Reliability and life data analysis
Failure rates, the bathtub curve, MTBF, Weibull analysis, censored data, B10 life and demonstration tests.
Reliability is the probability that a product performs its function for a stated time under stated conditions.
The bathtub curve
Failure rate over a product's life often has three phases: decreasing early failures (infant mortality, often from manufacturing defects), a low and roughly constant rate of random failures during useful life, and a rising rate as parts wear out.
Key measures
- Reliability R(t): probability of surviving to time t.
- Failure rate (hazard): instantaneous rate of failure among survivors.
- MTBF / MTTF: mean time between failures (repairable) or to failure (non-repairable). For a constant failure rate λ, and . At t = MTBF, only 36.8% of units survive.
- B10 life: time by which 10% of units have failed.
Weibull analysis
Introduced by Waloddi Weibull in 1951, the Weibull distribution fits most life data.
- β (shape): below 1 early failures, about 1 random, above 1 wear-out (2 to 4 typical for mechanical wear).
- η (scale, characteristic life): time by which 63.2% have failed.
Censored data
Units still running when the test ends (right-censored) carry information: they survived at least that long. Maximum likelihood estimation uses them correctly; ignoring them makes life look shorter.
Other tools
- Kaplan-Meier: nonparametric survival curve (1958).
- Accelerated life testing: test at higher stress (temperature, voltage) and extrapolate with models such as Arrhenius.
- Reliability growth (Crow-AMSAA, Duane): track improvement during development.
- Demonstration testing: zero-failure test to show reliability R at confidence C needs units; for R = C = 0.90 that is 22 units.
Questions and answers
Our MTBF is 10,000 hours. Will most units last 10,000 hours?
No. With a constant failure rate, only about 37% survive to the MTBF. Use B10 or a reliability at a stated time for customer commitments.
