Predictive maintenance
What is predictive maintenance? A practical guide for plant teams
What predictive maintenance is, how it differs from preventive maintenance, which technologies it uses and how to launch a pilot that pays for itself.
Predictive maintenance in one paragraph
Predictive maintenance (PdM) means servicing equipment based on its measured condition rather than on a calendar or after it breaks. Sensors track signals that change as a machine degrades, mainly vibration, temperature, electrical current and oil condition. Software compares those signals with limits and with the machine's own history. When the data shows a developing fault, the maintenance team plans the repair for a convenient window, with parts and people ready.
Reactive vs preventive vs predictive maintenance
| Reactive | Preventive | Predictive | |
|---|---|---|---|
| Trigger | Failure | Time or running hours | Measured condition |
| Downtime | Unplanned, often long | Planned, sometimes unnecessary | Planned, only when needed |
| Parts usage | Emergency purchases | Healthy parts replaced early | Parts used close to full life |
| Best for | Cheap, non-critical, redundant items | Wear that tracks time closely (filters, belts) | Critical rotating equipment with random failure timing |
In practice, good plants use all three. Predictive maintenance is not about monitoring everything. It's about putting continuous monitoring on the assets where a surprise failure is expensive, and where failure timing is too unpredictable for a fixed schedule.
The P-F curve: why timing is everything
The P-F curve plots equipment condition over time. P is the point where a potential failure first becomes detectable. F is functional failure. The gap between them, the P-F interval, is the time you have to act.
Different detection methods sit at different points on that curve. Changes in high-frequency vibration typically appear first. Overall vibration rises later, then audible noise and heat, then failure. The earlier your detection method, the longer your P-F interval and the more options you have: order parts at normal prices, combine the job with a planned stop, or run at reduced load until the repair window.
Rule of thumb: inspect at intervals shorter than half the P-F interval. Monthly vibration routes miss faults that progress in two weeks. Continuous monitoring doesn't.
Condition monitoring technologies
Vibration analysis
The backbone of PdM for rotating equipment. Imbalance, misalignment, looseness, bearing defects and gear wear each leave a signature. Wireless tri-axial sensors now make it affordable to monitor continuously. See our vibration monitoring guide for the details.
Temperature
Useful for bearings, motor windings and electrical connections. Temperature usually rises late in a mechanical failure, so pair it with vibration rather than relying on it alone.
Electrical and power monitoring
Current, power and power factor reveal overload, mechanical drag and efficiency loss. Power monitoring also gives you the energy view, so one installation serves both maintenance and energy goals.
Oil analysis, ultrasound and thermography
Valuable for gearboxes, hydraulics, steam traps and electrical panels. They're usually periodic, route-based checks that complement continuous sensors.
How to run a predictive maintenance pilot that proves value
- Rank criticality. List assets where failure stops production, is a safety risk or costs a lot to repair, especially those without a standby.
- Pick 5–20 assets. Enough to see real events within a few months, few enough to manage closely.
- Record the "before". Downtime hours, emergency repair costs and maintenance labour on those assets over the past year. Our ROI calculator helps you structure it.
- Install and baseline. Mount sensors consistently on bearing housings and let the system learn normal behaviour.
- Build the routine. A short weekly review of the risk-ranked asset list, with a named owner for every action.
- Close the loop. After every repair, record what you found. Confirm vibration returned to baseline. Those stories build the business case for scaling.
Five mistakes to avoid
- Monitoring everything at once. Start where failures hurt most.
- Alarms without actions. Every alert needs a recommended next step and an owner, or people learn to ignore it.
- Trusting silent sensors. A sensor that stopped reporting is not a healthy machine. Track data freshness.
- Averaging measurement points. Roll machine health up from the worst point. An average hides a failing bearing.
- No feedback loop. If you don't record what you find during repairs, you can't tune thresholds or prove ROI.