Measuring Temperature and the Four Ways It Goes Wrong

Why this matters

A tech who does not trust a temperature reading re-takes it. Then re-takes it in a different spot, swaps probes, borrows someone else's instrument, and half an hour later has four numbers and no decision. The re-measure loop is expensive and it never resolves, because every one of those attempts was aimed at an unnamed problem.

There are four routes by which a temperature reading ends up different from the temperature you wanted, and they need different fixes. One of them fixes itself if you wait, one gets no better with waiting, and one is not an instrument problem at all. A tech who names the route first fixes it on the second reading instead of the fifth.

Before a probe goes into anything

Hot surfaces and hot fluids burn on contact, and the burn is worse through a wet glove. Read hot pipe, flue and cabinet surfaces with a non-contact instrument or a probe on a handle, and if you must touch, use gloves rated for the temperature by their own manufacturer's marking, not general work gloves.

Do not drill a test port into anything that carries pressure, gas, fuel or refrigerant. A duct wall and a plenum are drillable. A pipe, a line set, a vessel and a header are not; they get a fitting installed under isolation or a thermowell designed in. An immersed reading in a pressurized line with no well is a shutdown job under 29 CFR 1910.147, with the line isolated, drained and residual energy relieved and verified before anything is opened.

Drilling or opening duct insulation or pipe lagging releases fibers into your breathing zone. Man-made mineral fiber needs a respirator selected and fitted under a program meeting 29 CFR 1910.134, not a nuisance dust mask. Where the material is unidentified and the building old enough that it could be asbestos-containing, do not drill: it is presumed asbestos-containing until sampled, with handling at 29 CFR 1910.1001 for general industry or 29 CFR 1926.1101 for construction, so name which Part your job falls under before anyone cuts. A glove does not address an airborne route.

Four routes, and why the route is the thing to name

Every bad temperature reading is one of these, and the symptom that distinguishes them is how the number behaves over time and how it responds to being moved.

1. Coupling. The sensor is not at the temperature you want, because heat is not getting between the target and the sensing element well enough.
2. Loading. The sensor changed the thing it is measuring, by conducting heat into or out of it.
3. Response time. The sensor is on its way to the right value and you read it before it arrived.
4. Wrong quantity. The sensor is correct, settled and honest, and it is reporting a temperature other than the one your decision needs.

Route 1: coupling

A sensing element reports its own temperature. Nothing more. Everything else is an assumption that the element and the target are at the same temperature, and that assumption is only as good as the thermal path between them.

Surface work is where this bites hardest. A probe tip resting on a pipe with an air gap under it, or held at an angle so only an edge touches, sits in a thermal divider between the pipe and the room air washing over it, and it reads a plausible number somewhere between the two. The fixes are all about the path: clean the surface, use a purpose-made surface probe with a compliant tip, add thermal compound or tape over the tip, shield the tip from moving air. Immersion work has the same problem in a different shape: a probe in a thermowell with no compound and no contact at the bottom is reading the air trapped in the well.

The direction is not always the same, which is why it cannot be corrected out. On a target hotter than the room, poor coupling reads low. On a target colder than the room, poor coupling reads high. A tech who assumes surface readings always run a bit low will be wrong on every refrigerant line they touch.

Route 2: loading

The probe is a heat path in its own right. It touches the target at one end and the room at the other, and the metal between conducts. On a hot target it drains heat out of the contact area and reads low; on a cold target it feeds heat in and reads high. The error is largest where the target has little mass or little ability to replace the heat you are stealing: a small component, a thin fin, a low-velocity air stream.

Stem conduction is the immersion version. Heat travels along the stem from the fluid, out through the wall or fitting, and into the room, so the tip never fully reaches fluid temperature. The remedy is immersion depth. The common field guideline is at least ten probe diameters, and where the probe manufacturer states a minimum immersion depth that stated value governs, with the ten-diameter rule of thumb only a sanity check on it. A rule of thumb and a manufacturer's specification are two different kinds of number.

The tell that separates loading from response time: a loading error settles. The number stops moving and sits at a steady offset, and waiting longer changes nothing. Response time looks identical for the first few seconds and then behaves completely differently.

Route 3: response time

A sensor approaches the true temperature exponentially. Its time constant is the time to cover about 63 percent of the gap between where it started and where it is going. After three time constants it has covered about 95 percent, after five about 99 percent.

The time constant on a data sheet is specified under stated conditions, and those are almost always favourable: moving fluid or moving air at a defined velocity. The same probe in still air is several times slower, because the heat has to get into it and still air is a poor delivery mechanism. A probe parked in a dead zone behind a turning vane is not operating at its rated time constant, and no amount of confidence in the specification changes that.

