A sensor that's sending a signal isn't necessarily sending the right signal. That distinction is the one that breaks processes, corrupts batches, and puts control systems into slow, invisible failure modes that nobody catches until something expensive goes wrong.
Most PM programs treat sensors like light switches. Working or not working. On or off. But that binary thinking misses how sensors actually degrade — and it misses the majority of the failure modes that matter.
The dangerous sensor isn't the dead one. It's the one still transmitting. Still green on the dashboard. Still logging data your control system is trusting completely.
Start here if you haven't already: The full picture of what sensor PM is supposed to catch
The Signal Isn't the Measurement
There's a difference between a sensor producing a signal and a sensor producing an accurate measurement. Most people treat them the same. They're not.
A 4–20 mA loop doesn't care if the reading is right. It just carries whatever the sensor says. A thermocouple with a degraded reference junction still produces a voltage. A pressure transmitter with a plugged impulse line still reports a pressure — it just reports whatever pressure was there when the line plugged. A flow meter with buildup on the sensing element still produces a number.
The signal lives. The accuracy dies.
This is the core problem with sensor failure modes in industrial environments: most of them produce a plausible-looking output right up until they don't. The reading drifts. The span narrows. The zero shifts. And because the control system keeps responding, and the trend on the historian looks roughly right, nobody goes looking.
Drift: The Failure Mode That Hides in Plain Sight
Drift is the most common sensor failure mode and the hardest to catch with a standard PM checklist.
A sensor drifts when its output moves away from the true value of the process variable — slowly, continuously, in a direction that often looks like a process change rather than a sensor problem. Temperature sensors drift upward as thermocouple wire oxidizes. Pressure transmitters drift as their sensing diaphragms fatigue. pH electrodes drift as the glass membrane ages and the reference junction clogs. Electrochemical gas sensors drift downward as their sensing cells deplete.
The rate of drift varies by sensor type and environment. A temperature sensor in a clean, stable environment might drift imperceptibly over years. The same sensor type in a high-vibration, high-humidity, thermal-cycling environment might be off by several degrees within months.
What makes drift genuinely dangerous is that it doesn't trigger alarms. The process looks like it's running. Setpoints are being hit — because the control loop is compensating for a reading that's shifted, not because the process is actually at the right condition. By the time drift is significant enough to cause an obvious problem, the real process has been running off-spec for a long time.
Catching drift requires trending, not inspection. A point-in-time check tells you nothing. What you need is a record of readings over time, compared against known references or cross-referenced against redundant sensors or independent verification. That's not what most PM programs build. That's the gap that sensor drift needs its own PM structure to address.
Span Error and Zero Shift: Calibration Failure Without a Fault Code
Span error and zero shift are related failure modes. They're also both perfectly invisible at the sensor's normal operating point.
A zero shift means the sensor reads incorrectly at the low end of its range — but if you're almost never operating at the low end, you'll never notice. A span error means the sensor reads correctly at one reference point but loses accuracy as the measured variable moves away from that point. Run the process at 70% of range and the error looks small. Run it at 95% and the error is significant.
Both failure modes are calibration-related but not always calibration-caused. A sensor can be perfectly calibrated and then develop zero shift or span error as its sensing element ages, as its electronics drift with temperature cycles, or as its installation environment changes. Calibration restores it temporarily. The underlying degradation continues.
This is why calibration intervals matter — but also why calibration alone isn't enough. What most PM programs get wrong about calibration is that they treat calibration as the fix rather than the check. A sensor that needs frequent re-calibration is a sensor that's degrading. The recalibration interval should be getting shorter, not staying the same.
Environmental Contamination: The Failure Mode That Announces Itself Slowly
Industrial sensors spend their working lives in environments that want to destroy them. Heat. Moisture. Vibration. Chemical exposure. Particulate buildup. Electromagnetic interference. These aren't unusual conditions. They're the operating environment.
Environmental contamination shows up differently depending on sensor type.
Optical sensors — photoelectric eyes, laser sensors, vision systems — fail when their lenses accumulate dust, oil mist, or condensation. The signal degrades as transmission drops. Often the sensor doesn't go full-fail. It develops a timing error, a false-trigger rate that creeps up, a detection margin that erodes. The sensor is still working. It's just less reliable every week, and nobody's watching that margin.
Mechanical sensors — pressure transmitters with impulse lines, flow meters with process connections, level switches with float assemblies — fail when their process interface plugs, scales, corrodes, or mechanically binds. The reading freezes at whatever the value was when the blockage occurred, or begins responding to a ghost of the actual process condition. Again: signal present, accuracy gone.
Electrochemical sensors — O₂ sensors, H₂S sensors, LEL sensors, CO sensors — have finite sensing cell life that accelerates dramatically with exposure to contaminants. A combustible gas sensor exposed to silicone vapors can be permanently poisoned in hours. The sensor still reads. It just reads low — potentially fatally low.
Contact sensors — thermocouples, RTDs, immersion probes — fail as their protective sheaths corrode, as moisture ingress changes the thermal path, as the measuring junction itself oxidizes or separates. The degradation is gradual until it isn't. A thermocouple that's slowly separating will drift, then become noise-heavy, then jump erratically before it finally fails open. Three of those four stages still produce a reading.
