Examples of predictive maintenance

Eight concrete cases — pump, motor, gearbox, fan, compressor, heat exchanger, steam trap and switchgear — each traced the same way: the failure mode, the technique that sees it, the signal that appears first, and the decision it should trigger.

Key takeaways at a glance
TopicKey point
How to read these examplesEvery case below follows the same four steps, because that is the sequence a working programme actually uses: failure mode → technique → first signal → decision .
Rotating equipmentAsset Failure mode Technique First signal Decision it triggers Centrifugal pump Rolling-element bearing degradation Vibration analysis, ultrasound ear
Static and process equipmentAsset Failure mode Technique First signal Decision it triggers Heat exchanger Fouling Performance monitoring — heat balance from existing instrumentat
A case that should not be predictiveThe honest example most articles leave out.
What makes these examples work in practiceA healthy baseline exists. Every signal above is a change from a reference.
Standards behind these casesThe cases above are method, not thresholds.

How to read these examples

Every case below follows the same four steps, because that is the sequence a working programme actually uses: failure mode → technique → first signal → decision. If you cannot complete all four, the monitoring is not predictive maintenance. It is data collection.

The last step is the one programmes skip. A detected fault that does not change a schedule has cost money and prevented nothing.

time in service →conditionPFP-F intervalfirst detectablefunctional failureinspect more oftenthan this window
The P-F interval is a property of the failure mode, not of the technique. Inspect less often than this window and the fault arrives between checks. Axes are unscaled — the shape is the point, not the values.

Rotating equipment

AssetFailure modeTechniqueFirst signalDecision it triggers
Centrifugal pumpRolling-element bearing degradationVibration analysis, ultrasound earlier stillBearing defect frequencies appear in the spectrum; friction rises in ultrasound before thatOrder the bearing, schedule into the next planned window rather than a breakdown
Induction motorRotor bar crackingMotor current signature analysisSidebands around line frequency in the current spectrumPlan a rewind or replacement; check whether starting duty caused it
GearboxTooth wear and pittingOil and wear-debris analysis, plus vibrationWear-metal concentration trends upward; gear-mesh sidebands growChange the oil, shorten the sampling interval, plan inspection at the next shutdown
FanImbalance from deposit build-up or erosionVibration analysisRising once-per-revolution amplitudeClean or rebalance before bearing and structural loads accumulate
CompressorValve wear, internal leakagePerformance monitoring, ultrasoundSpecific power drifts up at constant dutyOverhaul decision based on lost energy, not on run hours

Note what the pump and the motor have in common: the technique is chosen by the failure mode, not by the asset. The same pump needs a different technique for seal failure than for bearing failure.

Static and process equipment

AssetFailure modeTechniqueFirst signalDecision it triggers
Heat exchangerFoulingPerformance monitoring — heat balance from existing instrumentationFalling heat-transfer coefficient at comparable dutyClean on condition instead of on a calendar; see the fouling guide
Steam trapFailed open, blowing live steamUltrasound, supported by temperatureContinuous flow signature where cycling is expectedReplace immediately — the loss is continuous and invisible
Switchgear and MCCLoose or corroded connectionInfrared thermographyTemperature rise at a joint relative to the same phase elsewhereRe-torque or replace at the next safe isolation; severity sets the urgency

These three are worth separating from rotating equipment because their economics differ. A failed-open steam trap and a hot electrical joint both leak continuously — every hour of delay has a cost, so the detection-to-action time matters more than the detection sensitivity.

A case that should not be predictive

The honest example most articles leave out. A small, redundant, readily-available circulation pump with an installed spare, a low-cost replacement and no safety or quality consequence on failure does not justify monitoring. The correct engineering answer is run-to-failure with a spare on the shelf.

Applying condition monitoring here costs sensor, installation, data and analyst time to avoid a consequence that was already covered by redundancy. Programmes that instrument everything dilute the analyst attention that the genuinely critical assets need.

What makes these examples work in practice

  • A healthy baseline exists. Every signal above is a change from a reference. Without a known-good starting point the trend measures nothing.
  • Operating context travels with the reading. Speed, load and process state. A compressor's specific power means nothing without the duty it was delivering.
  • The interval is shorter than the P-F window. Monthly readings on a fault that develops in three weeks detect failures after they happen.
  • Somebody owns the alarm. A named route from finding to work order, with authority to change the schedule.
  • Two channels agree before a big call. Instrument drift rarely fakes the same fault in two different physical measurements.

Standards behind these cases

The cases above are method, not thresholds. Where a limit or a severity classification is needed, it belongs to a standard — go to the standard rather than to a blog, including this one.

NeedStandard to consult
Setting up a condition-monitoring programmeISO 17359
Vibration measurement and severity evaluationISO 20816
Diagnostics and data interpretationISO 13379
PrognosticsISO 13381
Analyst qualification by techniqueISO 18436
Failure and maintenance data recordingISO 14224
Maintenance terminologyEN 13306

Frequently asked questions

What is a simple example of predictive maintenance?

Trending vibration on a centrifugal pump. As a rolling-element bearing degrades, characteristic defect frequencies appear and grow in the vibration spectrum. The bearing is ordered and replaced in a planned window, before the failure takes the pump — and often the coupling and seal — with it.

What industries use predictive maintenance most?

Those where unplanned downtime is expensive or hazardous and assets run continuously: power generation, oil and gas, chemicals, pulp and paper, cement, metals, water utilities and large-scale food and beverage. The common factor is continuous operation with costly interruption, not the industry label.

Can predictive maintenance work without new sensors?

Often yes. Performance monitoring uses instrumentation that already exists for process control — flows, temperatures, pressures, motor current. Heat-exchanger fouling and compressor efficiency loss are both detectable from existing measurements before anyone installs a condition-monitoring sensor.

Which assets should not get predictive maintenance?

Cheap, redundant, quickly replaceable assets with no safety, environmental or quality consequence on failure. Run-to-failure with a shelf spare is the correct answer there, and it preserves analyst attention for critical equipment.

How do I know the programme is working?

Measure findings that changed a schedule, not alarms raised. A programme producing many alerts and no rescheduled work is not preventing failures. Recording failures and interventions consistently — ISO 14224 gives a structure — is what lets you tell the difference.

Related guides

Software that helps