ABSTRACT
In accelerated life testing, products are exposed to higher-than-use levels of stress (e.g. higher voltage, temperature, or pressure) to produce failure earlier than at use stress level. Data collected at such accelerated conditions are extrapolated by means of a model to estimate quantities of interest, e.g. hazard rates, average lifetime, at use condition. The inference on ALT usually assumes that there exists only a single failure mode comes from a pre-specified parametric family of distributions such as Weibull, lognormal, and so on. However, this assumption may not be appropriate in some products which have more than one failure mode actually. For example (Nelson 1990; Pascual 2006), assemblies of ball bearings can fail due to failure of the race, or the ball. A circuit board fails if any of its joints fail. In this case, the failure of product is caused by the earliest failure of any of the failure modes, which is commonly referred to as competing failure modes. Therefore, it’s necessary to analyze accelerated life testing when two or more failure modes are present.