Vibration and Mechanical Stress: The Failure Nobody Attributes to Sensors
Vibration kills sensors in ways that get diagnosed as something else.
A pressure transmitter mounted on a high-vibration pipe without an isolation valve or flexible connection experiences thousands of stress cycles per minute. The sensing diaphragm fatigues. The process connection work-hardens at the weld. The electronics develop cold solder joints that are intermittent enough to create random spikes in the signal that the historian just averages out.
The maintenance record says the transmitter is fine. The historian shows some noise. The process engineer thinks the noise is process-related. Nobody looks at the mounting.
Proximity sensors and photoelectric sensors mounted on vibrating machinery develop intermittent electrical connections at their cable entry points. The signal becomes unreliable in exactly the conditions where reliability matters most — when the machine is running at speed. The sensor passes every bench test. It fails in service.
Encoders on high-speed shafts fail when their bearings deteriorate or their coupling to the shaft loses rigidity. The feedback signal develops error that looks like process variation. The VFD compensates. The mechanical load shifts. Nobody looks at the encoder.
The pattern is consistent: vibration-induced failure produces symptoms that look like process variation, control system response, or intermittent electrical problems. The sensor is the last place anyone looks.
Reference and Ground Issues: The Failure Mode That Lives in the Wiring
Sensors don't operate in isolation. They're part of a signal circuit, and that circuit can corrupt an accurate sensor reading without touching the sensor.
Ground loops — created when a sensor circuit has more than one connection to ground at different potentials — inject noise into the signal that shows up as erratic readings, offset errors, or unusual variability that correlates with other equipment running. A 60 Hz hum component in a 4–20 mA loop produces an average reading that looks reasonable but contains errors that smear across the historian.
Reference errors in thermocouple circuits — specifically, incorrect cold junction compensation — produce a consistent offset that shifts with ambient temperature. In summer the reading is off one way. In winter it's off another way. The offset looks like a seasonal process change. It's a wiring error that's been there since installation.
EMI from variable frequency drives, large contactors, welding equipment, or poorly shielded cables corrupts sensor signals in patterns that track the interference source. The sensor itself is fine. The signal it's transmitting is not.
These failure modes don't show up on a sensor PM checklist. They require loop-level verification — checking signal integrity at the receiving end, not just at the transmitter. How bad sensor data becomes bad decisions starts here, in the gap between what the sensor is sending and what the control system is receiving.
Sensor Age and Lifecycle Failure: The Failure Mode Everyone Ignores Until It's Expensive
Sensors have finite service lives. Not manufacturer warranty periods — actual service lives in actual industrial conditions. Those are often shorter than anyone budgets for.
Electrochemical gas sensors typically last 1–3 years in service, often less in harsh environments. After that, the sensing cell output drops below threshold and the sensor reads low — or stops responding to the gas entirely. Many facilities run these sensors past their service life because they're still sending a signal. Still sending a signal is not the same as still detecting gas.
pH electrodes typically last 6–18 months in process service depending on the application. As the glass membrane ages and the reference junction clogs, span error and zero shift accumulate faster than calibration intervals can correct. The calibration procedure makes the sensor look acceptable at the calibration points. Between those points, the error is larger than anyone is measuring.
Pressure sensing diaphragms — particularly in corrosive, high-temperature, or high-cycle applications — have fatigue life limits. The sensor doesn't announce when it's approaching them. It just begins to drift more, require more frequent recalibration, develop increasing hysteresis between rising and falling process conditions.
Age-related failure accelerates with exposure. Every thermal cycle, every overpressure event, every chemical splash, every vibration hour shortens the remaining service life. A PM program that doesn't track sensor age and exposure history has no way to predict when this failure mode becomes likely — and no early warning when it arrives.
What Your PM Program Should Be Catching
None of these failure modes require expensive tools to identify early. They require structured observation and trending.
A proximity sensor PM that just checks "sensor detects target" is checking for total failure only. Add indicator response time to the check — is the indicator coming on instantly or with a lag? — and you're checking for sensing margin degradation.
A pressure transmitter PM that just records the reading is a checkpoint, not a PM. Add a cross-reference against a calibrated reference gauge, or against the reading from a nearby transmitter on the same process, and you're trending for drift.
A gas sensor PM that just confirms the sensor is powered is almost worthless. Add a bump test with known concentration gas and you're checking actual response. Record the response value over time and you're trending for sensing cell depletion.
The pattern is the same across sensor types. Point-in-time checks confirm existence. Trending confirms function. Most PM programs are built around the first. The sensor failure modes that matter require the second.
Start building the trending function here:
- Proximity Sensor PM Checklist
- Photoelectric Sensor PM Checklist
- Vibration Sensor / Accelerometer PM Checklist
- Load Sensor / Load Cell PM Checklist
- Flow Meter / Flow Sensor PM Checklist
- Level Sensor / Switch PM Checklist
- Hydrogen Sulfide (H₂S) Sensor PM Checklist
- Combustible / LEL Sensor PM Checklist
- Carbon Monoxide (CO) Sensor PM Checklist
The green light on the dashboard means the sensor is alive. It doesn't mean the sensor is right. That's the distinction your PM program needs to be built around.