RAMSynapse
Log inSign up

Maintainability Prediction · Chapter 3

The Method

How the analysis actually runs, step by step.

The procedure is short and its inputs are not. Most of the effort is in deciding what a repair actually consists of, and most of the error is in deciding it optimistically.

1. Fix the maintenance concept first

Nothing can be estimated until four things are settled, and they are programme decisions rather than analysis ones:

DecisionWhy the estimate depends on it
The level of repairReplacing a card and replacing the cabinet it sits in are different tasks
Who repairs, and with what trainingThe same task has different times for a technician and a specialist
Where, and with what accessOn the vehicle, in a workshop, at a depot: different disassembly entirely
What counts as completePowered up and checked out, or swapped and released

2. Choose the procedure the design can support

MIL-HDBK-472 offers five, and the choice is a statement about how much design exists. A task-by-task synthesis on a block diagram is false precision; an apportionment on a detailed design throws away what is known. Record which procedure was used, because two predictions produced by different procedures are not comparable.

3. List the repairs, from the items that fail

The repair list is the replaceable-item list, with a failure rate on each from the prediction. Where one item has genuinely different repairs depending on the mode, which the FMECA will already have identified, it becomes two rows with the mode's share of the rate.

4. Time each elemental activity

The seven activities, per repair, from a task-time library, a comparable design's measured times, or a walk-through on a mock-up. Two disciplines: estimate the activities separately rather than the total, because a total is a guess and seven parts are an argument; and record the basis of each, because a time from a measured predecessor and a time from imagination should not look alike in the table.

The isolation time deserves its own sentence. It is where the testability analysis lands: a fault isolated to one item takes minutes, and one isolated to a group of three takes the swap-and-retry cycle the ambiguity implies. Taking that number from the dependency model rather than inventing it is what keeps the two analyses consistent.

5. Weight, and produce both numbers

Mct = Σ λᵢ tᵢ ÷ Σ λᵢ for the mean, and the lognormal fit on the logarithms for the percentile. Report both, with the percentile's confidence level named. A single MTTR figure answers only half of what a maintenance concept needs to know.

6. Read the contribution list, which is the actual output

The same seven items ranked by λ·t rather than by failure rate. The fan fails most often and is out in eight minutes; the power module fails a third as often and is a quarter of the cabinet's total repair time.
The same seven items ranked by λ·t rather than by failure rate. The fan fails most often and is out in eight minutes; the power module fails a third as often and is a quarter of the cabinet's total repair time.

Sort by λᵢ tᵢ, and separately sum each elemental activity across the whole equipment. Those two sorts answer different design questions:

SortThe question it answers
By item, λ·tWhich item's repair should be made easier, or which item should fail less
By activity, summedWhether the design problem is access, diagnosis, alignment or checkout

The second is the one that changes enclosures. If disassembly and reassembly dominate, no amount of built-in test will help, and the finding belongs to the mechanical designer.

7. Compare against the requirement, and act

What a perfect diagnostic would buy on the worked example: Mct from 61.7 to 57.7 minutes, a 6.5 per cent cut, because diagnosis is a fifth of the time and access is two fifths.
What a perfect diagnostic would buy on the worked example: Mct from 61.7 to 57.7 minutes, a 6.5 per cent cut, because diagnosis is a fifth of the time and access is two fifths.

A prediction over the requirement has four honest responses, and they are ordered by cost: change the access, change the item boundary so less has to come apart, change the diagnostic so isolation is shorter, or renegotiate the requirement with the evidence attached. Guessing lower task times is not on the list.

8. Demonstrate, then correct

The prediction is a hypothesis. MIL-STD-471 turns it into evidence: a sample of induced faults, timed on hardware, by people representative of the ones who will do it. The comparison is worth as much for the task-time library as for the programme, because a library corrected by one demonstration predicts the next design better.


Want to see this on a live system model? Request a walkthrough.