In traditional reliability life testing and accelerated life testing, usally only the failure mode and the lifetime upon failure or censoring, are recorded. The information obtained from such testing may be signicantly improved by aslo recording one or more measures of deterioration, such as wear depth, crack length, leakage rate, etc. Failure occurs when a measure of deterioration, or a combination of them, reach a critical value. This approach calls for refined reliability models and estimation techniqes.
Established derministic models for some physical deterioration mechanisms are described. Alternative appraches for stochastic models are studied, such as cumulative stochastic processes, regression models with random cofficients, Wiener processes with random drift, and the Bernstein distribution and related models. Special attention is attached to models where a parameter set in the stochastic process is considered as a stochastic variable with one realization per specimen. Maximum likelihood estimatiors are studied and compared for the Wiener process model, and the Wiener process with random drift.