Predictive, preventive, corrective: running the numbers for a mid-sized plant

Every predictive maintenance presentation comes with a savings figure. Almost none of them apply to your plant, because the math depends on three things only you know: what an hour of line downtime costs, how often each machine fails, and what the part plus labor costs. This article has no market statistics. It gives you the formulas to build your own comparison of corrective, preventive and predictive maintenance with your plant's data.

It is written for the manager who has to justify a budget, and for the owner who wants to know whether the argument holds up.

The three numbers everything rests on

1. Cost of an hour of downtime (Ch). Add up what is lost per hour when the machine stops production: the contribution margin of the lost output (not revenue: what is left after the inputs you also stop consuming), idle labor, contractual penalties for late delivery, rework of the material that was in process. If the machine has a standby unit or the line has buffer stock, Ch is lower, sometimes close to zero for a few hours. Do this calculation per machine, not for the whole plant.

2. Mean time between failures (MTBF). For each critical machine, pull the work order history for the last two or three years and count the failures that stopped the machine. MTBF is operating hours divided by the number of failures. If there is no history, the plant's first maintenance action is to start recording it.

3. Cost of repair (Cr). Parts plus labor plus freight. Separate the planned repair (part in stock, crew on regular hours, shutdown agreed with production) from the emergency repair (air freight, overtime, outside contractors, and the part that broke along with it). The difference between the two is, on its own, a good part of the argument.

The corrective math

Corrective maintenance means running to failure. The annual cost of one machine is:

Annual cost = (operating hours per year ÷ MTBF) × (emergency Cr + unplanned downtime hours × Ch)

Unplanned downtime hours include diagnosis, waiting for the part and mobilization, not just the repair itself. A pump that takes two hours to get a new bearing can sit idle for two days waiting for the bearing to arrive. That number, not the workshop time, is what multiplies Ch.

Then add the cascading damage: a bearing that disintegrates takes the shaft with it, the shaft takes the seal, and the parts bill triples. Put that into the emergency Cr based on what has already happened at your plant.

The preventive math

Preventive maintenance means replacing at a fixed interval (T hours), whether the part is still good or not.

Annual cost = (operating hours ÷ T) × (planned Cr + planned downtime hours × Ch) + residual failures × cost of one corrective event

Two hidden terms:

• Discarded service life. If the component's actual average life is L and you replace it at T, the fraction of life thrown away is (L − T) ÷ L. A component that lasts 30,000 hours on average, replaced every 8,000, wastes more than 70% of its life at each replacement. That is money in parts and in downtime.
• Residual failures. A fixed interval does not eliminate failures: the scatter in service life means some parts break before T. And every intervention carries the risk of introducing a defect (a badly fitted bearing, lost alignment). Your plant's history tells you how many failures keep happening despite preventive maintenance; use that number.

The predictive math

Predictive maintenance means intervening when measurements show wear.

Annual cost = monitoring cost + condition-based replacements × (planned Cr + planned downtime hours × Ch) + undetected failures × cost of one corrective event

Where:

• Monitoring cost is what you pay per year for sensors, gateway and platform, plus the time someone spends looking at the alerts.
• Condition-based replacements tend toward the number of real failures the machine would have had (hours ÷ MTBF), because you replace what was genuinely about to fail, at the end of its actual service life. Only planned.
• Undetected failures do exist. No system catches everything: sudden failures, defects outside the sensor's range, ignored alerts. Be honest with this term. A system that warns based on trends anticipates progressive wear (bearings, misalignment, looseness), not the bolt that works loose.

Side by side

Build a table for each critical machine with the three rows. What usually decides it:

• If Ch is high and unplanned downtime is long (imported parts, equipment with no standby), corrective loses by a wide margin and predictive beats preventive on avoided downtime.
• If Ch is low (machine with an installed standby, line with buffer stock), predictive can still pay off through recovered service life and planned repairs, but the return is smaller and it makes sense to start with the machines where the math is obvious.
• If MTBF is high and the part is cheap, the machine may need nothing more than inspection. Predictive on everything is waste; predictive where it hurts is an investment.

Simple payback is:

Payback (years) = initial investment ÷ (current annual cost − annual cost with predictive)

An example with made-up numbers

Just to show the mechanics, with round values that come from no real plant: a machine with no standby, Ch of 5,000 (in your currency), MTBF of 6,000 hours, 6,000 operating hours per year (one failure per year), 16 hours of emergency downtime, emergency Cr of 12,000 and planned Cr of 6,000, 4 hours of planned downtime.

• Corrective: 1 × (12,000 + 16 × 5,000) = 92,000 per year.
• Predictive: monitoring (say 4,000 per year for this machine) + 1 × (6,000 + 4 × 5,000) = 30,000 per year.

Difference: 62,000 per year, on a single machine. With Ch of 300 instead of 5,000, the difference would drop below 10,000. Replace every number with your own.

What to do now

1. Pick the five machines whose breakdowns hurt the most over the last two years.
2. For each one, work out Ch, MTBF from the history, and both repair costs.
3. Fill in the three formulas. Where a number does not exist, note it as an estimate and mark it in a different color.
4. Compare with an actual monitoring proposal, machine by machine.
5. Start with the two where the difference is largest, and use the first year to calibrate the math with real data.

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

---

Sources: EN 13306 (Maintenance — Maintenance terminology) for the maintenance terms; IEC 60300-3-3 (Dependability management — Life cycle costing) for the cost structure over the life cycle. The formulas come from maintenance cost accounting and do not depend on market statistics.