Optimizing component solution spaces for systems design

Abstract

Systems design is concerned with breaking down large and complex problems on the system level into smaller and simpler problems on a component level. This can be accomplished by decomposing quantitative system requirements into requirements for each component. These new requirements can be expressed by component solution spaces, i.e., regions of permissible component design variables. They serve as design goals for component development and support decisions regarding component design. Component solution spaces should be chosen such that components can be designed independently of each other. This is the case when satisfying all component requirements implies satisfying the higher level system requirements. In addition, component solution spaces should be as large as possible to provide maximum flexibility for component design, and to encompass uncertainty. Motivated by related work, their shapes can be predefined as boxes, i.e., multi-dimensional intervals, which enables a simple visual and numerical representation. Unfortunately, this will make solution spaces small and associated requirements unnecessarily restrictive. In a new approach, the shapes of component solution spaces are also optimized to further enlarge their size. This is accomplished by decomposing the system performance function as a sum of predefined component performance functions and optimizing their individual contribution. For both the old and new approach, optimization schemes are presented which focus on linear system performance functions. Their effectiveness is demonstrated for a crash design problem.