How to Read Vibration Readings: mm/s, g, Microns and What the Numbers Mean

A technician hands you a sheet with one number on it: 4.8. Whether that is fine, marginal or alarming depends on at least five things that are not written on the sheet.

Four point eight of what? Millimetres per second or metres per second squared? RMS or peak? Measured over what frequency range? On the bearing housing or on the shaft? On which machine, mounted how? Change any one of those and the same figure moves from comfortable to urgent.

This is the gap that trips up most plants starting out with vibration monitoring. The measurement itself is straightforward and the instruments are good. The interpretation is where the value is, and it depends on understanding what the number is actually describing. This article works through that: the three quantities, the units, the averaging, the difference between an overall level and a spectrum, and how to read what you are looking at.

It is the companion to our article on vibration analysis for rotating equipment, which covers fault diagnosis, and our guide to ISO 20816 vibration limits, which covers what you are allowed to have.

The Three Quantities, and Why There Are Three

Vibration can be described as displacement, velocity or acceleration. These are not three different measurements. They are the same motion expressed three ways, mathematically related by differentiation, and each one emphasises a different part of the frequency range.

Displacement measures how far something moves, in micrometres or mils. It emphasises low frequencies. Displacement is the natural quantity for shaft measurements taken with proximity probes inside fluid-film bearings, because what matters there is physically how much the shaft is moving within its clearance. It is also the right quantity for slow-turning machinery, where velocity readings become uninformatively small.

Velocity measures how fast it moves, in millimetres per second. It is broadly flat across the mid frequency range where most mechanical faults live, which is exactly why it became the default for general machine condition. Unbalance, misalignment, looseness and most structural problems produce their strongest signature in the range velocity handles well. When somebody quotes a machine vibration level without qualification, they almost always mean velocity in mm/s.

Acceleration measures the rate of change of velocity, in metres per second squared or in g. It emphasises high frequencies. Rolling element bearing defects and gear mesh problems generate high-frequency energy that barely registers in velocity terms but is obvious in acceleration. This is why bearing analysis is done in acceleration or in one of the derived high-frequency techniques rather than in mm/s.

The practical consequence is that a machine can look perfectly healthy in velocity and have a badly damaged rolling element bearing. The energy is there; velocity measurement just does not weight it heavily. Choosing the wrong quantity is the most common reason a monitoring programme misses a fault it was nominally watching for.

RMS, Peak and Peak-to-Peak

This is where a great many misunderstandings live, because the same physical vibration can be quoted as three different numbers.

RMS is the root mean square, a measure of the energy content of the signal averaged over time. It is the standard basis for machine vibration severity, and it is what the ISO standards use. When a limit is quoted in mm/s without qualification, it is almost always RMS.

Peak is the maximum excursion from zero. For a pure sine wave, peak is RMS multiplied by the square root of two, approximately 1.414. Some instruments derive peak from RMS using exactly that factor, which is fine for a clean sine and misleading for anything else. Real machine signals are not sine waves.

Peak-to-peak is the full excursion from one extreme to the other, twice the peak for a symmetrical signal. Shaft displacement measurements are conventionally quoted peak-to-peak, because what matters is the total movement within the bearing clearance.

The trap is comparing a reading in one convention against a limit in another. A machine reading 6.5 mm/s peak is at roughly 4.6 mm/s RMS, and those two figures fall in different severity zones. Before comparing any reading against any limit, confirm both are on the same basis. Instruments can usually be configured either way, and the configuration is not always what somebody assumed.

The same motion expressed three ways, each weighting a different part of the frequency range. Choosing the wrong quantity is why programmes miss faults they were nominally watching for. Sources: ISO 13373-1:2002; ISO 20816-1:2016.

Frequency Range Matters As Much As the Number

Every overall vibration reading is the summed energy within a defined frequency band, and the band is part of the measurement.

The conventional band for general machine vibration severity is 10 Hz to 1,000 Hz, and this is the band the severity standards are built around. For machines running below 600 revolutions per minute the lower limit moves down to 2 Hz, because otherwise the running speed component itself falls outside the measurement.

