Optimization of reliability design problems considering uncertainty in component reliability and time-to-failure

Optimization algorithms were developed to solve complex reliability design problems which explicitly consider uncertainty in component reliability and time-to-failure. The developed algorithms allow for the determination of design solutions for very complex and difficult problems, which previously could not have been solved. Additionally, the research is very practical and will provide sophisticated design optimization methods which can be used for many new and evolving engineering designs when there is uncertainty in component reliability values. This research is characterized by a fundamentally different and improved perspective of the reliability optimization process.; The objective of the redundancy allocation problem is to determine a design configuration and to select components which collectively optimize some objective function (usually maximization of system reliability or minimization of system cost) without violating system-level constraints. The problems were formulated in a more general framework than had previously been accomplished, allowing for variability in component reliability and time-to-failure. Component and system reliability were considered as random variables and procedures were developed to accurately estimate a lower bound of system reliability based only on the means and variances of component reliability values. Since variability of system reliability was considered, it was also necessary to incorporate the element of user risk; the probability that the actual system reliability will be lower than some specified value or a lower-bound estimate. The extent of risk associated with a particular system design project was included in the problem formulations and influenced the final design configurations. These formulations were intended to more faithfully address the actual concerns and considerations of the engineering design community.; A genetic algorithm (GA) was used to determine the optimal design configuration and an adaptive penalty function was developed to satisfy constraints. The GA was tested on many different variations of the redundancy allocation problem with excellent results. The GA proved to be very effective in terms of solution quality, robustness, repeatability and computational effort.