How to Measure Vibration Correctly: Point, Mounting, Condition

The fan starts humming after cleaning, you press a vibrometer against the guard and see 14 mm/s. An hour later someone else measures on the bearing housing and gets 2.8 mm/s. Both readings are "real," and both are useless, because nobody recorded the point, the direction, the condition, or the speed. Let's break down what a trustworthy measurement is built from, and where it falls apart most often.

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

What You're Measuring: Velocity, Acceleration, or Displacement

The instrument can describe the same housing motion with three parameters. Vibration displacement in µm, vibration velocity in mm/s, vibration acceleration in m/s² or g. The relationship is simple: v = ω·d, a = ω²·d, where ω = 2π·f. The formulas show the key point. Displacement emphasizes low frequencies, acceleration emphasizes high ones, velocity sits right in the middle.

That's why vibration velocity is used for the overall assessment of a machine's condition. One number in mm/s RMS (root mean square) in the 10–1000 Hz band weighs the running-speed component (vibration at the rotor's rotation frequency, denoted 1x), its first harmonics, and moderate high-frequency energy more or less evenly. Zones A, B, C, D in ISO 20816 are built on this parameter, it's the one you compare against the alarm setpoint (the threshold above which you're supposed to act), and it's usually the value a contract's tolerance is written against.

Take acceleration when you're looking for impacts and high frequencies: rolling-bearing defects, gear meshing, rubs. An early-stage defect looks like this: vibration velocity RMS is still within normal range, while the vibration acceleration peak has doubled over a month. Velocity averages out and hides such short impulses; acceleration shows them.

Leave vibration displacement for slow-turning machines and shafts: spindle runout, clearances, shaft position in a plain bearing from proximity probes. At 3000 rpm, one micrometer of displacement gives a negligible velocity, so assessing a high-speed machine by displacement makes no sense.

RMS, peak, and peak-to-peak are three different numbers for the same signal. For a pure sine wave, peak = 1.41·RMS, and peak-to-peak = 2·peak: 4.5 mm/s RMS is 6.4 mm/s peak and 12.8 mm/s p-p. Compare a new reading's peak against a historical RMS value, and you'll get "a 40% increase" where nothing actually changed. Check your own instrument's band too. The Balanset-1A calculates vibration velocity RMS in the 5–200 Hz range: enough for the running-speed component and first harmonics on most industrial machines, but not the full 10–1000 Hz band. If contractual acceptance specifically requires 10–1000 Hz, nail down the requirement and the instrument in advance, not on handover day.

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

Where to Mount the Sensor: The Bearing Housing, Not the Guard

The centrifugal force from unbalance travels through the bearing into the housing, from the housing into the frame, and on into the foundation. So you need to measure where that force actually passes: on the bearing housing, as close to the bearing as possible, in a loaded zone of the casting. Anything further down that path no longer shows you the machine — it shows how the structure passed the vibration through and reshaped it.

A guard, a protective mesh, a thin cover, a sheet-metal panel, a pipe — these are separate vibrating systems with their own natural frequencies. Press a sensor against a fan's guard and you'll easily get 12 mm/s where the housing reads 2.5 mm/s. The level didn't rise because the machine got worse — it rose because the panel hit resonance at the rotation frequency. The reverse happens too: a soft panel damps the signal, you see a decent-looking number, and you miss a real defect.

On a "motor plus driven machine" unit there are usually four bearing housings, and you can't afford to mix them up. Agree on labels once and keep the same route every time: the motor housing on the cooling-fan side, the motor housing on the coupling side, the machine housing on the coupling side, the machine housing on the impeller side. A reading logged against the wrong housing ruins the trend (the history of readings at that point) permanently, and this usually only surfaces a year later.

