How do I calculate MTBF and MTTR from equipment logs?

Calculate an observed MTBF estimate by dividing the equipment's recorded operating hours by its qualifying failure count. Calculate mean time to repair by dividing the total time inside your explicitly defined repair or restoration boundary by the matching completed events. Keep the source log, observation window and unfinished events alongside both numbers.

The division is usually the easy part. The harder part is making sure that “operating,” “failed” and “repaired” mean the same thing on every row. Otherwise, a better-looking average may describe a changed report rather than better equipment.

Define the machine and the clock

Start with one identifiable repairable asset, or a deliberately defined comparable group. Record the operating condition, required function, observation period and failure criterion. A warning light that leaves the required function intact may belong in a different category from an event that stops the defined operation.

Do not treat the purchase date as the start of operating exposure if the machine spent part of that period stored, idle or unavailable. Equally, do not erase a genuine failure because the problem was easy to fix.

NIST's reliability handbook gives total system operating time divided by total failures as the MTBF estimate under its constant-rate model. That model condition matters. A logged ratio can describe the observed record without establishing a constant future failure rate, remaining service life or a safe maintenance interval.

MTTR needs its own boundary. Does the clock begin at failure, detection, maintenance notification or hands-on work? Does it end when the replacement part is fitted or when the machine passes the required checks and is available again? Write the answer next to the abbreviation.

IBM's MTTR explanation describes the average and notes the importance of clear starting and ending points. For the original case below, the chosen boundary is failure through verified restoration of the required function. Hands-on repair time is recorded separately.

A complete original equipment log

Consider one fictional machine observed for 1,020 elapsed hours. It is required to operate throughout that defined window. There are no excluded breaks, planned maintenance periods, idle intervals or other states. Every hour is either operating time or downtime following one of the recorded failures.

The machine is operating at the start. Four qualifying failures occur, and all four restorations are complete before the observation ends. Time is measured on one continuous elapsed-hour clock; these are teaching inputs, not field measurements or a recommended schedule.

• It operates from hour 0 to 100, fails at 100 and is verified restored at 104. Downtime is four hours.
• It operates from 104 to 304, fails at 304 and is restored at 310. Downtime is six hours.
• It operates from 310 to 610, fails at 610 and is restored at 612. Downtime is two hours.
• It operates from 612 to 762, fails at 762 and is restored at 770. Downtime is eight hours.
• It operates from 770 to 1,020 without another observed failure.

The operating intervals total 100 + 200 + 300 + 150 + 250 = 1,000 hours. The four downtime intervals total 4 + 6 + 2 + 8 = 20 hours. Together they account for all 1,020 elapsed hours.

That reconciliation is worth doing before calculating the averages. A gap or overlap in the event log can otherwise become a plausible-looking but unsupported denominator.

Calculate MTBF from operating exposure

The observed operating-hours-per-failure estimate is 1,000 ÷ 4 = 250 hours. Report it with both inputs: 1,000 operating hours and four qualifying failures during this observation.

Do not use 1,020 ÷ 4 and call the resulting 255 hours the same measure. That calculation includes the twenty hours during which the machine was down. This article's operating-time definition excludes them.

Also keep the final 250 operating hours after the last restoration. They are genuine observed exposure even though no further failure appears before cutoff. Dropping them because they do not end with a failure would remove useful evidence from the record.

The five operating intervals are not five failures. Nor should their simple average, 200 hours, replace the defined total-exposure-per-failure estimate. One interval ends at the observation cutoff with the machine still running; the event count remains four.

The 250-hour result does not tell you that the next failure will occur exactly 250 hours from now. It does not establish that the machine should be replaced at that age or that operating longer is safe. Those decisions need the equipment's actual requirements, condition, failure causes and an appropriate technical assessment.

Calculate full restoration and active repair separately

Under the stated failure-to-restoration boundary, MTTR is 20 ÷ 4 = 5 hours. The individual observations remain four, six, two and eight hours. No event in this small record lasted exactly five hours.

Now suppose the complete hands-on records are two, three, one and four hours respectively. One qualified technician performs each hands-on interval, with no overlapping technician work in this fictional record. Total hands-on time is 10 hours, averaging 2.5 hours per completed repair.

