How to Verify a Sensor Against the Real Quantity

Why this matters

Comparing a sensor to a second sensor tells you the two disagree. It does not tell you which one is right, and half the time the reference you brought is the one that moved. A real verification puts the element into a condition whose value is known from physics rather than from another instrument, and it does it at two points, because a single point of agreement cannot tell an offset apart from a slope error. Get that wrong and you send a good element to the truck while the loop keeps running against a number that is off by more at the operating point than it was at the bench.

The steps below are ordered by what a shop loses when it skips each one, not by the order your hands do them. Two of them happen before you leave the desk.

1. Decide which link in the chain you are testing

This is the most expensive step to skip, and it is the one skipped most often. A reading is the end of a chain: element, conditioning, transmission, receiving-input scaling, display. A verification that compares a reference against the displayed value passes or fails all five links at once and tells you nothing about which one moved.

Decide up front, and write it on the ticket: are you testing the element and its conditioning (compare the reference to the transmitted signal, read at the transmitter), or the whole chain end to end (compare the reference to the display)? Both are legitimate tests with different meanings.

Skip this and the failure looks like competence. The reference agrees with the display, you mark the sensor good, and the actual fault is a scaling parameter that happens to be right in the middle of the range where you tested. A sibling card covers the chain itself; this procedure only requires that you name your link before you start.

2. Write the pass/fail tolerance, with its unit of analysis, before you take a reading

Second highest loss, and it costs nothing to do. Once you have a number in front of you, you will find a reason it is acceptable.

State it in this shape: the tolerance, the unit of analysis, and whether the two points are judged together or separately. For example: plus or minus 1.0 degree, on a single settled reading at each of two points, and both points must pass, not the average of the two. That Boolean matters. An average passes a sensor that is 2.5 degrees low at the bottom and 2.5 degrees high at the top, which is a slope error large enough to make a loop misbehave at both ends of its range.

Where the sensor feeds a protective function rather than a control loop, the tolerance is not yours to set from experience. Take it from the equipment documentation, because the trip point was selected with an assumed sensor accuracy inside it.

3. Bring the reference to the element's condition, not to the room the element is in

The reference has to experience the same physical condition as the element being tested, at the same moment. A reference laid against the outside of a well, or held in the airstream a foot away from the element, is measuring a different quantity, and the difference you record is not error.

For a temperature element this usually means removing the element from its well and putting the element itself into the known condition. Remove the sensor from the well, not the well from the line. The well is part of the pressure boundary; unthreading it opens the process. If the sensor is direct immersion with no well, the line has to be isolated, depressurized and drained before the element comes out, and it has to be cool enough to touch, because a hot water line scalds well below boiling and a 140 degree surface will do it on contact.

If the equipment is a combustion appliance and any part of this test involves it firing, wear a personal CO monitor for the duration and place it in your breathing zone, not on the equipment.

Skip this step and you get a plausible-looking few degrees of disagreement that is really a thermal gradient, and you will chase it into a parts swap.

4. Test at two points that bracket the operating range

One point tells you the sensor is right at that point. It says nothing about anywhere else, and it cannot separate the two error shapes that behave completely differently in service.

• An offset is a constant number of units added everywhere. Correct it with a single offset parameter.
• A slope or span error grows across the range. An offset correction fixes it at one point and makes it worse at the other.

Two points that bracket the operating range are the minimum that distinguishes them. For temperature, a well-stirred slurry of crushed ice and clean water reads 32.0 degrees at any normal atmospheric pressure, which is what makes it usable in the field. Boiling water is the other end, but its value moves with elevation: near sea level, water boils roughly 1.8 degrees lower per 1,000 feet, so the value must be computed for the site, not assumed. Barometric conditions move it further, so treat a computed boiling point as good to a degree or so unless you measure pressure.

5. Let both settle, against a stated criterion and step size

"Wait for it to stabilize" is not a criterion. Use one: hold the reading until it changes by less than the reference instrument's own resolution over 60 seconds, then record. State the step size, because on a thermowell-style element in a stirred bath that can take several minutes, and the reading approaches its final value asymptotically, which means it always looks nearly settled.

