The Sensor Failure Modes and What Each One Looks Like
Why this matters
Sensor faults do not present as wrong numbers, they present as wrong shapes, and the shape names the link that failed. A value pinned at the end of scale, a value that never moves, a value that jitters and a value that arrives late are four different repairs, and only one of them is a new sensor. Techs who read the value instead of the shape end up buying the part, because the value is on the screen and the shape only exists across time, which means it only exists if somebody wrote it down.
This card is a signature-to-link map plus the log that produces it. The chain those links belong to is covered by a sibling card and is not re-derived here.
The signature table
Read the left column as the shape you observe at the display over time, not as a single reading.
| Signature at the display | What is happening | Which link failed |
|---|---|---|
| Pinned at an end of scale, steady | Open or shorted element or wiring, or the input has entered its own declared fault state | Element or its wiring |
| Plausible value that never changes across a real change in conditions | Stale data: a digital point holding its last received value, a bus device not responding, a frozen conversion | Transmission or scaling |
| Correct average with rapid jitter on top | High-resistance joint, induced noise on a low-level signal, an unterminated or double-grounded shield | Conditioning or transmission |
| Correct value, arriving late | Coupling: fouled element, air gap in a well, added mass, probe out of the stream | The physical coupling |
| Correct shape, wrong value, stable and repeatable | Drift, self-heating, or a scaling range set wrong | Element or scaling |
| Moves in the wrong direction when conditions change | Reversed polarity, inverted scaling, an element type set wrong on the input | Transmission or scaling |
| Correct until a specific event, then jumps and recovers | Intermittent connection, mechanical or thermal | Transmission |
| Two readings both plausible, both attached to the wrong equipment | Sensors swapped at a terminal strip, or two identical labels | Transmission |
Two entries need their conditions attached in the same breath.
Which rail an open circuit produces is not universal. It depends on the input circuit's topology and on whether the element's resistance rises or falls with the measured quantity. The same broken wire reads full scale on one controller and bottom of scale on another. Learn your controller family's convention once, from its documentation, and write it on the panel; do not carry a rule of thumb across brands.
A steady rail value may not be a sensor at all. Some designs run a protective device in series with an input, so a limit or a pressure switch opening produces the same pinned reading a broken wire does. Before you touch the sensor, establish which of the two you are looking at, and if it is the protective device, establish why it opened. A device that opened because the process genuinely reached its trip condition is working, and replacing it reaches the same end state as jumpering it, one step slower.
The log is the instrument
None of the shapes above are visible in a single reading. The artifact that makes them visible is a five-column log, and it is worth carrying as a form.
| Field | What goes in it | Why it is there |
|---|---|---|
| Clock time | Real time of day, to the minute | Lets you line events up against anything else that is timestamped |
| Displayed value | What the controller shows, raw, no mental correction | This is the thing whose shape you are reading |
| Independent value | A handheld reading taken at the element, same moment | Separates a wrong reading from a real condition |
| Output state | What the controller is commanding right now | Tells you whether the loop is acting on this input at all |
| Event note | Anything that started, stopped, opened or was touched | This column is where intermittents get solved |
The event column is the one people leave blank and the one that carries the answer. An intermittent is defined by what coincides with it.
A filled-in log, and what it says
Complaint: a heating system that stops for a minute or two at random, several times a day, and runs normally the rest of the time. Input scaled minus 40 to 250 degrees.
| Time | Displayed | Handheld at element | Output | Event note |
|---|---|---|---|---|
| 09:12 | 118.4 | 118.1 | Heat stage 1 | Steady |
| 09:18 | 118.6 | 118.3 | Heat stage 1 | Steady |
| 09:24 | 250.0 | 118.4 | Heat off, input fault shown | Refrigeration compressor on the adjacent circuit started at 09:24 |
| 09:25 | 118.5 | 118.4 | Heat stage 1 | Recovered with nothing touched |
| 09:41 | 118.2 | 118.0 | Heat stage 1 | Steady |
| 10:07 | 250.0 | 118.1 | Heat off, input fault shown | Same compressor started at 10:07 |
| 10:09 | 118.3 | 118.1 | Heat stage 1 | Recovered with nothing touched |
Read the columns against each other and the diagnosis assembles itself.
