- November 19, 2026
Not enough time? Get the key points instantly.
Your maintenance lead wants sensors on 40 pumps and two gearboxes by next quarter. Three vendors have sent datasheets, all promising 100 mV/g, IP67, and years of battery life. Choosing a vibration sensor for machine health monitoring from those pages is harder than it looks. The specs that decide success rarely sit on page one.
Choose wrong and the sensor still produces clean, plausible numbers. It simply cannot see the fault you bought it for. You find out after the fleet rollout.
By the end of this post, you will be able to shortlist the right sensor for each machine. You will also know how to prove that choice on your own equipment before you commit a fleet budget.
Most teams compare a vibration sensor for machine health monitoring on price, battery life, and IP rating. Those numbers decide budget, not detection.
Call it the spec-sheet mismatch. The specs that decide detection get checked last:
Frequency response as mounted, because a magnet can erase the band where bearing damage shows first.
Sensitivity against real amplitude, because slow machines produce signals that vanish in sensor noise.
Output against your data path, because nobody can rebuild a spectrum from an overall 4–20 mA value.
Take a cooling tower fan at 120 RPM. Its running-speed vibration sits at 2 Hz, where a standard accelerometer rolls off. You get a number, not a trend you can act on. [verify before publishing: low-frequency limit of the specific sensor class]
Start every selection with four questions per machine: what usually fails, how fast it turns, where you can mount, and where the data must go. The datasheet answers none of them. Your machine does.
Vibration sensors fall into three families, and no tuning fixes the wrong one.
Sensor type | What it measures | Best for | Main limitation |
|---|---|---|---|
Accelerometer (piezoelectric or MEMS) | Casing acceleration | Rolling-element bearings, gearboxes, most motors and pumps | Weak output at very low frequencies |
Velocity sensor | Casing velocity | Overall severity on medium-speed machines | Misses high-frequency bearing and gear detail |
Proximity probe (eddy current) | Shaft motion relative to the casing | Turbines and compressors with sleeve bearings | Needs a fixed mount near the shaft |
Sleeve bearings change the rules. Their oil film damps rotor motion before it reaches the casing, so turbomachinery relies on proximity probes.
On rolling-element bearings, the default vibration sensor for machine health monitoring is an industrial accelerometer. It covers bearing defects, gear mesh, and imbalance, because firmware can integrate acceleration into velocity. A velocity sensor cannot recover high-frequency content it never captured.
Sensitivity is where a vibration sensor for machine health monitoring most often goes wrong on slow or violent machines. Quoted in mV/g, it sets signal per unit of vibration. Range sets where the sensor clips. An IEPE sensor (a two-wire interface that powers the sensor over its signal cable) swings only about ±5 V. The two trade off directly.
Sensitivity | Typical range | Use it for |
|---|---|---|
10 mV/g | ±500 g | High-speed turbines, gear-mesh harmonics, impact-heavy machines |
100 mV/g | ±50 to ±80 g | Most rotating equipment: motors, pumps, fans, blowers |
500 mV/g | ±10 g | Slow machines: cooling tower fans, wind turbine main shafts, agitators |
Three datasheet details deserve a second look:
Range headroom. Pick 80 g over 50 g where impacts occur, because clipping hides the spikes that signal bearing damage.
Sensitivity tolerance. Precision sensors hold ±5%; low-cost ones vary ±10–20%. Load each calibrated value into your system.
Settling time. Low-frequency IEPE sensors can need seconds to settle after power-up. A battery node sampling 2-second bursts records only settling noise. A MEMS vibration sensor has no such delay.
A vibration sensor for machine health monitoring performs only as well as its mount. Datasheets quote response on a polished fixture, not a painted, curved bearing housing.
