Bearing Failure, Stage by Stage: Why Ultrasound Hears It Before Vibration Feels It and Heat Shows It
A rolling element bearing fails when the thin film of lubricant between its moving surfaces breaks down and metal starts to wear against metal.
Almost every pump, fan, motor and compressor in a mechanical room turns on bearings, so a bearing fault rarely stays small. It takes the machine down with it, and often the zone, the loop or the redundancy that machine was protecting.
Most bearings that fail early are not worn out. Bearing maker SKF splits premature failures four ways: poor lubrication 36%, fatigue 34%, mounting and handling 16% and contamination 14% (ReliaMag summary of SKF data). Yet one of the costliest bearing failures hits a pump that has barely run. It is the standby unit, damaged while it sat still beside its running twin.
Key takeaways
- Bearings fail in four stages, and the warning signs arrive in a fixed order: airborne ultrasound first, then vibration analysis, then audible noise, and heat last.
- Ultrasound is judged against each bearing's own baseline. About 8 decibels (dB) up suggests lack of lubrication, 16 dB suggests damage, and 35 to 50 dB warns of catastrophic failure.
- Standby pumps can be damaged while idle by false brinelling, where vibration from nearby running units wears marks into the raceway at the rolling element spacing.
- Variable frequency drives (VFDs) can push electrical current through motor bearings, pitting the races until a washboard pattern called fluting forms and failure is close.
What is a rolling element bearing, and why does it fail?
A rolling element bearing lets a shaft turn with very little friction by carrying the load on balls or rollers instead of a sliding surface. Every one has the same four parts, and each part fails in its own way.
- Inner race: the ring pressed onto the shaft. It turns with the shaft, so a defect on it passes in and out of the load zone on every revolution.
- Outer race: the ring seated in the housing. It usually stays still, so a defect on it sits in one place and gets struck by every rolling element that passes.
- Rolling elements: the balls or rollers that carry the load between the two races.
- Cage: the separator that keeps the rolling elements evenly spaced. It is also called the retainer.
None of these surfaces is meant to touch. A film of oil or grease, thinner than the surface roughness of the metal, holds the rolling elements off the races. While that film is intact, a well mounted bearing can run for years.
Failure begins when the film is lost or the surface is overloaded. Poor lubrication starves the contact, contamination puts hard particles into it, and bad mounting loads the races unevenly. Fatigue is the natural end of life, when repeated stress cracks the metal below the surface and flakes break away.
That is why the SKF split matters. Lubrication, mounting and contamination together cause about two thirds of premature failures. All three are in the hands of the people who install and maintain the machine.
What are the four stages of bearing failure?
The four stages of bearing failure are an industry convention used by reliability engineers to describe how a defect grows and which instruments can see it. It is not a formal standard, but it explains why different techniques catch the same bearing at very different times.
- Stage 1: the defect exists only in the ultrasonic range. A normal vibration spectrum shows nothing, and the machine looks and sounds healthy.
- Stage 2: small impacts start to make the bearing components ring at their own natural frequencies. Vibration analysts can pick this up with high frequency methods.
- Stage 3: defect frequencies and the sidebands around them appear clearly in the vibration spectrum, and heat begins to rise.
- Stage 4: the discrete peaks dissolve into broad random noise, vibration at running speed rises, and the bearing becomes audible and hot.
Read top to bottom, the list gives the order of warning signs. Ultrasound comes first, then vibration, then audible noise, which acoustic listening picks up, and heat comes last. Each step later in the list means less time left to plan the repair.
The practical lesson is that the instrument decides the warning time. A team that relies on touch and temperature finds bearings in Stage 4. A team that trends ultrasound and vibration finds most of them in Stage 1 or 2, with weeks to order parts and schedule the outage.
This is the core argument for condition based maintenance on rotating equipment. The defect is present long before anyone can feel it, so the question is only which signal you are watching.
What do ultrasound decibel readings mean for a bearing?
An ultrasound decibel reading on a bearing only means something compared with that same bearing's own baseline. The absolute number depends on the machine, the speed, the load and where the probe sits, so the change is the finding.
