Engineering & Technology · Reliability, Safety, Failure & Maintenance

Preventive, condition-based and corrective maintenance

A machine reports a changing condition indicator. The graph becomes useful only when someone can interpret it and complete an appropriate response before the function is lost. Buying a sensor does not by itself create that response capacity or establish a valid forecast.

Use Quality Engineering, Experiments & Process Evidence; Engineering Design Methods & Trade-offs; and Material Selection, Degradation & Failure. Read probability as a fraction of comparable cases, distinguish a rate from a count and retain units in energy and inventory balances. Models use invented teaching records and supplied thresholds; they do not approve a real safety or maintenance decision.

Back to the chapter

After reading, you can

Choose a maintenance strategy around a failure mechanism, a useful observation and the time needed to respond.

Match the maintenance action to the function

Corrective maintenance restores a function after a fault is identified. A deliberate run-to-failure policy may be suitable for some low-consequence items when replacement and recovery are manageable; it is not a default for safety-critical functions. The system consequence, hidden failures and possible secondary damage matter as well as the item’s purchase price.

Preventive maintenance uses scheduled time, usage or another planned trigger. Condition-based maintenance uses evidence about current condition. Predictive work adds a model of future condition or remaining service. These approaches can coexist within one asset because its different failure modes have different patterns and detectability. Replacing everything more often is not automatically effective and can introduce new defects.

Connect observation to action

A condition measurement needs a known relationship to the failure mode of interest. Load, environment, sensor drift and measurement location can change an indicator without the assumed damage changing. A threshold therefore needs evidence for its use, along with uncertainty and the consequence of missed or false alarms. The figure uses arbitrary supplied thresholds to isolate timing.

A detectable interval must accommodate sampling delay, interpretation, planning, resources, intervention and verification. If the observation is made too late, even perfect diagnosis may not preserve the function. Conversely, intrusive work has its own exposure and quality requirements. A useful strategy includes returning the asset to a verified configuration, not merely closing a work order.

Fictional indicator z=2 t, with t in days. An action threshold is supplied at z=8, crossed at day 4; functional failure is separately defined at z=20, day 10. Scheduled samples at days 3, 6, 9 measure 6, 12, 18 units. First sample above action threshold is day 6, leaving 4 days before the assumed failure boundary. A 2-day response completes on day 8; a 4.5-day response completes on day 10.5 and misses it. These are arbitrary teaching thresholds with perfect measurements and constant trend; no equipment limit or forecast confidence is supplied.

Learn whether the strategy changes the outcome

Keep asset identity, operating exposure, mode, condition history, action and subsequent result connected. Repeated events after replacement may indicate that the mechanism or its enabling condition remains. A lower failure count during reduced operation does not demonstrate improved reliability; compatible exposure and uncertainty still matter.

Reliability-centred maintenance begins with the required function and operating context, then selects applicable actions. Some failure modes offer no useful advance signal or age relationship, so a proposed monitoring or interval policy may not address them. Alternatives can include redesign, a different service arrangement or recovery provision. Choosing among them needs consequences and evidence, not a universal claim that one maintenance category always wins.

Check whether the response fits the warning interval

Supply an ideal indicator z=2 t, where t is operating days. An action threshold z=8 is reached at day 4; functional failure is defined separately at z=20, day 10. Samples at days 3, 6 and 9 read 6, 12 and 18. The first threshold detection occurs on day 6.

Four days remain before the supplied failure boundary. A response completed two days after detection finishes on day 8, leaving two days in this model. A 4.5-day response finishes on day 10.5 and misses the boundary. Real trends and observations need uncertainty; these exact thresholds and linear timing are teaching inputs, not equipment criteria.

Try the reasoning

Does a condition sensor predict every failure mode?
Show the answer and reasoning
No. Its measured quantity must be connected to the mode, and some failures lack a useful detectable warning interval.
Why is a six-day threshold-to-failure interval not always six days available for work?
Show the answer and reasoning
Sampling and interpretation can consume part of it before the first detection, followed by planning and response delays.
Can unnecessary scheduled work reduce reliability?
Show the answer and reasoning
It can introduce damage, incorrect configuration or other defects. The action and its verification need evidence just as the failure model does.

What to carry forward

A maintenance strategy succeeds through a justified mechanism, timely evidence, response capacity and verified restoration.

References

Teaching resources and research supporting this lesson’s explanations and examples.