Sensor Selection & Placement Guide for Practical Predictive Maintenance

Make predictable signals, not noisy data

Predictive maintenance works when the signals you collect meaningfully reflect the failure modes you care about and when those signals are reliable enough to act on. This guide helps you choose which physical or logical measurements matter for a particular PdM decision, where and how to place sensors, how to set sampling and buffering to balance fidelity with cost and latency, how to name and tag signals so they map into MES/SCADA and analytics, and how to validate and monitor signal health during pilots and rollouts.

Match your decision to the signal

Start from the maintenance question you need to answer (for example: detect bearing degradation, identify shaft misalignment, spot lubrication problems, or detect motor electrical stress). For each decision, identify the failure signatures that are observable and useful to the team. Typical signatures include vibration patterns, temperature trends, motor current changes, pressure/flow deviations, acoustic emissions, and oil condition metrics.

Signal selection patterns (practical examples)

• Bearings & rotating elements: vibration (accelerometers), envelope analysis, and temperature. Vibration often gives early detection; temperature shows longer-term drift.
• Electric motors & drives: motor current signature analysis, supply voltage, and bearing vibration. Current can reveal electrical faults and load changes without touching the machine.
• Pumps & valves: vibration, pressure, flow, and temperature. Pressure and flow reveal hydraulic faults; vibration indicates mechanical issues.
• Gearboxes: vibration and oil analysis (particle counts, viscosity). Mechanical wear shows in vibration harmonics; oil shows wear products.
• Slow-changing conditions: temperature, humidity, and strain. These are long-cycle signals—lower sampling rates are usually adequate.

Sampling and fidelity trade-offs

Choose sampling to capture the phenomenon of interest while avoiding unnecessary data volume. High-frequency phenomena (such as early-stage bearing defects or ultrasonic events) need higher sampling rates and possibly edge preprocessing; slow trends (temperature rise over hours/days) can use much lower rates. Consider:

• Signal bandwidth: ensure sampling is adequate for the highest frequency component you need to observe.
• Cost vs. value: more channels and higher rates increase hardware, network, storage, and processing cost. Aim for the minimum viable fidelity that supports reliable detection.
• Edge processing: use local feature extraction (RMS, kurtosis, spectrum peaks) when bandwidth or latency is constrained.
• Buffering & latency: ensure transient events are buffered locally if connectivity is intermittent and ensure timestamps are accurate and synchronized.

Sensor placement & mounting best practices

Placement is often more important than the exact sensor model. Follow these patterns:

• Close to the source: mount sensors where mechanical or thermal energy coupling is strongest (bearing housings, motor frames, pump casings).
• Rigid mounting: use stud or screw mounting where possible for vibration sensors. Avoid soft adhesives unless they are proven for the application.
• Orientation and sensitivity: record axis orientation in metadata—radial and axial readings can reveal different faults.
• Avoid interference: keep sensors away from electrical noise sources, high-temperature surfaces beyond sensor specification, and places likely to be bumped during maintenance.
• Environmental protection: use rated housings for dust, moisture, and chemical exposure and plan for cable routing or wireless antenna placement accordingly.

Naming, tagging, and metadata

Design a predictable tag naming convention so signals map cleanly into MES/SCADA, analytics, and reports. A simple, useful pattern contains location, asset, sensor type, axis (if applicable), and unit. Example pattern and instance:

• Pattern: Plant.Line.Asset.Sensor.Axis.Unit
• Example: PlantA-Cell3-MotorM1-Accel-Rad-g

Include rich metadata for each tag: physical mounting point, orientation, calibration date, nominal ranges, data owner, and expected sampling rate. Document any unit conversions and time zone or timestamp sources.

Signal validation & health monitoring

Detecting bad data early prevents false alerts and wasted modeling effort. Implement automated checks that look for:

• Missing or intermittent timestamps
• Flatline or stuck values
• Values out of expected ranges or sudden unit changes
• Excessive noise, transient bursts, or unusually high variance
• Drift relative to calibration baselines

Flag failing sensors for on-site inspection and compare signals against redundant channels when available (for example, a temperature sensor vs. thermal camera readings during validation).

Pilot checklist (practical rollout steps)

1. Define the maintenance decision and acceptable lead time for detection.
2. Map likely failure modes to candidate signals and select 1–3 high-value channels per asset for the pilot.
3. Specify sampling, buffering, and edge-processing strategy for each channel.
4. Design tag naming, metadata fields, and where the data will land (historian, cloud, local database).
5. Install sensors on a small representative set of assets using rigid mounts and record mounting notes and photos.
6. Collect a labeled baseline period (including known healthy operation and any known faults, if available).
7. Run simple feature extraction and sanity checks; validate timestamps, units, and signal health metrics.
8. Iterate: adjust placement, sampling, or preprocessing if signal quality is insufficient before scaling up.

Operational handoff and sustainment

Include the maintenance and operations teams from the start. Produce a sensor registry that includes tag names, mounting photos, calibration records, maintenance owner, and replacement procedures. Plan periodic validation (for example, after maintenance work or seasonal changes) and set automated alerts for sensor health anomalies.

When to involve controls, electrical, and safety engineering

Coordinate with electrical and controls teams for power, grounding, network segmentation, and any control cabinet penetrations. Never bypass safety systems or violate electrical codes—this guide supports planning and selection, not site electrical or safety approvals.

Keep it simple and measurable

Start with a small set of high-value signals, validate that they reduce uncertainty for a real PdM decision, and only expand instrumentation after proving value. A successful pilot should leave you with validated sensor placements, reliable tags and metadata, baseline data for analytics, and an operational plan to monitor sensor health.

Quick reference checklist

• Defined decision and failure signatures
• Selected sensors and mounting points with photos
• Sampling & buffering plan documented
• Tag naming and metadata recorded in registry
• Calibration and baseline data collected
• Automated signal health checks configured
• Stakeholder handoff to operations and maintenance