System reliability optimization considering variability of empirical component reliability estimates

The reliability optimization problem of redundancy allocation involves the selection of components and the appropriate levels of redundancy to maximize some objective function, usually the system reliability, given constraints on system cost, weight, etc, It has generally been required that the analyst know explicitly the component reliability or the time-to-failure distribution for each available component. However, for most actual design problems, the component reliability values are not known and are difficult to measure. This paper describes a new approach to the problem which uses empirical reliability data and not does require the explicit knowledge of the component reliability values. The new approach uses a genetic algorithm to determine the design configuration which maximizes a lower confidence bound on system reliability.