Vibration Sensors: From MEMS Accelerometers to IEPE Piezo — The Complete Quality-Tier Buyer's Guide

Quick picks by tier. Hobby → a MEMS accelerometer breakout (ADXL345, MPU6050) for tilt and gross motion. Maker/Prosumer → a low-noise wide-bandwidth MEMS (ADXL355/1002, IIS3DWB). Commercial → an IEPE/ICP piezoelectric accelerometer + signal conditioner for FFT bearing analysis. Hardened industrial / audited → a 4-20 mA industrial vibration transmitter or an eddy-current proximity probe for machinery protection. Why each wins is below — but if you just need a name, start there.

The thing that actually bites you: bandwidth, not the g-rating

The classic vibration-monitoring mistake is buying an accelerometer by its g-range and sample rate, bolting it on with a magnet, and expecting it to catch a failing bearing. It won't — because the faults you're hunting live at frequencies the sensor and its mounting can't see. A rolling-element bearing announces its death at defect frequencies (BPFO, BPFI, ball-spin) up in the kilohertz range, often as small, high-frequency spikes riding on the rotation. A cheap MPU6050 is usable to maybe ~1 kHz with a noise floor that buries those spikes. So it dutifully reports "vibration normal" right up until the bearing fails.

The second silent killer is mounting. The same accelerometer has a wildly different usable frequency range depending on how it's attached: a stud mount might be good to 10+ kHz, an adhesive pad to a few kHz, and a handheld magnet to maybe 1–2 kHz. Mount a good sensor badly and you've thrown away the bandwidth you paid for.

And you have to measure the right quantity. Overall-level trending for machine health (ISO 10816/20816) is done in velocity (mm/s RMS); spectral diagnostics of bearings and gears is done in acceleration (g). A sensor and pipeline tuned for one don't automatically serve the other.

So the real question isn't "is it vibrating?" — it's "can this sensor see the frequency band where my failure mode lives, mounted the way I can actually mount it, in the units my standard uses?"

The decision axes

• Usable bandwidth — the frequency range with flat response and low noise. The single most important spec, and the one cheap MEMS quietly lack.
• Noise floor (µg/√Hz) — decides whether small early-stage bearing faults are visible above the sensor's own hash.
• Mounting method — stud > adhesive > magnet > handheld, in descending usable bandwidth. It gates everything above.
• Quantity & standard — velocity (mm/s RMS, overall health) vs acceleration (g, spectral diagnostics); enveloping/demodulation for early bearing detection.
• Output type — digital MEMS (I²C/SPI) → IEPE/ICP analog (mV/g, needs constant-current conditioner + DAQ) → 4-20 mA transmitter (overall velocity to a PLC) → eddy-current proximity (shaft displacement).
• Environment & rating — temperature, IP rating, and ATEX/IECEx for hazardous areas.
• Triaxial vs single-axis, and whether you need raw waveform (for FFT) or just an overall RMS number.

Walking up the ladder

Hobby — MEMS breakout (ADXL345, MPU6050) ($3–10). Excellent for tilt, orientation, impact/drop detection, and seeing that something shakes. Limited bandwidth and a high noise floor make it unfit for bearing diagnostics, but it's a great way to learn and to flag gross imbalance or "the machine fell over." Stop here if you want motion/tilt or a crude "running vs stopped" signal.

Maker/Prosumer — low-noise wide-band MEMS (ADXL355, ADXL1002, IIS3DWB) ($20–80). Modern MEMS with low noise and kHz-class bandwidth (the IIS3DWB reaches ~6 kHz). With a stud or solid adhesive mount and a proper FFT, these do real condition monitoring on a budget — the basis of most affordable predictive-maintenance nodes. Stop here if you're trending a fleet of motors/pumps and can mount and sample them properly.

Commercial — IEPE/ICP piezoelectric accelerometer + conditioner ($200–800 + DAQ). The industry standard for vibration analysis: a flat response from ~1 Hz to 10+ kHz, low noise, stud-mounted, mV/g output through a constant-current (IEPE) conditioner into a DAQ. This is what bearing and gear FFT/envelope analysis is actually done with. Stop here if you need credible spectral diagnostics and early-fault detection on critical rotating equipment.

Hardened industrial — 4-20 mA vibration transmitter or eddy-current proximity probe ($300–2000+). Loop-output velocity transmitters (Wilcoxon, PCB, Banner) feed overall mm/s straight to a PLC for trip/alarm; eddy-current proximity probes measure shaft displacement for sleeve-bearing turbomachinery to API 670 machinery-protection standards, with ATEX options. You need this tier if vibration must trip a machine, satisfy an auditor, or protect turbomachinery.

Interface & wiring notes (per tier)

• MEMS (I²C / SPI): SPI for the high sample rates real vibration needs — I²C often can't keep up. Set the full-scale range and output data rate deliberately, and stud- or adhesive-mount rigidly; a sensor flopping on a breadboard wire measures the wire, not the machine.
• IEPE/ICP (coax + constant current): needs a 2–4 mA constant-current source and a DAQ that samples fast enough (≥2.56× your top frequency of interest). Stud-mount on bare, flat metal; cable whip adds noise.
• 4-20 mA transmitter: loop-powered, reads as overall velocity through an ADC/4-20 mA HAT — great for trend/alarm, not for spectra.
• Eddy-current proximity probe: needs a driver/proximitor and careful gap setup; this is a specialist machinery-protection install, not a bolt-on.

A note on the spec that flatters cheap sensors: a high sample rate doesn't help if the noise floor and physical bandwidth are low — you'll sample noise faster. And a magnet mount caps your usable frequency no matter how good the sensor is. Match sensor bandwidth, sampling, and mounting to the defect frequencies you're trying to catch, or you'll get a confident "healthy" reading from a dying machine.

How LoopString reads every tier — in one dashboard

Here's the part that makes the whole ladder moot from a software standpoint: a Raspberry Pi running LoopString's Node-RED templates reads a $5 MEMS node and an industrial 4-20 mA velocity transmitter into the exact same dashboard — same live trend, same vibration-threshold alerts, same maintenance log and duty-cycle history. You can prototype condition monitoring on a wide-band MEMS and add a loop-output transmitter for the critical machine without touching your dashboards or alert rules. Pick the tier your machine actually needs using the table above, wire it to a Pi, and watch the trend at app.loopstring.io.

Useful next reads: the Raspberry Pi industrial monitoring guide, the Raspberry Pi automation guide, and the MQTT sensor dashboard overview.