Why Bearings Fail — And Why Most Plants Catch It Too Late
Bearings account for roughly 40-50% of all rotating equipment failures in industrial plants. That number has held steady for decades despite improvements in bearing manufacturing quality. The problem isn’t the bearings themselves. It’s how we monitor them — or more accurately, how we don’t.
Most maintenance teams still rely on operator rounds and “feel” to catch bearing problems. By the time a bearing is hot to the touch or audibly noisy, you’re in the final 10-15% of its remaining life. Vibration analysis changes that equation entirely. A properly configured vibration program can detect bearing defects months before functional failure, giving you time to plan the repair on your terms.
The Four Stages of Bearing Failure
Bearing degradation follows a predictable pattern. Understanding these stages is critical because your detection method and response should change at each one.
Stage 1: Subsurface Fatigue (Earliest Detection)
Microscopic cracks form beneath the raceway surface. No visible damage exists. Ultrasonic emission monitoring or high-frequency enveloping (HFE) techniques can detect this stage. Traditional velocity measurements will show nothing abnormal. This stage can last months or even years depending on operating conditions.
Stage 2: Surface Defects Emerge
Subsurface cracks propagate to the surface, creating small spalls or pits. Vibration signature shows bearing defect frequencies — BPFO, BPFI, BSF, or FTF — depending on which component is damaged. The defect frequencies will appear with harmonics and sidebands. Overall vibration levels may still be within normal ranges at this point. This is your optimal repair window.
Stage 3: Defect Growth and Interaction
Individual defects grow and begin interacting. Multiple bearing defect frequencies appear simultaneously. Broadband noise floor rises. Overall vibration velocity readings start trending upward. You’ll often see random, high-frequency energy across the spectrum. Plan your repair now — you have weeks, not months.
Stage 4: Advanced Degradation
Bearing geometry is compromised. Clearances increase dramatically. 1x RPM amplitude changes due to looseness. High broadband vibration across the spectrum. Audible noise and elevated temperature are now present. Catastrophic failure is imminent — days to weeks at most.
Bearing Defect Frequencies: The Math That Matters
Every bearing has four characteristic defect frequencies determined by its geometry. You need these to identify which component is failing.
- BPFO (Ball Pass Frequency Outer Race) — Generated when rolling elements pass over a defect on the outer race. Typically the most common bearing fault. For a standard bearing, expect BPFO at roughly 40-48% of the number of rolling elements multiplied by shaft speed.
- BPFI (Ball Pass Frequency Inner Race) — Generated by inner race defects. Usually appears with RPM sidebands because the defect rotates with the shaft, moving in and out of the load zone.
- BSF (Ball Spin Frequency) — Rolling element defects. Often appears at 2x BSF since the defect impacts both races per revolution of the rolling element.
- FTF (Fundamental Train Frequency) — Cage defects. Always subsynchronous, typically 0.35-0.45x RPM. Cage faults are rare but dangerous — they can lead to sudden catastrophic failure.
Don’t try to calculate these by hand. Every bearing manufacturer publishes defect frequency ratios. SKF, NSK, Timken, and FAG all have free online calculators. Enter the bearing number and shaft speed, and you get exact frequencies. Build a database of these for every monitored bearing in your plant.
Setting Up Vibration Alarm Thresholds
ISO 10816 provides general vibration severity guidelines for different machine classes, but these are too broad for effective bearing monitoring. You need a layered approach.
Overall Velocity Alarms
Set overall velocity alarm thresholds based on machine class per ISO 10816-3:
- Class I (small machines up to 15 kW): Alert at 1.8 mm/s, Danger at 4.5 mm/s
- Class II (medium machines 15-75 kW): Alert at 2.8 mm/s, Danger at 7.1 mm/s
- Class III (large rigid foundation): Alert at 4.5 mm/s, Danger at 11.2 mm/s
- Class IV (large flexible foundation): Alert at 7.1 mm/s, Danger at 18 mm/s
These catch Stage 3 and 4 failures. That’s too late for good planning. You need envelope alarms for early detection.
Envelope (Demodulation) Alarms
Envelope analysis strips away the low-frequency vibration from imbalance, misalignment, and structural resonance, leaving only the repetitive impacts from bearing defects. Set envelope alarms using baseline readings plus a multiplier. A 6 dB increase over baseline warrants investigation. A 12 dB increase means a confirmed defect requiring planned repair. A 20 dB increase indicates advanced damage — schedule the repair within the next planned outage.
Trending Is Everything
Absolute values matter less than rate of change. A bearing running at 3.2 mm/s overall velocity for five years is not a concern. A bearing that went from 1.1 mm/s to 3.2 mm/s in the last 60 days is a problem, even though 3.2 mm/s might be “acceptable” per ISO standards. Trend your data monthly at minimum. For critical equipment, weekly or continuous monitoring is appropriate.
Measurement Points and Sensor Placement
Where you mount the sensor determines what you can detect. Bad sensor placement is the single most common reason vibration programs fail to catch faults.
- Mount accelerometers as close to the bearing as possible. Every inch of metal between sensor and bearing attenuates the high-frequency signals you need for early detection.
- Radial measurements (horizontal and vertical) detect most bearing faults. Axial measurements catch thrust bearing issues and angular contact bearing problems.
- Use stud-mounted sensors for permanent installations. Magnetic mounts are acceptable for route-based collection but introduce a resonance around 2-5 kHz that can mask bearing defect signals.
- Mark your measurement points permanently. Repeatability matters. If the analyst puts the sensor in a different spot each month, trending is worthless.
Common Mistakes That Kill Vibration Programs
After two decades of building and rescuing vibration programs, the same mistakes keep showing up.
Collecting data without analyzing it. Too many plants invest in vibration hardware, train someone to walk routes, and then let the data sit in the software untouched. Data collection without analysis is just exercise.
Monitoring everything the same way. Not every machine needs monthly vibration data. Classify your equipment by criticality. Critical machines get continuous monitoring or weekly routes. Non-critical machines might only need quarterly checks or run-to-failure treatment.
Ignoring the analyst role. Vibration analysis is a skilled discipline. ISO 18436-2 defines four certification levels for a reason. A Category I analyst can collect data and identify basic faults. Complex diagnostics require Category II or III expertise. If you don’t have that in-house, partner with a firm that does.
Setting thresholds too high. If your alarm thresholds never trigger, they’re too high. You should see occasional alerts that get investigated and resolved. An alarm system that never alarms isn’t protecting anything.
Building Your Bearing Monitoring Program
Start with your critical assets. Identify every bearing on each critical machine. Get the bearing numbers from the OEM manuals or pull them during the next maintenance event. Calculate defect frequencies and enter them into your vibration software. Set up envelope alarms at reasonable thresholds based on initial baseline readings.
Run monthly routes for 3-6 months to establish stable baselines. Adjust thresholds based on actual operating data — not theoretical values from a textbook. Train your operators to understand what the vibration program can tell them so they support the effort rather than viewing it as another initiative that doesn’t affect their work.
The goal is simple: detect bearing defects in Stage 2, plan the repair, and execute it during a convenient window. Plants that do this consistently see 30-50% reductions in unplanned bearing-related downtime within the first two years. The technology isn’t complicated. The discipline to collect, analyze, and act on the data consistently — that’s the hard part.