Maintenance strategy is often framed as a single choice — preventive or predictive — when in practice most reliable operations run some blend of both. This note sets out the actual difference between the two approaches, and the conditions under which each one earns its cost.
The core difference
Preventive maintenance acts on a fixed schedule — time elapsed, cycles run, or hours logged — regardless of the equipment's actual condition at that moment. Predictive maintenance instead acts on measured condition: vibration, temperature, current draw, or other signals that indicate wear is developing, triggering intervention closer to when it's actually needed.
Neither is inherently superior. Preventive schedules are simple to plan around and don't depend on sensing infrastructure. Predictive approaches can reduce unnecessary servicing and catch developing faults earlier, but only if the underlying condition data is trustworthy and the monitored failure modes are actually detectable in advance.
“A predictive strategy is only as good as the signal it's built on. Bad condition data produces false confidence, not better decisions.”
When preventive still wins
- Failure modes that don't show a clear, measurable precursor
- Low-criticality equipment where unplanned downtime is cheap
- Environments without reliable sensing or the budget to install it
- Components with well-understood, consistent wear-out lifespans
In these cases, a fixed schedule — tuned from experience rather than live data — is often the more honest and more maintainable answer. The goal isn't to chase the more advanced-sounding strategy; it's to match the strategy to how the equipment actually fails.
Where condition monitoring pays off
On equipment with high downtime cost and failure modes that do give advance warning — bearing wear, developing imbalance, gradual thermal drift — condition monitoring can shift maintenance from reactive firefighting to planned intervention. This is the territory explored more directly in the CNC predictive-maintenance work referenced below: applying data-driven methods to CNC machine tools rather than assuming a model will work everywhere.
The practical takeaway is to treat this as a portfolio decision rather than a policy decision. Different assets on the same production line can reasonably sit on different strategies, chosen asset by asset rather than declared as a blanket policy.
