Where a Sensor Is Reporting From and Why It Matters

Why this matters

Sensor location sets the ceiling on how well a loop can ever control, and no amount of accuracy raises that ceiling. A perfectly calibrated element in the wrong place gives you a correct reading of the wrong thing, and the loop will drive the system to satisfy that reading exactly. The complaint that follows is never "the sensor is in a bad spot", it is "half the building is uncomfortable", and it survives two sensor swaps and a controller replacement before anybody walks the duct.

The case

A single-zone rooftop unit over an open commercial floor. Complaint: the east half is cold, the west half is warm, and it has been that way since the tenant moved in. Setpoint is 75. The controller reads 75.0 and holds it. Two techs before this visit had checked the sensor against a reference in the return plenum and found it within a half degree, which it was.

Before anything else on this call. The unit is on a roof. General industry work near an unprotected edge triggers fall protection at 4 feet under 29 CFR 1910.28(b)(1); the construction counterpart is 6 feet under 29 CFR 1926.501(b)(1), and a service call on an existing rooftop unit normally falls under the general industry Part. Set up the guardrail, warning line or personal fall arrest the site requires before you open a panel, not after. Reaching into any air-handling section past a guard means the blower is isolated, locked and tagged first under 29 CFR 1910.147, with the wheel confirmed stopped by sight, because a belt-drive wheel coasts long after the contactor drops out.

What the reading was actually reporting

The return sensor sat in the return plenum roughly three feet downstream of the point where two return paths merge. It was reading the mixture of both, correctly.

Traverse gave roughly a 60/40 split of return airflow, the larger share from the east side. Spot readings in the occupied space gave about 74 on the east and about 79 on the west. Mass-weighted mixing at those proportions gives 0.6 times 74 plus 0.4 times 79, which is 44.4 plus 31.6, or 76.0. The sensor read 76.0. It was not drifting, it was not badly coupled, it was not miswired. It was accurate.

That mixing arithmetic holds where both streams are at similar density and specific heat and where the sensor is genuinely past the point of full mixing. Where a sensor sits before full mixing, it reads whichever stream happens to wash over its element, and the error is not an average at all - it is the temperature of one zone, wearing the name of the whole system.

Where the loop drove the building

Now run the loop forward. The controller cools until the mixed reading reaches 75.0. Hold the 5 degree spread and the 60/40 split constant and solve for what the two halves must be sitting at when the mixture reads 75.0: 0.6 times x plus 0.4 times x plus 5 equals 75, which reduces to x plus 2 equals 75, so x is 73 and the other side is 78. Check it: 0.6 times 73 is 43.8, 0.4 times 78 is 31.2, and the sum is 75.0 exactly.

So at perfect control, with a sensor reading dead on setpoint, the east half sits at 73 and the west half at 78. Nobody in the building is at 75. The loop is not failing. It is succeeding at a target that does not correspond to any place a person stands.

Holding the spread constant is an assumption worth naming: it is true where the imbalance comes from a fixed cause such as distribution, solar gain or a duct that was never balanced, and it fails where the spread itself varies with load, in which case the error moves through the day and the complaint pattern moves with it.

Why a better sensor cannot fix this

The ceiling this places on the loop has a number. The loop can hold the mixed value to whatever its own stability allows, but the spread between the two halves is untouched by control action, because the control has exactly one output and the two halves respond to it in a fixed ratio. The best achievable outcome with this sensor is that the error is centered: 2 degrees below setpoint on one side, 3 above on the other. Improving the sensor from a half degree of accuracy to a tenth moves the mixed value by a fraction of a degree and moves the spread by nothing.

This is the general shape and it is worth carrying past this example. A sensor that reads an average controls the average. A sensor that reads one location controls that location. Neither controls the spread. Every location decision is a choice about which quantity the system will be made true.

