The sensor selection question comes up early in every PdM deployment conversation. The choice of vibration sensor technology, temperature measurement approach, and current monitoring method shapes the cost structure, installation complexity, and — critically — which failure modes you can actually detect. There is no universal right answer, but there are clear trade-offs that determine which sensor type fits which application. Here's the framework we use when specifying a monitoring program for a new facility.
Vibration Sensors: MEMS Accelerometers vs. Piezoelectric
Accelerometers are the workhorse sensor for vibration-based PdM. The fundamental choice is between MEMS (Micro-Electro-Mechanical System) accelerometers and traditional piezoelectric (ICP/IEPE) sensors.
MEMS accelerometers are silicon-chip sensors fabricated using semiconductor manufacturing processes. They're small, low-power, inexpensive (typically $15–$80 per unit at industrial-grade quality), and can be integrated directly into wireless IoT sensor nodes without requiring separate signal conditioning hardware. The trade-off is frequency response: most MEMS sensors are accurate to 1–5 kHz, and higher-frequency MEMS sensors become expensive. They're also more susceptible to temperature drift, which matters in environments where bearing housing temperatures vary significantly with load.
Use MEMS when: you need low-cost coverage across a large asset fleet, your target failure modes are low-to-mid frequency (imbalance at 1X, misalignment at 2X, early-stage bearing defects in the 1–5 kHz envelope), and you're deploying wireless nodes where battery life is a constraint. A 10 kHz MEMS sensor in a wireless node sampling at 1 kHz per second draws significantly less current than a piezoelectric sensor with a 25 kHz sample rate, giving you 2–5× longer battery life for the same form factor.
Piezoelectric accelerometers use a crystalline element that generates a charge proportional to applied force. They have wider frequency response — typically 0.5 Hz to 25 kHz and higher for specialized units — lower noise floors at high frequencies, and better long-term stability. The cost is higher ($80–$400+ per channel for industrial-grade ICP sensors) and they require charge amplifiers or IEPE power conditioning, adding to installation complexity and cost.
Use piezoelectric when: you need high-frequency bearing envelope analysis (5–20 kHz), gearbox gearmesh frequency analysis at high tooth counts, or you're instrumenting high-speed spindles or precision machinery where the MEMS frequency ceiling would miss the relevant signatures. For permanent wired installation on critical high-value assets, the diagnostic depth of a good piezoelectric sensor outweighs the cost premium.
Vibration Sensor Selection Summary
| Criterion | MEMS Accelerometer | Piezoelectric (ICP/IEPE) |
|---|---|---|
| Frequency range | DC to 1–5 kHz (typical) | 0.5 Hz to 25 kHz+ |
| Unit cost | $15–$80 | $80–$400+ |
| Power draw | Low (battery-friendly) | Higher (needs IEPE supply) |
| Best for | Fleet-wide screening, low-frequency faults | Critical assets, high-frequency analysis |
| Wireless-compatible | Yes, natively | Requires wired or powered node |
In a typical Gearcadence deployment, we use triaxial MEMS sensors for fleet-wide coverage on motors, pumps, and fans — typically 60–80% of the asset count — and reserve wired piezoelectric channels for gearboxes, high-speed spindles, and any asset where the failure history justifies deeper spectral analysis.
Temperature Sensors: Contact vs. Non-Contact
Temperature is the second essential measurement channel for bearing health monitoring. Bearing contamination, lubrication breakdown, and overloading all produce measurable temperature rises before they generate significant vibration signatures. Temperature monitoring adds fault-mode coverage that vibration alone misses.
Contact temperature sensors — thermocouples and RTDs (Resistance Temperature Detectors) — measure the temperature at the point of contact. For bearing monitoring, a Type K thermocouple or PT100 RTD mounted directly on the bearing housing gives you the most accurate reading of actual bearing temperature. RTDs (PT100 or PT1000) provide better linearity and lower drift than thermocouples and are the right choice for continuous monitoring applications where trending accuracy over months matters. Typical installed cost for a PT100 with transmitter: $30–$80 per point.
