Maintenance costs manufacturing plants 10–30% of their operating budget, and the way that money is spent — reactively, preventively, or predictively — shapes reliability, safety, and production output. The three classic strategies are not a ladder of sophistication but a portfolio to be applied where each fits: reactive for cheap, non-critical assets; preventive (PM) for assets with known wear patterns; predictive (PdM) for critical assets whose condition can be measured; and RCM as the methodology that decides the mix rationally.
The Strategy Definitions
| Strategy | Principle | When it fits |
|---|---|---|
| Reactive (run-to-failure) | Repair after failure | Low failure cost, no safety impact, no production impact, short repair time. |
| Preventive (time/usage-based) | Replace or service at fixed intervals before expected failure | Known wear-out pattern (bearings, seals, filters), failure is age-correlated. |
| Predictive (condition-based) | Maintain when condition data says it is needed | Critical assets with measurable degradation (vibration, temperature, oil analysis) and random-ish failure times. |
| RCM (reliability-centered) | Select strategies per failure mode based on consequence | The plant-wide decision framework itself. |
The industry's rough distribution target: 10–20% reactive, 40–50% preventive, 30–40% predictive for mature programs — but the correct mix is a consequence-driven calculation, not a ratio target.
Preventive Maintenance Done Right
PM fails when it becomes a calendar ritual: tasks performed without regard to condition, intervals set by vendor default, and work orders closed by signature. Effective PM has:
- Task validity — each PM task must prevent, detect, or delay a specific failure mode; if the task cannot be justified, it is waste (and its "completion" is theater).
- Data-capturing tasks — every PM includes measurements and observations (vibration readings, temperatures, visual findings) recorded in the CMMS, turning PM rounds into a data stream for PdM analysis.
- Interval optimization — intervals set from failure history and condition data, and adjusted as data accumulates; fixed forever intervals are a symptom of a static program.
- Findings handling — a defect found during PM is a work order with priority, not a note in a drawer; the loop from PM finding to scheduled repair is what makes PM effective.
Predictive Maintenance: From Vibration to Data
PdM detects the early stages of failure. The mainstream techniques: vibration analysis (bearings, unbalance, misalignment, resonance), oil analysis (wear metals, contamination), thermography (electrical and thermal faults), motor current analysis, and ultrasonic (leaks, early bearing damage). The deployment pattern that works:
- Route-based collection — portable data collectors on a defined route per period, building the baseline; then fixed online sensors for the most critical assets.
- Baselines and trends — the value is in the trend: a bearing that rises slowly for months, then accelerates, is the PdM success story. One reading means nothing; twelve readings mean everything.
- Alarm bands — set alerts on rate-of-change and level with asset-specific thresholds; generic factory defaults generate alarm fatigue.
- Integration with planning — a PdM alert must become a planned repair at the next window — the full value is captured only when the condition data changes the schedule.
Modern PdM adds machine learning on the historian data (see the AI in manufacturing article): models trained on normal operation flag deviations the human analyst would miss — but the human and the process remain the core.
RCM: The Decision Framework
RCM (SAE JA1011) asks seven questions per asset: functions and performance standards; failure modes; failure causes; failure effects; consequences (hidden, safety, environmental, operational, non-operational); and the applicable, effective task for each consequence. The output is a per-failure-mode decision: scheduled restoration, scheduled discard, condition-based task, failure-finding task (for hidden failures), or redesign. RCM is an analysis, not a schedule — it produces the rationale for the PM/PdM mix, and it is worth doing once, deeply, for the critical assets, rather than superficially for everything.
Making the Program Work
- Criticality first — score assets by consequence (production, safety, environment, repair cost); spend the analysis effort where the score is high.
- CMMS as the spine — the maintenance management system holds the asset register, the tasks, the history, and the KPIs; without it the program is anecdotes.
- KPIs that measure effectiveness — not just PM compliance (a compliance number can be perfect while the program is useless) but: breakdown frequency, MTBF/MTTR, maintenance cost per asset, and the trend of PdM findings converted to scheduled repairs.
- Continuous review — failure data revises tasks and intervals; the review meeting (monthly, per area) is where the strategy adapts.
Summary
Maintenance strategy is a portfolio decision: reactive for the trivial, preventive for the wear-out, predictive for the critical, and RCM as the method that chooses rationally by consequence. Make PM tasks justified and data-capturing, make PdM trend-based and schedule-changing, and run the whole program on the CMMS with effectiveness KPIs. The goal is not to eliminate failures — it is to decide, deliberately, which failures are affordable.