A maintainability prediction is read by four people who want four different things from it, and only one of them wants the headline number.
The two figures, stated together
| Figure | What it sizes |
|---|---|
Mct | The maintenance burden: technician hours per operating hour, and the labour estimate built on it |
Mmax(90) or Mmax(95) | The slot: how long a repair may take before the plan breaks |
Quoting one without the other leaves the reader to assume the distribution, and the assumption is always wrong in the same direction, because repair times are skewed and people imagine them symmetric. The worked example's percentile is 1.9 times its mean; a maintenance window sized on the mean is overrun by more than a third of the work.
State with both: the level of repair, the maintenance concept they assume, and the procedure used. Two predictions that differ in any of the three are not comparable and are routinely compared anyway.
The contribution list, which is what a designer can act on
Sorted by λ·t, and sorted again by activity. The first list names the items whose repair should be made easier or whose failure rate should come down; the second names the design discipline that owns the problem.
| If the activity sum is dominated by | The finding belongs to |
|---|---|
| Disassembly and reassembly | Mechanical design: access, fasteners, connector placement, item boundaries |
| Isolation | Testability: the diagnostic cannot name the item |
| Alignment | Systems design: a part that needs setting up after fitting is a part designed to be fitted twice |
| Checkout | The test procedure, and whether the equipment can prove itself without a full sequence |
That table is the reason the analysis is worth running early. All four findings are cheap while the design is a drawing and expensive afterwards, and none of them is visible in a single MTTR figure.
What it feeds
| Consumer | What it takes |
|---|---|
| Availability | Mct as the repair term, with the logistic delay added separately |
| Support planning | Technician hours per operating hour, from Σλt, and the skill mix behind it |
| Spares | Removal rates per item, which the ambiguity groups can make larger than the failure rates |
| The contract | The predicted mean and percentile the demonstration will be measured against |
What the prediction does not support
- Precision. Individual task times are judgements; a factor of two on any one is unremarkable. The mean is a planning number, not a measurement, and quoting it to a tenth of a minute is a presentation error the worked example commits deliberately so the arithmetic can be followed.
- A downtime claim.
Mctis active maintenance time. What the operator experiences isMDT, which includes waiting for the part, the permit and the person, and on most fleets those exceed everything this analysis computes. - Preventive maintenance. A corrective prediction says nothing about scheduled work, which has its own task list and its own analysis.
- The technician you will actually get. Every time assumes trained, equipped and unhurried. The demonstration is where that assumption meets a night shift.
- An availability improvement. Halving
Mcton a system with an 8,000-hour MTBF moves availability in the fifth decimal place. Maintainability prediction is a labour-and-logistics tool; when it is sold as an availability lever it disappoints precisely.