The handheld column is flat across the whole log: the element sat between 118.0 and 118.4 the entire time, a range of 0.4 degrees. The process never went anywhere. That single fact eliminates every explanation that starts with a real condition, including the one the customer already believed, which was that the heat was overshooting and shutting itself down.
The displayed column jumped straight to the top of scale, 250.0, with no intermediate values. A drifting element moves through the numbers. An open circuit does not, and 250.0 is the declared end of scale on this input rather than a plausible process value. That is the pinned-at-end-of-scale signature, which points at the element or its wiring, not at scaling and not at coupling.
The event column ties both excursions to a motor start on an adjacent circuit, and the recovery happens without intervention. A failed element does not repair itself twice. A connection that is mechanically marginal opens when something shakes the panel or when a start-current surge pulls on a shared raceway, and closes again when the disturbance passes.
Both excursions ended within roughly a minute: 09:24 to 09:25, and 10:07 to 10:09. Note the second one honestly. It ran two minutes, not one. The pattern is "brief and self-recovering", not "exactly one minute", and writing it as the latter would have created a false precision that the next tech would try to match.
The finding is a marginal connection on the sensor circuit, most likely at a terminal or crimp that shares mechanical support with something that moves on that start event. The confirmation is a wiggle test with the log running: disturb one termination at a time and watch the displayed value while the handheld stays flat. The termination that reproduces the excursion is the fault.
Doing that means having the input circuit powered while you are inside an enclosure. De-energize and lock out the line-voltage side and prove it dead with live-dead-live against a known source, NFPA 70E-2021, 120.5, so that only the low-voltage control circuit remains live. If the fault only reproduces with the equipment running, that is the narrow condition 29 CFR 1910.333(a)(1) permits energized troubleshooting under, and it requires a meter and leads rated for the circuit's category and available fault current plus the shock and arc-flash protection the assessment calls for.
What each signature needs before you order a part
- Pinned at end of scale: confirm whether the pin came from the element, the wiring, or a protective device in series. Three different repairs.
- Frozen plausible value: confirm the point is actually updating by forcing a real change and watching. A stale point is a communication repair, not a sensor repair.
- Jitter: measure the raw signal at the element and again at the controller. Jitter present at both is the element or its conditioning. Jitter present only at the controller is the run.
- Late: this is almost never the element. Pull it and look at the tip, the well and the immersion depth first.
- Stable wrong value: two-point check against a known condition, which separates offset from slope. A sibling card covers the procedure.
- Wrong direction: nothing is broken. Something is configured or landed backwards, and it has probably been that way since commissioning.
- Intermittent: do not order anything until the event column has at least three coincidences. Two is a coincidence; three is a pattern.
- Swapped: force one input to change by a known amount and see which point on the screen moves.
When the log is complete enough to act on
A log earns a decision when it covers at least three occurrences of the symptom, at least one full normal cycle between them, and at least one deliberate change you made yourself. Fewer than three occurrences and you cannot separate coincidence from cause. No normal cycle in between and you have no baseline to compare against. No deliberate change and you have only observed, never tested, which means the best you can say is what correlates rather than what causes.
If the symptom will not appear inside a visit, leave the log running as a controller trend rather than as a clipboard and come back to it. A trend sampled every minute across two days answers a question no amount of standing in front of the panel will.
References
- 29 CFR 1910.333(a)(1) - live parts to be de-energized before work, and the narrow conditions permitting energized troubleshooting
- NFPA 70E-2021, 120.5 - process for establishing and verifying an electrically safe work condition
- Manufacturer documentation for the controller's declared end-of-scale and input-fault values, and for its input circuit topology
- See related: What a Sensor Actually Reports; How to Verify a Sensor Against the Real Quantity; How Sensors Fail