Vibration Measurement Units: mm/s, g, µm, and What RMS Means

Your report shows “2.8.” Someone else's report on the same machine shows “0.11.” A third person insists it's 45. All three figures can be correct at once: the first is mm/s RMS, the second is ips (inches per second), the third is peak-to-peak displacement in microns. Units and amplitude type get confused more often than the measurement itself goes wrong, and the cost of the mistake is the same either way: you either rework a sound machine, or sign off on a level that's already eating through the bearings, as if it were normal.

Updated 27 August 2026 · by AXILINE · Vila Nova de Gaia

One motion, three quantities

A bearing support makes one motion, and you can describe it with three different quantities. Displacement answers the question “how far did the point move.” Velocity answers “how fast is it moving.” Acceleration answers “how sharply is its speed changing.” All three describe the same process, and all three are correct.

Frequency is what ties them together. For a single sine wave, v = 2π·f·d and a = (2π·f)²·d, where f is the frequency in hertz and d is the displacement amplitude. The main consequence follows from this. Moving to velocity multiplies by frequency; moving to acceleration multiplies by frequency squared. That means displacement emphasizes low frequencies, acceleration emphasizes high ones, and velocity sits right in the middle.

In practice, what you're physically holding is almost always one sensor: an accelerometer. It outputs a signal proportional to vibration acceleration. The instrument gets velocity by integrating once, displacement by integrating twice. On the Balanset-1A, the integrators sit in hardware inside the two-channel USB module, together with the preamplifiers and the ADC, so on screen you read vibration velocity in mm/s directly and work with the running-speed component 1x — vibration at the rotor's rotational frequency.

There's a consequence that follows from this and gets forgotten. Integration suppresses high frequencies and amplifies low-frequency noise. Below the sensor's lower cutoff, it turns that noise into a neat, false rise at the left-hand end of the spectrum — the graph that breaks vibration down by frequency; the English-language literature calls this rise the ski-slope. Converting acceleration into displacement where the sensor has already stopped working is pointless: you'll get numbers with no actual motion behind them.

• Vibration displacement: µm, or mils in imperial units, 1 mil = 25.4 µm. Usually stated as peak-to-peak. Working range below roughly 10 Hz.
• Vibration velocity: mm/s, or ips in imperial units, 1 ips = 25.4 mm/s. Usually stated as RMS. Working range roughly 10–1000 Hz.
• Vibration acceleration: m/s² or g, where 1 g = 9.81 m/s². Stated as either RMS or peak, often both at once. Working range above roughly 1000 Hz.

Sources: Balanset-1A operation manual

What applies where: three quantities, three jobs

Vibration velocity became the universal condition assessment for housings by calculation, not by tradition. Over the 10–1000 Hz band, one number in mm/s RMS weighs the running-speed component, its first harmonics (frequencies that are multiples of it), and moderate high-frequency energy roughly evenly. Zones A, B, C and D in ISO 20816 are built on exactly this quantity. It's also usually the one you write into a contract and into a monitoring system's setpoint.

Displacement stays relevant where frequencies are low, or where you care about clearance. A slow-running drum at 30 rpm, a spindle, shaft position in a plain bearing from proximity probes. What matters here isn't an abstract “vibration level,” but the physical range of motion in microns: it either fits inside the clearance or it doesn't. Note that shaft vibration is assessed against separate documents, not against tables written for bearing housings.

Acceleration is what you need when you're hunting for impacts. A rolling-element bearing defect, gear meshing, rubbing, cavitation. The impulse lasts a fraction of a millisecond and lives at kilohertz frequencies, so it almost vanishes in velocity: integration divides its amplitude by frequency. A typical early-stage picture looks like this: RMS vibration velocity has stayed put for three months, while peak acceleration has doubled over the same period.

