Vibration
Reading the Signature: Five Bearing Faults You Can Catch Early
Every bearing defect has a frequency. Learn the four defect frequencies, the four stages of failure, and how much warning each one gives you.
A rolling element bearing almost never fails without warning. It fails after months of increasingly obvious complaint, in a language most plants are not listening to. The useful thing about that language is that it is arithmetic: a defect in a specific part of a bearing produces impacts at a specific, calculable rate.
Why bearings have frequencies
Picture a bearing running with a small spall — a pit — on the outer race. Every time a rolling element passes over that pit, it drops into it and strikes the far edge. That impact happens at a rate determined entirely by geometry: how many rolling elements there are, their diameter, the pitch diameter of the bearing, the contact angle, and shaft speed.
Because it is geometry, it is predictable. And because it is predictable, an analyst who sees energy at that exact frequency knows which surface is damaged — not merely that "the bearing is bad".
The four defect frequencies
With N rolling elements, ball diameter Bd, pitch diameter Pd, contact angle φ and shaft speed S:
| Frequency | Defect location | Formula |
|---|---|---|
| BPFO Ball pass frequency, outer |
Outer race | (N/2) × (1 − (Bd/Pd)·cos φ) × S |
| BPFI Ball pass frequency, inner |
Inner race | (N/2) × (1 + (Bd/Pd)·cos φ) × S |
| BSF Ball spin frequency |
Rolling element | (Pd/2Bd) × (1 − ((Bd/Pd)·cos φ)2) × S |
| FTF Fundamental train frequency |
Cage | ½ × (1 − (Bd/Pd)·cos φ) × S |
You do not have to calculate these by hand — bearing manufacturers publish them, and analysis software holds databases by part number. But two things are worth internalising.
First, a rough sanity check. For many common bearings, BPFO lands near 0.4 × N × shaft speed and BPFI near 0.6 × N × shaft speed. If you know the shaft is turning at 25 Hz and the bearing has 9 rolling elements, expect outer-race energy somewhere around 90 Hz. That is enough to know roughly where to look.
Second, none of these are integer multiples of shaft speed. That is diagnostically valuable. Unbalance sits at exactly 1× running speed; misalignment typically at 2×; looseness produces a run of integer harmonics. Bearing frequencies are deliberately non-integer, so they cannot be confused with any of those.
The five faults, and what each looks like
1. Outer race defect (BPFO)
The most common defect you will find, partly because the outer race is usually stationary and takes the load in one place. Energy appears at BPFO with harmonics. Because the defect sits in a fixed position relative to the load zone, the impacts are of consistent amplitude — so you typically see no sidebands.
2. Inner race defect (BPFI)
The inner race rotates with the shaft, so the defect moves in and out of the load zone once per revolution. Impacts are therefore strong in the load zone and weak outside it — the signal is amplitude-modulated at running speed. The signature: energy at BPFI, with sidebands spaced at 1× running speed. Those sidebands are the giveaway.
3. Rolling element defect (BSF)
A spall on a ball or roller strikes both races as it rotates, so energy often appears at 2× BSF as well as BSF itself. The defective element is carried around by the cage, so you commonly see sidebands spaced at FTF.
4. Cage defect (FTF)
Least common as a primary failure, and usually a consequence of something else — severe lubrication failure, or a bearing that has already been damaged elsewhere. FTF is a low frequency, below shaft speed, and rarely appears alone. It more often shows as sidebands around other frequencies. A cage frequency in a spectrum is a late-stage warning, not an early one.
5. Lubrication failure
Not strictly a defect frequency, but the fault that causes most of the others and the one with the longest warning. Inadequate lubrication produces metal-to-metal contact, which generates high-frequency noise long before any race is measurably damaged. It shows up in ultrasonic and high-frequency measurements — not in a standard velocity spectrum, which is exactly why plants miss it.
Catching this one is the highest-value thing a condition monitoring programme does, because at this stage the fix is grease, not a bearing.
The four stages of bearing failure
Bearings degrade in a well-documented sequence. Knowing which stage you are in tells you how much time you have.
| Stage | What is measurable | Roughly how much life remains |
|---|---|---|
| Stage 1 | Ultrasonic and very high frequency energy only. Nothing visible in a normal velocity spectrum. Bearing sounds and feels fine. | The largest share of remaining life. Often still correctable by lubrication. |
| Stage 2 | Bearing natural frequencies begin to ring — typically in the region of 500–2,000 Hz. Sidebands start to develop around them. | Substantial. Plan the replacement; no need to rush. |
| Stage 3 | Defect frequencies and their harmonics appear clearly in the velocity spectrum, with sidebands. Damage may now be visible to the eye on strip-down. | Limited. Schedule replacement at the next available window. |
| Stage 4 | Broadband noise rises and the discrete defect frequencies may disappear into it. 1× running speed increases. Audible noise, heat. | Very little. Failure is imminent. |
The stage 4 trap. Analysts new to this sometimes relax when clear defect frequencies vanish from a spectrum they were tracking. It usually means the opposite of improvement: the damage has spread from a discrete spall to general surface destruction, and the signal has smeared into broadband noise. A rising noise floor with a falling defect peak is an urgent finding.
Why envelope analysis matters
Early bearing impacts are small, sharp and high-frequency. Alongside them a large machine produces enormous low-frequency energy from unbalance and misalignment. In an ordinary spectrum the bearing signal is buried — not absent, buried.
Envelope analysis (also called demodulation, or acceleration enveloping) solves this by filtering out the low-frequency content, keeping the high-frequency band where the impacts live, and extracting the rate at which those impacts repeat. The result is a spectrum in which bearing defect frequencies stand out plainly.
The practical implication: if your programme only looks at velocity spectra, you are systematically finding bearing faults late. Not never — late. Usually stage 3 rather than stage 1 or 2, which is the difference between planning a replacement and reacting to one.
What to do with all this
You do not need to become an analyst to benefit from it. The practical minimum:
- Know your bearing part numbers. Without them, defect frequencies cannot be calculated and every diagnosis is guesswork. This is the single most common gap we find.
- Measure the same points, the same way, at the same intervals. Trend matters more than absolute value. A reading in isolation tells you far less than six readings over six months.
- Include a high-frequency or enveloped measurement. Otherwise stages 1 and 2 pass unnoticed.
- Record shaft speed with every reading. Defect frequencies scale with it; a reading without speed cannot be properly interpreted.
- Act on stage 2 findings. The value of the programme is entirely in what you do between detection and failure.
The point
The reason bearing failures feel sudden is not that bearings fail suddenly. It is that the early stages are inaudible, invisible, and only detectable with instruments pointed at the right frequency band.
A bearing that destroys itself on a Tuesday afternoon was very likely telling you about it in March.