Mounting method | Effect on high-frequency response | When to use it |
|---|---|---|
Stud into a spot-faced hole | Widest response; matches the datasheet | Permanent installs on critical assets |
Epoxy or adhesive pad | Slightly below stud; usually negligible | Permanent installs where drilling is not allowed |
Magnet on a flat, greased surface | Can stay close to stud response up to about 20 kHz | Route-based data collection |
Magnet, dry or on a curved surface | Resonance can fall below 12 kHz | Quick checks only |
Output decides what your team can do with machine health monitoring data later.
Output | What you get | Fits best when |
|---|---|---|
IEPE (dynamic analog) | Full waveform for spectrum and envelope analysis | Wired online systems and route analyzers |
4–20 mA transmitter | One overall value, such as velocity RMS | Alarms inside an existing PLC, DCS, or SCADA system |
Digital MEMS (SPI or I2C) | Raw samples into a microcontroller | OEM equipment and custom monitoring products |
Wireless vibration sensor | Features or waveforms over Wi-Fi, BLE, or LoRaWAN | Large fleets where cabling costs more than sensors |
A 4–20 mA loop suits shutdown alarms. Its spectrum never leaves the transmitter, so many plants add IEPE or wireless for condition monitoring. Digital MEMS output also lets the node run anomaly detection, covered in our guide to running machine learning models on microcontrollers.
Every battery-life figure assumes a measurement schedule. SPM rates its Airius sensor at four years with four measurements per day. Analog Devices modelled a mesh node sending data every minute. Its batteries lasted about 19 days. At one minute per hour, they lasted 1.4–2.1 years.
A datasheet tells you what a sensor does on a test fixture. Only your machine tells you what it does in your plant. There, variable-frequency drives inject noise, steel blocks radio links, and crews mount sensors wherever they can reach.
The sensor decision should therefore end in a proof of concept (POC): a working setup on one or two representative machines, built to the design you would roll out. Nothing else shows how a vibration sensor for machine health monitoring behaves on your equipment.
POC layer | What it proves | What goes wrong without it |
|---|---|---|
Sensor and mount | The fault band survives on your bearing housing | A magnet or bracket erases the bearing band |
Node hardware and power | Signal-chain noise and real battery drain at your interval | Batteries go flat in months instead of years |
Firmware | Sampling, on-node features, and OTA updates work | Raw data floods the network |
Connectivity | Data reaches the gateway from inside your plant | Dead zones behind motors and steel walls |
Integration and dashboard | Alerts reach the PLC, SCADA, or app your team uses | Data nobody looks at |
Agree on pass criteria before the build starts:
The setup flags a known or seeded fault, such as a deliberate imbalance.
Measured battery drain projects to the target life at the chosen interval.
Data reaches your existing system without manual export.
A technician installs a node within the budgeted time.
A POC also exposes failures datasheets never mention, such as cables and connectors that fail before the sensor. Controller integration is usually the hardest layer; our industrial CNC drilling machine automation project shows what that work involves.
Match each machine to a starting point:
Machine | Starting sensor choice | Why |
|---|---|---|
Motor or pump, 1,000–3,600 RPM | 100 mV/g accelerometer, stud-mounted radial at the drive end | Covers bearings, imbalance, and misalignment |
Fan or agitator below 300 RPM | 500 mV/g low-frequency accelerometer | Lifts tiny low-frequency signals above the noise |
Gearbox | 100 mV/g or 10 mV/g with mounted resonance above 30 kHz | Gear-mesh harmonics sit high in the spectrum |
Kiln fan or dryer roll above 120 °C | High-temperature IEPE (about 150 °C) or charge-mode (about 260 °C) | Standard IEPE tops out near 120 °C |
Your own equipment product | Digital MEMS vibration sensor on the controller board | Lowest unit cost and direct raw-data access |
Before you commit a fleet budget, confirm six points:
The dominant failure mode and its frequency band.
Sensitivity and range that fit speed and shocks.
A mount that preserves that band.
An output your systems can ingest.
A power budget at your real measurement interval.
A POC on one representative machine that meets agreed pass criteria.
A vibration sensor for machine health monitoring that clears all six, POC included, is ready to support predictive maintenance across the fleet.