A piezoelectric vibration sensor with 100 mV/g nominal sensitivity is the most common industrial accelerometer and outputs a continuous voltage proportional to acceleration, which a downstream vibration meter integrates to velocity and displacement for ISO 10816 / ISO 20816 severity evaluation [S1].
A tachometer is a fundamentally different instrument: it produces a discrete pulse train (or analog voltage / 4-20 mA / frequency output) locked to shaft angular position, used for closed-loop speed control, phase reference for balancing, and overspeed protection — not for vibration severity.
What Each Device Actually Measures
The dominant industrial vibration sensor is a piezoelectric pressure-type accelerometer in which a preloaded mass compresses and stretches a piezoelectric crystal, generating a charge that the internal electronics convert to a low-impedance voltage output; the article gives a representative sensitivity of 100 mV/g with usable bandwidth typically spanning 1 Hz to 10 kHz [S1]. A vibration analyzer then processes that voltage into acceleration (g or m/s²), velocity (mm/s RMS), and displacement (µm peak-to-peak) for trending.
A tachometer measures rotational speed: optical, magnetic, Hall-effect, or proximity probes deliver 1 pulse per revolution (PPR) up to several thousand PPR; resolvers and encoders deliver quadrature signals (e.g. A/B/Z, 1024 PPR and up) giving both speed and absolute angular position. The output domain is RPM, Hz, or degrees, never g.
Core Spec Comparison: Accelerometer vs Tachometer
The two instrument classes diverge on almost every critical spec, which is why mixing them up on a P&ID is a common spec error. Below is a practical comparison based on what the reference materials support; values flagged with [S#] are sourced, others are stated as qualitative engineering norms. [S1]
Output signal: accelerometer = continuous AC voltage proportional to acceleration (≈100 mV/g typical, [S1]); tachometer = discrete pulse train, frequency output, or quadrature A/B. Frequency response: accelerometer is wideband vibration (≈1 Hz–10 kHz) [S1]; tachometer is DC to a few hundred Hz following mechanical shaft speed. Power and interface: accelerometers are passive two-wire IEPE / constant-current powered, or low-voltage MEMS; tachometers are often 5–24 VDC, NPN/PNP push-pull, or encoder line-driver (RS-422). Mounting: accelerometers stud-mount, adhesive-mount, or hand-held probe on bearing housings [S1]; tachometers target a target slot, keyway, gear tooth, or shaft encoder disk.
Selection rule of thumb: if the engineering question is "how much is this machine shaking?" or "is the bearing degrading?", specify an accelerometer. If the question is "how fast is this shaft turning, and when does it cross zero degrees?", specify a tachometer.
Where Each Device Belongs in a PdM Architecture
Condition-monitoring dashboards pair the two deliberately: accelerometers are wired to a multi-channel vibration analyzer (the Adash A3800, for example, accepts 4–16 channels of IEPE accelerometers and runs FFT, envelope, and trending [S1]); the resulting velocity RMS in mm/s is compared against ISO 10816-3 zone boundaries (e.g. 2.8 mm/s, 4.5 mm/s, 7.1 mm/s for medium machines, 300–600 kW) to drive alarm and trip setpoints.
A tachometer feeds the same controller a once-per-revolution pulse, which the analyzer uses as a phase / order reference so vibration spectra can be displayed as orders (1×, 2×, 3× shaft speed) rather than only frequency. Without the tachometer pulse you cannot tell whether a peak at 50 Hz is shaft imbalance (1× of a 3,000 RPM motor) or a structural resonance. This is also why order tracking is non-negotiable on variable-speed drives above ~600 RPM.
Application Fit: Who Should Buy Which
Buy a vibration sensor when the failure mode is bearing wear, imbalance, looseness, cavitation, or gear-mesh wear; when the asset is medium-to-high criticality (pumps, fans, motors, compressors, gearboxes, spindles); and when ISO 10816 / ISO 20816 compliance, alarm-to-trip documentation, or analytics-grade waveform capture is required. Piezoelectric accelerometers with 100 mV/g remain the workhorse for general rotating machinery, while MEMS digital-output triaxial accelerometers dominate IoT-class retrofits and low-cost monitoring of HVAC and small pumps [S1].