UE Systems, a maker of ultrasound instruments, publishes alarm steps above baseline that many programs adopt (UE Systems):
- +8 dB: lack of lubrication, or the very start of a problem.
- +12 dB: the beginning of a failure mode.
- +16 dB: damage, an advanced failure condition.
- +35 to 50 dB: catastrophic failure is near.
Decibels are a logarithmic scale, so each step is a larger jump in sound energy than the numbers suggest. A bearing up 16 dB is far louder in the ultrasonic range than one up 8 dB. Neither may make a sound a person can hear.
Using ultrasound to grease a bearing
The +8 dB step doubles as a greasing signal. When a bearing reads about 8 dB over baseline and the sound quality has not changed, the usual cause is a thinning lubricant film (UE Systems).
The method is simple. Add grease a little at a time while listening, and stop as soon as the reading falls back to baseline. A reading that stays high after a careful shot of grease points to something other than lubrication, and that bearing needs a follow up test.
Stopping matters as much as starting. Over greasing churns the excess, which raises friction and heat, and the pressure can push grease past the seals. Calendar greasing, the same amount on the same date regardless of need, is one of the most common ways good bearings are harmed. Our preventive maintenance guide covers which tasks belong on the calendar and which should move to condition.
What are bearing defect frequencies, and how do they separate a bearing fault from other vibration?
Bearing defect frequencies are the rates at which a damaged spot on a bearing gets struck as the shaft turns. Each part has its own frequency, set by bearing geometry and shaft speed, so a peak at one of them names the damaged part.
- BPFO, ball pass frequency outer race: how often rolling elements pass a fixed point on the outer race.
- BPFI, ball pass frequency inner race: how often rolling elements pass a point on the inner race.
- BSF, ball spin frequency: how fast each rolling element spins on its own axis. It points to a damaged ball or roller.
- FTF, fundamental train frequency: how fast the cage turns. It points to cage wear and sits below running speed.
When exact bearing geometry is not to hand, analysts use a rule of thumb. BPFO is about 0.4 times the number of rolling elements times shaft speed, and BPFI is about 0.6 times the same product (Fabrico). Shaft speed here is in hertz (Hz), or revolutions per second.
A worked example
Take a pump motor running at 1,780 revolutions per minute (rpm). Divided by 60, that is a shaft speed of 29.7 Hz. If its bearing has 9 rolling elements, the estimates work out like this:
- BPFO: 0.4 × 9 × 29.7 Hz, or about 107 Hz.
- BPFI: 0.6 × 9 × 29.7 Hz, or about 160 Hz.
Neither number is a whole multiple of running speed. That is the tell. Imbalance shows up at exactly 1× running speed and misalignment at 2×, because those faults repeat once or twice per revolution. A bearing defect repeats at a rate set by the rolling elements, so its peaks fall between the running speed harmonics.
An inner race defect adds one more clue. The inner race turns with the shaft, so its defect moves in and out of the load zone every revolution. That makes the BPFI peak carry sidebands spaced at running speed on either side. Those sidebands are what Stage 3 refers to.
This is where vibration analysis earns its place: it does not just say the machine is rough, it says which part is failing. Our vibration guide covers the spectrum, sensor types and why the same measurement point matters.
Why is heat the last warning sign for a bearing?
Heat is the last warning sign because a bearing only runs hot once friction is substantial, and by then the surfaces are already damaged. In the four stage model, heat begins to rise in Stage 3 and becomes obvious in Stage 4.
Several things hide the early rise. The housing, shaft and grease carry the heat away. Load, speed and room temperature also move the reading more than a small defect does. A slightly warm pump bearing is easy to dismiss and hard to act on.
Electrical connections behave the opposite way. A loose or corroded lug has higher resistance, and resistance turns current into heat immediately, so thermal imaging is often the earliest warning on panels and switchgear. On a bearing, the same camera is a late confirmation.
Thermal still has a job on rotating equipment. The most useful bearing comparison is against a sibling: the same bearing on the identical pump next to it, under the same load in the same room. A housing clearly warmer than its twin deserves a closer look. That is why we argue for a reference in the same frame on every thermal reading.