• The point is on the bearing housing or its massive cast section — not on the guard, not on a fence, not on the terminal-box cover.
• As close to the bearing as possible: the longer the signal's path through the metal, the more the structure distorts it.
• In a loaded zone of the housing, not on a thin rib or a thin casting wing.
• The point is marked: by a center-punch, paint, a sticker, or a welded pad for a stud. Next time the sensor goes in the same spot.
• Every housing has its own label, and the route doesn't change from one round to the next.
• The point is safe: rotating parts are out of reach, and the housing doesn't burn your hand.

Never reach the sensor, or your hand, through a guard. Skip a point you can't reach and record why, then arrange access separately. A reading taken at the risk of injury won't be repeatable anyway: you physically can't place the sensor the same way a second time.

Three Directions and What Each One Tells You

At every housing you take three directions: horizontal-radial (H), vertical-radial (V), and axial (A). The axis of maximum sensitivity is marked on the sensor's body, and it must line up exactly with the measurement direction. Mount the sensor at a 30° angle, and the instrument will show you a projection — a low-balled number — and you'll conclude the machine has calmed down.

The centrifugal force from unbalance acts radially, so unbalance lives in H and V. The horizontal direction usually reads higher: the foundation's horizontal stiffness is lower, so the housing there deforms more. For balancing that's an advantage — higher sensitivity. That's exactly why sensors are more often mounted horizontally, even though there's no fundamental difference.

The axial direction answers a different question. A pronounced running-speed component in the axial direction, plus a roughly 180° phase shift across the coupling, means misalignment, and balancing won't remove it: you need shaft alignment. But even here there's a caveat. Strong couple unbalance also produces axial vibration by "twisting" the housing structure. One direction alone doesn't make a diagnosis.

A phase difference of about 90° between H and V at the same housing is the typical picture of unbalance. It's a good sign, but a lone one. How to build the full picture from the spectrum, phase, and the share of 1x is covered separately in the articles on reading a vibration spectrum and on overall vibration, 1x, and phase.

Sources: ISO 13373-3:2015

Mounting the Sensor: Stud, Adhesive, Magnet, Probe

The sensor, the mount, and the surface form a mechanical chain with its own resonant frequency. Above that frequency you can't trust the reading: the mount starts ringing on its own and mixes its own peaks into the spectrum (the picture of how vibration is distributed across frequencies). Most often it's not the sensor but the mounting method that limits the usable frequency range.

A textbook example every vibration analyst knows. You take a bearing envelope reading (an analysis method that pulls out short impulses from bearing defects) with a magnet, a clean-looking peak appears around 1.8 kHz in the 2–8 kHz band, and it gets read as an outer-race defect. Remount the sensor on a stud, and the peak vanishes. It was the magnetic mount's natural frequency, landing right inside the analyzed band.

The probe deserves its own note. It's held by hand pressure, and the hand shakes, tires, and changes angle. An upper limit around 300–500 Hz means that at 1450 rpm you'll still see the running-speed component and the second harmonic, but the tenth harmonic will already be past the trustworthy range, and bearing frequencies won't show up at all. A bearing defect isn't visible with a probe, and that doesn't mean "the bearing is fine" — it means "the instrument isn't looking there." Plus low repeatability: the pressure, angle, and point all change every time, the spread easily runs past 30%, and a real rise in level drowns in the method's own scatter.

The Balanset-1A comes with two standard single-axis accelerometers up to 25×25×20 mm and up to 40 g, with both mounting options provided: magnet or M4 threaded stud. For balancing in the machine's own bearing housings, where you're working with the running-speed component and first harmonics, a magnet on a prepared pad is enough. For high-frequency diagnostics, use the stud. If a point is needed permanently, prepare it once at the next shutdown: clean the spot, glue or weld on a pad, and tap a hole for the stud.

These limits are a working guideline, not a standard. There's no universal table: take the actual usable band from the documentation for your specific sensor and mounting method. And the golden rule of trending: don't change the mounting method at an existing point. Spectra taken with a magnet and with a stud differ, and you'll read that difference as a change in the machine.

