Predictive vs Preventive Maintenance: What's the Difference?
Preventive maintenance and predictive maintenance differ in what triggers the work. Preventive maintenance follows a schedule based on calendar time, runtime hours or cycle counts. Predictive maintenance happens when measured evidence shows that an asset's condition is getting worse. Preventive maintenance needs good records and a sensible interval. Predictive maintenance needs reliable condition data, someone who can interpret it, and enough warning before failure to plan the work.
Most plants use both, and many assets are best left to run until they fail. The useful question is which approach suits each asset.
What triggers each type of maintenance?
Preventive maintenance is triggered by an interval agreed in advance. Typical examples are changing a gearbox oil every set number of runtime hours, replacing a filter every quarter, or inspecting a conveyor belt after a fixed number of cycles. The work happens whether or not the component needs it yet.
Predictive maintenance is triggered by a condition. A vibration level rises on a motor bearing, a temperature trend creeps upward on a gearbox, or a pump draws more current for the same duty. When the measurement crosses an agreed limit, or the trend shows a change, the team plans an intervention.
The two can be combined. A team might inspect a fan on a fixed interval and also watch its vibration, bringing the inspection forward if the trend changes.
What data does each approach need?
The data needs are the biggest practical difference between the two.
Preventive maintenance needs:
- An accurate asset register with the components to be serviced.
- A reliable trigger value, such as calendar dates, runtime hours or cycle counts.
- Work records showing what was done, when, and what was found.
- An interval based on manufacturer guidance and site experience.
Runtime-based intervals are often more realistic than calendar intervals for machines with variable utilisation. Our guide to using machine runtime for maintenance scheduling covers how to collect and validate those hours.
Predictive maintenance needs all of that, plus:
- A measurement that responds to the failure mode you care about.
- Sensors, or a route-based inspection, that can capture it consistently.
- A baseline for normal behaviour under known operating conditions.
- Context such as load, speed, product and machine state, so that a process change is not mistaken for degradation.
- Agreed alert limits and a named person who reviews and acts on them.
Predictive maintenance is only as good as the measurement behind it. A drifting sensor can create a false trend or hide a real one. Before trusting condition data, confirm that the instrument is accurate. Our guide to detecting sensor drift in plant data explains how to check this, and keeping instrument calibration records covers the supporting evidence.
How do the costs compare?
This section compares the type of cost, not the size of any return. The actual figures depend on your assets, labour and production consequences.
| Cost area | Preventive maintenance | Predictive maintenance |
|---|---|---|
| Set-up | Low: schedules in a maintenance system or spreadsheet | Higher: sensors, connectivity, data storage, analysis tools |
| Skills | Planners and technicians following task lists | Plus people who can interpret trends and set limits |
| Parts and labour | Some parts replaced before end of useful life | Work planned closer to actual need |
| Ongoing upkeep | Reviewing intervals | Sensor maintenance, calibration, data quality checks, limit tuning |
| Risk of unplanned failure | Remains for failures that are not age-related | Reduced only where the measurement gives enough warning |
| Risk of induced faults | Intrusive tasks can introduce errors | Fewer intrusive tasks, but false alerts cost investigation time |
Predictive maintenance moves cost from parts and routine labour to instrumentation, data and analysis. Whether that trade is worthwhile depends on the asset.
Which assets suit which approach?
Use the asset's failure behaviour and the consequence of failure to choose. The table below is a starting guide, not a rule. Confirm choices with your maintenance lead and the equipment manufacturer's guidance.
| Asset characteristic | Suggested approach | Reasoning |
|---|---|---|
| Wear tracks usage closely (filters, belts, lubricants, seals) | Preventive, on runtime or cycles | Predictable wear makes an interval effective and cheap |
| Rotating equipment with a detectable failure mode (motors, pumps, fans, gearboxes) and costly downtime | Predictive | Vibration, temperature or current usually changes before failure |
| Failure gives little or no measurable warning (many electronic components) | Preventive replacement, spares holding, or run-to-failure | Condition monitoring has nothing reliable to detect |
| Low cost, low consequence, quick to replace (lamps, small fans) | Run-to-failure | Monitoring or scheduled replacement costs more than the failure |
| Required by regulation, insurer or manufacturer warranty | Preventive, as specified | The interval is set externally; condition data may support but not replace it |
| Safety-related equipment | As defined with a qualified assessor | Test and maintenance intervals form part of the safety design |
| Critical asset, long lead-time parts, warning available | Predictive plus preventive inspection | Early warning protects lead time; inspection catches what sensors miss |
| Variable utilisation, wear follows use | Preventive on runtime, not calendar | Calendar intervals over- or under-service the asset |
For assets that fail without warning, the right answer is often the right spares strategy rather than more monitoring. See choosing critical control system spares.
What does a worked decision look like?
Illustrative example: a plant reviews three assets on one packaging line.
- Main drive motor and gearbox. A failure stops the line and the gearbox has a long replacement lead time. Bearing wear normally shows in vibration and temperature before failure. The team keeps a preventive oil change on runtime hours and adds vibration and temperature monitoring with agreed review limits.
- Line conveyor belt. It wears steadily with use and is quick to replace. The team uses a preventive inspection every set number of runtime hours and replaces the belt based on what the inspection finds.
- Panel cooling fan. It is cheap, a spare is held, and the panel has a temperature alarm. The team runs it to failure and replaces it when the alarm or an inspection shows it has stopped.
None of these choices is permanent. The team reviews each one after any failure, or once enough history has built up to test whether the interval or limit is right.
What mistakes should you avoid?
Monitoring everything. Sensors on low-consequence assets add data to manage without changing any decision.
Measuring the wrong thing. A temperature sensor will not warn of a failure that shows up first in vibration. Start with the failure mode, then choose the measurement.
Collecting data without an owner. An alert that nobody reviews gives no protection. Name who acts and how quickly.
Setting limits from a single good day. Build a baseline across normal operating conditions, and record what "normal" covered.
Ignoring data quality. Gaps, drift and unit errors produce false alerts or silence. Cleaning plant data before building dashboards applies here too.
Dropping preventive tasks too early. Keep existing intervals until the condition data has proved it gives reliable warning.
When does this comparison not apply?
Some intervals are fixed by regulation, insurers, warranty terms or a safety assessment. Condition data may add information but should not be used to extend those intervals without approval from whoever set them. For safety-related equipment, any change to testing or maintenance should be decided with a qualified assessor.
Frequently asked questions
Is predictive maintenance always better than preventive maintenance?
No. It suits assets whose failures give measurable warning and are costly. For predictable wear or low-consequence assets, a schedule or run-to-failure is often simpler and better.
Can we start predictive maintenance with existing PLC data?
Often, yes. Motor current, run status and some temperatures may already exist in the controller. Check that the signals respond to the failure mode and that their accuracy is known. See collecting data from an existing PLC.
Is condition-based maintenance the same as predictive maintenance?
The terms overlap and are often used interchangeably. Both trigger work from measured condition. Some teams reserve "predictive" for approaches that estimate remaining time before failure.
How do we know if a preventive interval is right?
Review what technicians find at each task. If parts are routinely in good condition, the interval may be too short. If failures happen between tasks, it may be too long or the failure may not be age-related.
If you want help choosing between these approaches for your own assets, see our services or contact us.