Read that again in practical terms: a slow-speed machine measured on a standard 10 Hz to 1,000 Hz setting will not include its own running speed vibration in the reading. The instrument will confidently report a low overall level for a machine that is shaking visibly. This is not an instrument fault. It is a configuration error, and it is common.

At the other end, a band that stops at 1,000 Hz excludes most rolling element bearing defect frequencies and all gear mesh frequencies on anything but a slow gearbox. An overall velocity reading in the standard band is simply not the tool for finding those faults.

So the first question about any reading is which band it was taken in, and the second is whether that band contains the frequencies where the fault you care about would appear.

Overall Level Versus Spectrum

An overall level is a single number summarising everything in the band. It is excellent for trending and nearly useless for diagnosis.

Its strength is that it is cheap, fast, repeatable and comparable. Walking a route with a hand-held meter and recording overall levels at fixed points produces a trend, and a trend is what detects change. For a great many plants this is the entire condition monitoring programme, and it catches a worthwhile proportion of developing faults.

Its weakness is that it cannot tell you what changed. A rise from 2.8 to 4.5 mm/s says something is different; it does not say whether that is unbalance, misalignment, looseness, a bearing, or the pump cavitating. Two machines with identical overall levels can have completely different problems, and the same machine can develop a serious localised fault with almost no movement in its overall level if the fault energy sits outside the band or is small relative to the total.

A spectrum decomposes the signal into its frequency components, and this is where diagnosis happens. The convention is to express frequencies as multiples of running speed, written as 1X, 2X and so on, because that makes the pattern comparable across machines running at different speeds.

Reading a spectrum starts with locating running speed and confirming it against the actual shaft speed rather than the nameplate. Then the pattern:

A dominant 1X component usually indicates unbalance, though it is also produced by a bent shaft, an eccentric rotor, and by misalignment in some configurations. A high 2X component alongside 1X commonly points at misalignment, particularly with axial energy present. A series of harmonics at 1X, 2X, 3X and beyond suggests looseness or a non-linearity. Sidebands around a central frequency indicate modulation, which is characteristic of gear and bearing faults. Non-synchronous components at fractions or non-integer multiples of running speed point at things not rigidly connected to the shaft, including rolling element bearing defects, belt problems and some fluid-related instabilities.

Our article on vibration analysis for rotating equipment works through these patterns in more detail. The point here is that the pattern lives in the spectrum, and no overall level contains it.

Phase, and Why It Settles Arguments

Phase measurement compares the timing of vibration at one point against a reference, usually a once-per-revolution trigger. It is under-used, and it resolves the diagnostic ambiguities that amplitude alone cannot.

Unbalance and misalignment both produce 1X energy. They differ in phase behaviour: unbalance typically shows a consistent phase relationship across the machine, while misalignment characteristically shows an axial phase difference across the coupling. A bent shaft and an unbalanced rotor look similar in a spectrum and differ clearly in phase.

Phase also identifies structural problems. Comparing phase between the machine foot and the baseplate, or across a joint, reveals relative movement that indicates looseness or a cracked structure. If two points that should move together are moving out of phase, they are not properly connected.

Taking phase requires a reference and a little more setup than an overall reading. It repays the effort on any machine where the diagnosis is genuinely uncertain, which is most of the interesting ones.

The conversion that quietly moves machines between severity zones. Confirm the reading and the limit are on the same basis before comparing them. Source: MIMAH engineering analysis.

The Change Criterion: The Part Most People Skip

Here is the single most useful idea in the whole subject, and it is written into the standards rather than being a matter of opinion.

Absolute vibration limits are a coarse severity scale derived from broad fleet experience. They tell you roughly where a machine sits relative to machines in general. They cannot tell you whether a particular machine has changed, because they know nothing about what it read when it was healthy.

The change criterion addresses that. A significant increase from an established baseline warrants investigation even when the absolute level remains within an acceptable zone. A machine that has always run at 1.2 mm/s and now reads 2.6 mm/s has more than doubled, and it needs looking at, despite 2.6 mm/s being unremarkable in absolute terms. Conversely, a machine that has read 4.5 mm/s for six years and still reads 4.5 mm/s is stable, and stability is informative.

