The Temperature You Measure and the Temperature That Matters

Why this matters

A sensor does not read the fluid. It reads itself. Whatever temperature the sensing element settles at is the balance point of every heat path running into and out of that element, and the fluid is only one of them. When the other paths are small, the two numbers agree and nobody thinks about it. When they are not small, the sensor is honestly reporting its own temperature while the system it controls runs somewhere else entirely.

That error is not noise and it does not average out. It is systematic, it always points the same direction, and its size can be worked out from the paths involved. The job below is one where a correctly functioning sensor, a correctly functioning controller, and a correct control strategy combined to run a building nine degrees off target for two seasons.

Before you touch a control or a hot pipe

• Working inside a boiler or air handler control panel. Line voltage is present alongside the low-voltage control terminals, and control work is electrical work under 29 CFR 1910.333(b)(2) rather than the mechanical lockout standard. Prove your meter on a known live source, prove the circuit dead, prove the meter again, which is the sequence at NFPA 70E-2021, 120.5. De-energize before landing test leads on control terminals; a slipped probe across an adjacent line-voltage terminal is the common injury here, not the sensor circuit itself.
• Strapping or clamping anything to a hot pipe. Heating water above 140 F scalds on contact within seconds, and the strap and the pipe will both be at fluid temperature. Wear gloves rated for the contact temperature, and let a section cool before working on it where the schedule allows.
• Lifting pipe insulation to reach the metal. Thermal system insulation in older buildings is presumed asbestos-containing until sampled, and tearing it puts fibers in your breathing zone. Read at an already-open section or have it sampled first. The duty sits at 29 CFR 1910.1001 in general industry and 29 CFR 1926.1101 for maintenance and renovation that disturbs it.
• Anything on a fuel-fired appliance. Do not adjust firing or air without a calibrated combustion analyzer, because you can make an appliance produce carbon monoxide while it looks and sounds normal. If a CO alarm sounds while you are on site, everyone leaves the building immediately, no switches are touched, nothing is used to make a call inside, and the call goes out from outside.
• Never jumper, bypass, or remove a high limit to test a control theory. The limit is the last protection between an overheating appliance and a failure, and a jumper left in place is a fire and scald hazard for whoever comes next.

The signal

Small commercial building, two-pipe hydronic heat with outdoor reset on a condensing boiler. Complaints in two directions at once: occupants too warm on mild days, and the boiler short cycling and occasionally locking out on high limit. The building had run this way for two heating seasons and had been visited three times.

First measurement was the cheapest one available. The controller displayed 152 F supply. An independent thermometer in an existing well three feet downstream on the same pipe, with no branch between the two, read 161 F.

Nine degrees apart on the same water. Everything after that was working out which of the two was lying and why.

What a sensor actually reports

Picture two paths at the sensing element. One conducts heat from the pipe into the sensor, and it has some resistance set by the surface finish, the clamping pressure, and whether there is thermal compound or an air gap in between. The other carries heat away from the sensor to the surrounding air, by convection off the sensor body and the strap and by radiation to the room, and it has its own resistance set mostly by how well the sensor is covered.

The element settles where those two balance. Write the pipe temperature as Tp, the room as Tr, and the fraction of the total path sitting on the contact side as f:

Sensor reading = Tp - (Tp - Tr) x f, where f = R(contact) / (R(contact) + R(loss)).

Everything useful follows from that one relationship.

• The error is proportional to how far the pipe is above the room. The same installation that is nearly right on a 90 F pipe in a 70 F room is badly wrong on a 200 F pipe.
• The error always reads low on a hot pipe and high on a cold one, because the room is always pulling the sensor toward itself. It never scatters both ways.
• Insulating over the sensor attacks the term that matters. It multiplies R(loss), and f falls in direct proportion.
• A poor contact and a missing cover compound each other, because one raises the numerator while the other lowers the denominator.

Three candidates, eliminated on evidence

The reset curve. Checked against the design documentation and correct for the outdoor temperature that day. A wrong curve would produce a building that is uniformly too warm or too cool, not a controller disagreeing with a thermometer on the same pipe. It also would not explain the high-limit lockouts. Eliminated.

The sensing element. A known-good sensor of the same type was fitted in the same location, in the same strap, under the same conditions. It read within a degree of the original, still about 8 F below the well. Two independent elements agreeing with each other and disagreeing with the fluid is not element drift. Eliminated.

The controller's own scaling. With the circuit de-energized, the sensor leads were lifted and a precision resistance substituted at the terminals corresponding to a known temperature. The controller displayed within a degree of that value. The input path is sound. Eliminated.

What was left was the installation. The sensor was strapped over a painted, slightly rough pipe surface with no thermal compound, and the pipe insulation stopped about two inches short of the strap, leaving the sensor body and its strap sitting bare in 68 F room air.

The number that located it

Pipe 161 F, room 68 F, so the pipe sits 93 F above the room. The observed error is 9 F.

f = 9 / 93 = 0.097.