There's a flip side too, and it's worth stating plainly. Measuring acceleration on a slow-running unit is pointless: at 50 rpm the running frequency is 0.83 Hz, and the acceleration from it drowns in noise. Assessing a fast machine by displacement is just as pointless: at 3000 rpm, 4.5 mm/s RMS gives a peak-to-peak range of only 40 microns, and a displacement sensor on the housing simply won't catch motion that small.

The last row is deliberately out of place here. Residual unbalance isn't vibration, it's mass multiplied by radius. It can't be compared to mm/s or to peak-to-peak microns, and “in tolerance” in balancing software means something entirely different from “in zone B” under ISO 20816. We have a separate article on these three different tolerances.

Sources: ISO 20816-1:2016 · ISO 21940-11:2016 · ISO 13373-3:2015

Conversion table: the same 4.5 mm/s

Converting between quantities takes one line, but only at a known frequency. From v = 2π·f·d you get d = v / (2π·f). Multiplying velocity by 2π·f gives you acceleration. Here f is the frequency of the specific component you're converting, not “vibration frequency in general.”

Let's see what this gives you in practice. Take a single value, 4.5 mm/s RMS, and work out what peak-to-peak displacement and what acceleration it corresponds to at different frequencies.

• A hundredfold spread across the two outer columns, and that's physics, not error.
• At 10 Hz, that 4.5 mm/s means 200 microns of peak-to-peak motion: you'll see the support move with your eye and feel it with your hand.
• At 1000 Hz, the same 4.5 mm/s means 2 microns: no motion visible or audible, but the acceleration is almost 3 g.
• The same number in mm/s describes completely different mechanical situations, and without the frequency you don't know which one.

Now for the limitation that means you can't apply this table mechanically. The formula works for a single sine wave, that is, for a single spectral line at a known frequency. Broadband RMS over the 10–1000 Hz band is built up from dozens of components at once, and it doesn't have a single frequency. Converting an overall 4.5 mm/s RMS into microns is impossible in principle: the question “at what frequency” goes unanswered. Convert individual components instead. And check the sensor's lower cutoff: a typical industrial accelerometer honestly works from around 1–2 Hz, and with a built-in integrator that cutoff can be even higher.

RMS, peak, peak-to-peak: four numbers for one signal

The quantity answers the question “what are you measuring.” The amplitude type answers a different question: “how did you collapse the oscillation into a single number.” These get confused more often than the units themselves, and the consequences are worse, because units are usually labeled on screen, while the amplitude type is buried in the settings.

RMS is the root-mean-square value. It's tied to the energy of the oscillation and responds weakly to single outliers, which is why machine-condition limits are written in RMS specifically. Peak is the maximum deviation from zero. Peak-to-peak is the full range, from minimum to maximum. Rectified average shows up in older instruments and in datasheets, and you need to recognize that one on sight too.

• 4.5 mm/s RMS is 6.4 mm/s peak and 12.7 mm/s peak-to-peak. Three different numbers, the same vibration.
• Compare today's peak reading with the RMS from last year's report, and you get a “40% rise” where the machine hasn't changed at all.
• The reverse mistake is symmetric: divide the measured peak by 1.41 and write the result down as an RMS value the instrument never actually showed.

The main limitation that trips up the 1.41 factor. It's derived for a pure sine wave. On an impact signal, the ratio collapses completely: for a bearing with spalling, peak acceleration can exceed RMS by eight to twelve times. Multiply RMS by 1.41 and call it the peak, and you've understated the actual peak by a factor of six — meaning you've missed exactly what you were looking for. If the limit is stated in RMS, record RMS. If the instrument can only output peak, say so in the report in plain words, and don't force the figure to fit the setpoint.

Crest factor: the number that catches impacts

Crest factor is the ratio of peak to RMS for the same signal: CF = peak / RMS. On its own it doesn't measure anything, but it tells you right away what shape of signal you've caught — harmonic, or impact-driven.

The point is that RMS climbs slowly, while peak reacts to impacts immediately. As long as the defect is localized, RMS barely moves while peak climbs, and crest factor rises. This is the earliest sign you get without spectral analysis at all — just from two numbers off the same acceleration reading.