The remaining ten elapsed outage hours are recorded waiting or other time outside those hands-on intervals: two, three, one and four hours. The record supplies those categories, but it does not yet establish why the waiting occurred.

Calling 2.5 hours the same full-restoration average would hide half the observed unavailability. Calling all twenty outage hours technician labor would make a different mistake. Elapsed downtime and labor hours are different quantities, even when they appear in neighboring columns.

If several technicians work simultaneously, labor hours can exceed elapsed hours. If no one can start because a part is missing, elapsed hours can grow without matching hands-on work. Preserve both where they matter to the operating decision.

Reconcile availability without turning it into a promise

For this fully accounted observation, the operating share of the required window is 1,000 ÷ 1,020 = about 98.04%. Downtime is about 1.96%. The figures describe this machine and this completed window under the supplied all-hours-required boundary.

Using the two matching averages gives the same arithmetic: 250 ÷ (250 + 5). That agreement follows from the common four-event counts and the complete time account. It is not an independent test proving future availability.

Substitute the narrower 2.5-hour hands-on average and the expression becomes about 99.01%. That higher percentage omits the ten elapsed hours outside hands-on work. No outage ended earlier; the report simply stopped counting part of it.

Actual businesses also have idle time, planned shutdowns, parallel equipment, partial service and changing required hours. Define those states before extending this calculation. A machine-level ratio is not automatically the proportion of customer orders delivered, an operation's revenue at risk or the availability of an entire redundant system.

Keep an unfinished repair visible at cutoff

Look at a separate report cut from the same original timeline at hour 766, while the fourth repair is still open. At that point, observed operating exposure is 750 hours. The first three outages total twelve hours, and the open outage has lasted another four hours, giving sixteen observed downtime hours and 766 elapsed hours in total.

The completed-restoration average is 12 ÷ 3 = 4 hours, explicitly for the three completed events. The fourth failure has happened, but its final restoration duration is not yet known at this earlier cutoff.

Do not quietly omit that open event and announce that recovery has improved from five hours to four. The smaller figure uses a different completion set and leaves the ongoing outage unresolved. Record its start, elapsed duration so far, present obstacle and next action.

Likewise, do not pretend its four elapsed hours are a final four-hour repair. A consistent report can show completed-event statistics alongside open-event exposure without inventing an ending. More advanced reliability analysis may handle incomplete observations explicitly; a spreadsheet shortcut is not a substitute for choosing an appropriate method.

Zero observed failures needs an honest label

Suppose a separate later window records 150 operating hours and zero qualifying failures. Report those facts. The ratio 150 divided by zero does not give a finite numerical MTBF estimate.

“No failures observed in 150 operating hours” is useful evidence. “The machine never fails” is an unsupported conclusion. A short quiet window cannot establish unlimited reliability.

NIST treats the zero-failure case through model-dependent statistical bounds rather than a finite point estimate from that division. If a contract or technical decision needs such a bound, establish the model, exposure, confidence requirement and analysis with appropriate expertise. No bound or failure probability is calculated here.

Use the log to choose an investigation

Return to the longest completed outage: eight hours, including four outside hands-on work. Find the actual reason for those four hours. A missing part, delayed authorization, unavailable specialist and necessary verification step would call for different responses.

Then review the failures themselves. Repeated symptoms may suggest a common cause, but a shared label in a dropdown menu does not prove one. Preserve the observations, affected components, operating conditions, repair performed and verification result.

Compare periods only after checking exposure, product or duty changes, failure definitions, restoration boundaries and open events. A lower failure count during much lighter use is not automatically improved reliability. A different mix of equipment can also change a pooled average without improving any particular asset.

The paid-off-machine guide addresses another decision: arranging and funding scheduled service around accepted work. These reliability averages do not replace that service requirement. They add evidence about failures and restoration inside a defined operating record.

Keep a compact review sheet: asset and duty; required window; operating exposure; qualifying failures; completed restoration durations; hands-on labor; open events; source records; and the next supported investigation. The useful result is a question someone can act on, with enough evidence to know whether the eventual change helped.