Rearranging, R(loss) / R(contact) = (1 - f) / f = 0.903 / 0.097, about 9.3.

The escape path to the room is only about nine times the resistance of the contact path. On a properly insulated installation with a clean, compounded contact that ratio should be very large, and the resulting error should disappear into the instrument's own tolerance. Nine is not a marginal installation. It is a sensor that is roughly a tenth in the room.

What nine degrees cost, in three places

The controller drives until its sensor reads the target. If the sensor reads 9 F low, the water runs 9 F high. At a reset target of 160 F the loop was actually running near 169 F, and that showed up in three separate places.

Distribution loss. Heat lost from piping scales with the difference between the water and the space around it. At 169 F in a 68 F space that difference is 101 F against the intended 92 F, so distribution loss ran roughly 10 percent higher than designed, all of it delivered into ceilings and chases rather than into occupied rooms.

Cycling margin. The high limit was set at 180 F. Against the intended 160 F supply that leaves 20 F of headroom for a light-load overshoot. Against the actual 169 F it leaves 11 F, barely half. On a mild day with most zones satisfied, the boiler ran into the limit and locked out, which is exactly the complaint that had brought techs out three times.

Condensing operation. Return water measured 138 F. A condensing boiler only recovers latent heat from the flue gas when the return water is below the flue gas dew point, which for natural gas at typical excess air sits around 130 F; the exact figure moves with fuel and with excess air, so use the boiler's own curve. At 138 F return, this boiler had not been condensing at all. The whole reason it was specified was sitting unused.

Predicting the fix, then confirming it

The correction was three parts, each attacking one term: clean the pipe to bright metal under the strap, apply thermal compound, and extend the insulation over the sensor and at least six inches past it in both directions at the same thickness as the run.

Predicting the result is worth doing before you do the work, even roughly. Covering the sensor with the same insulation as the pipe multiplies R(loss) by a large factor, and compound on a cleaned surface cuts R(contact) meaningfully. Using illustrative figures of an eight-fold increase in R(loss) and a halving of R(contact), f falls from 0.097 to roughly 0.007, which on a 93 F difference is an error of about 0.6 F. Those two factors are estimates, not measurements, so treat the prediction as an order of magnitude: the error should land under a degree rather than at some improved-but-still-wrong value like four degrees.

Measured after the work, same conditions: controller 160.4 F, well 161.0 F. Six tenths of a degree apart.

With the reading honest, the reset control drove the loop to its actual 160 F target. Return water settled at 129 F, back under the dew point, and the boiler began condensing. The high-limit lockouts stopped because the headroom was back to its designed 20 F.

Note what the verification is, and what it is not. It is a comparison against an independent thermometer at the same point on the same fluid, which is the only check that can catch this class of error. It is not "the building is comfortable now," and it is not the controller agreeing with itself.

Where the same error hides elsewhere

The mechanism is not specific to strap-on sensors. Any sensor with a competing heat path shows it, and four are worth knowing by shape.

Immersion wells. A well conducts heat along its own stem out to the fitting and the pipe wall, so a probe that does not reach far enough into the stream reads partway between the fluid and the pipe. Insert to the depth the well was designed for, and close the air gap between probe and well with thermal compound or a heat-conductive fill where the manufacturer permits it, because still air in that gap is a large resistance sitting exactly on the contact path.

Infrared readings on bright metal. An infrared instrument infers temperature from emitted radiation, and polished copper or bright steel emits very little, so the instrument reads far below true and picks up reflections of whatever is nearby. Put a patch of matte tape or flat paint on the surface, let it come to temperature, and read that instead. This is a different failure from the one above, because the reading is wrong before any heat path is involved.

Air-stream sensors near a coil or a burner. A sensor in an air stream also sees the surfaces around it by radiation. A duct sensor a short distance downstream of a hot heat exchanger reads above the true mixed air temperature, and one just downstream of a chilled coil reads below it. Move it far enough downstream for the stream to mix, or shield it from line of sight to the hot or cold surface.

Stratified air. A single point in a large duct or a tall space is not the average of it. Temperature stratifies vertically and across a duct after a turn or a coil, so a single sensor is reporting one streamline. Where the average matters, traverse the section once at commissioning and record how far the sensor's location sits from the average, then apply that offset knowingly rather than assuming it is zero.

References

• ASHRAE Standard 41.1, Standard Method for Temperature Measurement, for sensor placement, insertion depth, and stem conduction error
• ASHRAE Handbook, Fundamentals volume, chapters on measurement and instruments and on radiant exchange
• 29 CFR 1910.333(b)(2) for electrical work at control panels; NFPA 70E-2021, 120.5 for live-dead-live proving
• 29 CFR 1910.1001 and 29 CFR 1926.1101 for asbestos-containing thermal system insulation
• See related: How to Read a Temperature Profile Along a Run; What a Phase Change Does to a Temperature Reading