MTTF (mean time to failure): how to calculate it, and what it can't tell you

MTTF (mean time to failure) is the average operating time a non-repairable part runs before it fails. To calculate it, add up the operating hours of every unit you're tracking, including the ones still running, and divide by the number of failures. It's used for parts you replace rather than repair: bearings, seals, belts, fuses.

MTTF is one of the simplest reliability numbers, and one of the most often miscalculated. Plenty of guides divide by the number of units instead of the number of failures, and plenty treat the result as the age at which to replace a part. Both mistakes lead to bad decisions on the plant floor.

This guide covers the formula, a worked example with real-looking plant numbers, how to get MTTF from a failure rate or a datasheet FIT value, and what MTTF can't tell you.

What MTTF means

MTTF applies to non-repairable items. When one fails, you don't fix it, you remove it and fit a new one. Typical examples in a plant:

• rolling-element bearings
• mechanical seals and O-rings
• V-belts and timing belts
• fuses, contactor coils and solenoid coils
• proximity sensors, lamps and filter elements

The machine those parts sit in is a repairable asset. When a conveyor stops because a bearing seized, the conveyor gets repaired and goes back into service. For the conveyor you track MTBF (mean time between failures). For the bearing, you track MTTF.

So one machine has one MTBF, and each replaceable part inside it has its own MTTF.

How to calculate MTTF

MTTF = Total operating hours of all units ÷ Number of failures

Two details matter:

1. Total operating hours includes the units that haven't failed. A bearing that has run 5,000 hours without failing is evidence about how long bearings last. Leave it out and your MTTF comes out far too low.
2. You divide by failures, not by units. The two are only the same number when every unit you're tracking has already failed, which is true in a lab life test run to the end and almost never true in a working plant.

Worked example: 12 conveyor drive bearings

Illustrative numbers. A line has 12 identical drive bearings, all fitted new at the same time. The line runs about 5,000 hours a year. You follow those 12 bearings for one year:

Total operating hours = 1,400 + 2,600 + 3,500 + 4,500 + (8 × 5,000) = 12,000 + 40,000 = 52,000 hours.

Number of failures = 4.

MTTF = 52,000 hours ÷ 4 failures = 13,000 hours

Now compare the two common shortcuts:

The shortcuts understate the bearing's life by a factor of three or four. If you used 3,000 hours to plan spares, you'd buy roughly four times as many bearings as you need.

Two honest caveats about the correct answer:

• Four failures is a small sample. Treat 13,000 hours as a first estimate and update it as more failures come in. One extra early failure would move it a lot.
• The formula assumes a roughly constant failure rate. That's the standard assumption behind MTTF, and it's exact when failures are random. If these bearings are clearly wearing out (failures bunching up later in life), the average still describes them, but it hides the shape. More on that below.

When a failed bearing is replaced, the new one becomes another unit to track. Its hours count toward the total from the day it's fitted.

The common MTTF formula mistake

You'll often see MTTF written as total operating hours ÷ number of units (or "number of assets"). That version only works for a batch where every unit has run to failure, like a component life test carried through to the last unit.

In a plant you're nearly always looking at a mix: some parts have failed, most are still running. Divide by units in that situation and you mix two different things: the hours of the parts that failed and the hours of the parts that are still fine. The answer drifts depending on how many units happen to be in the group, which is not a property of the part.

The definition used in reliability engineering is time per failure. Texas Instruments' application note on MTTF, failure rate and life testing defines MTTF as the average time, in device-hours, per failure observed. Divide by failures and the problem goes away.

MTTF vs MTBF vs MTTR

MTTF and MTBF use the same arithmetic. The difference is what you're counting. For MTTF, each failure ends that unit's life. For MTBF, the asset comes back and keeps accumulating hours. Our MTTR vs MTBF guide covers how the two repairable-asset metrics work together, and MTTR has its own deep dive.

How to calculate MTTF from a failure rate or FIT

Equipment and component datasheets often give a failure rate instead of an MTTF. If the failure rate is constant, the conversion is simple:

MTTF = 1 ÷ failure rate (λ)

Example: failure rate. Say the supplier's data gives a solenoid coil a failure rate of 40 failures per million hours, which is 0.00004 failures per hour. MTTF = 1 ÷ 0.00004 = 25,000 hours.

Electronics are often rated in FIT (failures in time): failures per billion (10⁹) device-hours. The TI note above describes FIT as the failure rate scaled to failures per billion device-hours.

MTTF (hours) = 1,000,000,000 ÷ FIT

Example: FIT. Say a drive control board is rated at 500 FIT. MTTF = 1,000,000,000 ÷ 500 = 2,000,000 hours, about 228 years of continuous running.

