Guide

Sensor & Signal Selection Guide for Predictive Maintenance

Practical, field-ready guidance for picking sensor types, mounting approaches, sample rates, and capture practices that actually reveal failure precursors — plus a technician checklist and example failure-mode → signal mappings for realistic predictive-maintenance pilots.

Why signal selection matters

Predictive maintenance succeeds when the sensors and capture practices reveal the physical signatures that precede failure. The wrong sensor, poor mounting, low sample rates, or missing baseline captures can make a promising pilot produce noisy data, wasted effort, and false vendor promises. This guide helps you choose sensors, sampling settings, and mounting so your pilot measures meaningful signals without wasting budget.

Quick decision principles

Match the sensor to the failure physics. Measure what changes when the asset begins to fail.
Capture the right frequency band. Use sampling and filters that include the frequencies where failures appear.
Control mounting and wiring. A loose sensor or poor cable routing ruins data more often than the wrong sensor model.
Collect baselines under stable load conditions and document metadata (asset state, load, speed, time, operator).

Common sensor types and when to use them

Accelerometers (vibration)

Best for rotating equipment faults: imbalance, misalignment, looseness, bearing defects, gear mesh problems. Choose a piezoelectric ICP accelerometer for wide frequency response and robust field use. Key specs: sensitivity (mV/g), frequency range, mounting method, and temperature rating.

Velocity sensors and velocity processing

Useful for overall machine health and where lower-frequency vibration (below a few hundred Hz) dominates. Many systems compute RMS velocity from acceleration data — accelerometers are more versatile.

Temperature sensors (RTD, thermocouple, thermistor)

Good for bearing and winding overheat detection, lubrication issues, and electrical overheating. Use RTDs for stability and accuracy where small delta changes matter; thermocouples for high temperatures.

Current and voltage sensors (CT, hall-effect, clamp-on)

Ideal for electrical-to-mechanical fault detection: motor stator/rotor faults, stalled conditions, rotor bar defects, and load changes. Current signature analysis (CSA) can reveal mechanical faults indirectly through electrical behavior.

Proximity probes (eddy-current)

High-sensitivity measurement of shaft displacement, rubs, and orbits. Use when you need subsynchronous/synchronous shaft motion data or when absolute displacement matters.

Acoustic sensors / ultrasound

Effective for early lubrication issues, air or gas leaks, electrical arcing, and bearing distress at high frequencies not captured well by low-cost accelerometers.

Speed / tachometer

Always useful when diagnosing rotating machinery. Speed reference lets you map vibration components to running speed harmonics (1x, 2x, etc.).

Sampling rate and frequency-band guidance

Choose sample rates to capture the highest frequency you care about. The Nyquist theorem requires sampling at at least twice the highest frequency, but practical condition monitoring commonly uses higher multiples to preserve waveform fidelity and support analysis (envelope detection, spectral kurtosis).

Rule of thumb: sample at 5–10× the highest frequency of interest for reliable spectral and time-domain analysis.
Example: if bearing fault energy appears up to 5 kHz, set a sample rate of 25–50 kHz.
For slow-speed equipment where faults show below 1 kHz, sample rates of 5–10 kHz are often adequate.
Use anti-aliasing filters ahead of ADCs and ensure ADC resolution is sufficient (16-bit or better for many applications).

Mounting rules that matter

Prefer stud mounting for accelerometers when possible — it gives the cleanest coupling into the structure.
Adhesive mounting is acceptable for short-term checks; beware temperature and surface prep requirements.
Place sensors close to the root cause (bearings, gearbox housing, motor casing) rather than a convenient bracket.
Align directional sensors with the axis of interest and label orientation clearly.
Keep cable runs short, secure, and shielded; avoid routing alongside power cables to reduce electrical noise.
Torque mounting hardware to manufacturer recommendations and use locking features to prevent loosening.

Baseline capture sessions — do these every time

Baselines give you the 'healthy' signature for comparison. Record under representative, steady-state operating conditions and capture these items:

Asset ID, location, date/time, operator, process state (load, speed, temperature).
Sampling rate, filter settings, sensor model and serial, mounting method and torque.
Duration: capture long enough to include several revolutions at operating speed. Typical durations range from 30 seconds up to several minutes depending on speed and analysis needs.
Capture both raw waveforms and summary statistics (RMS, peak, kurtosis) and store them with metadata.

Avoid aliasing and poor analysis

Aliasing occurs when high-frequency energy folds into lower frequencies due to insufficient sampling or missing anti-alias filters. Prevent it by choosing sample rates that cover the highest significant frequencies and by engaging analog anti-aliasing filters just before digitization. Validate capture chains with a known signal where possible.

Signal → failure-mode examples

Use these common mappings when selecting sensors and analysis techniques:

Bearing defects: high-frequency vibration energy; envelope analysis on accelerometer data often reveals fault frequencies.
Imbalance: strong 1× running speed vibration, visible in low-frequency spectra and time waveform amplitude.
Misalignment: 1× and 2× components, axial vibration, and harmonics; shaft coupling checks and phase analysis help confirm.
Looseness: broadband increase in vibration, structural resonances, and sidebands; often visible across frequencies.
Gear mesh: gear mesh frequency and sidebands, require accelerometers with sufficient upper-frequency response.
Electrical faults (motors): current signature changes, negative sequence components, and unusual temperature rises in windings.
Lubrication issues: rising acoustic/ultrasound levels and increasing high-frequency vibration before temperature rises.

Practical data-quality checklist for pilots

Confirm the failure mode you want to detect and choose the sensor type that directly observes its signature.
Record metadata for each sensor: model, serial, mounting method, orientation, calibration date.
Select sampling rate and anti-aliasing filter to cover expected frequencies (see sampling guidance above).
Mount securely (stud where possible), torque per spec, and verify orientation and cable strain relief.
Capture baseline under stable operating conditions and save raw waveforms plus summary metrics.
Check for noise: switch off nearby non-essential equipment and compare with a known-good sensor or duplicate channel.
Label cables and sensors on drawings and in the CMMS/asset registry so future technicians repeat the same setup.
Plan periodic verification captures and re-calibration intervals for sensors in the field.

Acceptance criteria for a pilot signal

Decide how you will know the signal is useful before scaling:

A consistent difference between baseline and degraded-state captures (statistical or spectral).
Ability to reproduce signal on repeat captures and across similar assets.
Actionable thresholds or trending behavior that correlate with maintenance records or inspections.
Reasonable data volume and bandwidth that fit the planned ingestion/storage setup.

Quick field technician checklist (print-friendly)

Use this at the machine before you walk away:

Confirm asset ID and target failure mode.
Sensor type, model, serial documented on the tag.
Mounting: stud/adhesive used and torque recorded.
Orientation arrows labeled; directional axes noted.
Sample rate and filter settings recorded in capture device.
Baseline capture complete under steady-state load; raw data saved.
Cables secured and shielded; connector strain relief in place.
Photos of mounting location and cable routing stored with metadata.
Notes on operating conditions (load, speed, recent maintenance) added.

Next steps for teams

For a practical pilot, map 5–10 critical assets, document the specific failure modes you want to detect, and run baseline captures on each. Validate signals against known faults (historical failures or controlled tests) before investing in wide sensorization. Track acceptance criteria and iterate sensor placement, sampling, and analysis until you get reliable, actionable alerts.