Vibration analysis is the closest thing condition monitoring has to a stethoscope: it hears a defect months before the bearing can be seen, and it names the damaged component with surprising precision. The reason is that a rolling bearing is a mechanical calculator. Every raceway and every rolling element imposes its own rhythm on the vibration signature, and those rhythms can be worked out from geometry and running speed alone. Read the rhythms correctly and a spectrogram stops being noise and becomes a sentence: the inner ring of the drive-end bearing is spalling, and it has been getting worse for six weeks.
Why Each Component Has Its Own Rhythm
Consider a bearing with a rotating inner ring, a stationary outer ring and a loaded zone. The balls pass any fixed point on the outer raceway at one rate, they pass a point on the rotating inner raceway at a faster rate, each ball spins about its own axis at another rate, and the whole ball set advances at the slower cage rate. Those four rates - ball pass frequency outer (BPFO), ball pass frequency inner (BPFI), ball spin frequency (BSF) and the fundamental train or cage frequency (FTF) - are fixed multiples of shaft speed for a given bearing. Manufacturer tables and monitoring software compute the multipliers from the part number and the speed, so a 6205 running at 1,500 rpm has a 25 Hz shaft rate and a set of defect frequencies that scale directly from it.
The kinematic ranges are worth remembering because they make signatures easier to trust. For most deep-groove bearings, BPFO lands at roughly 3 to 6 times shaft speed, BPFI at roughly 4 to 8 times, BSF at roughly 1.5 to 3 times, and FTF at roughly 0.35 to 0.45 times; a large spherical roller such as a 22312 shifts these values with its many long rollers. BPFO is usually the richest in energy because the outer raceway is stationary and its defect passes through a roughly constant load, giving a steady train of impacts. BPFI is higher in frequency and is modulated by the rotating load zone, so it grows characteristic sidebands spaced at shaft speed. BSF is often seen near twice its fundamental because a defect on a ball contacts both raceways in one revolution, and FTF is low, weak and tied to the fraction of speed at which the ball set advances.
Reading the Signature of Each Damaged Part
A defect on a specific component modulates vibration at that component's frequency, so the peak family identifies the part, provided the analyst reads the pattern rather than a single line. Amplitude alone is a poor guide because the same energy looks different depending on load, sensor position and machine stiffness; the shape of the family is far more reliable.
Amplitude versus pattern
A single tall peak at a defect frequency can be a spur, a harmonic of something else, or a resonance, so the experienced analyst confirms a diagnosis from the whole family - the fundamental, its harmonics and the sidebands around them - before writing anything down.
| Frequency family | Damaged component | How the peak behaves |
|---|---|---|
| BPFO with harmonics | Outer raceway | Steady amplitude, broadly constant through the shaft revolution |
| BPFI with shaft-rate sidebands | Inner raceway | Sidebands widen and multiply as the fault grows |
| BSF, often near twice the fundamental | Rolling element | Weak and unstable; a defect contacts both raceways per turn |
| FTF or cage frequency | Cage or lubrication | Low and weak; a rise usually means lubrication, not fracture |
| Broadband floor with no distinct peaks | Generalised wear | Late stage: roughness has replaced discrete impacts |
| Sub-synchronous shaft harmonics | Not the bearing | Points to looseness or misalignment - check the mounting first |
Envelope Analysis and the Early-Defect Problem
An early defect produces tiny, high-frequency impacts that disappear into the noise floor of an ordinary velocity spectrum. Envelope analysis recovers them by demodulating a high-frequency band and examining the repetition rate of the modulation that the bearing defect imposes on it. This is why a bearing can be screaming in the envelope spectrum while looking innocent in the raw spectrum.
Demodulating a high-frequency band
The signal is band-pass filtered around a structural resonance of a few kilohertz that the impacts excite, then rectified and enveloped so the low-rate repetition becomes visible as a peak at BPFO, BPFI or BSF.
Why raw velocity hides early defects
Velocity integrates acceleration and therefore emphasises lower-frequency motion, so the faint high-frequency impacts of an incipient defect are buried. A trend of overall velocity will lag the truth by weeks, which is why the envelope is read alongside it.
Harmonics and sidebands as a severity gauge
As a fault develops, more harmonics of the fundamental appear and sidebands spread around them. The widening of that family, more than any single amplitude, tracks how far the damage has progressed.
Lubrication Signatures and Shaft-Current Harmonics
Not every bearing problem is a crack. Starved or over-thick grease changes the contact dynamics: cage-frequency energy rises, broadband high-frequency content climbs and temperature follows. This signature is reversible - re-lubricate correctly and the spectrum cleans up within hours - which is why the vibration and lubrication programmes should be one system rather than two departments. A rising cage and high-frequency energy should trigger a grease check before it triggers a work order for a bearing change. Electrical erosion from shaft currents leaves its own trace: distinctive harmonics spaced around the defect frequencies, plus the fluting geometry that appears on the raceway itself. A spectrum that shows those harmonics should trigger an electrical investigation, not simply a new bearing.
Setting Up Routes, Sensors and Speed Data
A programme that works starts small and disciplined. Tag every bearing position with its part number and speed, compute the defect frequencies once, and collect routes on a sensible period - monthly for most machinery, more often for critical or already-deteriorating positions. Four basics decide whether the data is usable.