The reverse is not true. A bearing that feels normal to the touch can still be in Stage 1 or 2, so a cool housing is not proof of a healthy bearing.
Why do standby pumps fail when you finally need them?
Standby pumps often fail at start up because their bearings were damaged by false brinelling while they sat idle next to running equipment. The bearing never turned, yet it wore.
Here is how it happens. A bearing that does not rotate never rebuilds its lubricant film. Vibration from the duty pump next door travels through the shared base, piping and floor. It makes the rolling elements rock in tiny arcs against the races.
Without a film, those small oscillations wear away the metal at each contact point. The marks end up spaced exactly at the rolling element pitch, one under each ball or roller. The name comes from true brinelling, which is denting from an overload; false brinelling looks similar but is caused by wear.
The damage grows with the severity of the vibration and the length of the idle time. It is often found only when the standby unit is started and runs noisy. A bearing marked this way may still spin smoothly by hand and pass a casual check, then fail early under load (ReliaMag).
Why this matters most in a data center
Data centers design cooling plants for N+1 redundancy, meaning one more pump than the load needs. That spare often sits idle on a shared base with the duty pumps, in exactly the conditions that cause false brinelling.
The countermeasure is standard practice: rotate duty and standby units on a schedule. Each run rebuilds the film and moves the rolling elements to new contact points. The same storage guidance recommends periodic shaft rotation for stored spares and keeping them away from running machines (ReliaMag).
Rotation only works if it is recorded. A run date on the pump tag or in the maintenance management system is the only proof that the spare has been exercised. Redundancy that has never been tested is closer to redundancy on paper than redundancy in practice.
Can a variable frequency drive (VFD) damage motor bearings?
Yes. A variable frequency drive can damage motor bearings by pushing electrical current through them, a problem that affects VFD driven motors of every size (EST AEGIS).
Modern drives use pulse width modulation (PWM), switching the output voltage on and off very fast to shape the motor's power. That switching creates a common mode voltage, which couples onto the motor shaft. The voltage builds until it discharges through the thin lubricant film to the frame, as a tiny spark, possibly tens of thousands of times per second.
Each spark melts a microscopic crater in the race. The process is called electrical discharge machining (EDM), the same principle used to cut metal on purpose. Over time it leaves a recognizable sequence:
- Pitting: thousands of microscopic pits across the race surface.
- Darkened grease: the lubricant is burned and degraded by the discharges.
- Frosting: the race takes on a dull, matte, frosted look.
- Fluting: a washboard pattern of ridges across the race. Once fluting develops, failure is close.
Larger motors add a second path. Above about 100 horsepower (hp), or 75 kilowatts (kW), shaft voltage can also drive a circulating current through both bearings, damaging them at once (EST AEGIS).
How bearing currents are controlled
There are two main defenses. A shaft grounding ring gives the current a low resistance path to ground around the bearing. An insulated bearing blocks the path through the bearing itself.
They are usually paired. Insulating one bearing without grounding the shaft can push more current through the other bearing. For motors above 100 hp, manufacturer guidance pairs a grounding ring at one end with an insulated bearing at the other (EST AEGIS). Smaller motors commonly use a grounding ring alone.
VFDs are everywhere in modern plants, on pumps, cooling tower fans and air handlers. Any motor on a drive that has needed repeat bearing replacements deserves a look for this cause before the next bearing goes in.
Where are the bearings in a mechanical room?
Every pump, fan, motor and compressor in a mechanical room turns on bearings. The most critical ones sit in the machines that move water and air. Each location has its own access, load and failure pattern.
- Chilled water pumps: motor bearings and, on many designs, pump bearings, often in duty and standby pairs.
- Condenser water pumps: the same layout, carrying heat to the cooling towers.
- Heating water pumps: the hot side of the plant, often running hard through the winter and idle through the summer.
- Air handling unit (AHU) fans: pillow block bearings on the fan shaft of belt driven units, plus the motor bearings.
- Computer room air handler (CRAH) fans: the fans that move cold air across the data hall.
- Cooling tower fans and gearboxes: long shafts and gear sets on the roof, exposed to weather and hard to reach while running.
