How to Tell a Sensor Fault From a Real Condition

Why this matters

A reading that looks wrong is the most expensive kind of ambiguity on a service call, because both answers are plausible and both have a repair attached. Treat a real condition as a sensor fault and you replace a good part and leave a system running toward whatever the condition was heading for. Treat a sensor fault as a real condition and you chase airflow, staging and refrigerant for two hours on evidence that was never true.

The separation that works is not a better meter. It is a prediction written down before you act, with a direction and a number and a window, so that the observation can fail it. A reading you can only confirm is a reading you cannot test.

The sheet you fill in before you touch anything

Seven fields. Fill fields one through five with the equipment untouched, then act, then fill six and seven. The discipline is that field seven is written before field six exists.

Field 1 - the reading in question, with time and units. Copy it exactly as displayed, no mental correction. If you round it here you will compare against the rounded number later.

Field 2 - a quantity physically coupled to it that you can measure independently. Physically coupled means the two must move together by a mechanism you can name out loud. Fan current and duct static are coupled through the fan curve. Temperature rise and firing rate are coupled through the energy balance. Two temperature sensors in the same airstream are coupled through the air. If you cannot name the mechanism, you have picked a quantity that merely correlates, and it will not settle anything.

Field 3 - the direction the coupled quantity must move if the reading is real. Direction first, on its own line, because direction is where the strongest evidence lives and it is the field people skip straight past on the way to a magnitude.

Field 4 - the magnitude threshold and the settling window. State it as a change greater than a stated amount, measured after a stated number of minutes at steady operation, using an instrument whose resolution is finer than the threshold. Without the threshold you will accept any movement as agreement.

Field 5 - what you will do to force the change. Best case is one input you control, changed once, in a direction you choose. If the process is already changing on its own, say so here and accept that your evidence is weaker, because you did not control the timing.

Field 6 - what actually happened. Raw, both quantities, with times.

Field 7 - the verdict rule. Written before field 6. Use these, and add your own case-specific one if you have it:

That last row is the one that saves the most visits and the one nobody writes down. Two instruments agreeing does not mean the reading is right, it means they share whatever is wrong.

Working the sheet: a duct static reading nobody could explain

A single-zone air handler. The controller shows duct static pressure at 1.0 inches of water column. The commissioning record for this unit shows 0.8 inches at 1,800 feet per minute average duct velocity and 8.2 amps of fan motor current. Nothing has been modified. The complaint is nuisance high-static shutdowns.

Before the meter comes out. Any current measurement means opening an enclosure that carries line voltage. De-energize and prove dead with live-dead-live against a known source, NFPA 70E-2021, 120.5, if the reading can be taken with the unit off, which for a fan current reading it cannot. Since this measurement genuinely requires the fan running, that is the narrow condition 29 CFR 1910.333(a)(1) permits energized troubleshooting under: use a clamp meter and leads rated for the circuit's category and available fault current, plus the shock and arc-flash protection the assessment calls for, and keep your body and the meter's leads out of the plane of any rotating component. Do not open an access door into a section where the wheel is exposed while it turns. If a duct pressure port has to be added rather than used, wear eye protection for the drilling, and confirm what you are drilling into: never put a bit into a duct you have not positively identified as supply or return rather than a flue or vent connector.

Field 1. 1.0 inches water column, 10:20, duct static input.

Field 2. Fan motor current. The mechanism: at a fixed fan speed, a higher system resistance moves the operating point up the fan curve, airflow falls, and shaft power falls with it. On a forward-curved centrifugal wheel that fall is steep; on a backward-inclined or airfoil wheel it is shallower near design. Either way the direction is the same, which is what the prediction needs.

Field 3. If the static is genuinely 25 percent above the commissioned value at unchanged fan speed, motor current must fall below 8.2 amps.

Field 4. At least 0.3 amps below 8.2, so under 7.9 amps, read after 10 minutes of steady operation with a clamp meter resolving to a tenth of an amp.

Field 5. No forcing action available without changing the system. Steady-state observation with the unit at its normal operating point, which is a weaker test and is noted as such on the sheet.

