The sensor said the temperature was 340°F. The actual temperature was 412°F. The process ran for six more hours before the product was condemned.
The sensor wasn't dead. It was reading. It was just wrong — and nobody in the system knew the difference between a sensor sending bad data and a process actually running at 340°F.
That is the specific problem with process measurement errors. The failure mode isn't silence. It's convincing misinformation.
Why Bad Sensor Data Is Worse Than No Data
No data is visible. An engineer notices a dead transmitter. A screen shows "no signal." An alarm fires. Someone goes to investigate.
Bad data is invisible. The screen shows a number. The alarm doesn't fire. The control loop accepts the reading, makes adjustments based on it, and logs it into the historian as if it were true. Nobody goes to investigate because nothing looks wrong. Everything looks normal. It just isn't.
This is the part that most PM programs — and most discussions of sensor maintenance — don't adequately address. Calibration drift, measurement bias, installation error, and environmental interference don't announce themselves. They corrupt the data stream quietly, and the control system treats corrupted data exactly the same way it treats accurate data.
For a deeper look at how this cluster of issues affects your entire instrumentation program, the full instrumentation PM framework is the right starting point.
The Five Ways Process Measurements Go Wrong
Process measurement errors don't come from one cause. They come from several, and they compound.
1. Zero and Span Drift
A sensor that was accurate at commissioning slowly shifts its output. The zero point moves. The span compresses or expands. The reading is plausible — close enough to expected that no alarm fires — but it's been wrong for months. This is calibration drift, and it affects every sensor type: pressure transmitters, temperature sensors, flow meters, analytical instruments.
2. Systematic Bias
An installation problem, a wiring error, or a configuration mistake that was baked in from day one. The sensor isn't drifting — it never started in the right place. Systematic bias produces readings that are consistently off by a fixed amount or percentage, in one direction, forever. These are some of the hardest errors to catch because the trend looks stable. Trending stable garbage looks exactly like trending stable truth.
3. Process-Induced Error
The sensor is working correctly. The measurement point is not representative of the process. A temperature sensor mounted downstream of a mixing point reads something different than what's happening at the actual control point. A pressure tap that's partially obstructed reads low. A flow meter installed with insufficient upstream straight run reads a turbulent flow profile and reports flow that doesn't match reality.
The instrument is healthy. The measurement is still wrong. And sensor calibration won't fix it, because calibration only tells you if the sensor is reading accurately at the test point — not whether the test point represents the process.
4. Environmental Interference
Electromagnetic interference from nearby VFDs, high-voltage cables, or relay switching corrupts signal transmission. Temperature extremes outside the sensor's rated range shift its output. Vibration loosens connections. Moisture ingress causes erratic readings. None of these are sensor failures in the conventional sense. The sensor body is intact. The wiring looks fine. The reading on the screen is just noise that passes through all the normal sanity checks.
5. Signal Chain Errors
The sensor is accurate. The transmitter is calibrated. The reading that arrives at the DCS is still wrong — because something in between introduced an error. Incorrect scaling in the transmitter configuration. A range mismatch between the transmitter and the controller input. An analog-to-digital conversion error in an old I/O card. Process measurement errors don't have to originate at the sensing element. They can be manufactured anywhere in the signal chain.
What Happens to the Process When the Data Is Wrong
Control loops don't question their inputs.
A PID controller running a temperature loop does not know the difference between a correct reading of 340°F and a drifted sensor reporting 340°F when the actual temperature is 412°F. It accepts the reading, calculates the error from setpoint, and adjusts the output accordingly. If the sensor says the temperature is low, the controller adds heat. If it says the temperature is high, it backs off. The controller is doing exactly what it was designed to do. It's just doing it based on a lie.
Temperature control loops running on drifted sensors can drive a process into conditions that are dangerous, produce out-of-spec product, or damage equipment — all while the HMI screen displays a process that looks like it's running fine. The trend graph is smooth. The alarms are quiet. The historian will record every bad reading as truth.
For pressure measurements and differential pressure transmitters, the consequence is often flow miscalculation. Differential pressure-based flow measurement is particularly vulnerable — a blocked impulse line, a partially plugged sensing port, or a reference leg filled with process fluid introduces a systematic error that accumulates silently. The flow totalizer keeps running. The mass balance never closes. Nobody goes looking because the flow reading looks plausible.
Flow meters built on other technologies carry their own vulnerabilities. Coriolis meters can report incorrect density if air entrapment affects tube vibration. Magnetic flow meters require full pipe — a partially filled pipe reports flow lower than reality, and the meter gives no indication that the reading condition is abnormal. Radar and micropulse level transmitters can false-trigger on foam, vapor, or agitation and report a level that has no relationship to actual inventory.
Analytical instruments present a different version of the problem. A pH electrode that's coating, desiccating, or past its service life doesn't fail in a way that produces an obvious error. It drifts toward a reading that looks like a slightly off-spec process. The decision to adjust chemistry gets made. The wrong amount of reagent gets added. The actual problem — a dying electrode — goes unaddressed.
