Sensor Calibration and Offset Record

Purpose

This standing instruction decides whether a control sensor needs an offset, applies one only when the evidence supports it, and writes down everything needed to undo it. An offset is a correction to a measurement error. It is not a fix for a sensor in the wrong place, and it is not a way to make a customer's complaint agree with a display.

This needs a standing instruction because offsets stack invisibly. A thermostat carries one, a zone panel another, an equipment board a third, and none announces itself. Two years later the house runs cold, the next tech finds a display that agrees with his thermometer, and the two degrees hiding in a panel he never opened are doing the damage. Every one of those entries was made by someone solving a real complaint. What was missing was the line saying what was corrected, against what reference, on what date.

Safety actions that gate this procedure

• Where a step opens a panel or an equipment cabinet to read or enter an offset, open and lock the disconnect and prove dead first, live-dead-live on a known live source before and after per NFPA 70E-2021, 120.5, with work practices at 29 CFR 1910.333(b)(2), written for a qualified person under 1910.332.
• Relocating a sensor means opening a wall or a ceiling. Do not cut blind, and treat old plaster, textured coatings and joint compound as a disturbance rather than an access: unidentified material of that age goes to an assessment before a blade touches it.
• Do not simulate a sensor value, force an output or override a point on a protective loop - a limit, a freeze sensor, a pressure switch - to create a test condition. The extreme that drives the equipment to is the same hazard as moving the element by hand.

Scope

Covers verifying and correcting temperature and humidity sensors that control residential and light-commercial HVAC equipment: thermostat internal sensors, remote room sensors, outdoor sensors, and supply or plenum sensors.

Does not cover laboratory calibration or issuing a traceable certificate. Does not cover sensor placement design on a new install, or zone panel commissioning. Does not cover diagnosing a sensor that is open, shorted or reading wildly, which is a replacement and not a calibration question.

Roles and handoffs

Procedure

1. Establish what your reference instrument is actually worth, and check it. Acceptance: the accuracy figure written with its basis - percent of reading, percent of span, or a fixed value in degrees - and its character stated as a worst-case bound rather than a spread; plus a single-point check in a well-stirred ice-water slurry, which sits at 32 F, read and recorded. Wrong looks like "my thermometer is accurate," which supports no decision band. Stop rule: an ice-point check outside the instrument's own bound and it goes out of service until replaced or re-certified. Hazard: none here, bench work.

2. Find and read every place an offset can already exist on this system before you change one. Acceptance: a written list of offset locations - thermostat, each remote sensor, zone panel, equipment board - with the current value at each, including the zeros. Wrong looks like a correction entered at the thermostat while a previous tech's undocumented entry sits in the panel, so the system carries two fixes for one complaint. Stop rule: an offset you find but cannot explain is recorded and left in place until step 5 can judge it. Hazard: reading a panel or board value means an open enclosure, so the electrical gate above applies.

3. Verify the sensor's placement before you consider an offset at all. Acceptance: the sensor confirmed off exterior walls, out of direct sun, out of any supply airstream, and not over a cavity or chase carrying duct, flue or plumbing. Wrong looks like an offset entered to cancel a sensor above a warm chase, which makes the display agree with a thermometer at the same wrong place and does nothing for the person in the room. Stop rule: a placement fault gets the sensor moved, and no offset is entered on that sensor until it has been re-measured at the new location. Hazard: opening the wall is governed by the cut-and-disturbance gate above, and the old cable is capped and labeled rather than left live in the cavity.

4. Co-locate the reference with the sensor and take the comparison after a stated settle time. Acceptance: both readings at the same point after the settle time the reference manufacturer states or 15 minutes, whichever is longer, repeated once a few minutes later to prove stability; record every value to 0.1 F. Wrong looks like a reading taken as the probe is placed, which measures the probe warming up. Stop rule: two comparisons disagreeing by more than the reference's own bound means nothing has settled, so wait and repeat rather than averaging. Hazard: none here, instrument work in an occupied room.

5. Compute the difference, compare it against the combined bound, and only then decide. Difference is the sensor reading minus the reference reading, so a positive difference means the sensor reads high, and the offset entered is the negative of it. Acceptance: the difference printed, and the combined bound printed as the reference's bound plus the sensor's published bound added together, because two worst-case bounds from two different instruments add rather than combining in quadrature. Wrong looks like a half-degree offset when neither instrument resolves half a degree, which is noise written into a control. Stop rule: a difference inside the combined bound and no offset is entered; you cannot correct what you cannot resolve. Hazard: none here, arithmetic at the wall.

6. Repeat the comparison at a materially different room condition before applying any offset. Acceptance: a second difference computed the same way at a temperature well away from the first, agreeing with the first within the reference's bound. Wrong looks like a fixed offset against a sensor whose error grows as the room gets colder, correcting one condition and worsening the other. Stop rule: a difference that changes with condition means the sensor is replaced, not offset. Hazard: driving the building to a second condition is a command - do not take a space below the temperature that protects plumbing in the coldest room served, and do not create the condition by simulating a value on any protective loop.

