Accelerated Life Testing: start with the failure mechanism

Before selecting a temperature or duration, define the reliability question. A life projection, a comparison between designs and a demonstration against a requirement need different test designs and different analyses.

Record the intended operating profile, the failure criterion, the relevant population and how failed and surviving specimens will be monitored. Sample size and stress levels must be selected together, rather than treating test duration as the only design variable.

Increased stress may shorten the time to a relevant failure, but it can also introduce a different failure process. An appropriate acceleration model should be supported by the failure mechanism, suitable literature and the test evidence.

Arrhenius-type behaviour can be appropriate for thermally activated processes. Fatigue models relate cycle-driven damage to cycle conditions. Neither should be applied as a universal component-life equation. A fit to several stress conditions helps assess the proposed relationship, but does not remove uncertainty in extrapolation.

The report should state the selected life distribution, acceleration model, data treatment and parameter estimates. Include surviving specimens as censored observations, and distinguish a point estimate from an appropriate confidence bound.

Converting cycles into calendar time requires the expected service cycle frequency and the applicable damage relationship. Time under stress and the number of stress reversals should not be treated as interchangeable quantities.

A useful deliverable identifies what further evidence would materially change the conclusion.