Stir the bath. An unstirred ice bath develops a warmer layer where the element sits, and an unstirred pot of boiling water is not uniform either. Skip the stir and you have introduced the exact gradient error step 3 was written to eliminate.

6. Record raw and uncorrected, both points, before applying anything

Write down what the sensor said, what the reference said, and what correction was already in the controller for that input. Skipping this is a slow loss rather than a fast one: without the previous correction value on paper, nobody can ever tell whether the sensor drifted or somebody else corrected it last season, and the second time the same call comes in you start from zero.

The worked check

Site elevation 3,000 feet. Boiling point computed at 212 minus 1.8 times 3, or 5.4 degrees, giving 206.6 degrees. Ice bath at 32.0. Tolerance stated in advance at plus or minus 1.0 degree per point, both points must pass, testing the element and its conditioning at the transmitter.

Wear heat-resistant gloves for the hot point, keep the vessel on a stable surface at working height rather than on the floor, and check the element's lead insulation temperature rating before immersion - a lead rated for process temperature at the tip is not always rated for immersion along its length.

Results after settling to the stated criterion:

Both points fail the stated 1.0 degree tolerance, so the sensor needs correction. Now separate the two error shapes.

Indicated span is 209.0 minus 33.8, or 175.2 degrees. True span is 206.6 minus 32.0, or 174.6 degrees. The ratio is 175.2 divided by 174.6, or 1.0034 - a slope error of 0.34 percent of span. Across 174.6 degrees that contributes 0.6 degrees at the top. Add it to the 1.8 degree offset measured at the bottom and you get 2.4 degrees, which is exactly the error observed at the top.

So this is an offset of plus 1.8 degrees with a negligible slope error. The correction is a single offset of minus 1.8 degrees, applied at the link you tested. After correction, the top point sits at 209.0 minus 1.8, or 207.2 against a true 206.6: 0.6 degrees high, inside the 1.0 degree tolerance. Both points pass. Done.

Now the case that makes the two-point rule non-negotiable. Suppose the ice bath had matched perfectly at 32.0 and the boiling point read 218.0. Indicated span 186.0 against a true 174.6 is a ratio of 1.065, a 6.5 percent span error, contributing 11.4 degrees at the top. A single-point check at the ice bath passes that sensor with a clean bill of health, and it is 11.4 degrees wrong where the loop actually operates. If the loop drives a limit that trips at a fixed value, 11.4 degrees is the difference between a system that runs and a system that trips, or worse, one that should trip and does not.

Where this method does not apply, and what replaces it

Not every quantity has a field-reproducible known point.

• Pressure has one: atmospheric, at a vented port, which verifies zero but gives you no second point without a deadweight tester or a calibrated reference gauge teed in. Zero-only is a one-point check and carries the one-point limitation exactly as described above.
• Airflow, current and most derived quantities have no known point at all. Here you verify by substitution, not by reference condition: install a second sensor of the same type in the same location, confirm agreement, then swap positions and confirm the readings follow the position rather than the device. That distinguishes a device fault from a location effect, which is the question a reference check would have answered.
• Any sensor whose element cannot be removed without breaking a pressure boundary or a combustion path is not a field-calibration candidate at all. Verify it in place by corroboration against a physically coupled quantity, and replace rather than adjust if it fails.

Where a failed check does not mean a failed sensor

If the sensor passes both points on the bench and the installed reading still disagrees with the process, the element is fine and the coupling is not: an air gap in the well, a fouled tip, a probe that is not in the stream, a location that is reading a different quantity than the one you care about. That is a placement and installation problem, and a sibling card covers it. Correcting a controller offset to paper over a coupling problem builds in an error that is only right at one load.

References

• 29 CFR 1910.333(a)(1) - live parts to be de-energized before work, and the narrow conditions permitting energized troubleshooting, where the check requires reading a signal at a live terminal
• Manufacturer documentation for the sensor's stated accuracy, immersion depth and lead temperature rating, and for the protective device's required sensor tolerance
• Trade-standard practice for ice-point and boiling-point field reference checks, including elevation correction of the boiling point
• See related: What a Sensor Actually Reports; Where a Sensor Is Reporting From and Why It Matters; Why a Drifting Sensor Is Worse Than a Dead One