How to Decide Whether the Problem Is Supply or Demand

Why this matters

Every compressed air complaint arrives in the same words: "we do not have enough air." That sentence covers two completely different faults with completely different bills. Either the compressor and its storage cannot put enough air into the header, or they can and the header cannot get it to the machine. Sell the wrong one and you have installed capacity into a system whose real fault is a plugged filter, or you have re-piped a building whose compressor was undersized the day it landed. The split is decidable in one visit with two pressure readings taken at the same instant, and the reason shops get it wrong is that they take one.

Before any of this: adding a gauge to a live compressed air line means isolating that section at its upstream valve, locking and tagging the isolation and the compressor disconnect under 29 CFR 1910.147, opening a downstream vent and confirming zero on a gauge open to the line before you cut threads or break a joint. If the room's noise makes normal speech impossible at arm's length, wear hearing protection under a program meeting 29 CFR 1910.95 for the whole time you are in there.

Step 1: Get an event you can watch more than once

The split only shows up under the load that causes the complaint. Pressure at 6 a.m. with the plant idle tells you nothing. Find the specific act that produces the symptom, name the machine, and get the operator to run it on a cycle you can time. Skip this and you will measure a static system, conclude everything is fine, and be back in a week.

Stay outside the machine's motion envelope while it cycles and never defeat a guard or a light curtain to get a probe closer; run the sensing line out to where you are standing instead. If the event involves a blow-off, do not stand in the discharge path and do not let the stream reach skin.

Step 2: Fix your instruments and write down what their error is made of, before you read anything

This step is the one that decides whether your finding survives a challenge, and it is the one that gets skipped.

Two labels go on every gauge before it is used in this comparison.

Basis. Is the stated accuracy a percent of the reading, or a percent of full scale or span, or a fixed count? Percent of full scale does not shrink as the reading shrinks. Say a mechanical gauge's plate states 1 percent of full scale on a 200 psig span, which is illustrative and you must read your own plate: that is plus or minus 2 psi at 180 psig and also plus or minus 2 psi at 20 psig.

Character. Is it a fixed systematic offset, an independent random spread, or a worst-case bound? A manufacturer's stated accuracy is normally a worst-case bound, and worst-case bounds add linearly and are reported as bounds, never as a plus-or-minus interval. Two of those illustrative gauges, one at each end of your measurement, give a bound of 4 psi on the difference between them. A repeatability spread you measured yourself across repeated events is an independent random spread, and two independent spreads combine in quadrature, so two equal terms multiply by the square root of two, not by two.

There is one move that removes most of this. Use the same gauge at both points, moved between them across two runs of the same event, rather than two gauges at once. A fixed offset in one instrument cancels in a difference, leaving only that offset's effect on the difference itself, which for a fixed count is nothing. You pay for it by requiring the event to repeat, which step 1 already gave you. Where the event cannot be repeated and you must read two gauges at once, cross-check them against each other at the same tap first, note the disagreement, and carry the 4 psi bound into every conclusion.

Step 3: Establish the quiet header pressure

Read the receiver's discharge pressure with the plant running normally but with the event not firing. This is the number the event drops away from, and it is not the compressor's cut-out setting. Write it down with the time.

Step 4: Take three pressures on one time base, through at least five events

• P1, receiver discharge, just before the event.
• P2, receiver discharge, at the trough of the event.
• P3, the affected machine's own inlet, downstream of its filter, regulator and hose, at the trough.

Record all five events, take the median of each set, and record the spread. The spread is your measured repeatability and it is the only error term in this procedure you did not get off a plate. If the spread is larger than the difference you are about to call, you have not measured an event, you have measured a plant doing several things at once.

Step 5: Apply the gate

Two quantities, both per event, both taken on the median of at least five events:

• Supply-side loss equals P1 minus P2. This is the compressor and its storage failing to hold the header.
• Path loss equals P2 minus P3. This is the distribution, the filters, the dryer, the drops, the hose and the couplers failing to deliver.

The gate:

• If path loss is at least twice supply-side loss, it is a demand-side and distribution fault.
• If supply-side loss is at least twice path loss, it is a supply and storage fault.
• If neither is twice the other, both are real, and you fix the cheaper one first and re-measure rather than guessing which dominates.

And a floor that overrides all three: a loss you report as a finding must exceed your combined instrument bound. A 3 psi path loss read on two gauges whose bounds add to 4 psi is not a small path loss, it is no measurement at all.

