Abstract
The paper develops a systematic procedure for constructing probability models for unit lifetimes. It identifies the smallest class of such models that suffice for selecting a cost-optimal maintenance or operating policy. Lifetime-cost functions are introduced, replacing the classical failure rate functions for specifying model classes. It is shown how these classes can be represented through Bayesian parametric models. To illustrate, several common lifetime models and their finite population analogs are derived, including the exponential, the Weibull, Freund’s multivariate exponential, and a new multivariate Weibull.