7. Enter the offset, write the log line, and re-verify anything protective that the entry touched. Acceptance: a log entry carrying sensor identity and location, reference instrument and its accuracy basis, both raw readings at both conditions, the computed difference, the offset AS ENTERED, the running total across all layers, the date and the tech; plus a re-read confirming the corrected display now agrees with the reference within the combined bound. Wrong looks like an offset entered and no line written, which is the condition this procedure exists to stop creating. Stop rule: where the sensor feeds a protective function such as a supply air limit or a freeze cutout, the offset has moved that threshold, so re-verify the trip margin on a live run before the visit closes, or do not enter the offset. Hazard: that verification run puts the equipment back in service with the customer nearby, so doors fastened, stand to the hinge side of the panel, and nobody in the blower compartment before the call.

The record this produces

One log line per sensor touched, filed to the equipment record:

• Sensor identity and location, including whether it was relocated on this visit
• Reference instrument, its accuracy figure with the basis named, and the ice-point result
• Raw sensor and raw reference readings at both conditions, to 0.1 F
• The computed difference at each condition, and the combined bound it was judged against
• The offset AS ENTERED, and the running total of every offset on the system
• Any offset found and removed, with the reading that justified removing it
• The protective-function re-verification where a limit or cutout sensor was involved

The running total earns this record its place. A single entry looks harmless in isolation, and the only way anyone sees a system carrying several is a line that adds them up.

Worked pass: thermostat reads warm, occupants report the house is cold

Step 1: reference thermometer's certificate states plus or minus 0.5 F over the range in use, expressed as a worst-case bound. Ice-water slurry check reads 32.3 F, inside that bound, so the instrument stands. The thermostat manufacturer's manual publishes a sensor bound of plus or minus 1.0 F, so the combined bound for this comparison is 0.5 plus 1.0, which is 1.5 F.

Step 2: three offset locations exist. The thermostat's own offset reads 0.0 F. The remote bedroom sensor carries a minus 2.0 F offset entered at an unrecorded date. The equipment board has no offset field. All three recorded as found.

Step 3 FAILS. The thermostat is mounted on an exterior wall directly above the chase carrying the supply run to the room over it. Stop rule taken: no offset is entered on that thermostat, and it is relocated to an interior partition wall on the same level, clear of the chase and out of the supply throw. The old cable is capped and labeled at both ends.

Step 4, at the old location before the move: reference 68.4 F, thermostat 72.0 F. At the new location after a 20 minute settle and a repeat five minutes later: reference 70.2 F, thermostat 70.9 F, repeat matching.

Step 5: difference at the old location is 72.0 minus 68.4, which is plus 3.6 F, well outside the 1.5 F combined bound and the reason for the complaint. Difference at the new location is 70.9 minus 70.2, which is plus 0.7 F, inside the 1.5 F bound. No offset is entered on the thermostat, because the fault was a location and correcting it with a number would have left the occupant exactly where they started.

Step 6: second condition the following morning, 62.5 F on the reference against 63.1 F on the thermostat, a difference of plus 0.6 F. That agrees with the 0.7 F from the warmer condition within the reference's 0.5 F bound, so the residual is fixed rather than drifting, and inside the band regardless.

Step 6 also settles the remote sensor. With its minus 2.0 F offset still in place it displays 68.1 F against a reference of 70.2 F at the same point, so its raw reading is 68.1 plus 2.0, which is 70.1 F. Raw against reference is 70.1 minus 70.2, a difference of minus 0.1 F, inside the combined bound. The offset does nothing except make a good sensor read two degrees cold, so it is removed and logged with the readings that justified it.

Step 7: log lines written for both sensors. Running total across the system is now 0.0 F. No supply air limit or freeze sensor was touched, and that is recorded as not applicable rather than left blank. A full cooling call is run with the doors on and the interlock proved.

Checking this pass: step 5's acceptance asks for the difference and the combined bound both printed, and every difference above shows its subtraction inline against the 1.5 F band from step 1. Step 4's stated precision is 0.1 F and no figure finer than that appears anywhere in the run. The remote sensor's raw reading is reconstructed as a printed line rather than asserted, so the removal decision can be checked by anyone reading the log.

References

• Reference instrument manufacturer's accuracy statement or calibration certificate, which supplies the basis and the bound every decision above rests on
• Thermostat, sensor and control manufacturer's manuals, for published sensor accuracy and where offsets can be entered
• NFPA 70E-2021, 120.5 and 29 CFR 1910.333(b)(2) with 1910.332, for de-energizing and proving dead before opening a panel or board in step 2
• See related: the smart thermostat setup SOP, which places remote sensors and defers offsets here; the zoning panel commissioning SOP, which owns the supply air limit sensor