The Pad, the Sensitivity Axis, and the Cable

The mounting pad matters more than it seems. A magnet on a layer of paint gets a springy cushion between the sensor and the metal, and the mount's resonance drops to around 1 kHz. A magnet on a convex surface rests on two points instead of a flat plane and rocks. A magnet on swarf, rust, or oily grime holds worse and can come loose while the machine is running.

The cable is part of the measurement too. A freely hanging cable whips along with the machine and generates triboelectric noise — a stray charge from the shield flexing. For charge-output sensors this is critical; for IEPE sensors with built-in electronics the effect is smaller, but not zero. Tie the cable down to the housing so the connector has strain relief, and don't run it parallel to power cables.

• The pad is cleaned down to bare metal: paint, filler, rust, and grime removed.
• The pad is flat. A flat magnet needs a flat surface; for a cylinder, use a two-pole magnet or machine a flat spot.
• The sensor's sensitivity axis is exactly along the measurement direction, and the sensor isn't tilted.
• Bring the magnet in by rolling it into place, not by dropping it. An impact overloads the input and causes "ringing" in the first seconds of the recording.
• The cable is secured, has strain relief at the connector, and doesn't run alongside power cables.
• Connectors are clean and fully latched.

An implausible number almost always means a problem in the measurement chain, not the machine. 250 mm/s on a pump housing instead of the usual 3 mm/s means a broken cable, a magnet that's come off, or an overloaded input. Check the chain and remeasure before raising an alarm.

The Tachometer and the Phase Marker: Why and How to Set Them Up

A tachometer serves two purposes. First: an exact rotation frequency. Without it you can't find the running-speed component in the spectrum or tell it apart from mains interference. On a two-pole motor at 2970 rpm that's 49.5 Hz, while mains interference sits right next door at 50 Hz. Without a speed reading, you're just guessing which line belongs to what.

Second, and the one that matters for balancing: the reference phase. The instrument synchronizes the vibration signal with the pulse from the marker, extracts exactly the running-speed component from the overall vibration, and measures its phase: the angle that shows where the rotor's "heavy spot" sits relative to the marker. The Balanset-1A has a phase measurement error of ±1° and a speed range of 100–100,000 rpm. Without a pulse from the marker there's no phase and no weight calculation — only the overall vibration level is left.

1. 1

   Stick on one marker

   Stick the reflective strip directly onto the rotor or shaft, on a clean, degreased spot. One marker per revolution, exactly one. The instrument will read two markers as double the speed, and the whole picture will drift along with the phase.

2. 2

   Get contrast by reflection

   The marker needs to stand out from the background by reflectivity, not color. On a polished, shiny shaft the tape works poorly: darken the area around it first, then stick on the marker.

3. 3

   Mount the laser sensor on a stand

   A magnetic stand is included in the kit. Mount it on a fixed, rigid surface — not on a vibrating guard, not on a fence. A shaky tachometer produces a jumping phase, and you'll end up looking for the cause in the rotor.

4. 4

   Set the angle and distance

   Aim the beam at the marker nearly perpendicular to the surface, and stay clear of mirror-like reflections from the housing. Check that the speed reading is stable and matches what you expect. Jumps to multiples of the expected value mean stray reflections or an accidental second marker.

5. 5

   Don't re-stick the marker between runs

   The marker's position is the phase reference point. Move the marker, and the phase changes even though the rotor is the same. There's a separate article on where to reference the weight-mounting angle from afterward.

Without a tachometer you get only the overall vibration. That's enough to compare the level against a setpoint, but not enough for balancing, or for working out how much of the level is due to unbalance.

Sources: Balanset-1A operation manual

Measurement Condition and Repeatability

A number without its condition is meaningless. Vibration depends on speed, load, damper or valve position, machine temperature, and what else is running nearby. A reading at 60% load and a reading at rated load are two different machines — you can't compare them against each other.

Speed is especially sensitive near resonance. In a resonance zone, a change in rotation frequency of just 100 rpm can raise vibration tenfold and flip the phase by 180°. On a VFD-driven machine, record the frequency setpoint as a number, not as "about the usual."

