Mean Time Between Failures: Calculate and Use MTBF

Home » Resources » Mean Time Between Failures: Calculate and Use MTBF

Estimated reading time: 11 minutes

Mean time between failures (MTBF) is the average operating time per failure of a repairable asset. For a production machine, calculate it by dividing its total operating hours by the number of equipment failures recorded over the same observation period:

MTBF = total operating time ÷ number of failures

A machine with 160 operating hours and four failures has an observed MTBF of 40 operating hours. That does not mean it will break down every 40 hours, or that servicing it just before hour 40 will prevent the next failure.

The useful question is what the pattern tells you to investigate and change. Once maintenance and production agree an intervention, its unavailable period also needs to reach the order schedule. SkyPlanner APS connects that capacity change to the work and delivery dates it affects, instead of leaving you to reconcile a maintenance list with separate machine queues.

Decide what your MTBF calculation actually measures

The division is easy. Getting the same meaning into the hours and the failure count takes more care. Before extracting a report, agree four things with the people recording the data.

Define the asset

Start with a named machine or clearly bounded system. A machining centre, its spindle and the entire machining department are different measurement boundaries.

If you measure the whole machine, a spindle failure that prevents it from performing its required work is a machine failure. Replacing the spindle component and returning the machine to service does not make the machine a non-repairable item.

A line-level measure is possible, but define a line failure as a loss of the line’s required function. Adding every component alarm together can count several records for the same line interruption.

Define a failure, not just a stop

Specify the required function and the event that counts as its loss. An equipment fault that prevents the machine from meeting its required operating condition can qualify, even if it has not completely powered off.

Planned servicing, a normal product changeover, a missing operator and waiting for material are not automatically equipment failures. Keep those losses visible, but separate them from the failure count.

Decide how to handle brief faults and repeated resets. Several alarm entries during one unresolved breakdown should not become several independent failures. A new failure after a genuine restoration is a new event. Record the original detail so you can investigate recurring faults rather than make them disappear through a reporting threshold.

Choose the exposure clock

This guide uses operating-time MTBF: hours in which the asset is actually operating within the measurement boundary. Repair time and other non-operating time do not enter the numerator. The result is not the average calendar interval from one breakdown timestamp to the next.

State what your running signal includes. A powered-on hour meter, a cycle-running signal and a staffed shift calendar can produce different totals. If your equipment’s relevant exposure is cycles or starts rather than running hours, track that measure alongside the hours instead of assuming they are interchangeable.

Keep the observation window consistent

Use the same reporting start and end for the operating hours and failure events. Count failures by their occurrence within that window; do not count a repair finishing this month as a new failure if the breakdown began last month.

Include all actual operating time within the window, including running time that does not end in a failure before the reporting cutoff. You are measuring the exposure observed during the period, not selecting only the completed runs that happened to fail. For a period’s downtime total, count only the unavailable hours inside that window; retain the full event and restoration record separately when a breakdown crosses the cutoff.

For each failure, retain the machine ID, occurrence time, restoration time, observed symptom or failure mode and relevant operating conditions. That detail makes the average explainable.

How to calculate MTBF: a manufacturing example

Imagine one repairable machining centre observed over a reporting period containing 200 staffed hours. Its time records reconcile as follows. The categories do not overlap, and the four failures are distinct equipment failures under the agreed definition.

Time category	Hours in the period	Treatment in this calculation
Planned maintenance	12	Not operating time; not a failure
Normal setup and changeovers	8	Not operating time; not a failure
Waiting for material, with the machine not running	10	Not operating time; not an equipment failure
Unavailable because of the four equipment failures	10	Not operating time; retain for downtime analysis
Actual operating time	160	MTBF numerator
Total staffed time	200	Reconciliation total, not the numerator

The calculation is:

160 operating hours ÷ 4 failures = 40 operating hours per failure

If you begin with the staffed-time total, you first remove the non-operating categories: 200 − 12 − 8 − 10 − 10 = 160 hours. If your source already reports 160 actual operating hours, use that directly. Subtracting downtime again would understate MTBF.

The example also shows why a shift calendar alone is insufficient. Dividing 200 staffed hours by four would produce 50 hours per failure, but that would include time when this machine was not operating.

Report the result with its context: “Machine MC-01, reporting period, 160 operating hours, four failures, observed MTBF 40 hours.” The numerator and count matter as much as the headline value when someone compares it with another period.

Handle zero failures and multiple machines without distorting the result

What if there were no failures?

Do not divide by zero or report an infinite MTBF as proof of perfect reliability. Write “no failures observed in 160 operating hours,” using the actual exposure in your report.

That is useful evidence, but it is not the same evidence as several years of failure-free running. A formal lower confidence bound requires a statistical model and a chosen confidence level. For an ordinary production review, preserving the zero count and exposure is more honest than inventing a point estimate.

What if several machines contribute data?

For a justified group of comparable machines, use total machine operating hours ÷ total failures. Do not divide one shared calendar period by all the machines’ failures, and do not take an unweighted average of their individual MTBF values.

