Condition-Based Maintenance: Sensor Selection & Placement Quickguide

Why sensor choice and placement matter

Condition-based maintenance succeeds when sensors reliably capture the physical changes that precede failure. Choosing the right sensor type and putting it in the right place maximizes signal-to-noise ratio (SNR), lowers data costs, shortens time-to-insight, and avoids wasted installation effort and budget.

Quick approach

1. Identify the specific failure modes you care about (bearing wear, lubrication loss, misalignment, overheating, electrical imbalance, clogging, corrosion, etc.).
2. Map failure modes to measurable signals (vibration, temperature, acoustic emission, current, voltage, humidity, pressure, etc.).
3. Choose sensors that are sensitive to those signals and feasible to install where the signal originates.
4. Design a sampling and connectivity plan that balances detection needs with data volume and cost.
5. Pilot on a small, representative set of assets, measure baseline SNR, and iterate placement before scaling up.

Sensor types: what they detect and trade-offs

Sampling & data guidance (practical rules of thumb)

• Match sampling frequency to the physics you want to capture. Vibration for bearing defects often benefits from kHz-range sampling to see high-frequency energy; temperature and slow process variables can be sampled from once-per-minute to once-per-hour for trend detection. Acoustic sensors often require kHz sampling as well.
• For trending vs. diagnostics: trending (daily averages, RMS) needs far less data than diagnostic waveform capture needed for FFT or envelope analysis. Combine both approaches: low-frequency trending with occasional high-resolution captures when anomalies appear.
• Consider edge processing to compute summary metrics (RMS, kurtosis, crest factor, spectra peaks) and only send those metrics to reduce bandwidth and storage costs, while keeping occasional raw captures for model building or forensic analysis.
• Plan for synchronization needs when correlating signals (e.g., vibration and electrical). Timestamping and consistent clocks matter.

Mounting and installation best practices

• Place sensors as close as practical to the fault source. For bearings, mount accelerometers on the bearing housing, not on nearby panels.
• Use solid mounting: stud-mounted accelerometers yield better coupling and SNR than adhesive mounts; magnetic bases can be convenient but check for movement and resonance.
• Respect orientation: some sensors are directional — note axis orientation and record it in metadata.
• Avoid indirect mounting through air gaps or painted surfaces; remove paint or provide a machined mounting pad where possible for thermal/accelerometer sensors.
• Protect cables and connectors from mechanical damage, coolant, and high temperatures. Route cables to avoid crossing moving parts and EMI sources.
• For temperature sensors, ensure good thermal contact (thermal paste, proper insertion depth for thermowells); for IR sensors, ensure stable emissivity and clean line-of-sight.
• Log physical context in metadata (asset, location, mounting method, orientation, date installed) so future analysis can account for installation differences.

Sensor-selection rubric (practical scoring tool)

Score each candidate sensor 1 (poor) to 5 (excellent) on these dimensions, then weight them according to your priorities (default weights suggested in parentheses):

1. Signal relevance to failure mode (weight 0.30)
2. Expected signal-to-noise ratio at candidate location (0.25)
3. Installation cost & complexity (0.15)
4. Data volume & processing needs (0.10)
5. Reliability / maintenance needs (0.10)
6. Integration & safety constraints (0.10)

Example: Vibration sensor for bearing defect might score 5/5 on relevance, 4/5 on SNR, 3/5 on installation cost, 2/5 on data volume, 4/5 on reliability, 5/5 on integration = weighted composite. Use the composite to compare alternatives (e.g., vibration vs temperature) and to justify pilot installations.

Pilot design checklist

• Pick a small, representative asset set (3–10 similar machines) and a control group if possible.
• Define success criteria up front (detection lead time, false positive rate, required SNR, reduction in unplanned downtime, etc.).
• Record baseline failure history and inspect assets to confirm expected failure modes.
• Install sensors with careful notes on placement and mounting. Capture at least 30–90 days of baseline trending before relying on models where possible.
• Collect both summary metrics and periodic raw captures for validation of algorithms.
• Define governance: who reviews alerts, how to triage, and how to close the loop with maintenance actions and labeled outcomes.

Common pitfalls and how to avoid them

• Over-instrumentation: instrument for clear, testable use cases rather than adding every sensor available.
• Poor placement: avoid mounting that degrades SNR; pilot and measure instead of assuming placement is correct.
• Insufficient metadata: without installation metadata, sensor data is hard to interpret and maintain.
• Misaligned expectations: define what predictive success looks like and what actionable lead time is required.
• Weak data governance: ensure labeling, storage, access controls, and retention policies are in place before scaling.

Next practical steps

1. Run the rubric on 2–3 candidate sensors for a representative machine and pick the best-scoring option.
2. Pilot with both trend metrics and occasional raw captures; evaluate SNR and detection capability after 30–90 days.
3. If pilot shows poor SNR, iterate placement (closer to source, different mounting) before changing sensor type.
4. Document lessons and standardize successful placements as 'standard work' for future rollouts.

References & further reading

For specialized diagnostics (envelope analysis, wavelet transforms, motor current signature analysis), consult domain experts or established vibration and electrical signature textbooks and vendor application notes. Use this guide as a practical starting point and adapt it to your assets, environment, and risk tolerance.