Predict how long repairs will take

A maintainability prediction answers one question early, while a design can still change: how long will repairs take? Not for one part, but for the whole system, weighted by what actually fails. It is the maintainability twin of a reliability prediction: that one adds up how often things fail, this one adds up how long they take to put right.

You get three numbers and a list:

• Mct, the mean corrective maintenance time: the average repair, over all the failures the system will have.
• M_max, the time by which most repairs (95% by default) are done: the figure a support contract or a turnaround target is written against.
• Inherent availability, MTBF ÷ (MTBF + Mct): the share of time the system is up, counting only repair time, with no waiting for parts or people.
• What takes the time: the lines, largest share first, so you know which repairs to make quicker, or which parts to make fail less often.

Start from a parts list

Open Modelling → Maintainability prediction and press New prediction. The lines table is the parts list: one row per thing that can fail.

• With Reliafy Maintenance, Start from a product fills the rows from the product's parts list: every lowest-level part with its part number, how many go into one product and the assembly it sits in. The times are yours to add.
• Paste a parts list takes rows from a spreadsheet: part number, name, quantity, or just names. A header row names the columns, so a sheet with failure rate, MTTR or the elements fills those too.
• Or type the rows. A blank row is always waiting at the end.

Quantity is how many of that line are in one system. Four motors are one row with quantity 4.

Enter each line's repair time

Each line takes one MTTR (its mean time to repair), in the unit chosen at the top (Repair times in: minutes or hours). If you know how a repair breaks down, tick Split repair times into elements and enter the elements the way handbooks split them, so they can be compared and the slow ones found:

Element	What it covers
Isolate	Finding what failed: diagnostics, fault isolation.
Access	Getting to it: covers off, disassembly.
Replace	Taking the failed part out and fitting the new one, or repairing it in place.
Restore	Putting everything back together.
Check out	Proving it works: tests, alignment, calibration.

The line's MTTR is then the sum of its elements. Every time in the table and in the results is in the unit chosen at the top.

Most done by is optional: the time 95% of that line's repairs are done by, say twice its MTTR when repairs vary a lot, 1.3× when they're routine. It is the line's spread, and it's what makes M_max possible (see below).

Give each line a failure rate

Each line counts by how often it's the one being repaired: its failure rate × quantity. A part that fails ten times as often weighs ten times as much in the average, however long or short its repair.

• Type a failure rate per line, in the unit chosen at the top (Failure rates in: per million hours by default, or per hour).
• With a product chosen, a saved reliability prediction for that product supplies the rates where a line's part number or name matches, and the result says so.
• With no rates at all, every line counts the same. The result says that too, and availability isn't available: it needs the system's MTBF, which comes from the rates (1 ÷ their sum, times quantities) or from the System MTBF box.

Read the answer

Press Compute. The answer comes first, in one sentence, then the tiles: Mct, M_max, inherent availability and the MTBF behind it.

Below it:

• Where the repair time goes: each element's share of Mct across every line. If most of it is Access, the design needs better access, not faster technicians.
• What takes the time: the lines, largest share first, with the running total. The top few lines are where a shorter repair, or a more reliable part, brings Mct down most.
• By assembly, when the lines name their assemblies (Propulsion › Arm 1 counts under Propulsion and under Arm 1): each assembly's own Mct and share.
• Every line's figures, folded away: the elements, MTTR, spread and M_max per line.

Numbers are shown to three or four figures with their unit. Open the sample, Survey quadcopter — maintainability prediction (sample), to see a finished one. With Reliafy Maintenance, adding the Survey quadcopter on Products also saves the same prediction among your own, as Survey quadcopter — maintainability, attached to that product.

How the numbers are worked out

In plain words: the average repair is the average of the lines' repair times, each counted as often as that line fails.

Mct = Σ(λ · q · MTTR) ÷ Σ(λ · q)

where λ is the line's failure rate, q its quantity and MTTR the sum of its elements. With no failure rates, λ is 1 for every line.

M_max is the time by which 95% of repairs are done (change the percentage in Most repairs done by). Each line's repair time is taken as lognormal, the usual shape for repair times: most are near the typical value, a few take much longer. Its mean is the line's MTTR and its spread comes from Most done by, the 95% point you gave. The system's repair time is the mix of the lines' distributions, each weighted by λ · q, and M_max is the 95% point of that mix, found numerically.

If no line has a spread, M_max isn't available and the result says so; Mct still is. If some lines have one and others don't, the others count as always taking exactly their MTTR, which makes M_max a little low: the result says how many.

Inherent availability is

A_i = MTBF ÷ (MTBF + Mct)

with both in hours. MTBF is 1 ÷ the summed failure rates (× quantities), or the System MTBF you typed.

What it leaves out

• Logistics and administrative delay. Waiting for a spare, a technician or a permit isn't repair time. Inherent availability is the best the design can do; operational availability is lower.
• Preventive maintenance. Only corrective repairs are counted. Planned work belongs in a replacement interval or an RBD's maintenance policy.
• Redundancy. A failed part in a redundant pair may not take the system down. For availability with redundancy, model it in a repairable block diagram.
• Guessed spreads. M_max is only as good as the "most done by" figures. Where you have repair records, fit them: the lognormal's spread (σ) is typically 0.3 to 1.0 for repair times.

Save it

Press Save prediction under the answer and give it a name. The lines are kept with the result, recomputed on the server when saved, and the prediction appears on the Maintainability prediction page. Open one and press Edit a copy to change its lines and save a new version.

The Free plan keeps 1 saved maintainability prediction (the shared sample doesn't count); computing without saving has no limit. To save another, delete the one you have or upgrade to Pro, which removes the limit.

From an agent or the command line

The same calculation is the analyse tool's maintainability action (and change maintainability_save to keep it); over REST it's POST /api/v1/analyse/maintainability, and on the command line reliafy analyse maintainability --lines '[…]'. See use the API and command line.