Failure rate, usually written as the Greek letter lambda (λ), is how often an asset fails per unit of operating time: λ = number of failures ÷ total operating time over a defined period. A pump that failed 3 times in 6,000 operating hours has a failure rate of 0.0005 per hour. Lower is better, and during an asset's useful life λ is simply the reciprocal of MTBF.
Failure rate and MTBF are two views of the same fact. MTBF tells you the average hours between failures; λ tells you the failures per hour. Reliability engineers reach for λ because it adds up cleanly across components and converts into the tiny, comparable units the electronics world uses. This guide covers the calculation, a worked example, the FIT unit, why λ is not constant over an asset's life, and how to turn the number into maintenance decisions.
How do you calculate failure rate?
Divide the number of failures by the total operating time the population accumulated in the same period. The arithmetic is trivial; the discipline is in the definitions, exactly as it is for MTBF.
- Operating time is time the asset was running or available to run. Exclude planned downtime, changeovers, and repair time. If you count idle or repair hours, λ looks artificially low.
- A failure is a loss of the asset's intended function. Fix the definition before you count, does a two-minute jam an operator clears qualify? Whatever you decide, apply it the same way every time.
- Population time matters when you pool identical units. Twenty motors each running 1,000 hours give 20,000 unit-hours; three failures across that fleet is λ = 3 ÷ 20,000 = 0.00015 per hour. Pooling like units is how you get a stable estimate for a rare failure.
A worked example
Hypothetical: a fleet of 8 identical roller conveyors each ran 2,500 operating hours over a year, for 20,000 unit-hours combined. The gearboxes failed 4 times across the fleet. The fleet failure rate is λ = 4 ÷ 20,000 = 0.0002 per hour two failures for every 10,000 hours of running. The reciprocal, MTBF = 5,000 hours, says the average gearbox runs about five thousand hours between failures.
Two things this buys you. First, because λ is expressed per hour, you can multiply it by planned run hours to forecast work: 0.0002/h across 20,000 planned hours next year predicts about 4 gearbox failures to staff and stock for. Second, λ adds. If the same conveyor also has a drive-belt failure rate of 0.0003/h and a bearing rate of 0.0001/h, the conveyor's overall rate is roughly 0.0002 + 0.0003 + 0.0001 = 0.0006/h, the series-system sum that a single MTBF number hides.
What is FIT (failures in time)?
FIT, failures in time, is failure rate expressed in a very small unit: 1 FIT = one failure per 1,000,000,000 (10ⁿ) device-hours. It exists because electronic components are so reliable that per-hour rates would be strings of leading zeros. A capacitor rated at 20 FIT is expected to fail 20 times per billion hours of operation. The unit is standard across semiconductors and electronics, and it makes component rates addable across a board with thousands of parts.
Is failure rate constant over an asset's life?
No, and this is the trap. The reciprocal shortcut λ = 1/MTBF only holds when the failure rate is constant, which describes just the flat middle of the bathtub curve. Early in life, infant-mortality failures from bad installs and manufacturing defects push λ up. Late in life, wear-out drives it up again. Assuming a single constant λ across all three regions is how a machine sliding into wear-out still posts a comfortable lifetime average while its recent rate climbs.
Practically, compute λ over a recent, rolling window rather than the asset's whole life, and watch its trend per asset. A rising short-window λ is an early warning the lifetime number will smother. It also tells you which maintenance strategy fits: random, constant-rate failures respond to condition-based and predictive triggers, while wear-out failures with a rising rate are candidates for scheduled replacement before the knee of the curve.
How do you turn failure rate into maintenance decisions? A 5-step loop
- Standardize the failure definition and the clock. One definition of failure, operating hours only, captured the same way every time. Automatic capture from machine signals beats memory, because unlogged minor stops quietly drag λ down. This is the same data hygiene that failure coding in the CMMS is built to enforce.
- Compute λ per asset and per failure mode. A machine has one MTBF but several failure modes, each with its own rate. Break λ down by mode so you attack causes, not just assets. Pull the recurring modes into a reusable failure mode library.
- Pool identical units for rare failures. One motor rarely gives a stable rate. Twenty identical motors do. Combine unit-hours across the fleet to get an estimate you can trust.
