Reliability Engineering: MTTF, MTBF, Failure Rate and the math behind it

Every engineered system fails eventually. Reliability engineering is the discipline of quantifying when and how often they fail, turning vague intuitions about durability into numbers you can design against. This post walks through the core terms in reliability/ safety engineering and their mathematics: the failure rate, the reliability and failure functions, the meaning of MTTF and MTBF, and how to combine component numbers up into a system-level estimate.

Failure Rate

The failure rate, denoted λ (lambda), is the frequency at which a component fails per unit of time. If a population of components experiences 5 failures over 100,000 operating hours, the failure rate is 5 / 100,000 = 0.00005 failures per hour.

Because these numbers are tiny, electronics reliability often uses FIT (Failures-In-Time):

1 FIT = 1 failure per 10⁹ device-hours.

A component rated at 50 FIT fails, on average, 50 times per billion hours of operation.

Reliability Function

The reliability function R(t) is the probability that the component is still working at time t:

By definition

R(0) = 1 (everything works at the start)

R(∞) = 0 (nothing lasts forever).

R(t) is monotonically decreasing from the beginning (t=0) to the end (t=∞).

We can derive a general formula for R(t) from

the hazard rate h(t) = −R′(t) / R(t) = λ

where λ is a constant failure rate.

Solving the differential equation, we get

R(t) = exp(−∫λ dt) = exp(−λt)

This exponential law is the workhorse of reliability engineering. Its defining property is memoryless-ness: A component that has run for 1,000 hours is exactly as likely to survive the next hour as a brand-new one. That is what makes the "constant failure rate" assumption both convenient and, for the useful-life region, reasonable.

Example: 

A pump has λ = 0.0001 failures/hour.

Its reliability at 5,000 hours is R(5000) = exp(−0.0001 × 5000) = exp(−0.5) ≈ 0.607.

There is roughly a 61% chance it survives that long.

Failure Function

The failure function F(t), also called unreliability or the cumulative distribution function is simply the complement of reliability:

F(t) = P(T ≤ t) = 1 − R(t)

It gives the probability that failure has occurred by time t. Its slope is the failure probability density f(t): the quantity the hazard rate and the MTTF integral are both built from:

f(t) = dF/dt = −dR/dt

For the exponential case,

f(t) = λ·exp(−λt).

These three functions R(t), F(t), and f(t) are just different views of the same underlying time-to-failure distribution, and any one determines the other two.

MTTF and MTBF

Two metrics summarize a component's life in a single number, and they are frequently confused.

MTTF (Mean Time To Failure) applies to non-repairable items: a fuse, a lightbulb, a disposable sensor. When it dies, you replace it, you don't fix it. Mathematically, MTTF is the expected value of T, which equals the area under the reliability curve:

MTTF = ∫ R(t) dt, (0<t<∞)

For an exponential distribution this integral collapses:

MTTF = 1 / λ

So our pump with λ = 0.0001 failures/hour has an MTTF of 10,000 hours. The failure rate is the only input, yet it fixes both the reliability curve and the MTTF at once:

R(t) = exp(−0.0001·t), and

MTTF = 1/λ = 10,000 hours.

You can then read survival probabilities straight off the curve: A new pump has about a 61% chance of reaching 5,000 hours (R = exp(−0.5) ≈ 0.607).

Alternately, if there are 100 new pumps,

about 61 of them will survive after 5,000 hrs (R(5000))

about 37 of them will survive after 10,000 hrs (R(MTTF=10000))

It is important to note that R(MTTF) is the mean time to failure, and not the median time to failure. Median time represents the time it takes half the components to fail!

MTBF (Mean Time Between Failures) applies to repairable systems: a server, a motor, an aircraft subsystem. It measures the average uptime between consecutive failures. In principle, MTBF = MTTF + MTTR, where MTTR is the mean time to repair. In practice, repair time is small compared to operating time, so MTBF ≈ 1/λ as well, and it is estimated in the field as:

MTBF = total operating time / number of failures

The crucial takeaway: MTBF is not a guaranteed lifespan. An MTBF of 10,000 hours does not mean every unit lasts that long. Because reliability at t = MTBF is e⁻¹ ≈ 0.368, only about 37% of units survive to their own MTBF. It is a statistical average of a population, not a promise about any single device.

System MTBF and Reliability Calculation

Real systems combine many components. How they are arranged determines overall reliability.

Series systems need every component to work. Because the survival events are independent, reliabilities multiply:

R_system(t) = R₁(t) × R₂(t) × … × Rₙ(t)

For exponential components, this means failure rates simply add:

λ_system = λ₁ + λ₂ + … + λₙ,   so   MTBF_system = 1 / Σλᵢ

A series system is always less reliable than its weakest link.

Put three of those pumps in series, so all three must run for the system to work.

Their rates add, λ_system = 3 × 0.0001 = 0.0003, so the system MTBF drops to 1 / 0.0003 ≈ 3,333 hours, a third of a single pump's 10,000.

The survival odds fall too: at 5,000 hours,

R_system = exp(−0.0003 × 5000) = exp(−1.5) ≈ 0.223, versus 0.607 for one pump on its own. i.e about 22% survive after 5,000 hrs of operation instead of 61%

Parallel systems use redundancy: the system works if at least one path survives. Here it is easier to multiply the failure probabilities:

R_system(t) = 1 − [(1 − R₁(t)) × (1 − R₂(t)) × … × (1 − Rₙ(t))]

Run two identical pumps in parallel so one can carry the load if the other quits. A single pump has R(5000) ≈ 0.607, so R_system(5000) = 1 − (1 − 0.607)² = 1 − 0.393² ≈ 0.846 i.e. redundancy lifts the 5,000-hour survival odds from 61% to 85%. For n identical parallel units, the system MTBF works out to a harmonic series:

For the two pumps that gives MTBF = 10,000 × (1 + 1/2) = 15,000 hours, up from 10,000 for one. Note the diminishing returns, though: the second unit adds far more than the tenth.

MTBF_parallel = (1/λ) × (1 + 1/2 + 1/3 + … + 1/n)

Wrapping Up

The mathematics of reliability rests on one distribution, the exponential, and one guiding idea: failures are probabilistic, so we describe them with functions, not fixed dates. With those tools, "how reliable is it?" stops being a guess and becomes a calculation.