Bearing Diagnostics from Vibration: Signs, Stages, and What to Do

The pump starts humming, the bearing housing temperature climbs ten degrees, and the vibrometer stubbornly shows a decent 3.5 mm/s. That's how it goes. A rolling-bearing defect lives at frequencies that a standard vibration velocity RMS reading (root mean square — essentially the overall vibration level) barely sees. Below, we break down what signs appear at each stage, exactly what to look at on the instrument, and how a bearing defect differs from looseness, rubbing, an electrical cause, and poor lubrication.

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

A Bearing Speaks in Impacts, Not in a Sine Wave

A healthy rolling bearing vibrates quietly and broadband. The rolling elements roll over smooth raceways, the lubricant holds its film, and the spectrum stays flat. A defect changes the picture qualitatively, not just quantitatively. The moment a dent, a spall, or flaking appears on a raceway, every rolling element produces a short impact as it passes over the defect.

The impact lasts tens of microseconds. It carries almost no low-frequency energy, but it excites everything nearby that's capable of ringing: the bearing races, the bearing housing wall, the accelerometer's own body. From this comes the key practical consequence. An early defect lives in the range from 2 to 20 kHz and up, while the familiar vibration velocity RMS reading works in the 10–1000 Hz band. One range barely sees the other.

A second consequence: the impacts are periodic. The period is set by the bearing's geometry and speed, and it's exactly this periodicity that turns into a diagnostic signature. From there, the only question is what instrument you use to pull it out.

Your instrument's measurement band matters more than its brand. Check the datasheet for what band it calculates vibration velocity RMS in, and whether it has a separate high-frequency acceleration channel. Without that, you don't actually know what your reading is capable of seeing.

Why a Bearing Defect Isn't Fixed by Balancing

Balancing reduces one specific force: the centrifugal force from an unbalanced mass. That force acts exactly once per revolution, so in the spectrum it sits at the 1x frequency, and that's the only thing that gets reduced. A raceway defect produces impacts at its own frequencies, which aren't multiples of the running speed. There's physically nothing a correction mass can do to remove them.

There is a reverse relationship worth knowing, though. Unbalance loads the bearing with an extra rotating force and shortens its life, so balancing is useful as prevention. But it doesn't restore a raceway that already has a spall on it.

There's a practical trap here too. A worn bearing with excessive clearance behaves like looseness inside the assembly: the 1x phase jumps around from reading to reading, the influence coefficient comes out unreliable, and the balancing result is unstable. If the phase doesn't repeat, deal with the mechanics — don't increase the trial weight. We have a separate article going into more detail on the relationship between overall vibration, 1x, and phase.

• Balancing reduces 1x. It doesn't touch bearing frequencies, the comb of harmonics, or the noise floor.
• Replacing a bearing without fixing the cause is pointless: the new one will die the same way. Look for misalignment, looseness, unbalance, overload, the lubrication regime.
• The "within tolerance" message in the software means only one thing: residual 1x is below the target you entered yourself. It says nothing at all about the bearing's condition.

Four Stages: From Ultrasound to Audible Noise

1. 01

   Onset. The vibration velocity spectrum is empty

   A subsurface micro-crack, the start of flaking. Impacts are weak and very short, with the energy going into roughly the 20–60 kHz range. The standard spectrum is clean, RMS is normal, and the crest factor has barely changed. This can be seen with an ultrasonic method or acoustic emission. You'll miss this stage with a route vibrometer, and that's normal.

2. 02

   Resonances wake up. The envelope earns its keep

   The defect has grown into a distinguishable spall. Impacts have gotten stronger and excite structural resonances of the assembly and the sensor, typically from 500 Hz to 2 kHz and above. The calculated defect frequencies and their harmonics show up in the envelope spectrum. The crest factor moves from the usual 3–4 up to 6–8, while vibration velocity RMS can still sit in zone B — "normal" on the A–D zone scale. This is the best point to make a decision: order the bearing, schedule the replacement, shorten the monitoring interval.

3. 03

   Visible in the standard spectrum. Audible to the ear

   The characteristic frequencies and their harmonics climb into the vibration velocity spectrum, with sidebands appearing around them — extra peaks flanking the main line — spaced at the running frequency. Audible noise shows up, housing temperature often rises, and the lubricant starts acting like an abrasive. From this stage to failure is usually weeks, not months.

4. 04

   Pre-failure state. A false improvement

   The defect has grown into distributed wear. The noise floor — the solid background in the spectrum — rises, discrete lines blur together, the crest factor falls back down to 3–4, while RMS and 1x climb: clearance has grown, and the rotor wanders. It's easy, and expensive, to get this one wrong. A dropped crest factor gets read as an improvement, when it's actually the last step before the cage fails. The machine gets shut down per site procedure.