- Forecast work from the rate. Multiply λ by planned run hours to predict failures, then size crews, spares, and the PM schedule to match. This turns a backward-looking metric into a forward plan.
- Re-measure on a rolling window. Track short-window λ per asset, quarter over quarter, on a fixed definition. A falling rate with flat maintenance cost is reliability work paying off; a rising rate is a wear-out or install-quality problem to chase with root cause analysis.
Failure rate, MTBF, or MTTF, which do you use?
These three metrics answer different questions, and mixing them is the most common reporting error after miscounting failures. Failure rate is per-hour and additive; MTBF is the reciprocal for repairable assets; MTTF is the analogous average for items you replace instead of repair.
| Metric | What it measures | Unit | Use it for |
|---|---|---|---|
| Failure rate (λ) | Failures per unit of operating time | per hour (or FIT) | Adding component rates, forecasting work, reliability math |
| MTBF | Average operating time between failures | hours | Repairable assets: fillers, pumps, conveyors, packaging lines |
| MTTF | Average time to failure of a non-repairable item | hours | Throwaway parts: belts, seals, filters, bearings you replace |
Rule of thumb: reach for λ when you are doing arithmetic (summing series components, projecting failures, sizing spares) and for MTBF when you are reporting a repairable asset's reliability trend to plant leadership. Use MTTF for consumable components, and never quote MTBF for a part you throw away.
How do plants misuse failure rate?
Assuming a constant rate everywhere. The clean λ = 1/MTBF relationship only lives in the flat part of the bathtub curve. Applying one lifetime rate to a machine in wear-out understates its current risk. Use a rolling window.
Counting calendar time instead of operating time. A standby pump that sits idle 90% of the year has almost no operating hours; dividing failures by calendar hours makes it look far more reliable than it is. Match the clock to the exposure, operating hours for duty equipment, and be explicit when you use calendar time for standby units.
Pooling units that are not really identical. Combining unit-hours only gives a valid rate if the machines, duty cycles, and environments genuinely match. Pool a chilled-room conveyor with an ambient one and the average describes neither.
Treating a database number as your number. Published reliability data is a starting estimate under stated conditions. Your maintenance quality and environment can move the real rate by a wide margin, so replace borrowed rates with your own history as soon as you have enough failures to compute it.
Where does failure-rate data come from?
The honest answer is: from a maintenance system that captures every failure with a consistent code and a real operating clock. A CMMS that records work orders, failure modes, and downtime is the source of the failure counts; machine counters or connected PLC data supply the operating hours. Where a plant has too little of its own history for a rare, high-consequence asset, published reliability databases give a starting estimate. Just remember those are population averages under stated conditions, your duty cycle, environment, and maintenance quality move the real number.
What the numbers say
- The oil and gas industry's reliability data standard, ISO 14224 exists precisely to make failure and maintenance data comparable across operators, and it defines failure rate as failures per unit of time (calendar or operational) for a given population (ISO 14224:2016, Collection and exchange of reliability and maintenance data for equipment). Its standardized taxonomy is what makes pooled λ estimates meaningful.
- In electronics, 1 FIT = 1 failure per 10ⁿ device-hours is the industry convention, and MTBF is defined as the reciprocal of the constant failure rate (MTBF = 1/λ) during useful life (Failure rate, engineering overview). Both ideas rest on the same assumption you must check: a constant λ only applies in the flat region of the bathtub curve.
- Reducing λ pays in labor as much as parts: the U.S. Bureau of Labor Statistics projects 13% growth (2024–2034) in employment of industrial machinery mechanics and millwrights, far faster than average, with roughly 54,200 openings a year (BLS Occupational Outlook Handbook). Every failure you prevent is technician time you did not have to find.
Failure rate is the reliability metric that scales, it adds across components, forecasts future work, and converts straight to MTBF. Keep it per asset and per mode, compute it on a rolling window, and feed it from failure data clean enough to survive an audit. That data discipline is the whole game, and it is the theme of our equipment reliability guide and the operator-led care behind total productive maintenance. For how one plant got floor data trustworthy enough to compute metrics like these, see the CLS case study.