What each location choice actually buys

• In the space, at the worst-case location. The loop protects the worst spot and over-serves everywhere else. Choose this when the risk is one-sided, for example a freeze risk or a process that must not exceed a limit.
• In the space, at a representative location. The loop centers the error. Choose this when the spread is small and comfort is the only stake. Verify "representative" by measuring, not by where the conduit was easy to run.
• In the return path, after full mixing. The loop controls a flow-weighted average. Choose this for load-following on a system whose zones are genuinely similar, and never on a system with a known spread.
• In the discharge. The loop controls what the system is producing, not what the space is experiencing. This is a supply-side quantity and it needs a space-side reset above it or it will hold a fixed discharge into a changing load.
• On the equipment itself. Almost always a protective location rather than a control one. A limit, a freeze protection device or a rollout switch is placed where the hazard is, not where the control needs to see.

That last line is the distinction that matters most on a service call. A protective device's location was chosen to catch a condition. A control sensor's location was chosen to represent a quantity. When a protective device opens, do not treat it as a badly located control sensor and move it or bypass it. Establish what condition reached it: a limit that opens because the process genuinely arrived at its trip point is working, and relocating it to a spot where it stops opening reaches the same end state as jumpering it.

How this was confirmed on site

Three measurements, in this order, because each one eliminates a class of explanation the next one would otherwise still have to consider.

Independent space readings at both ends, at the same time, with the same instrument. Two calibrated readings taken minutes apart in a building with a swinging load prove nothing; take them together or take them repeatedly. This established that the spread was real and not an artifact.

A flow split at the return branches. Without this the mixing arithmetic has no weights, and 74 and 79 mixed to 76.0 only works at 60/40. At 50/50 the same two temperatures give 76.5, which would have pointed at a sensor error of a half degree instead of at the sensor being correct. The weights are what made the finding conclusive.

A reference reading beside the sensor's element. Half a degree of agreement. That closes the door on drift and coupling, and it is the check the previous two techs performed on its own, where it was necessary but told them nothing about placement.

Finding out where a sensor actually is when nobody wrote it down

On most of the systems you will service, no drawing exists, the drawing that exists is wrong, or the sensor was moved during a remodel and nobody updated anything. Three methods, in the order that costs least, and the third one is the only one that is conclusive.

Trace the cable. Slow, reliable where the run is accessible, useless above a hard ceiling. Worth starting because it costs nothing but time and it often gets you within one room.

Read the response timing. Force a change somewhere you can reach and watch how long the point takes to respond. A point that moves within seconds is in the airstream near the disturbance. A point that takes minutes is either far away, heavily damped, or inside a well. This narrows the search; it does not end it, because a slow response and a distant location look identical from the panel.

Force a local change at each candidate and watch which point moves. This is the conclusive one. Warm one candidate location gently and see which displayed value responds and which do not. Gently is the operative word: use a warm damp cloth or a hand held near the element, not a heat gun, because a heat source held against a sensor can exceed the element's rated temperature and damage it permanently, and a heat gun inside an enclosure or a duct also risks the insulation on nearby wiring. If the space is occupied, tell somebody first, because you are about to make a controller respond.

Label what you find, on the sensor and in the record, the moment you find it. The whole reason this section exists is that the last person who knew did not write it down, and the next visit repeats every one of these steps from zero.

Where the answer changes

On a system with real zone control, the same spread is a distribution problem to be solved by the zone dampers and their own sensors, and a poorly placed system sensor matters much less because it is no longer the only voice.

On a process with a hard limit rather than a comfort target, centering the error is the wrong choice entirely. Move the sensor to the worst case and accept the over-service, because the cost of the excursion is not symmetric with the cost of over-serving.

Where the spread varies with load rather than sitting fixed, no single location centers the error across the day. That is the condition that justifies adding a second sensor and controlling from the greater of the two, or the lesser, depending on which side carries the risk.

References

• 29 CFR 1910.28(b)(1) - general industry fall protection at 4 feet from an unprotected side or edge
• 29 CFR 1926.501(b)(1) - construction fall protection at 6 feet, the counterpart duty for work classified as construction
• 29 CFR 1910.147 - control of hazardous energy, for isolating and locking out a blower before reaching into an air-handling section
• Manufacturer documentation for the controller's sensor placement requirements and for the protective devices installed on the equipment
• See related: What a Sensor Actually Reports; How to Tell a Sensor Fault From a Real Condition