The same gate, two shops, opposite answers

Shop A, a sheet metal shop with a slow press feed. Quiet header P1 reads 100 psig. Across five events the receiver troughs read 98, 98, 97, 98 and 98 psig, median 98, spread 1 psi. The machine's inlet troughs read 68, 67, 69, 68 and 68 psig, median 68, spread 2 psi. Both spreads are inside the 4 psi two-gauge bound, so the event is repeatable.

Supply-side loss is 100 minus 98, which is 2 psi. Path loss is 98 minus 68, which is 30 psi. Path loss is fifteen times supply-side loss, comfortably past the twice-as-large gate, and 30 psi is far outside the 4 psi bound. This is distribution. The compressor is holding the header within 2 psi while the machine starves, and 30 of the 32 psi the machine lost never left the compressor room. The finding is a restriction, and the first three places to look are the machine's own point-of-use filter, its regulator setting, and the coupler and hose feeding it, which the article on what a hose does to the pressure the tool actually sees covers in detail.

Note the honest limit on the 2 psi supply-side figure: on two gauges bounded at 4 psi combined, 2 psi is not a measurement. You cannot report "the compressor lost 2 psi." You can report that the compressor's loss was smaller than the instruments could resolve, which is all the gate needs.

Shop B, a wood shop with the same complaint on a clamp table. Quiet header P1 reads 100 psig. Receiver troughs read 76, 75, 77, 76 and 76 psig, median 76, spread 2 psi. Machine inlet troughs read 71, 70, 72, 71 and 71 psig, median 71, spread 2 psi.

Supply-side loss is 100 minus 76, which is 24 psi. Path loss is 76 minus 71, which is 5 psi. Supply-side loss is 4.8 times path loss, past the gate in the other direction. This is supply and storage: the compressor and receiver cannot hold the header through the event, and by the time the deficit reaches the machine only a fifth of what it lost happened in the pipe.

Here the instrument rule bites in the useful direction. That 5 psi path loss, read on two gauges with a combined bound of 4 psi, is barely outside the bound and is not reportable as a number. Re-run the event with one gauge moved between the two taps, so the offset cancels, and the 5 psi becomes a real reading rather than an artefact. It does not change the gate's answer, which was decided by the 24 psi, but it is the difference between telling a customer "there is about 5 psi in your piping" and telling them something you cannot defend.

What would flip either answer: a P2 tap taken upstream of the dryer and filters rather than downstream of them. Those components sit between the receiver and the plant, and their drop belongs to the path, not to supply. Tap on the wrong side of them and Shop A's 30 psi could include a plugged filter that you have just charged to the compressor. State where your P2 tap is, every time.

Step 6: Confirm the side before you quote anything

For a supply finding, convert the receiver's decay into a flow using the vessel's standard cubic feet per psi and confirm the resulting deficit is a plausible size for the equipment that fired. The storage article in this library carries that conversion. A deficit far larger than the machine's rated consumption means something else fired at the same time.

For a distribution finding, walk the path with a gauge and split the loss into segments, so that the segment losses add to roughly the total you measured. If they do not add up, there is a restriction you have not found yet, often a partly closed valve nobody remembers touching. The article on how to measure pressure drop across a compressed air system owns that walk.

Step 7: Change one thing, then re-measure the same event

Re-run the identical event with the same instrument set and the same five-event median. A fix that moves the number it was supposed to move, and does not move the other one, is proven. A fix that improves the machine's inlet pressure while the receiver's trough gets worse has moved the problem, not solved it, and that pattern shows up when someone answers a path restriction by turning the compressor's setpoint up.

How to verify you got this right

Three checks, all cheap.

The spread test. Your five-event spread on each reading should be smaller than the loss you are calling. Where it is not, you measured plant noise.

The addition test. Supply-side loss plus path loss should equal the quiet header pressure minus the machine's trough inlet pressure, exactly, because they are defined that way. In Shop A: 2 plus 30 is 32, and 100 minus 68 is 32. In Shop B: 24 plus 5 is 29, and 100 minus 71 is 29. If yours does not close, one reading was taken at a different moment than you thought.

The bound test. Read your own gauge plates again, write the combined bound, and confirm every number you are about to put in front of a customer is larger than it. Anything smaller gets described as below the resolution of the instruments, not as a value.

References

• 29 CFR 1910.147, control of hazardous energy, general industry: the basis for the isolate, lock, tag and verified-zero sequence before installing a gauge or test port in a pressurised line.
• 29 CFR 1910.95, occupational noise exposure, general industry: the basis for hearing protection under a program where compressor room noise reaches the action level.
• See related: How to Measure Pressure Drop Across a Compressed Air System; What Storage Buys You That Horsepower Cannot; What a Hose Does to the Pressure the Tool Actually Sees; and this library's articles on pressure measurement and instrument error, and on pressure and flow fundamentals.