Field 6. Current measured 8.4 amps at 10:31, after 11 minutes steady. Static reading held at 1.0 throughout.

Field 7 applied. The predicted direction was down. The observation is 8.4, which is 0.2 amps above the commissioned 8.2, not 0.3 below it. The prediction failed on direction, which is the strongest of the failures. The static reading is not tracking the process.

Now the second instrument. An independent manometer teed to the same pressure tap reads 1.0 as well. Under the last row of the verdict table, two instruments agreeing on the same port does not clear the reading, it moves the fault upstream of both. Inspect the tap: the pickup is oriented facing into the airstream rather than perpendicular to it, so it is sensing total pressure, which is static plus velocity pressure, not static alone.

Check whether the numbers support that. Velocity pressure at the commissioned 1,800 feet per minute is 1,800 divided by 4,005, squared, which is 0.449 squared, or 0.202 inches of water column. The constant 4,005 assumes standard air at 0.075 pounds per cubic foot near sea level and around 70 degrees; at altitude or at elevated duct temperature the air is lighter and the velocity pressure for the same velocity is lower, so this reconciliation would have to be recomputed on a hot deck or at a high-elevation site.

True static of 0.80 plus a velocity pressure of 0.20 is 1.00, which is exactly what both instruments read. Re-orienting the pickup perpendicular to the flow returns 0.80. The transmitter was never faulty, the shutdowns were never real, and no airflow work was needed.

Why the prediction has to be written down

You can run every one of those measurements without a sheet and reach the wrong answer, because the failure mode is not measurement, it is memory. Standing in front of a panel with a reading you already suspect, you will accept 8.4 amps as "about the same, so nothing conclusive" and move on to a refrigerant gauge set. Written in advance as "must be under 7.9", 8.4 is not inconclusive. It is a failed prediction, and it points somewhere specific.

The same discipline catches the opposite error. If current had come back at 7.4 amps, the prediction would have passed, the static would have been real, and every hour spent on the transmitter would have been wasted. The sheet does not favor either answer, which is the whole reason it works.

What to do when nothing is coupled tightly enough to predict

Some readings have no clean partner. Three fallbacks, weakest last.

Substitution in place. Fit a second sensor of the same type at the same location, confirm the two agree, then swap their positions. If the readings follow the position, the location is producing the value and both devices are fine. If the readings follow the device, you have found the bad one. This is the only method here that separates a device fault from a location effect.

Corroboration by the rest of the system. Ask whether every other observable is consistent with the reading being true. A sibling card covers this route in detail for fault codes; it is weaker than a prediction because it can only ever say "consistent with", never "must be".

Forced excursion. Change one input you control by a known amount and watch which displayed values move. This confirms a point is live and participating, which is not the same as confirming it is accurate, and it is the right test when the suspicion is a stale or unused point rather than a wrong one.

Where a failed prediction does not mean a sensor fault

Two cases, both worth naming before you order anything.

A protective device in the loop. If the reading in question sits in a string with a limit, a pressure switch or an interlock, an end-of-scale value may be that device opening rather than a sensor failing. Establish which, and if it is the device, establish what condition reached it. A protective device that opened because the process arrived at its trip point is doing its job, and replacing it or adjusting around it reaches the same end state as jumpering it.

A coupled quantity that is itself faulted. Your prediction rests on the second measurement being trustworthy. If the fan current reading had come off a network point rather than a clamp meter, a failed direction test would prove only that two numbers disagree. Take the coupled quantity with your own instrument wherever you can, and note it on the sheet when you cannot.

References

• 29 CFR 1910.333(a)(1) - live parts to be de-energized before work, and the narrow conditions permitting energized troubleshooting
• NFPA 70E-2021, 120.5 - process for establishing and verifying an electrically safe work condition
• Manufacturer documentation for fan performance data, commissioning values and any protective device trip points on the equipment
• Trade-standard practice for static pressure tap orientation and for the standard-air velocity pressure relationship
• See related: Codes That Lie: Sensor Failure vs Real Condition; What a Sensor Actually Reports; The Sensor Failure Modes and What Each One Looks Like