- Sensor placement. Mount the accelerometer on the bearing housing as close to the load zone as the structure allows.
- Sampling rate. Use a rate high enough to cover the demodulation band, commonly tens of kilohertz, so the envelope has something to work with.
- Enough revolutions. Collect at least a few shaft revolutions so short-lived impacts are captured consistently.
- Speed with every reading. A defect-frequency trend plotted against the wrong speed produces nonsense, and forgetting to log speed is the most common error in new programmes.
Turning a Spectrum Reading into a Work Order
The last step is the one most programmes get wrong: converting a diagnosis into a scheduled action with a defined response rather than a discussion. The table below sets the mapping, and the habit is to apply it step by step so that a reading triggers a defined response rather than a debate.
| Damage stage | Where the evidence sits | What to schedule |
|---|---|---|
| Incipient defect | Visible only in the envelope spectrum | Plan a replacement at the next opportunity and verify the spare |
| Established discrete defect | Clear peaks in velocity, growing sidebands | Order the part, set a date, increase the monitoring frequency |
| Advanced damage | Harmonics spreading, overall level climbing | Schedule promptly; remaining life is weeks, not months |
| Cage or lubrication signature | Rising cage and high-frequency energy | Act on the grease first, then re-measure before ordering |
| Broadband noise with no peaks | Overall level with little structure | Treat as urgent; the discrete-defect window has passed |
Trends, Baselines and Common Analysis Errors
Two habits separate a programme that pays for itself from one that generates false alarms. The first is a baseline: record a clean reading on a new or freshly overhauled machine so every later number has a reference, and never compare a new route against a number taken under different conditions. The second is patience with the obvious - a defect frequency that appears but stays flat is a defect that is not growing, and the correct action may be to keep watching rather than to open the machine. Conversely, a broad rise with no distinct peaks is later-stage than a clean peak, not earlier. Alarm philosophy matters less than consistency, and consistency is what eventually lets a team calibrate its own judgement against its own machines.
- Baseline first. A reading on a new or freshly overhauled machine gives every later number a reference to be judged against.
- Trend, not snapshot. A single measurement says little; three routes at the same speed and load say a great deal.
- Do not over-react to a flat peak. A defect frequency that appears and stays flat is a watch item, not necessarily a work order.
- Log the speed every time. At 1,500 rpm the shaft rate is 25 Hz, and every defect frequency is a fixed multiple of it.
The stage model, the ISO 10816 severity zones - the familiar 1.4, 2.8, 4.5 and 7.1 mm/s boundaries - and the ISO 15243 vocabulary used to describe the outcome are discussed further in the failure analysis guide and the condition monitoring notes. Where the spectrum points at an electrical cause, the mechanism is covered in the insulated bearing guide. Deep groove and cylindrical duties are the most commonly monitored, so see deep groove ball bearings and cylindrical roller bearings, and sensorised packages are widely available from SKF. When a replacement is planned, geometry and suffix options can be pre-checked in the bearing selector, and the underlying concepts are collected in the technical library.
- Every component has a computable frequency. BPFO, BPFI, BSF and FTF each name a part and its rhythm.
- Read the family, not one peak. Harmonics and sidebands spread as damage grows, so the pattern is the trend.
- Envelope analysis hears early defects. Demodulating a high-frequency band exposes faults that raw velocity hides for weeks.
- Some signatures mean lubrication. Rising cage and high-frequency energy usually call for a grease check, not a bearing order.
- Consistency beats cleverness. Fixed routes, fixed sensors and logged speed make the trends trustworthy.
Frequently asked questions
How do I know which bearing component a frequency belongs to?
Match the measured frequency to the calculated defect frequencies for that bearing: BPFO for the outer raceway, BPFI for the inner raceway, BSF for a rolling element and FTF for the cage. The sideband pattern confirms it, because inner-ring defects grow shaft-rate sidebands while outer-ring defects stay steady.
Why does an early defect show in the envelope but not the velocity spectrum?
Early impacts are tiny and high in frequency, so they sit under the noise floor of a velocity reading. Envelope analysis demodulates a high-frequency band around a structural resonance and reveals the repetition rate of those impacts, which is why it detects incipient defects weeks before overall velocity moves.
What do sidebands around BPFI tell me?
They confirm the defect is on the rotating inner ring. The load zone passes over an inner-ring defect once per shaft revolution, so that fault is amplitude-modulated at shaft speed and grows sidebands spaced at shaft rate. A widening sideband family indicates the fault is developing.
Can a defect frequency appear without a real defect?
Yes. A defect frequency at a low, flat level with no growth may reflect a harmless geometric feature or a false trigger, which is why the trend matters more than any single reading. A peak that stays flat over several routes is usually a watch item, not a work order.
How often should I collect bearing vibration data?
Monthly suits most machinery and is a common starting cadence, with more frequent routes for critical or already-deteriorating positions. The fixed period matters more than its exact value, because comparable readings taken at a steady interval build the trend that a single measurement cannot.
Why does the spectrum change when the speed changes?
Because every defect frequency is a fixed multiple of shaft speed. When speed changes, every peak moves, and a trend plotted against the wrong speed is meaningless. Always record the running speed with each reading so the frequencies stay comparable route to route.