Stage boundaries are approximate. A slow-turning assembly under heavy load moves through them differently than a motor at 3000 rpm. Rely on the trend at your specific point, not on absolute numbers from someone else's table.

Calculated Defect Frequencies: What's Behind the Four Letters

Every bearing element produces its own impact repetition frequency when damaged. You don't need to derive the formulas — the instrument or a database calculates them from the bearing's part number. What you need to understand is the meaning: how many times per second the defective surface meets whatever is striking it.

• These frequencies are calculated from the geometry: the number of rolling elements, element diameter, pitch diameter, contact angle, and shaft rotation frequency.
• The frequencies come out fractional and not multiples of the running speed. That's exactly what sets them apart from the 1x harmonics that looseness sprays around.
• The calculation is approximate. Rolling elements slip, so the actual frequency typically drifts from the calculated one by a few percent.
• Under axial load the contact angle changes, and the entire set of frequencies changes with it. You can't carry the formula for a radially loaded bearing over to a thrust bearing.

Don't force-fit an observed peak to the calculation. If a line lands close to BPFO but happens to be an exact multiple of the running speed, you're almost certainly looking at a harmonic of looseness or misalignment, not a bearing.

Sources: ISO 13373-3:2015

What to Look at Besides the Vibration Velocity Spectrum

Vibration velocity RMS in the 10–1000 Hz band answers the question "how hard is the machine as a whole shaking." It answers the question "what's wrong with the bearing" late. The toolkit here is different, and almost all of it lives in acceleration.

• Vibration acceleration, m/s². Work in acceleration, not velocity. Integrating into velocity suppresses exactly the top of the range you need.
• The envelope spectrum, also called enveloping. A bandpass filter isolates the high-frequency resonant zone, usually somewhere within 2–10 kHz. The instrument then builds the envelope — the contour the amplitude pulses along — and its spectrum (FFT) shows the impact periodicity. This is the primary method for early diagnosis.
• Crest factor, also called peak factor: the ratio of the peak value to the RMS of the same signal. About 1.4 for a pure sine wave, around 3 for noise, 6 and above for an impulsive process. It rises at the second stage and falls at the fourth.
• Eyeball the time waveform. Look for individual impulses, their period, their repeatability. Record at least 4–6 shaft revolutions: at 1500 rpm that's a fraction of a second; on a 60 rpm roll, it's already seconds.
• Narrowband RMS in the bearing zone, say 500–5000 Hz. A localized defect raises it several-fold, while it shifts the overall level by only a few percent.
• Housing temperature, noise, lubrication condition. Data of different kinds corroborates itself better than three parameters pulled from the same spectrum.

Sensor mounting decides everything. A magnet on a flat pad honestly works up to about 1.5–2 kHz; beyond that, the mount's own resonance creeps into the reading. Enveloping needs a stud or adhesive. The classic mistake looks like this: a "defect" at 3 kHz, picked up with a magnet, vanishes after switching to a stud.

Looks Like a Bearing, But It Isn't

Impulsiveness and an elevated high-frequency background don't only come from a raceway defect. Before ordering a bearing, rule out the neighboring causes. Each one can be checked quickly, without disassembly.

Until you've ruled out these causes, "bearing" stays a hypothesis, not a diagnosis. We cover situations where vibration is high but there's no unbalance in detail in the piece on why balancing doesn't help.

Plain Bearings: Oil Whirl and Oil Whip

A plain bearing has no rolling elements, no BPFO and no BPFI. The physics there is different: the shaft floats on an oil wedge that it creates itself by rotating, and there's no direct metal-to-metal contact during normal operation. The fault signs are different too.

Oil whirl looks like a steady subsynchronous line at roughly 0.42–0.48 of the running frequency. It tracks speed: raise the rpm, and the line moves up proportionally. The causes are usually operational, not mechanical: an underloaded bearing, excess clearance, the wrong oil viscosity, low pressure.

Oil whip is more dangerous. When, during a run-up, the whirl frequency reaches the rotor's first critical speed, the whirl locks onto the resonance: the frequency stops rising with speed and pins itself at the critical speed. This is self-excited vibration with large amplitudes, capable of destroying the machine quickly.

The practical takeaway is simple. This isn't unbalance. Unbalance sits exactly at 1x, and a subsynchronous line below the running frequency won't go away with any correction weight. Attempting to balance here just burns a shift and leaves the machine in the same state.