- Chiller compressors: high speed bearings inside the machine that carries the plant's biggest load.
- Coolant distribution unit (CDU) pumps: the pumps that move coolant to liquid cooled racks.
- Air compressors: the units that supply controls and pneumatic equipment.
Office buildings carry most of this list too, as part of their heating, ventilation and air conditioning (HVAC) plant. Our guides to HVAC maintenance in office buildings and data center cooling loads cover how these machines fit into a maintenance program.
A useful first exercise is to list every bearing on site. Mark which ones have a running twin, which run on a VFD and which can only be reached during an outage. Those three marks predict most of the risk in this guide.
What looks like a bearing fault but isn't?
Several common faults look or sound like a failing bearing but have different causes and different fixes. Replacing a bearing for one of them wastes a part and leaves the real problem in place, and that problem often destroys the new bearing as well.
- Imbalance: a peak at exactly 1× running speed, mostly in the radial direction. Covered in Part 2 of this series.
- Misalignment: a peak at 2× running speed, high axial vibration and heat at the coupling rather than the bearing. Also covered in Part 2.
- Cavitation: a broadband crackle, often described as gravel in the pump, that changes as flow changes. Covered in Part 3.
- Belt squeal: a loud squeal at start up from a slipping belt, not a bearing. Covered in Part 4.
- A loose guard or panel: a rattle from sheet metal that vibrates with the machine but has nothing to do with its bearings.
The defect frequencies are what separate them. Imbalance and misalignment sit on whole multiples of running speed, while bearing defects fall between them. Cavitation tracks flow, and a loose panel goes quiet when you press on it.
This is also why no single sensor tells the whole story. Ultrasound, vibration, acoustic listening, thermal and the pump's own gauges and panel readouts each rule out different lookalikes. A rising bearing reading with steady discharge pressure means something different from the same reading with pressure swinging.
Repeat failures on the same machine are the strongest hint that a lookalike is involved. If a pump keeps eating bearings, the bearing is probably the victim, a pattern covered in our guide to recurring HVAC failures.
What does a bearing failure cost in an office building versus a data center?
A bearing failure costs an office building comfort, tenant goodwill and often a much larger repair than the bearing itself. In a data center it can cost cooling capacity and redundancy, sometimes without anyone noticing until a second failure.
In an office building
- Lost airflow: when an AHU fan bearing fails, the zone it serves loses its air, and the tenant calls start.
- Seal leaks: play in a worn pump bearing lets the mechanical seal faces open. The pump leaks, and water damage can follow.
- A destroyed motor: a collapsing motor bearing can let the rotor drop and rub the stator, turning a bearing replacement into a motor replacement.
In a data center
- Liquid cooling at risk: a failed CDU pump bearing threatens secondary loop flow to liquid cooled racks. Our guide to liquid cooling inspection covers what else to watch on that loop.
- Redundancy spent silently: when a duty pump fails, the standby takes over. The plant has used its spare and is one fault from a cooling loss.
- A standby that cannot take over: a standby pump falsely brinelled while idle may not take over when it is called on.
The common thread is that the bearing is the cheap part. The cost is in what fails next, which is why early detection buys so much. Bearing faults left to Stage 4 also feed the backlog of deferred maintenance that turns into capital spending.
What can a walk past prove, and what needs a follow up test?
A walk past can prove that something is visibly or audibly wrong with a bearing, and it can flag a rising trend. It cannot say which part of the bearing is damaged or what condition the lubricant is in. It helps to label every finding by how much it actually shows.
- Directly observable: a bearing housing running hotter than the same bearing on its sibling unit, grease purged from the seals, and a visibly wobbling shaft. These are facts anyone on the walk can record with a photo or a reading.
- Strong indicator: an ultrasound reading rising against the bearing's own baseline, and defect frequencies appearing in a vibration spectrum. These point clearly at a bearing problem and justify planning work.
- Suspected: which race is damaged, and the condition of the lubricant. Both need a contact vibration measurement, an oil or grease sample, or a teardown to confirm.
These tiers mark the limits of any route based check. A walk past and a few readings will not replace a contact vibration measurement, a lubricant lab report or a teardown. Their job is to catch the change early and send the right machine for the right test.
