Vibration, temperature and current signatures can flag a failing fan, pump or UPS module weeks before it trips a protection relay — but only if the sensors were designed into the asset from the start, not bolted on as an afterthought once a platform vendor is chosen. Predictive maintenance for ELV assets is a sensor-placement and data-pipeline problem well before it's a dashboard problem.
What Each Sensor Type Actually Tells You
Accelerometer-based vibration analysis is the workhorse of rotating-equipment predictive maintenance: raw acceleration data run through an FFT (fast Fourier transform) reveals frequency peaks tied to specific fault modes — bearing defect frequencies calculated from the bearing's geometry, shaft misalignment showing up at 2x running speed, rotor imbalance at 1x running speed. Current signature analysis on motor windings picks up electrical faults — broken rotor bars, stator winding degradation — that vibration alone won't reveal cleanly. Thermal imaging or point thermal sensors catch a different failure class again: loose electrical connections, degrading insulation, bearing friction heat. No single sensor type covers every failure mode, which is why a genuine condition-monitoring programme layers two or three sensor classes on any asset judged critical enough to warrant the investment.
Choosing a Telemetry Network for the Site
- LoRaWAN suits low-rate, battery-powered sensors (temperature, vibration summary statistics rather than raw waveform) spread across a large plantroom, basement or campus footprint, with multi-year battery life and low per-node cost.
- NB-IoT is the better fit for dispersed or remote assets — rooftop plant, isolated pump stations — where carrier network coverage is more practical than deploying a private LoRaWAN gateway.
- WiFi is necessary where raw high-frequency vibration waveform needs streaming continuously for advanced spectral analysis, at the cost of much higher power draw and correspondingly shorter battery life or a mains-powered sensor.
- Edge preprocessing (computing an FFT or RMS summary on the sensor itself rather than streaming raw data) dramatically reduces bandwidth and battery demand, and is now standard on condition-monitoring sensor hardware rather than an advanced option.
Design takeaway: Specify sensor type, mounting location and network technology per asset class in the ELV design brief, rather than leaving "add predictive maintenance" as a single generic line item — a chiller motor, a lift machine and a UPS module each need a different sensor combination and network fit.
Retrofit vs New-Build Sensor Placement
New-build ELV designs can specify sensor mounting points on equipment schedules before plant is even ordered, coordinating directly with the mechanical package. Retrofits are messier: sensors need to be added to running equipment without shutting it down, which usually means magnetic-mount or clamp-on accelerometers rather than the more accurate stud-mounted sensors used in new installations, and a compromise on precision that should be documented rather than silently accepted.
Avoiding Alert Fatigue
The fastest way to kill a predictive maintenance programme's credibility is to set alarm thresholds from generic manufacturer defaults rather than each asset's own baseline behaviour — a chiller's "normal" vibration level varies enough between installations that a one-size-fits-all threshold either misses real faults or floods the facilities team with false positives. Tiering alerts by trend severity — a slow 90-day drift versus a sudden step change — lets a facilities team distinguish a genuine early warning worth scheduling around from an urgent dispatch, which is the entire commercial case for predictive over reactive maintenance.
Frequently Asked Questions
What does vibration analysis actually detect before a motor fails?
An accelerometer's raw signal, run through an FFT (fast Fourier transform), reveals frequency peaks corresponding to specific fault modes — bearing defect frequencies, shaft misalignment, rotor imbalance — often weeks before the fault is severe enough to be audible or trip a protection relay.
Should predictive maintenance sensors use LoRaWAN, NB-IoT or WiFi?
LoRaWAN suits low-rate, battery-powered telemetry (temperature, vibration summary data) across a large plantroom or campus footprint at low cost; NB-IoT suits remote or dispersed assets needing carrier network coverage; WiFi suits higher-bandwidth needs like continuous raw vibration streaming, at the cost of much shorter battery life.
How do you avoid alert fatigue on a predictive maintenance platform?
Set alarm thresholds from each asset's own baseline behaviour rather than a generic manufacturer default, and tier alerts by trend severity (a slowly worsening 90-day trend versus a sudden step change) so facilities teams can distinguish an early warning from an urgent dispatch.