Mean downtime (MDT): how long production really loses, not just how long the repair takes

Two plants run identical compressors, and both report the same average repair time: four hours. Yet one loses eight hours of production every time one goes down, and the other loses five. The repair is the same. The difference is everything that happens around it, the wait to hand the machine over, the time to isolate and lock it out, the run to the storeroom, the start-up sequence at the end.

Mean downtime is the metric that refuses to ignore all of that. It measures the full window the asset was lost, not just the stretch when someone was turning a wrench.

What it actually measures

MDT is the average total time an asset is out of service per downtime event, from the moment it becomes unavailable to the moment it is fully back in production. It sweeps in everything: the waiting and administrative delay, the logistics of parts and permits, the operations work such as lockout and cleaning, the repair itself, and the test before handback. If mean time to repair is the time the repair took, MDT is the time production actually lost, and the two are rarely the same number.

How to work it out

MDT = Total downtime (hours) / Number of downtime events

An asset has 10 downtime events. The scheduled ones run 3, 9, 15, 8 and 6 hours; the unscheduled ones 7, 14, 2, 4 and 8.

MDT = [(3 + 9 + 15 + 8 + 6) + (7 + 14 + 2 + 4 + 8)] / 10 = 76 / 10 = 7.6 hours

By pre-staging tools and spares and tightening up the lockout routine, the team trims the average event to 6.1 hours. The asset fails exactly as often as before, but the plant wins back fifteen hours of production across the year, purely by shrinking the time wrapped around the repair.

Why availability is really built on MDT

This is the quiet reason MDT matters more than it first appears. When you calculate the operational availability you actually experience, the honest version uses mean downtime, not bare repair time, because the asset is just as unavailable while it waits for a permit as while it is being fixed. In practice many plants record the entire lost window but file it under "repair time", which means a great many quoted repair times are really mean downtimes in disguise, and an availability figure built on the smaller number quietly overstates how dependable the plant is. If you want availability to tell the truth, it has to be built on the full window MDT captures.

Decompose it, and the fix appears

The real power of MDT is that it comes apart. A single seven-hour average tells you little, but split it into its pieces, time to mobilise, time to diagnose, time waiting for parts, the active repair, and the test before handback, and the biggest slice is your biggest opportunity, pointing straight at the thing to fix. A long wait-for-parts says the problem is the storeroom; a long wait-for-permit says it is the work-control process; a long mobilise says it is staffing or escalation. None of that shows up while the number stays lumped together, which is why a plant serious about winning back hours tracks not just the size of MDT but its shape.

Where it can mislead you

• MDT is not repair time. It deliberately includes all the waiting, logistics and admin around the physical work, which is usually the larger part of the total, so never confuse a tidy repair time with the hours production actually lost.
• If an asset is not needed every hour, MDT matters less, because a long outage during idle time costs you nothing. Read it in the light of how much the asset was actually wanted.
• Keep the components, not just the total. The lumped figure is a symptom; the breakdown is the diagnosis.
• Compare across sites only where the asset types and contexts genuinely match.

Mean time to repair tells you how good your people are with a wrench. Mean downtime tells you how long the plant actually stood still, and the gap between the two, all that waiting and walking and handing over, is usually where the easiest hours in the whole plant are hiding.