The three strategies, honestly defined
| Preventive (PM) | Condition-based (CBM) | Predictive (PdM) | |
|---|---|---|---|
| Trigger | Time or run-hours elapsed | A measured value crosses a threshold | A trend says failure is developing, with lead time |
| Needs | A schedule and discipline | Sensors + alert thresholds | Sensors + trending/analysis over time |
| Fails when | Failure isn't age-related (most aren't), or the schedule drifts from real usage | Thresholds set wrong; alarms ignored | Applied to assets too cheap to justify it |
| Cost shape | Steady labour + parts, some wasted | Modest monitoring cost, big surprise-reduction | Highest setup, highest payoff on critical assets |
| Best for | Statutory items, lubrication, filters, cheap components | Motors, pumps, fans, conveyors — the broad middle | Bottleneck and high-consequence assets |
The dirty secret of the "vs" debate: decades of reliability studies keep finding that most failure modes are not age-related — they're random in time, triggered by lubrication, contamination, misalignment or electrical stress. A calendar can't see any of that. That's the fundamental case for putting measurement on equipment, whichever label you use for the strategy.
Where preventive maintenance is still exactly right
- Statutory and safety items — pressure vessels, lifting gear, fire systems: the law schedules these, not the vibration trend.
- Cheap, replaceable, failure-random parts — belts, filters, seals: replacing on interval costs less than instrumenting them.
- Lubrication — the highest-return maintenance activity in most plants, and inherently schedule-driven (though run-hours beat calendar).
- Anything whose failure is cheap and harmless — if failure costs less than monitoring, run to failure or PM it; predictive there is hobbyism.
Where predictive earns its keep
Predictive maintenance pays where three things coincide: the asset's failure stops production (or creates danger), failure gives measurable warning (bearing wear, imbalance, overheating, rising current draw), and the lead time buys something — a planned weekend repair instead of a 2am breakdown with no spares. Classic candidates: the bottleneck line's drive motors, critical pumps and fans, compressors, conveyors. Our conveyor case study is the pattern: weeks of bearing warning converted a catastrophic stop into a scheduled swap.
One honest caveat from the field: predictive projects fail more often from programme design than from sensors — alarms without owners, thresholds nobody trusts, monitoring bought before anyone decided who acts on it. We wrote up the failure modes in why most vibration-monitoring projects fail; read it before buying anything, including from us.
The South African twist: load shedding lies to your schedule
Calendar-based PM assumes equipment runs a predictable duty. Load shedding broke that assumption: a machine that spent 30% of the month dark doesn't need its monthly service yet — and servicing it anyway wastes parts and wrench-time. Worse, the start-stop cycling that outages force onto motors and compressors is itself a failure accelerant the calendar never sees. The practical fixes, in order of effort: switch PM triggers from calendar to measured run-hours (a current clamp per machine is enough); add condition alerts on the assets that cycling punishes; and keep the records flowing through outages — monitoring that survives load shedding is a design requirement here, not a nice-to-have.
Choosing per asset: a 20-minute exercise
- List your top 10 assets by consequence-of-failure (production stop × repair cost × safety).
- Top of the list → condition monitoring now, predictive trending as the data accumulates. This is our asset & condition monitoring scope — vibration, temperature, current, run-hours, usually as a monthly subscription rather than capex.
- Middle of the list → CBM thresholds plus run-hours-based PM.
- Bottom of the list → preventive on run-hours, and stop feeling guilty about it.
Small plants sometimes ask whether they need "preventive maintenance software" first. Software organises intent; it doesn't create knowledge. A small operation gets further with sensors on its three critical machines and a disciplined schedule for the rest than with a CMMS full of guessed intervals — add the software when the asset list outgrows the whiteboard.