The tell: a response-time error is still moving. If the display is drifting in one direction, you are watching route 3 and the fix is to wait, not to move the probe.

Route 4: the wrong quantity

The instrument is right and the question was wrong. Air temperature, radiant temperature, surface temperature and the bulk mean temperature of a flowing stream are four different quantities that can differ by a large margin in the same location. A globe thermometer and a dry-bulb probe in one room disagree by design, and the sibling card on how air temperature and radiant temperature differ owns that comparison. A single-point reading in a duct is not the stream's mean unless the stream is well mixed, which the sibling card on stratification covers.

The tell: the reading is steady, repeatable, survives every fix for the other three routes, and still does not agree with what the equipment is obviously doing. That is the signature of a correct measurement of the wrong thing.

Routing table

Worked example: one duct probe, two of the four routes

Measuring supply air temperature in a duct to get a temperature split across a coil. Return air measures 75 F and is stable. The probe's data sheet lists a time constant of 12 seconds in moving air at the velocity it specifies; the probe was in a 72 F equipment room before insertion; true supply air is 55 F. Those figures are illustrative.

First reading, taken at 10 seconds. The gap the sensor has to cross is 72 minus 55, which is 17 F. Ten seconds is a little under one time constant, so about 43 percent of the gap remains: the remaining gap is 17 multiplied by 0.43, or about 7.3 F, and the display reads about 62.3 F.

Compute the split from that number and it is 75 minus 62.3, or 12.7 F. The true split is 75 minus 55, or 20 F. The reading understated the split by 7.3 of 20 F, about 36 percent, and 12.7 F on a coil that should give 20 F is exactly the size of error that sends a tech looking for a charge problem that does not exist.

Second reading, same spot, held longer. At three time constants, 36 seconds, about 5 percent of the gap remains: 17 multiplied by 0.05 is about 0.9 F, so the display reads about 55.9 F. At five time constants, 60 seconds, about 1 percent remains, and the display reads about 55.1 F. Route 3 is now closed, and it closed by doing nothing except waiting.

Except this probe settled at 57.5 F and stopped. It held there for a further minute without moving. That is the loading tell from route 2: response time is still moving, loading has settled. Waiting longer will not recover the remaining 2.5 F.

The probe was inserted through a short test port with only a small part of its length in the air stream, so heat was travelling up the stem, through the duct wall, and into the 72 F equipment room. Re-inserted to the manufacturer's stated minimum immersion depth, with the port collar sealed so room air was not being drawn along the stem, it settled at 55.3 F.

Recomputing the split against the honest number: 75 minus 55.3 is 19.7 F, against the 12.7 F the first reading supported. The coil was fine.

Why the sequence had to run in that order. Waiting first was correct, because until the number stops moving you cannot tell a loading offset from a partially-arrived reading. A tech who starts adjusting immersion while the probe is still travelling is chasing a number that would have moved anyway. Route 3 has to be closed before route 2 can be seen.

The failure mode. The tech who reads at 10 seconds gets 62.3 F, computes 12.7 F of split, and reports a coil problem. The one who waits but does not check immersion gets 57.5 F, computes 17.5 F of split, and reports a marginal coil. Both numbers are stable enough to look like measurements, and neither is wrong by an amount that trips anyone's suspicion.

What would change the answer. Had the settled offset persisted after full immersion and a sealed port, routes 2 and 3 are both closed and route 4 is the remaining candidate: an honest reading at one point in a stratified stream, answered by a traverse rather than a deeper probe. And had the target been colder than the room, both the coupling and the loading errors reverse sign, so a tech carrying "surface readings run low" as a rule would have corrected the wrong way.

How to verify you got this right

• Watch the number for ten seconds before you write it. Moving means you are not done. This catches route 3 every time and costs nothing.
• Change one thing and re-read. Deeper, better contact, shielded from air. If the number moves, you found a real error. If it does not, that route is closed.
• Take a second reading by a different method on the same target. Two methods agreeing rules out most of routes 1 through 3 at once.
• Say out loud which temperature the decision needs. Air, surface, radiant, or the mean of a stream. If you cannot say it, route 4 is open regardless of the instrument.

References

• 29 CFR 1910.147 - the general lockout standard, governing isolation and relief of residual energy before a pressurized or fluid-carrying line is opened
• 29 CFR 1910.134 - respiratory protection program requirements where insulation fiber is disturbed
• 29 CFR 1910.1001 (general industry) and 29 CFR 1926.1101 (construction) - asbestos requirements, including the presumption applied to unidentified thermal system insulation
• Probe manufacturer's data sheet for time constant, its specified test condition, and minimum immersion depth
• See related: How Air Temperature and Radiant Temperature Differ; The Stratification That Fools a Single Reading; How to Read a Surface Temperature Honestly