Ammeters used for motor load monitoring produce a quieter version of the same failure. A current measurement that reads low leads a maintenance manager to believe a pump is running unloaded when it's actually cavitating or has worn impeller clearances. The decision to not intervene gets made based on data that said everything was fine.
The Difference Between Sensor Failure and Measurement Error
These are not the same thing, and treating them as if they are causes PM programs to protect against the wrong problem.
Sensor failure means the sensor has stopped functioning — it's outputting a constant value, a rail voltage, a broken-wire condition that trips a diagnostic alarm, or nothing at all. These are visible. Most systems catch them. Most PM programs include a check for them.
Measurement error means the sensor is functioning but the value it's producing does not accurately represent the process variable it's supposed to measure. The sensor passes every functional check. The loop diagnostics show no fault. The output is within range. The calibration record is current. And the reading is still wrong.
Industrial sensor failure modes (Industrial Sensor Failure Modes: Why Sensors Stop Telling the Truth Long Before They Stop Sending a Signal) covers the mechanics of how sensors transition from accurate to inaccurate without triggering any failure indication — and why the conventional approach of checking for signal presence is not the same as checking for signal accuracy.
The gap between these two categories is where most process measurement problems live. The sensor isn't broken enough to flag. It's just wrong enough to matter.
How Measurement Error Propagates Through Operations
A single wrong reading doesn't stay in its lane.
A flow meter that reads 8% high drives overfilling, over-dosing, or incorrect mass balance calculations. A pressure transmitter with a blocked sensing port causes the control system to think the system is underperforming and drives the pump to work harder — increasing wear on equipment that doesn't need to be pushed. A temperature reading that's 15°F low causes a thermal process to run long, over-treat, or under-treat.
But the more insidious consequence is the decision trail that follows.
Operations makes adjustments based on what the data says. Maintenance gets work orders generated by what the historian shows. Quality correlates failures to process variables that were never accurately measured in the first place. Engineering bases capacity calculations on flow totals that were accumulated from a meter running with a systematic error.
Every decision that uses bad data is a bad decision. Even good decisions made from bad data are actually bad decisions — they just don't know it yet.
Sensor drift (Sensor Drift: Why It Happens, What It Costs You, and How to Catch It Before It Lies to Your Control System) addresses the specific mechanism by which accurate sensors become inaccurate sensors over time — and why the pace of drift is rarely visible until the error has grown large enough to affect process outcomes.
What a PM Program Needs to Catch Measurement Errors
Maintenance rounds that verify a sensor has a signal don't catch measurement errors. Neither does waiting for the calibration cycle to come around.
The PM checks that actually catch these problems are specific.
Redundancy checks and cross-verification. Where a process variable is measured by more than one sensor, or where an independent check is available, comparing readings regularly catches drift before it grows large. A temperature sensor whose reading doesn't track with a nearby thermowell can be investigated. A flow meter that disagrees with a tank level change can be questioned.
Process sanity checks. Does the flow reading make sense given the pump curve and the inlet and discharge pressures? Does the temperature reading make sense given the heat input and the expected outlet? Does the mass balance close? Sensors that are wrong usually disagree with something else in the process — the problem is that nobody is looking.
Inspection of impulse lines and process connections. Blocked sensing ports, partially filled reference legs, and plugged impulse lines don't announce themselves through a transmitter output. They require physical inspection. Most PM programs don't include this with enough frequency or specificity.
Calibration verification against known references — not just previous readings. Calibrating a sensor against a standard and recording a found-as reading before adjustment is what produces the data needed to catch systematic drift. A calibration that only checks whether the sensor is in tolerance and documents the adjusted values tells you the sensor passed — not whether it was drifting, how fast, or in which direction.
Environmental condition checks. The conduit is sealed. The junction box doesn't have condensation. The cable tray routing doesn't run the signal wire next to the VFD output cables. These are inspection tasks, not calibration tasks, and they catch error sources that no calibration frequency will ever address.
The Practical Takeaway
A sensor that's reading and a sensor that's accurate are two different things.
Your PM program almost certainly has checks for the former. Whether it has the right checks for the latter depends on what it's actually looking for during each visit — and whether anyone is cross-verifying the data that comes out the other end.
Bad sensor data doesn't fix itself. It stays in the historian. It shapes decisions. It gets used in reports. By the time someone figures out the reading was wrong, the trail of decisions made from it is usually long enough to cause real problems.
The task lists below are the starting point for building PM checks that go after measurement accuracy — not just sensor presence.
- Temperature Control Loop / PID Controller PM Checklist
- Pressure Sensor / Transmitter PM Checklist
- Differential Pressure Transmitter PM Checklist
- Coriolis Flow Meter PM Checklist
- Magnetic Flow Meter PM Checklist
- Radar / Micropulse Level Transmitter PM Checklist
- pH / Conductivity / Analytical Instrument PM Checklist
- Ammeter PM Checklist
The sensor isn't broken. It's just wrong. That distinction is your PM program's problem to solve.