When a machine suddenly stops one morning, the failure actually began weeks earlier. A microscopic crack on a bearing surface, degraded lubrication, a loosening bolt… Each of these processes leaves measurable traces in vibration, temperature and sound. Predictive maintenance tracks those traces systematically and schedules intervention at the right time.
Three maintenance strategies
- Corrective (reactive): fix after failure. Simple, but unplanned downtime, secondary damage and emergency parts are expensive.
- Preventive (time-based): replace parts on a calendar. Prevents some failures, but healthy parts are replaced too and faults developing off-schedule can be missed.
- Predictive (condition-based): the real condition of the equipment is measured and intervention is planned when it is actually needed.
Predictive maintenance does not eliminate the other two; it helps choose the right strategy for each asset. Running a cheap, non-critical, redundant machine to failure can make sense; a mill fan that stops production must be monitored.
The P-F curve: how early can we see it?
The P-F curve describes the time between the moment a defect becomes detectable (P) and the moment of functional failure (F). The longer this interval, the more time you have to plan.
A typical sequence for a bearing failure:
- Ultrasound and high-frequency vibration (envelope analysis): the earliest warnings — surface fatigue and lubrication problems.
- Bearing frequencies in the vibration spectrum: characteristic peaks such as BPFO and BPFI become visible.
- Rising overall vibration level: damage has progressed.
- Audible noise and heat: very close to failure.
- Functional failure.
Touching the machine and listening only catches the last stage. Measurement-based monitoring can stretch the warning time from weeks to months.
Which faults does vibration analysis reveal?
| Fault | Typical vibration signature |
|---|---|
| Unbalance | Dominant radial 1X, stable phase |
| Misalignment | 1X and 2X, high axial vibration, 180° phase shift across the coupling |
| Mechanical looseness | Many harmonics of 1X, sometimes 0.5X sub-harmonics |
| Bearing defect | BPFO, BPFI, BSF, FTF and harmonics; clear in the envelope spectrum |
| Gear defect | Gear mesh frequency and sidebands |
| Resonance | Excessive amplitude and phase change near a natural frequency |
To calculate bearing frequencies for your own bearing, use our Bearing Fault Frequency Calculator.
How to set up a predictive maintenance programme
Following a flow similar to the ISO 17359 condition monitoring standard:
1. Asset criticality analysis
Each asset is scored by production impact, redundancy, failure history and repair cost. Monitoring effort goes to the most critical assets first.
2. Measurement points and parameters
Horizontal, vertical and axial points are defined at each bearing. Overall level (mm/s RMS), spectrum, envelope and, where needed, temperature are specified. Points are marked so every reading is taken in the same place.
3. Baseline and alarm limits
The healthy signature of each asset is recorded. Alarm limits start from the ISO 10816-3 zones and are refined over time from the asset's own trend.
4. Periodic measurement and trending
The interval depends on criticality and how fast faults develop; monthly is a common starting point for general industrial equipment. The real value comes not from a single reading but from the trend.
5. Action and feedback
After every round, findings are prioritised (green / amber / red) with root cause and recommended action. The result of each intervention is measured again, so the programme keeps improving.
Conclusion
Predictive maintenance is less about buying an instrument and more about discipline: the right points, regular measurement, correct interpretation and actions that are followed through. Once that discipline is in place, unplanned downtime and unnecessary part replacements drop noticeably. See our predictive maintenance service or contact us to set up a starter programme for your plant.