Buy a tachometer when the engineering requirement is rotational speed measurement, overspeed shutdown, phase reference for balancing or order analysis, or closed-loop speed feedback on a VFD. Do not specify a tachometer for vibration severity — it lacks the bandwidth and the acceleration-domain physics; the RMS of a tachometer pulse train correlates with neither ISO 10816 velocity nor with bearing envelope demodulation.
Buy both when the asset is critical and variable-speed: the accelerometer carries the alarm/trip logic, the tachometer provides the order reference and overspeed interlock. For a full buying spec walk on the vibration side see the Vibration Sensor Buying Guide 2026: Types, Specs, and Selection Map — it lines up piezoelectric, MEMS, and IEPE choices against sensitivity, bandwidth, mounting, and hazardous-area ratings.
Limits, Failure Modes, and Common Spec Errors
Accelerometer failure modes are dominated by mounting and cable: a hand-held probe on a 50 mm cast-iron housing reads roughly 5–10× lower than a stud-mounted sensor at the same point; loose studs clip the high-frequency response above 5 kHz and mask early bearing-fault signatures. Cable noise and ground loops saturate low-frequency (<5 Hz) channels, so DC-coupled velocity from an accelerometer is rarely trusted below 1 Hz. Piezoelectric sensors also cannot measure true static acceleration — they require AC excitation. [S1]
Tachometer failure modes are dominated by target geometry and air gap: optical and inductive probes miscount when tooth spacing is uneven or the gap exceeds the probe's rated stand-off (often 1–3 mm for inductive, 10–50 mm for laser tach). At high RPM, optical tachometers lose pulses due to ambient light or reflective surface finish; magnetic and Hall-effect probes are more robust but require a ferromagnetic target. Encoder-line tachometers add failure modes around quadrature decoding, EMI on long cable runs, and missing Z (index) pulses that break absolute position on cold start.
Common spec errors worth flagging on a review: specifying a tachometer where an accelerometer is required because the wrong sensor "is already in stock"; buying a 100 mV/g accelerometer when the IEPE input of the vibration analyzer is configured for 10 m/V (a 10× sensitivity error); and ordering a hazardous-area accelerometer without confirming the Ex marking matches the zone (Ex d IIC T4 vs Ex ia IIC T4 carry different installation rules). For a parallel decision-matrix view on a different but structurally similar pairing, the Dynamometer vs Hopper Scale: Spec Map, Capacity Ranges, and Selection Logic piece follows the same selection discipline and is worth reading alongside this one.
ML-Assisted Vibration Decisioning on Top of the Sensor Layer
Open-source Decision Tree classifiers such as agb2k/Vibration-Sensor-Classification-Model (SciKit-Learn based) demonstrate how a 100 mV/g-class accelerometer output, captured as labelled normal / fault CSV, trains a fault-vs-healthy classifier in tens of lines of Python [S2]. A related IoT reference, fmilburn3/Vibration_Sensor_SW-180xxP, uses an SW-180xxP spring-mass vibration switch to fire a digital trigger LED on a low-cost MCU — useful for threshold alarms but incapable of spectral diagnostics [S3].
Production-grade classification still runs in the vibration analyzer itself: envelope analysis on bearing-tone sidebands, kurtosis, and crest factor remain the first-pass features before any ML model is loaded. Tachometer-pulsed order tracking is the prerequisite for variable-speed ML, because the same physical fault shows up at different frequencies at 1,500 vs 4,500 RPM — without an order reference, training data labels are non-stationary.
Track the following two signals over the rest of 2026 before committing to a vendor refresh: (1) emergence of IO-Link vibration sensors with embedded FFT and alarm thresholds, which compress the accelerometer + analyzer pair into a single IP67 node, and (2) the next round of encoder-less VFD tachometer feedback, which pushes the tachometer function into the drive firmware via back-EMF estimation. Either shift changes the BOM count and the engineering effort line on a predictive-maintenance rollout; for adjacent selection work, the Capping & Sealing Machine TCO: 10-Year Cost Driver Stack and Spec Map article shows how a 10-year TCO discipline is built when sensor and instrumentation decisions feed downstream OPEX.