A good walkthrough record captures each finding with its tier, its location and a photo, so the follow up test starts from evidence rather than memory. What a camera can catch on that walk, such as purged grease or a missing guard, belongs in the same record.
Walk up checklist: what to check at each bearing
- Listen at each bearing housing for grinding, rumbling or a change in tone since the last round.
- Look for grease purged from the seals, grease on the floor or the guard, and discolored housings.
- Compare each housing's temperature with the same bearing on the sibling unit, not with a fixed number.
- Check the tag or the log for the last date each standby pump ran, and for how long.
- On VFD driven motors, check whether a shaft grounding ring is fitted.
- Confirm the guards are in place and nothing is touching the shaft.
How FacilityOps detects it
Coming Soon
Questions to ask about your own equipment
- 1. When did each standby pump last run, and for how long?
- 2. Do you have a baseline ultrasound reading for each critical bearing?
- 3. Are your VFD driven motors above 100 hp fitted with shaft grounding?
- 4. Is greasing done by calendar, or by condition?
- 5. Which bearings on your site cannot be reached safely while running?
Frequently asked questions about bearing failure
What are the stages of bearing failure?
Reliability engineers describe four stages. In Stage 1 the defect shows only in the ultrasonic range, and in Stage 2 the components ring at their natural frequencies. In Stage 3 defect frequencies appear in the vibration spectrum and heat starts to rise. In Stage 4 the bearing is noisy, hot and close to failure.
How early can ultrasound detect a failing bearing?
Ultrasound can detect a failing bearing in Stage 1, before the defect shows in a normal vibration spectrum and well before heat or audible noise. Readings are compared with the bearing's own baseline, with about 8 dB above baseline commonly treated as the first alarm.
Does a hot bearing mean it is failing?
A bearing running clearly hotter than the same bearing on an identical machine is a warning, and usually a late one. Heat is the last of the four warning signs, so a hot bearing is often already damaged. A cool bearing is not proof of health, because early stage defects add little heat.
What causes most premature bearing failures?
In SKF's breakdown of premature bearing failures, poor lubrication causes 36%, fatigue 34%, mounting and handling 16% and contamination 14%. Most of those causes are in the hands of the people who install, grease and protect the bearing.
What is false brinelling?
False brinelling is wear damage to a bearing that is not turning but is exposed to vibration, often from nearby running equipment. The lubricant film breaks down at the contact points, and small oscillations wear marks into the raceway spaced exactly at the rolling element pitch. It is a common reason standby pumps are noisy or fail when started.
Can a VFD damage motor bearings?
Yes. A pulse width modulated variable frequency drive creates a common mode voltage that couples onto the motor shaft and discharges through the bearing as tiny sparks. Over time this pits the race, darkens the grease and forms a washboard pattern called fluting, after which failure is close.
How do you know when a bearing needs grease?
With ultrasound, a bearing that reads about 8 dB above its own baseline with no change in sound quality usually needs grease. Add grease slowly while listening and stop when the reading returns to baseline. Greasing by calendar alone risks both under and over greasing.
Can you over grease a bearing?
Yes. Too much grease churns inside the bearing, which raises friction and temperature, and the pressure can push grease past the seals. Greasing a little at a time while monitoring with ultrasound helps avoid it.
Should standby pumps be rotated, and why?
Yes. Rotating duty and standby pumps on a schedule is the standard countermeasure to false brinelling. Running rebuilds the lubricant film and moves the rolling elements to new contact points. It also proves the standby will start and carry the load when needed.
Sources
- ReliaMag: bearing failure cause statistics (SKF premature failure split)
- UE Systems: bearing condition monitoring with ultrasound
- UE Systems: bearing inspection and ultrasound guided lubrication
- EST AEGIS: shaft voltage and bearing currents
- EST AEGIS: bearing protection best practices for motor repair
- ReliaMag: preventing bearing damage during storage
- Fabrico: bearing defect frequencies and rule of thumb estimates
Everything in this piece, as a checklist you can take on your rounds
Turn the key points from this article into a practical inspection checklist.