Choose a maintenance strategy for each way equipment can fail, rather than giving the whole machine one label. Scheduled work fits some age or usage patterns. Condition-based work responds to an observed state. Prediction earns its place when a credible forecast changes a useful maintenance decision. Most assets need a mixture.
Start with the function you need, the consequence of losing it, and the evidence that an intervention helps. A forecast needs to improve the action or its planning enough to justify its added effort. Equally, a convenient schedule is weak protection against a failure mechanism it does not address.
Start with a failure mode, not a machine label
A failure mode is a specific way a function becomes unacceptable: a drive belt breaks, for example, rather than simply saying that a machine is unreliable. The consequence describes what follows, including disruption to dependent work. Equipment purchase price alone does not establish that consequence.
NASA's 2008 reliability-centered maintenance guide starts its analysis with functions, failures and consequences. It also asks whether a maintenance task applies to the failure mode and is effective. Those questions are useful before comparing technologies.
Hypothetical worked scenario
Consider RW-06, a small rewinder preparing blank paper rolls for internal production paperwork at a separate support bench. Assume an owner review has established that it has no safety or product-release function. Its loss causes paper scrap and interrupts roll preparation. This invented situation is not a client case or an equipment assessment. Its existing guarding and safe-work requirements still apply; a non-safety function does not make physical maintenance risk-free.
- Required function
- Prepare usable rolls for the production support team.
- Candidate failures
- Belt breakage, bearing wear, electrical board failure.
- Decision record
- STR-06, linked to RW-06 in the maintenance register.
- Accountable owner
- Maintenance manager, with operations and qualified technical review.
Separate the trigger from the work performed
For this comparison, preventive means a task triggered by elapsed time, runtime or cycles. Condition-based means the observed condition triggers review or work. Predictive means a forecast of future condition or failure informs the decision. Industry terminology overlaps, so ask a supplier what actually triggers its recommendation.
| Approach | Decision basis | Question to ask |
|---|---|---|
| Scheduled preventive | Time or accumulated use. | Why does this task fit this interval? |
| Condition-based | An observed condition. | What finding warrants action? |
| Predictive | An estimated future state. | What changes because of the forecast? |
| Corrective | A detected fault to restore. | What repair and verification are needed? |
A scheduled inspection can reveal a condition that prompts corrective work before the required function is lost. A forecast can lead to a planned replacement. Corrective work therefore need not mean an emergency after complete breakdown, and scheduled inspection can support a condition-based strategy.
Do not treat these names as successive maturity grades. A mixed program is coherent when each task has a reason. Installing sensors does not automatically turn inspection into prognosis, the forecasting of future condition.
Keep schedules where they have a defensible basis
Scheduled work can be a sensible choice when an applicable service requirement exists or wear has a credible relationship with age or use. It offers predictable workload and purchasing. Its weakness appears when the chosen interval has no demonstrated relationship to the failure being managed.
For RW-06, a belt is a candidate for scheduled attention only after reviewing the actual belt, duty and applicable manufacturer guidance. The scenario provides no belt-life evidence. A calendar date cannot prove remaining life, and replacing a belt will not resolve an unrelated electrical-board fault.
- Keep visible
- Applicable task, current interval and source requirement.
- Challenge
- Which failure mechanism does the task prevent or reveal?
- Evidence gap
- Belt age relationship remains unestablished in STR-06.
- Decision boundary
- Retain the current approved plan pending authorized review.
Mandatory requirements, original equipment manufacturer (OEM) instructions and qualified safety requirements are not casually removed because monitoring looks promising. Any proposed interval change needs the applicable authorized process and competent review. This comparison prescribes no new legal or manufacturer maintenance interval. Intrusive work also has costs: access, reassembly and possible introduced defects deserve consideration alongside breakdown risk.
Use observed condition when inspection can change the outcome
Condition-based maintenance does not require a remaining-life model or continuous sensing. An inspection finding can be enough when a qualified reviewer understands its relevance. The important question is whether the condition can be examined meaningfully and whether intervention can improve the outcome.
