Four maintenance strategies
| Strategy | When work is done | Works well for | Weakness |
|---|---|---|---|
| Reactive (run to failure) | After it breaks | Cheap, redundant, non-safety items such as light fittings or small duplicated pumps | Unplanned downtime, secondary damage, safety risk |
| Preventive (time-based) | At fixed intervals of time, hours or cycles | Failures that really are age-related: wear liners, filters, lubricant top-up, belts | Replaces good parts, can introduce faults during intrusive work |
| Predictive (condition-based) | When measured condition shows a developing failure | Failure modes with a detectable warning period: bearings, gears, misalignment, electrical connections | Needs sensors, skills and a workflow to act on findings |
| Proactive (root cause) | Before degradation starts, by removing its cause | Repeat failures: precision alignment, balancing, lubrication cleanliness, correct installation | Needs discipline and evidence from the other strategies |
A mature program uses all four. Predictive maintenance finds the developing failures; proactive maintenance removes their causes so they stop coming back.
Most failures are not age-related
Studies of airline equipment in the 1960s and 70s, later confirmed in other industries, found six patterns of conditional failure probability over the life of a component. Only a minority show a clear wear-out age.
| Pattern | Shape | Share in the original study |
|---|---|---|
| A · Bathtub | High early failures, constant, then wear-out | about 4% |
| B · Wear-out | Constant, then rising sharply at a certain age | about 2% |
| C · Gradual ageing | Slowly rising, no clear wear-out age | about 5% |
| D · Low when new | Low at first, then constant | about 7% |
| E · Random | Constant probability at all ages | about 14% |
| F · Infant mortality | High after installation or overhaul, then constant | about 68% |
The shares vary by industry, but the lesson holds: for most failure modes, replacing parts at a fixed age does not reduce failures and can increase them (pattern F). What does work is detecting the failure as it develops, which is exactly what predictive maintenance does.
The P-F curve
Most failures do not happen instantly. A defect starts, grows, and eventually stops the machine from doing its job. The P-F curve plots equipment condition against time along that path:
- P (potential failure) is the earliest point at which you can detect that the failure is developing, with a given technique.
- F (functional failure) is the point where the asset no longer does what you need it to do.
- The P-F interval is the time between them: your warning period.
Different techniques detect the same failure at different points. Ultrasound and vibration usually see a bearing defect months before it gets hot; temperature and noise see it days or hours before failure. The earlier the technique, the more time you have to plan.
Setting the measurement interval
The classic rule from reliability-centred maintenance is: measure at intervals no longer than half the P-F interval. If a defect can appear just after a measurement, a half-interval rule still gives you at least one detection and half the P-F interval to plan and act.
In practice:
- Use the shortest credible P-F interval for the failure mode, not the average, especially on critical assets.
- Subtract the time you need to act: analysis, approval, parts, scheduling a stop. The usable warning is the P-F interval minus one measurement interval minus that lead time.
- Fast-developing failure modes (lubrication loss, cavitation, process upsets, electrical faults) can have P-F intervals of days or hours. Only online monitoring catches them reliably.
- Variable or intermittent duty shortens effective intervals: a machine that runs one shift a day only produces useful data when it runs.
| Technique | Typical P-F range (indicative) | Monthly route enough? | Online (wireless) helps when… |
|---|---|---|---|
| Ultrasound | 3 to 9 months | Often, for bearings and lubrication | Bearings are slow, remote or lubricated automatically |
| Vibration | 1 to 9 months | Often, for standard-speed balance-of-plant | Asset is critical, hard to reach, unsafe to approach, or runs intermittently |
| Oil analysis | 1 to 6 months | Monthly to quarterly sampling is common | Large gearboxes, hydraulics and turbines with high contamination risk |
| Motor current | 1 to 6 months | Yes, for rotor and eccentricity faults | Motor is remote, submerged or in hazardous areas; load varies |
| Thermography | 1 week to 2 months | Quarterly to annual electrical surveys are common | Critical switchgear, kilns, conveyors with fixed cameras |
| Temperature | 1 day to 2 weeks | No | Always online: it is the late safety net |
Where wireless changes the economics
A monthly route is limited by the walking time of a technician. Wireless sensors make hourly or daily measurement cheap, so the measurement interval stops being the weak link. The P-F interval then mostly limits how far ahead you can plan, not whether you detect the fault.
The cost of waiting
The further a failure progresses down the curve, the more it costs: early detection means a planned bearing change; late detection means a damaged shaft, a burnt winding or secondary damage to the driven machine, plus unplanned downtime. That cost curve is why early techniques are worth the effort for critical assets.
Key takeaways
- Only a minority of failure modes are age-related, so condition-based maintenance fits most of them.
- The P-F interval is the warning time between detectable and failed. It depends on the technique.
- Measure at intervals no longer than half the shortest credible P-F interval.
- Wireless online monitoring removes the measurement interval as the limiting factor.
Try it in the apps
The Program Setup app suggests a measurement interval per asset and technique, based on criticality and the half-P-F rule.