This Faculty Early Career Development (CAREER) grant provides funding for the development of optimization algorithms with the goal of maximizing system reliability when critical components, which may be used in the design, have uncertain reliability. The objective is to determine the most reliable system design by the determination of a system-level configuration and selection of components, given system-level constraints, e.g., cost budget, space requirement. Traditional approaches have generally required explicit knowledge of reliability for all available components, thereby ignoring the variability and the risk implicit in using uncertain estimates. This research is explicitly considering uncertainty and variability with component and system reliability estimation, and is working to produce stochastic optimization algorithms that can be tailored and used to recommend design solutions for different user risk profiles. Analytical methods are being developed to propagate uncertainty to the system-level. For the design of modern electronic systems (and others), this is an accurate design model because the recent development of many high-technology components and materials precludes extensive testing needed to explicitly determine component reliability. Problem-specific integer programming and Genetic Algorithms (GAs) are being used as optimization methods to search and consider possible design solutions. The immediate benefits of this research will be theoretical insights into the propagation of uncertainty and the effectiveness of design improvement strategies. This research will provide tools which can lead to safer and more robust engineering designs. Additionally, there will likely be important insights concerning the use of stochastic optimization methodologies in a risk-averse environment. Considering risk aversion, it may be too risky to maximize an expected objective function without regard to variability. The immediate benefits will be directed towards the reliability engineering community. However, there may likely be far-reaching implications leading to improved solution strategies for other problem domains, such as facility layout, which are characterized by stochastic objectives and risk-averse decision-makers.