For the rewinder bearing, changed vibration might justify investigation, but it does not establish bearing damage. Mounting, operating load or the measurement itself could explain a change. Inspectability includes suitable access and a valid method; it does not mean approaching moving equipment to obtain a reading.
- Recognize: identify a condition plausibly linked to bearing deterioration.
- Examine: confirm qualified, practical inspection under applicable procedures.
- Judge: establish whether meaningful warning could support intervention.
- Act: assign an authorized response and retain unresolved findings.
NASA's task descriptions distinguish work directed by time from work directed by condition. Applying that distinction here does not establish a sensor threshold. The owner still needs evidence that the chosen observation and response can address this failure mode. Where that relationship remains unknown, record it rather than treating a quiet dashboard as evidence of health.
Require a forecast to improve a real decision
A condition trigger says something warrants attention now. A prognostic forecast estimates a future condition or the time to a defined event. Remaining useful life is an estimate of time until a defined loss of acceptable function, under stated operating assumptions.
NASA's ProgPy glossary defines prognostics around future performance and event timing. Its predictor definition includes a model, initial state and estimated future load; its time-to-event definition connects a failure event to remaining useful life.
- Condition question
- Does this bearing observation justify inspection?
- Forecast question
- When might the defined bearing failure occur?
- Value question
- Would that estimate change an authorized planning choice?
- RW-06 finding
- No validated forecast or remaining-life estimate is supplied.
If a condition finding already leads to the same justified intervention and planning choice, prediction may add little. If a forecast could change a purchasing or service choice, compare its added value with model upkeep and the consequences of an incorrect estimate. A date without uncertainty or operating assumptions offers false precision. Predictive maintenance is not universally the best option.
Compare a mixed plan against the current work
Use one row per failure mode rather than a score for the whole rewinder. The following worksheet records candidates, not approved maintenance instructions. Each row connects a consequence to evidence and a possible useful intervention.
| Failure and consequence | Evidence question | Candidate and value |
|---|---|---|
| Belt breaks; roll preparation stops. | Does age or use predict deterioration? | Retain required tasks; review scheduled replacement basis. |
| Bearing wear permits shaft movement; uneven winding spoils rolls. | Can inspection reveal meaningful change? | Condition review could focus investigation before unusable rolls result. |
| Board fails; rewinder stops. | No useful precursor established. | Review restoration options; prediction has no demonstrated basis. |
Compare expected work avoided with the effort and disruption added by observation, inspection, repair and upkeep. Include nuisance investigations and missed deterioration. Claimed savings need a defensible comparison with what the current plan would have done; a forecast count is not a business result.
Repair after failure is one possible strategy for some assets, requiring its own consequence review. If no maintenance task addresses an unacceptable consequence, refer the problem for engineering consideration, including redesign. More frequent observation cannot compensate for an unmanageable failure mechanism. An unestablished board precursor means the evidence is absent here, not that every board failure is inherently unpredictable.
Record the choice and what would change it
Keep STR-06 in the maintenance register as the system of record. Link the selected task to its failure mode, supporting finding, authority and review status. Store the worksheet with the approved plan so a later reviewer can distinguish a proposal from the work actually authorized.
- Propose: name the task and evidence still missing.
- Review: maintenance manager obtains relevant qualified input.
- Authorize: record the permitted task and applicable constraints.
- Revisit: compare findings with outcomes after work or changed duty.
Conventional software can track due tasks and link reviewed findings. AI can summarize those records and highlight conflicting statements, with references. Neither a generated summary nor a model score authorizes physical work, changes an interval or closes an unresolved decision.
For each open row, the maintenance manager assigns a reviewer, the missing decision evidence and a review date in STR-06. Repeated spoilage, a changed operating duty or a confirmed fault that the selected task missed should reopen the choice. An incomplete inspection remains incomplete; it is not a healthy finding.
The useful next step is a single failure-mode review with the maintenance owner. For RW-06, the belt basis, bearing inspectability and board restoration choice remain open. Resolve those questions before buying prediction. The business objective is dependable roll preparation with justified maintenance effort, judged from recorded outcomes rather than the sophistication of the strategy name.