Predictive, preventive and corrective maintenance: the difference in practice

Three words come up in every maintenance conversation, almost always in the same order, as if they were rungs on a ladder: corrective at the bottom, preventive in the middle, predictive at the top. That is not how it works. They are three different decision regimes, and a well-run plant uses all three at the same time, each on the right machine.

This article explains what each regime is, with a shop-floor example, and gives the criteria for choosing machine by machine. If you need the comparison in money (cost of an hour of downtime, MTBF, payback), read the numbers for a mid-size plant, which has the formulas. Here the subject is concepts and decisions.

Corrective: act after the failure

Corrective maintenance is the work done after an item has stopped performing its function. It has two very different faces:

• Unplanned: the machine stopped, nobody saw it coming, and the crew scrambles. This is where the high cost lives: rushed diagnosis, parts out of stock, expedited freight, overtime and collateral damage.
• Planned, by decision: run to failure on purpose, because the failure is cheap, does not stop production and does not hurt anyone. Warehouse lighting, a dosing pump with an installed spare, the small motor on a secondary exhaust fan.

The mistake is not having corrective maintenance. It is having unplanned corrective maintenance on a machine that is expensive to stop.

Preventive: act at set intervals

Preventive maintenance is carried out at predetermined intervals or according to prescribed criteria, to reduce the probability of failure. The interval can be calendar time (every six months), operating hours (every 8,000 h), cycles (every 100,000 actuations) or production volume.

Examples: gearbox oil, compressor air filter, V-belts, heat exchanger cleaning, retightening panel terminals, and the periodic inspection of boilers and pressure vessels required by regulation.

Preventive maintenance is not wrong. It is the right choice when:

• the part is cheap and the job is quick;
• wear is predictable by time or use (lubricant that oxidizes, a filter that clogs, a seal that dries out);
• there is a legal or safety requirement with a fixed deadline;
• there is no measurable symptom before the failure.

Where it wastes: when the part you replaced still had life left. And every intervention reintroduces risk, from a badly seated bearing to lost alignment. Replacing more does not necessarily mean failing less.

Predictive: act on condition

Predictive maintenance is intervention driven by measurement. You track a parameter that changes before the failure (vibration, temperature, current, particles in the oil, ultrasound, differential pressure) and intervene when the measurement shows that wear has started, not when the calendar says so.

In the EN 13306 maintenance terminology it sits under preventive maintenance as condition-based maintenance: the goal is still to avoid the failure, but the trigger is the actual state of the item.

The gain is twofold: the part is used until close to the end of its real life, and the shutdown is planned, with the part already purchased and the crew on shift.

What Nowlan and Heap found

The technical case for predictive maintenance predates any sensor. In the reliability-centered maintenance report published in 1978 by F. Stanley Nowlan and Howard F. Heap of United Airlines, the items studied were sorted into six patterns of failure probability versus age. Only a small minority followed the classic bathtub curve, and the great majority showed no age-related wear-out zone: the chance of failure was practically the same whether the item was new or old.

The consequence is straightforward: replacing by age only reduces the chance of failure where the failure mode really is progressive wear over time. For everything else, you either measure the condition or accept the failure.

The P-F interval: why frequency matters

A progressive failure does not happen all at once. There is a point P, where the failure is already detectable by some technique, and a point F, where the machine stops performing its function. The distance between the two is the P-F interval, and it is the window you have to act.

The established rule of thumb: inspection frequency must be at most half the P-F interval. If a bearing gives eight weeks of warning before it seizes and the vibration route comes by once a quarter, the route will miss the failure most of the time.

Two things change that interval. The technique: ultrasound and oil analysis usually warn earlier than the overall vibration value, which warns before temperature, which warns before audible noise. And the measurement frequency: a monthly route gives you three points per quarter to draw the trend; readings several times a day show the slope within days. It is not that the sensor sees better than the technician with a data collector: it simply looks more often.

Five questions to decide machine by machine

1. What happens if it stops without warning? Lost production, a safety risk, environmental damage, or nothing beyond the inconvenience?
2. How does it fail? Is the dominant mode progressive (bearing wear, misalignment, fouling) or sudden (a bolt that works loose, an electronic component that fails open)?
3. Is there a measurable symptom beforehand? Vibration, heat, current, noise, particles in the oil.
4. How much do the part and the job cost?
5. Is there a legal or safety deadline? If there is, it rules, and predictive maintenance comes in as a complement.

With the answers in hand:

• Sudden failure, low consequence, cheap part: deliberate corrective maintenance. Write it into the plan, so it does not look like neglect.
• Wear predictable by time, cheap part, or a legal requirement: preventive.
• Progressive failure, high consequence, measurable symptom: predictive.
• Unacceptable consequence, for safety or the environment: the issue is not the maintenance regime, it is redundancy or redesign.

What to do now

1. List your assets and rank each one by the consequence of an unplanned stop.
2. For each one, write down the failure mode that shows up most often in the work-order history.
3. Assign the regime using the five questions, and note the rationale next to it.
4. Start measuring only where the answer was predictive. Monitoring everything is waste; monitoring what hurts is an investment.

If you want to know what your plant needs, talk to one of our engineers.

---

Sources: EN 13306, Maintenance — Maintenance terminology (definitions of corrective, preventive and condition-based maintenance); F. S. Nowlan and H. F. Heap, Reliability-Centered Maintenance, United Airlines / US DoD, 1978 (the six failure patterns); ISO 17359:2018, Condition monitoring and diagnostics of machines — General guidelines (procedure and P-F interval).