• A pure sine wave: 1.41. Anything close to this number points to a harmonic signal — more likely unbalance or misalignment than impacts.
• Broadband random noise from a sound unit: roughly 3–4.
• A localized bearing or meshing defect: 5–10 and higher.
• A late stage of failure: back down to 3–4, because the impacts have merged into continuous noise.

Sources: ISO 13373-3:2015

Six lines without which a number means nothing

“We've got 7 mm/s” isn't a measurement result, it's half of one. The other half is made up of the supporting details, and without them the figure has nothing to be compared against: not a limit, not a previous reading, not a colleague's data.

A separate trap is the system of units. Instruments of American origin default to ips and mils. The route has an Alarm setpoint of 7.1 mm/s, the instrument shows 0.28, the technician sees a “small number” and moves on. But 0.28 ips is exactly the same 7.1 mm/s, and the setpoint has already been hit. Keep three conversions in your head: 1 ips = 25.4 mm/s, 1 mil = 25.4 µm, 1 g = 9.81 m/s².

The frequency band trips people up more quietly, but just as reliably. Two instruments at the same point give 3.1 and 4.6 mm/s, and both are right: the first has an upper cutoff of 200 Hz, the second 1000 Hz, and the difference is high-frequency energy the first instrument simply doesn't count. On the Balanset-1A, RMS vibration velocity is measured over the 5–200 Hz band, across a 0.02–80 mm/s range. That's enough for the running-speed component and the first harmonics of industrial machines, but it's not the full 10–1000 Hz band. If contractual acceptance specifically requires 10–1000 Hz, agree on the instrument in advance, not on handover day.

• Quantity: displacement, velocity or acceleration. Without it, the figure can't be read at all.
• Units and system: mm/s or ips, µm or mils, m/s² or g.
• Amplitude type: RMS, peak, peak-to-peak, or rectified average.
• Frequency band, for example 10–1000 Hz. For machines below 600 rpm the lower cutoff is usually dropped to 2 Hz, but check the applicable part and edition of the standard for your own machine.
• Point and direction: a specific bearing support, and H, V or A. Plus the sensor mounting method, because that's exactly what cuts off the upper end of the band.
• Regime: speed from the tachometer, load, temperature, what else was running nearby.

Sources: ISO 20816-1:2016 · Balanset-1A operation manual

Common mistakes that cost you a shift

What to write in the report, and what to do if the numbers don't match

A report doesn't exist to look nice. It exists so that a year from now, someone else can repeat your measurement and compare their figures to yours. The test is simple: hand the report to a colleague and ask whether they can reproduce the reading without asking you anything. If not, the report is missing a line.

Below is a sample entry for a single point. The figures are illustrative; what matters is the structure.

• Machine, unit, point number and measurement direction.
• Quantity, units and amplitude type in one line: “vibration velocity, mm/s, RMS.”
• Frequency band, and for a spectrum, also Fmax, the number of lines, the window, and the number of averages.
• Sensor type and mounting method: stud, adhesive, magnet, probe.
• Speed from the tachometer, load, temperature, running time before the reading.
• Instrument, serial number, date of the last sensor sensitivity check.
• The limit you're comparing against: document, part, edition, machine class, support type.
• Three repeat readings and the spread between them, not a single number.

If the figures from two readings still don't match and the report is complete, what follows is diagnostics, not arithmetic. We're engineers who design and manufacture the Balanset instruments and balance with them on site ourselves, and in our experience, discrepancies come from three things: a different band between the instruments, different sensor mounting, and a different machine regime. Send us the two reports and the speed, and we'll work out where the discrepancy is coming from. And if the numbers show the running-speed component dominates, the question moves onto practical ground: balancing the rotor in the field, in its own bearings, without removing the impeller. Advisory support on choosing the quantity, band and measurement point is part of the job, not something sold separately.

Sources: Balanset-1A manufacturer specification

Frequently asked questions

Related content