This is why the baseline matters so much, and why it has to be established under repeatable conditions: same measurement points, same mounting, same load, same speed, same temperature state. A baseline taken on a cold machine at half load is not comparable with a reading taken hot at full load, and the difference between them will be interpreted as a fault by somebody eventually.

ISO 13373-1 sets out how to make these measurements consistently, covering transducer selection, mounting, location and operating conditions. ISO 13373-2 covers the processing and presentation of the resulting data. Between them they describe the discipline that makes readings comparable, which is the entire foundation of the change criterion.

How Transducer Mounting Changes the Answer

The way an accelerometer is attached sets the upper frequency limit of everything it measures, and this is not a small effect.

A stud-mounted accelerometer on a properly prepared surface is usable across the full frequency range of the sensor. An adhesive mount is somewhat lower. A magnetic mount is lower again, commonly usable to a few kilohertz. A hand-held probe with a stinger tip is the worst, often unreliable above a few hundred hertz, and it is also the least repeatable in position and pressure.

The consequence: a route walked with a hand-held probe cannot detect rolling element bearing faults reliably, because the frequencies where those faults appear are above what the mounting can transmit. The instrument reports a number, the number looks fine, and the bearing fails anyway.

If bearing condition matters, the mounting has to support it. This is usually the cheapest single improvement available to a struggling monitoring programme.

Common Misreadings

Comparing peak against an RMS limit. Inflates the reading by around 40 per cent and moves machines into zones they do not belong in.

Using the standard frequency band on a slow machine. Excludes running speed entirely and reports a reassuring number for a machine with a real problem.

Reading velocity and expecting to find bearing faults. The energy is at frequencies velocity does not weight. Use acceleration or a dedicated high-frequency technique.

Trending readings taken at inconsistent points or conditions. Produces scatter that gets interpreted as machine behaviour. Most apparent variability in vibration trends is measurement variability.

Treating a limit table as a verdict. The limits are a severity scale, not an acceptance specification, and the standards say so explicitly.

Assuming the nameplate speed. Spectra are read in multiples of running speed, and running speed on a loaded induction motor is not the synchronous speed on the plate. Getting this wrong shifts every identification along by a few per cent, which is enough to mislabel components.

How the transducer is attached sets the ceiling on everything it can measure. A route walked with a hand-held probe cannot reliably find bearing faults. Source: ISO 13373-1:2002; MIMAH engineering practice.

Frequently Asked Questions

What does 4.5 mm/s actually mean? Almost certainly 4.5 millimetres per second RMS, measured in the 10 Hz to 1,000 Hz band on a bearing housing. Whether it is acceptable depends on the machine's size, its support structure and its own history. On its own the figure describes energy, not health.

Should I measure in velocity or acceleration? Velocity for general machine condition, unbalance, misalignment and looseness. Acceleration for rolling element bearings and gears. Displacement for shaft measurement in fluid-film bearings and for slow-turning machines. Most complete programmes use more than one.

Why do my readings vary so much between visits? Usually measurement variability rather than machine variability. Check that the point, mounting, load, speed and temperature state are the same each time. Inconsistent measurement produces scatter that looks like machine behaviour and destroys the value of the trend.

Is a rising trend always a problem? Not always, but it is always worth investigating. A rise following a change in operating conditions, load or ambient temperature may be entirely explicable. A rise with nothing else changed is the earliest warning most plants ever get.

The Number Is the Beginning of the Question

Vibration measurement is not difficult and the instruments available today are genuinely good. What separates plants that get value from it from plants that generate paperwork is whether anybody understands what the readings describe: which quantity, which convention, which band, taken how, compared against what.

The plants that catch problems early are rarely the ones with the most expensive equipment. They are the ones where somebody established a proper baseline, kept the measurement conditions consistent, and treated a doubling from that baseline as worth investigating even when the absolute number looked fine. Across four decades of rotating equipment work in Sudan, Nigeria and Egypt, that habit has produced more early catches than any single piece of hardware.

Setting up monitoring, or trying to make sense of readings you already have? Talk to our engineering team. We will check the measurement basis, establish defensible baselines, and tell you what your existing data is actually saying.