes now pass 22 percent times 1.044, which is 23.0 percent of capacity. Year two production demand is 0.71 minus 0.230, which is 0.480, and 0.480 against 0.400 is a rise of 20 percent.

Read what the correction did. The leak load did not grow at all in physical terms: not one new hole, and the entire 1 point rise in leak share is the setpoint somebody moved. And the production growth is 20 percent rather than the 22.5 percent the half-corrected version claimed and the 14.5 percent the naive version claimed. Three defensible-looking numbers from one pair of visits, and only one of them survives having its denominators written down.

Two corrections have been applied at this point, splitting leaks out and putting the leak figure onto year two's pressure. A third is still outstanding, and the record supports it being named rather than guessed. Compressor room ambient rose 16 F between visits, warmer inlet air is less dense, and a machine delivering less mass per revolution loads longer to meet the same demand. That pushes the same way as the setpoint correction, so the true production growth is below 20 percent, and 20 percent is an upper bound rather than a finding. Route the magnitude to the machine's inlet condition curve. Writing that sentence is only possible because somebody wrote down a room temperature, which is the field most likely to be dismissed as irrelevant on the day it is taken.

References

• Compressor controller documentation for running hour and loaded hour counters and their reset behaviour
• Manufacturer data for the dryer's rated pressure dew point and its rated inlet conditions, and for each filter element's rated clean differential
• Instrument manufacturer specifications for accuracy basis (percent of reading, percent of full scale, or a fixed bound) and calibration interval
• 29 CFR 1910.95, occupational noise exposure, and 29 CFR 1910.147, control of hazardous energy, where a field requires a component to be opened
• See related: How to Baseline a Compressed Air System in One Visit; How to Measure Pressure Drop Across a Compressed Air System; Documenting a Maintenance Baseline