The System That Passed Every Check and Still Starved One Machine

Why this matters

A compressed air survey that comes back clean and a machine that still faults on low air is not a contradiction, and treating it as one is how the next three visits get wasted. Every standard system check is an average taken over seconds or minutes. A single point of use can starve on an event that lasts a fraction of a second. Those two facts can both be true at the same time, in the same building, with no instrument reading wrongly anywhere. The finding in this case was not a fault the checks missed; it was a class of fault no average could have contained.

Stop before anything is adjusted

The machine in question was faulting on its own low-air-pressure switch. That switch is a protective device, and two things get done to it in the first hour of a call like this: someone jumpers it, and someone raises the header setpoint until it stops complaining. Neither is available here.

The switch does not get jumpered or replaced. Establish why it opened before treating it as the fault. A correctly operating pressure switch that is replaced reaches the same end state one step slower and with a part number on the ticket, and the machine it was protecting now runs a stroke on air it does not have.

The header setpoint does not get raised. Raising discharge pressure raises compressor power draw and simultaneously raises the flow out of every leak and every unregulated drop, because those are fixed restrictions whose flow rises with upstream absolute pressure. Both effects push the same way.

For the instrumented tests below, the machine has to cycle while a transducer is on its inlet. Nobody stands inside the stroke envelope, no guard comes off, no interlock is defeated, and the person watching the logger is outside the machine's working area with the operator running it from its own controls. To fit the transducer in the first place, close the drop shutoff, 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 and mechanical energy. Hearing protection on at the compressor room door, since these rooms commonly reach the 85 dBA eight-hour time-weighted average at which 29 CFR 1910.95 attaches.

Six checks, all passed

The plant had already paid for a survey and it was a competent one.

Six passes and a machine that faulted several times a shift. The survey was not wrong. It was answering a different question than the one the machine was asking.

What each passing check could not have seen

This is where the diagnosis actually happened, and it happened on paper before anyone brought another instrument.

An averaged capacity figure cannot contain a sub-second event. A machine drawing a large volume in half a second and nothing for the next half minute has a small average and a large instantaneous demand. Averaging is the operation that destroys exactly that information.

A logger's sample interval and its accuracy are separate specifications, and only one of them was checked. A transducer accurate to a fraction of a psi tells you nothing about an event shorter than its sample interval, because it never sampled during the event. Bandwidth is not accuracy, and a survey that quotes an instrument's accuracy as evidence of its coverage has confused the two.

A pressure drop reading taken at steady flow is a statement about steady flow. The drop from receiver to point of use under a transient is a different quantity, driven by the acceleration of the air in the pipe as well as by friction, and it does not appear in a steady reading at all.

The regulator was set correctly at the wrong flow. Set under flow is the right method, and the flow it was set at was the machine's steady draw. The machine's steady draw and the machine's stroke draw are two different numbers.

Leak load and dew point were answering unrelated questions. Both had to be excluded and both were, correctly. Neither could ever have explained a fault tied to a stroke.

The reading that was not an average

They put a pressure transducer on the machine's own air inlet, downstream of its regulator, logging fast enough to resolve the stroke rather than once a minute, and ran the machine through twenty cycles.

Inlet pressure sat at 90 psig between strokes and dipped to 71 psig for about four tenths of a second on every stroke, recovering fully before the next one. The machine's low-air fault switch is set at 75 psig. The switch was doing its job perfectly, on every single stroke, and the plant's minute-interval logger had recorded nothing because the event occupies about 1.4 percent of the 28 second machine cycle.

Do the sampling arithmetic once, because it is the part worth taking away. An event lasting 0.4 seconds inside a 28 second cycle is present 1.4 percent of the time. A logger sampling once a minute has essentially no chance of landing inside it, and even a logger sampling once a second lands inside it on about 1.4 percent of its samples, and reports the wrong depth when it does, because the sample is one instant somewhere on the curve rather than the minimum.

Where the air goes in four tenths of a second

The machine drives a large single-acting cylinder. Free air consumed per stroke is the swept volume multiplied by absolute pressure over atmospheric.