Non-contact infrared (IR) thermometers measure surface temperature from a distance using emitted thermal radiation. Handheld IR guns are a standard route-based tool — a technician can take hundreds of surface temperature readings per hour across a machine fleet without any installation cost. Fixed IR sensors mounted on the machine can do continuous non-contact monitoring, but they're sensitive to surface emissivity (painted vs. bare metal reads differently) and viewing angle, which adds calibration complexity.
For permanent condition monitoring, contact RTDs on bearing housings are the standard recommendation. For route-based surveys and thermal imaging of electrical panels, motor windings, or cooling systems, IR is the right tool. The two approaches are complementary, not competitive.
One practical note: a single temperature channel per bearing is often insufficient for diagnosis. On a horizontal motor, measure both the drive-end (DE) bearing and the non-drive-end (NDE) bearing. An asymmetric temperature rise between DE and NDE is diagnostic — it points to load distribution problems, misalignment, or a bearing defect on the hotter side. Both channels together give you information that one cannot.
Current Sensors: Motor Current Signature Analysis
Motor current provides a third monitoring channel that's often underutilized in PdM programs. Motor Current Signature Analysis (MCSA) uses a current transformer (CT) clipped around one motor phase lead to capture the current waveform, which encodes information about both the electrical health of the motor and the mechanical load it's driving.
The technique works because any mechanical fault in the driven load — a bearing defect, a pump impeller imbalance, a gear tooth crack — modulates the motor's torque demand at fault-characteristic frequencies. That torque modulation appears as a sideband in the current spectrum at the motor supply frequency (typically 60 Hz in US facilities) ± the fault frequency. MCSA can detect bearing defects, broken rotor bars, winding faults, and driven-load mechanical problems without a vibration sensor on the load side.
Split-core current transformers — the type that clips around an existing conductor without requiring a wire break — cost $15–$50 each and install in minutes. They're the least intrusive sensing option available. For a facility where vibration sensors can't be conveniently mounted on certain assets (inside guards, in hard-to-reach locations, in food processing environments with high-pressure washdown requirements), CT-based MCSA is often the practical answer.
The limitation: MCSA sensitivity degrades with variable-frequency drives (VFDs). When a motor is driven by a VFD, the supply frequency is dynamic, which spreads the fault sidebands and makes spectral analysis harder. In VFD-controlled assets, vibration sensors on the motor and driven equipment are generally more effective than current monitoring.
Wireless vs. Wired: The Installation Trade-off
Wireless sensor nodes eliminate cable runs — which are often the most expensive and disruptive part of a permanent monitoring installation in an existing facility. A wired triaxial accelerometer with conduit-run cable from a motor to a central DAQ box might cost $400–$800 in installation labor alone. A wireless node with a MEMS sensor clipped magnetically to the motor housing deploys in 15 minutes at near-zero installation labor cost.
The trade-offs for wireless are battery life, sampling rate constraints, and network reliability. Battery life depends on sampling frequency and radio protocol: a node sampling at 100 Hz continuously will drain a standard AA battery in 2–4 months, while a node sampling once per 15 minutes at burst mode can last 2–3 years. Most practical wireless PdM systems use triggered or scheduled burst sampling rather than continuous streaming.
Plant Wi-Fi is often insufficient for IIoT sensor networks — too many dead zones, too much contention with production systems, and IT security policies that restrict device onboarding. Industrial wireless protocols (LoRaWAN, WirelessHART, ISA100) provide more predictable coverage and better determinism for sensor data. Gearcadence's edge gateway supports OPC-UA and Modbus TCP for wired sensors alongside standard Wi-Fi or Ethernet backhaul, letting facilities mix wired and wireless deployments based on asset accessibility and criticality without requiring a single network standard.
The practical recommendation: start with wireless MEMS nodes for initial fleet-wide coverage and baseline establishment. Identify the highest-consequence assets during the first 90 days. Then upgrade those specific assets to wired piezoelectric with continuous sampling, while leaving the rest of the fleet on wireless periodic sampling. This hybrid approach delivers broad coverage at manageable cost while concentrating diagnostic depth where it matters most.
