Measurement and indicators

MTBF and MTTR

In French: MTBF et MTTR

MTBF is the mean time between two failures, MTTR the mean time to restore service. The first measures reliability, the second the ability to repair — and confusing them leads to treating the wrong problem.

Two quantities, two levers

Mean time between failures is calculated by dividing the cumulative operating time by the number of failures that occurred over the period. It rises when the equipment breaks down less often, which is a matter of design, upkeep and operating conditions.

Mean time to repair is calculated by dividing the cumulative downtime by that same number of failures. It falls when repairs are faster, which is a matter of maintenance organisation, parts available and ease of access to the components.

The two combine into an availability, but they do not call for the same actions. A machine that rarely breaks down for long periods and a machine that stops often and briefly can show the same availability while posing opposite problems.

The definitions that make the figures diverge

Everything depends on what is called a failure. Minor stoppages, adjustments, changeovers are counted in some workshops and not in others — and two sites of the same group then become incomparable without anyone noticing.

Everything also depends on the boundaries of the repair time. Is it counted from the breakdown or from the arrival of the technician? Until restart or until the return to nominal rate? Each choice is defensible, and a gap in definition produces variations larger than all the improvements of a year.

The practical rule is therefore to write the definitions before publishing the indicators, and to compare only periods during which they did not change. A spectacular improvement often coincides with a silent redefinition.

What the mean hides

Time to repair is almost always very asymmetric: many short interventions and a few very long ones. Its mean then describes no real intervention, and it is the tail of the distribution that disrupts production.

The list of failures, for its part, is more instructive than its mean. Classified by cause and by frequency, the failures immediately point to the two or three subjects that weigh — which a single indicator, however well calculated, will never do.

Finally, a mean time between failures calculated over too short a period is unstable: two breakdowns more or fewer make it vary sharply. It is an indicator of annual trend, not a weekly steering indicator.

The two mean times, and availability

MTBF = cumulative operating time ÷ number of failures

MTTR = cumulative downtime ÷ number of failures

availability = MTBF ÷ (MTBF + MTTR)

The availability obtained this way takes only failures into account: it ignores planned stoppages, changeovers and rate reductions. That is why it is not directly comparable with the availability rate used in the OEE calculation, whose scope is wider.

The common mistakes

Comparing figures drawn from different definitions. It is the dominant flaw of these two indicators, and it survives every reorganisation because nobody rereads the definitions.

Steering on availability alone. It mixes two distinct problems and lets the easier one to treat be chosen — almost always repair speed, whereas breakdown frequency is what costs.

Using mean time between failures as a weekly indicator. Over a week, it varies mostly at the whim of chance and makes people react to noise.

On the shop floor — Paper mill

The availability of a line had improved by several points in a half-year, entirely thanks to a fall in repair time obtained through an advanced parts stock. The number of breakdowns, for its part, had risen. The summary indicator was improving while the equipment was degrading, and nobody was looking at the two quantities separately.

Frequently asked questions

What is the difference between MTBF and MTTF?

Mean time between failures applies to repairable equipment, where the intervals between successive breakdowns are counted. Mean time to failure applies to a component that is replaced instead of repaired: there is then only one lifetime per unit. The confusion is frequent and makes comparisons between sources inconsistent.

Should the aim be to increase MTBF or to reduce MTTR?

Reducing repair time gives faster results and costs less, which explains why one almost always starts there. But only increasing the time between failures treats the cause: equipment that breaks down as often as before keeps disrupting production, even if repaired fast. Both are pursued, in that order of priority.

Over what period should these indicators be calculated?

Over a period containing enough failures for the mean to mean something — at least about ten, which often represents several months on reliable equipment. Calculated over a week, they vary at the whim of chance and provoke reactions to variations that are not variations.

Do these indicators apply outside production?

Yes, and they are commonly used for IT services and the technical equipment of a building. The vocabulary changes — incident rather than breakdown, recovery rather than repair — but the two quantities and their traps remain identical, starting with the sensitivity of the figures to the choice of measurement boundaries.