That doesn't mean one board lasts 228 years. It's a population figure. If the plant has 40 of those boards running around the clock (8,760 hours a year), you'd expect about 40 × 8,760 ÷ 2,000,000 ≈ 0.18 failures a year, roughly one every five to six years across the whole fleet.

One caution: datasheet failure rates and FIT values are measured or predicted under stated conditions (temperature, voltage, duty) and cover the part's useful life, before wear-out. Heat, vibration, dust and power quality on your floor can make real life much shorter. Use them for comparison and rough planning, and let your own failure records overrule them.

What MTTF doesn't tell you

It's an average, not a guaranteed life

This is where MTTF gets misused most. A part with an MTTF of 13,000 hours does not typically run 13,000 hours and then fail.

If failures are random (a constant failure rate, which is the assumption behind the formula), about 63% of parts fail before reaching the MTTF. Only about 37% make it past. Half have already failed by about 69% of the MTTF (the median life). That's standard exponential-distribution math: the chance of failing by time t is 1 − e^(−t/MTTF), and at t = MTTF that's 1 − e⁻¹ ≈ 0.632.

So "replace it before it reaches its MTTF" doesn't prevent most failures. And for a part that fails randomly, replacing it early doesn't help at all: a new part is just as likely to fail next month as the old one was.

Whether to replace on a schedule depends on the failure pattern

The useful question isn't "what's the MTTF?" but "does this part wear out?" Parts fail in different patterns over their life:

• Wear-out: failures are rare at first, then climb with age. Belts, brushes, some seals and anything that erodes or fatigues. A scheduled replacement can work here, if you know the age where failures start to rise.
• Random: the chance of failure is about the same at any age. Much of the electronics and many complex components behave this way. Scheduled replacement adds cost without reducing failures. Condition checks, or a spare on the shelf, are better answers.
• Early-life (infant mortality): failures are highest just after installation, often from fitting errors, contamination or a bad batch. Here a scheduled replacement makes things worse, because every replacement restarts the risky period.

Our reliability-centered maintenance guide covers these patterns and how RCM picks the right task for each. The short version: MTTF can't tell you which pattern you have. Two parts with the same MTTF can need completely different maintenance.

For a wear-out part, base the replacement interval on the age where failures start in your own history, with some margin, not on the mean. Reliability engineers use Weibull analysis to see the shape of the failure data: its shape parameter (beta) comes out below 1 for early-life failures, around 1 for random failures, and above 1 for wear-out. You don't need to run Weibull yourself to use the idea. Even a simple chart of "age at failure" for one part type will show you whether failures bunch up late or are spread evenly.

Bearings are rated differently

Bearing manufacturers don't usually publish an MTTF. They publish a rating life (L10), defined in ISO 281 as the life that 90% of a large group of identical bearings is expected to reach or exceed under a given load and speed. It's a different statistic from MTTF (a 10% failure point, not an average), so don't compare the two directly.

How maintenance teams actually use MTTF

MTTF is still worth tracking. It's good at answering population questions:

• Spare-parts stocking. Expected failures per year ≈ number of units installed × operating hours per year ÷ MTTF. For the 12 bearings above: 12 × 5,000 ÷ 13,000 ≈ 4.6 failures a year. That, plus the supplier's lead time, tells you how many to keep on the shelf.
• Comparing suppliers or part types. Two brands of mechanical seal on the same duty, over the same period, compared on MTTF. Keep the duty the same, or the comparison means nothing.
• Spotting a problem early. If a part that usually lasts around 13,000 hours starts failing at 2,000, something changed: the supply batch, the installation method, the lubrication, or the load. MTTF moving sharply is the signal to do a root-cause analysis.
• Feeding an RCM or PM review. Your own MTTF figures and failure ages are the evidence behind every interval in your preventive maintenance schedule. They're also part of a wider maintenance KPI set.

Where a CMMS helps

MTTF is only as good as the failure records behind it. You need two things per part type: a count of failures, and the hours each unit ran. The second usually comes from production or runtime records. The first is where most plants lose data, because failures get fixed and never written down consistently.

In MachDatum, every breakdown is a work order against the asset it happened on. Work order workflows can require a failure mode and root cause before a job moves on, so "bearing seized" is recorded the same way every time instead of as free text. Each replacement is logged in the asset's history, so you can count failures per part type and see when each one happened. That gives you the failure count for the formula, and the failure ages you need to see whether a part is wearing out or failing at random.

We're onboarding our first group of manufacturing teams now. If your failure history lives in notebooks and memory, see how MachDatum CMMS handles assets and work orders.