The cylinder is 6 inch bore by 24 inch stroke. Swept area is about 28.3 square inches, swept volume about 679 cubic inches, which is 0.393 cubic feet. At 90 psig the absolute ratio is 104.7 over 14.7, which is 7.12. So free air per stroke is 0.393 times 7.12, about 2.8 cubic feet.

As an average, that is nothing. At one stroke every 28 seconds, 2.8 cubic feet times 60 over 28 gives 6.0 acfm, which is exactly the figure on the machine's data sheet and exactly what the drop was sized against.

As an instantaneous rate, it is the whole problem. 2.8 cubic feet delivered in 0.4 seconds is 7.0 cubic feet per second, which is 420 cubic feet per minute of free air. That is 70 times the machine's own average consumption, and 4.2 times the entire compressor's rated 100 acfm. No drop leg sized on 6 acfm delivers 420 acfm, and no compressor in that room could either.

The fix is local, and the reason is in the arithmetic above

Nothing about the plant's average demand needs to change, so nothing about the compressor needs to change. What has to change is where the air for the stroke comes from. Put storage at the machine, on the machine side of a check valve, so the stroke draws from a local reservoir and the drop leg refills that reservoir over the following twenty seven seconds at a rate it can comfortably supply.

The check valve is not optional. Without it the stroke pulls the header down as before and the local vessel simply becomes more pipe.

Size the storage with the same relationship that governs any receiver: the free air held between two pressures is the vessel volume times the pressure difference divided by 14.7. The receiver article carries that relationship and its use in sizing against a demand event.

Here the demand event is 2.8 cubic feet of free air. The vessel starts at the 90 psig the regulator holds, and the deepest it may fall is the 75 psig fault setpoint plus margin, so allow it to fall to 80 psig and no further. That is a 10 psi working band. Vessel volume is 2.8 times 14.7 divided by 10, which is 4.1 cubic feet, or about 31 gallons.

Then check the refill. 2.8 cubic feet of free air over the 27.6 seconds between strokes is 6.1 acfm, which is the same order as the 6.0 acfm average the drop was already sized for. The average never changed. Only the source of the air during the event did.

This vessel is a pressure vessel, not a length of capped pipe. It carries its own indicating gauge, a spring-loaded safety valve and a drain at its lowest point, which 29 CFR 1910.169(b) requires of an air receiver, and it is built and stamped to the ASME Boiler and Pressure Vessel Code, Section VIII, as adopted by the jurisdiction's boiler and pressure vessel law in the edition that jurisdiction has adopted. A fabricated pipe reservoir on a plant air header is the version of this fix that gets someone hurt.

How they confirmed it

Same transducer, same drop, same twenty cycles, machine running its normal product. Minimum inlet pressure through the stroke came up from 71 psig to just above 80 psig, which is 5 psi clear of the 75 psig fault setpoint, and the faults stopped.

Two things were checked deliberately rather than assumed. First, the recovery: inlet pressure had to be back at 90 psig before the next stroke started, because a reservoir that has not refilled turns a single-stroke problem into a progressive one after four or five cycles. It was. Second, the header: with the check valve in place, header pressure during the stroke no longer moved outside its normal band, which confirmed the event was now being served locally rather than pulled through the plant.

The generalisable part is a habit rather than a number. Before accepting that a system is adequate for a machine, ask what the machine's largest single air event is and how long it lasts, then ask what the instrument that cleared the system samples at. When the event is shorter than the sample interval, the instrument has not passed the machine, it has failed to observe it.

References

• 29 CFR 1910.147, control of hazardous energy, for isolation of stored pressure energy before fitting a transducer
• 29 CFR 1910.169, air receivers, including the requirement at (b) for a drain at the lowest point, an indicating pressure gauge and one or more spring-loaded safety valves
• 29 CFR 1910.95, occupational noise exposure
• ASME Boiler and Pressure Vessel Code, Section VIII, as adopted by the jurisdiction's boiler and pressure vessel law in the edition that jurisdiction has adopted
• See related: What a Receiver Is For and What Happens Without One; What a Point of Use Regulator Is Protecting; How to Measure Pressure Drop Across a Compressed Air System