How should a plant verify the quality of condition monitoring sensor data?
An anomaly algorithm cannot repair a wrong tag mapping. Record the asset ID, bearing or measurement position, sensor type, engineering units, range and intended failure modes. Trace the point from the field through input cards, PLC or DCS configuration and historian. Distinguish motor and driven equipment measurements. A plausible tag name is not evidence of a correct mapping.
Inspect the complete measurement chain
- Physical installation: vibration sensor mounting, orientation, surface and cable strain.
- Process interface: probe location, thermowell, sample port or impulse line condition where applicable.
- Electrical: wiring, shielding, grounding, supply voltage and signal saturation.
- Configuration: range, engineering units, sample rate, filters, timestamps and alarm logic.
- Measurement assurance: required calibration, functional checks and suitability for the intended decision.
Use approved procedures and safe isolation. Do not disturb operating process connections or electrical equipment solely to investigate a questionable trend without authorization.
Compare equivalent operating conditions
Capture speed, load, flow, process temperature, product, ambient conditions and startup state. Compare steady state with steady state. Use separate baselines when duty materially changes the signal. An oil result, bearing temperature and vibration trend respond to different physical mechanisms, so agreement among channels is helpful but not mandatory. Verify a surprising signal without automatically discarding it.
Flag bad data instead of interpreting it as healthy
Useful quality flags include stale timestamps, missing samples, flatlined values, sudden scaling changes, out of range readings, sensor replacement and communication loss. Define who owns each alert type and what independent monitoring is needed while a channel is unreliable. Keep an audit trail when a signal is suppressed or a threshold is changed. A missing measurement should never silently become a zero risk conclusion.
Example: vibration increases after instrument replacement
A pump vibration value steps upward immediately after a sensor change, but load and process conditions remain stable. The team checks transducer type, mounting, channel scaling and a suitable independent reading. If the measurement chain caused the step, it corrects the configuration and documents the event. If the change is real, the equipment risk assessment continues. The timing of replacement alone does not prove either explanation.
Track the trustworthiness of monitoring
Measure unresolved sensor defects, time to restore critical channels, percentage of verified asset mappings and alarms traced to measurement problems. Review sensor incidents alongside confirmed equipment defects. NIST explains that traceability of a measurement result does not by itself establish fitness for use; measurement uncertainty and the intended application matter. DOE guidance offers a broader framework for managing maintenance and condition monitoring.
Explore KANE consulting
Discuss practical maintenance and reliability improvements for your facility.
Request a consultation →Related answers: condition monitoring conflicting signals · condition monitoring baseline
Technical references
NIST metrological traceability · DOE Operations and Maintenance Best Practices