For example, imagine two comparable machines under the same recording rules. One contributes 160 hours and four failures; the other contributes 240 hours and two failures. The pooled result is 400 ÷ 6, or about 66.7 operating hours per failure—not the simple average of 40 and 120, which would be 80.

Keep the individual results as well. A pooled figure can hide the machine causing most of the production trouble. Different equipment designs, ages or duties may warrant separate groups rather than a single factory-wide reliability score.

Interpret MTBF before choosing the action

A higher observed MTBF means fewer counted failures per operating hour within the stated boundary. Whether it represents an improvement depends on what else changed.

Compare like with like

Compare periods with consistent failure definitions, exposure clocks and asset boundaries. Then check the work being done. A machine running a lighter product mix under different loads is not necessarily evidence that a maintenance change worked.

Show operating exposure and failure count alongside the trend. One short period with a single event can move the average sharply. Longer observations help, but combining years of different operating conditions can conceal a recent deterioration. Review the recent pattern and the conditions behind it, not just the accumulated average.

There is no universal “good MTBF” for every production machine. A useful target reflects comparable history, required duty and the consequences of failure. A high value on an easily bypassed machine may matter less to delivery performance than frequent interruptions on the only machine capable of a critical operation.

Do not turn the average into a countdown

An MTBF of 40 hours does not identify the next failure time or the machine’s remaining life. Failures can cluster, and some operating runs can be much longer than the average.

Nor does MTBF alone justify a preventive-maintenance interval. The right action depends on the failure mechanism, condition evidence, equipment guidance and the consequence of failure. Age-related wear and intermittent contamination, for example, need different investigations; an overall average cannot tell you which one you have.

Use a worsening result to locate the events and ask a more specific question: which failure mode is recurring, under which conditions, and what change would address it? Then review whether that problem returns after the intervention under comparable operating conditions.

Read failure frequency together with lost time

MTBF does not tell you how long production remains unavailable after a failure. Two machines with the same MTBF can have very different effects on the order book if one returns quickly and the other waits for a specialist or part.

Review total failure-related downtime and the longer individual interruptions alongside the average. Also follow the affected work downstream. A short interruption just before a critical operation’s only available slot can matter more to a delivery than a longer stop with ample recovery time.

MTBF vs MTTR vs MTTF

These measures answer different questions. Keep the item boundary and the time definition visible rather than treating the acronyms as interchangeable scores.

Measure	Question it answers	Practical distinction
MTBF: mean time between failures	How much operating time did a repairable asset accumulate per failure?	The machine returns to service after repair.
MTTR: mean time to repair or restore	How long does the defined repair or restoration take on average?	State whether the clock includes waiting, diagnosis, testing and restart; reports use different boundaries.
MTTF: mean time to failure	How long does a non-repairable item last before failure?	Relevant to an item replaced rather than repaired, not automatically to the whole machine containing it.

For production planning, the full unavailable period matters even when a repair report counts only hands-on work. A component may take little time to replace, yet the machine may remain unavailable much longer while the team obtains it and completes the checks needed to return to production.

That distinction prevents a common planning error: treating the average repair duration as a promise that every future breakdown will be resolved within that time.

Turn the reliability finding into a production decision with SkyPlanner

Suppose the investigation leads maintenance to agree a three-hour intervention on the machining centre next Wednesday. This is a chosen maintenance window based on the technical decision—not a failure date predicted from its 40-hour MTBF.

That agreement changes the capacity available to customer orders. If it stays only in the maintenance calendar, the production plan can continue assigning work to a machine that will not be available. The planner then discovers the conflict while trying to protect the week’s deliveries.

In SkyPlanner, record the agreed time range as a workstation exception. The exception makes the resource unavailable for that period. Its scheduling AI, Arcturus, can then recalculate schedulable operations around the changed capacity, using the configured workstations, shifts, operation durations, dependencies, priorities and locks. The Gantt shows the resulting sequence.

For the Wednesday intervention, review whether the waiting machining operations still leave time for their following inspection and assembly work. If an eligible alternative takes a job, check the commitments already using that resource too. A free slot is valuable only when the operation can genuinely run there and the remaining route still fits.

This is why SkyPlanner is a useful next step once reliability work changes machine availability. You retain the technical and commercial decisions; Arcturus handles the connected scheduling calculation. Less time goes into rebuilding queues, and production and sales gain a common basis for the delivery decisions that remain.

Keep the maintenance window current if its agreed timing changes. Recalculate and review the affected commitments rather than carrying an outdated availability assumption into the next shift.

See what one machine’s availability means for your deliveries

A useful MTBF review ends with a clearer action, not just a number. Define the evidence, investigate the failures that matter, and put the agreed capacity change into the plan that production actually follows.

Choose SkyPlanner when you need to coordinate that change across customer orders without manually repairing every connected machine queue. SkyPlanner integrates with any ERP or MES through its open REST API, so the scheduling improvement can build on your existing order and production information.

Book a SkyPlanner demo with one machine, its agreed unavailable period and the orders passing through it. Ask to see which operations move, what happens downstream and which delivery promises need a decision. If you prefer to explore first, try SkyPlanner with demo data and see how a capacity change reshapes the schedule.