Temperature works more slowly, but just as reliably. A cold machine has different clearances and different actual shaft alignment than a warmed-up one: thermal growth shifts the feet. Let the unit settle into steady-state operation and record how long it ran before the measurement. On fans, also note the impeller's condition: dust buildup changes both the level and the phase within weeks, so the measurement date here isn't a formality.

• Speed measured with a tachometer, not taken from the nameplate.
• Load is the same as last time: for a pump, flow and pressure; for a fan, damper position; for a motor, current.
• Machine is warmed up, and the run time before the measurement is recorded.
• It's noted which neighboring units were running. A nearby machine transmits vibration through the foundation and through piping.
• Instrument settings are unchanged: parameter, amplitude type, band, Fmax (the spectrum's upper frequency), number of lines, averaging.

1. 1

   Take the reading three times

   Not three readings in a row from the same spot — three full measurements: remove the sensor, put it back, measure. This way you're testing the mount and the pad, not the instrument's electronics.

2. 2

   Work out the spread

   For a velocity round on a healthy machine, treat a spread within 10–15% as normal. More than that means a problem in the chain: the pad, the magnet, the cable, a tilted sensor, or an unstable operating condition.

3. 3

   Check phase stability

   For balancing, phase matters more than amplitude. If the 1x phase drifts by tens of degrees between runs at an unchanged condition, the reading can't be used. The cause is either in the measurement or in the machine: resonance, a loosened impeller fit, a crack.

4. 4

   Record a baseline level

   While the machine is healthy, take a baseline across all points and directions at a known condition. From then on you're comparing against this machine's own history, not against some table found online.

Trend matters more than the absolute number. 2.7 mm/s rising by 0.3 mm/s a month is more dangerous than a steady 4.2 mm/s the machine has held for three years running. That said, take the numeric zones from the applicable, current part of the standard for your specific machine type, power, and support type — not from a one-size-fits-all table.

Sources: ISO 13373-5:2020 · ISO 20816-1:2016

Common Mistakes and What to Log

Keep one table per machine and don't change its structure. If you had to change the point, the mount, or the settings, don't edit the old entries. Open a new baseline and note alongside it what changed and why: the old data stays in the archive and keeps its meaning.

If You Need a Measurement the Customer Will Trust

A measurement isn't an end in itself. It answers one practical question: repair, or balance. That's why on site we start not with weights but with measuring at the housings: overall vibration and the running-speed component in every direction, the spectrum, phase, repeatability. If 1x accounts for most of the level, we balance the rotor on site, in its own bearing housings, in one or two correction planes (rotor cross-sections where weights are mounted), and hand over a report with before-and-after numbers. If 1x is small, we say so directly: balancing won't help, and you need to deal with fasteners, shaft alignment, bearings, or resonance.

We design and manufacture Balanset instruments and do our own on-site balancing with them. If you want to measure on your own, the Balanset-1A kit covers the whole measurement chain: two accelerometers with magnet or M4 stud mounting, a laser phase sensor on a magnetic stand, a two-channel USB module with preamplifiers, integrators, and an ADC, digital scales for the trial weight, and Windows software. The instrument measures speed, vibration velocity amplitude and phase — both overall and 1x — shows the time waveform and the FFT spectrum, acquires both channels at once, stores readings and influence coefficients (the machine's remembered response to a trial weight, which speeds up repeat balancing), and files results into an archive for reports. The same system also serves as the measurement core of a soft-bearing balancing machine.

• Machine type, power, speed, support type, and bearing type.
• Overall level and 1x at each housing, with the point and direction noted.
• Sensor mounting method and the condition the numbers were taken at.
• What changed on the machine over the last month: cleaning, a blade replacement, a repair, realignment.
• Whether there's a measurement history and a baseline for the healthy condition.

With this set of data, working through the case takes minutes instead of a second site visit. Consulting support on measurement method and on the instrument is included.

Sources: Balanset-1A manufacturer specification

Frequently asked questions

Related content