• Determine where the line sits: exactly at 1x, or below the running frequency. This is the first and cheapest test.
• Check whether the frequency tracks the speed. If it does, it's whirl. If it freezes at one value as the speed keeps rising, it's already whip, and that's a reason to act immediately.
• Check the oil parameters: pressure, temperature, viscosity. Then the actual load on the bearing and the clearance.
• Don't confuse this with partial rub: partial rub also produces subharmonics. They're distinguished by the shaft center's orbit, not by the spectrum from a housing-mounted sensor.

A housing-mounted accelerometer only helps so much here. Proper work with plain bearings requires eddy-current probes on the shaft, orbit plots, and monitoring the shaft center's position. See a steady subsynchronous line — the honest move is to bring in a specialist with that equipment, not to balance the rotor.

In Practice: Trending, Setpoints, When to Replace

One measurement shows a condition. A decision is made from two or more. It's the trend that tells "this machine has always been like this" apart from "it doubled over six weeks."

Set setpoints — warning and alarm thresholds — on three levels at once. The overall vibration velocity RMS level in the 10–1000 Hz band catches the machine as a whole and rests on zones A, B, C, D per the applicable part of ISO 20816. A narrowband level in the bearing zone catches a localized defect months earlier. Component setpoints on specific lines, including the calculated defect frequencies, catch it with precision. A working guideline for narrowband and component thresholds: warning at twice the baseline, alarm at four times it. The baseline is taken from a healthy machine at the same condition.

Tie the interval to how fast things develop. For rolling bearings, months usually pass between the first sign and failure, so set the measurement interval to no more than half that time. On a critical assembly, once second-stage signs appear, shorten the interval to weeks.

Comparability of the data matters more than its quantity. Same point, same direction, same mount, same Fmax (the spectrum's upper frequency limit), same number of lines, same condition and load. Change a setting without logging it, and years of trend data turn into a pile of incomparable numbers.

• Second-stage signs, the trend rising: order the bearing, put the replacement on the schedule, shorten the monitoring interval.
• Third stage, characteristic frequencies with sidebands already in the vibration velocity spectrum, temperature rising: replace at the next shutdown window.
• Fourth stage, a raised noise floor, blurred lines, a dropped crest factor, audible noise and heat: shut down per site procedure.
• Signs are present, but the trend has held steady for months: watch more closely and don't rush to disassemble. Not every defect found develops fast.

Being honest about service life. Vibration lets you assess condition and urgency, but not remaining life in hours. An ISO 281 rated life calculation is a design figure for selecting a bearing under given loads, not a forecast for an assembly where a defect has already appeared. Any phrase like "X weeks left" is a planning scenario assuming an unchanged rate of progression, not a guarantee. Check the applicable part and edition of the standard for your specific machine.

Sources: ISO 281:2007 · ISO 20816-1:2016 · ISO 13373-3:2015

What AXILINE Brings, and Where the Method's Limits Are

First, about the limits, so you don't build up false expectations. The Balanset-1A measures vibration velocity RMS in the 5–200 Hz band, shows the FFT spectrum, the time waveform, and 1x amplitude and phase simultaneously on two channels, and logs everything to an archive. That's enough to answer questions like "is this unbalance or not," "which housing is worse," "what happens on coast-down," and "did things improve after the work." It doesn't have an envelope spectrum or a high-frequency channel for early bearing diagnosis, and we don't pretend it does.

So here's how we work. We come to your site, mount sensors on your bearing housings, compare overall vibration against 1x, take spectra at both housings, and tell you straight. If 1x dominates and the mechanics are sound: we balance the rotor on site, in its own bearing housings, with no removal. If the picture is different: you'll hear that before paying for a balancing job that wouldn't have helped, and you'll get a clear list of next steps, including cases where you need an envelope analyzer or eddy-current shaft probes.

The instrument can be bought and used on your own. The kit includes two accelerometers, a laser phase and speed sensor, a two-channel USB module with preamplifiers, integrators, and an ADC, and Windows software: balancing in one and two planes, measuring speed, amplitude, and phase, overall vibration and 1x, FFT, a polar diagram, splitting weight across fixed positions and calculating drilling, working from saved influence coefficients, trim balancing, arbor eccentricity calculation, tolerance calculation against balance quality grades G, and an archive for reports.

AXILINE's engineers design and manufacture Balanset instruments and do their own on-site balancing with them. Consulting support is included: not sure how to read a spectrum — ask, and we'll work through it together.

Sources: Balanset-1A operation manual · Balanset-1A manufacturer specification · ISO 20816-1:2016 · ISO 21940-11:2016

Frequently asked questions

Related content