Interactive uncertainty allocation and tradeoff for early-stage design of complex systems

Abstract

A common probabilistic approach to perform uncertainty allocation is to assign acceptable variability in the sources of uncertainty, such that prespecified probabilities of meeting performance constraints are satisfied. However, the computational cost of obtaining the associated tradeoffs increases significantly when more sources of uncertainty and more outputs are considered. Consequently, visualizing and exploring the decision (trade) space becomes increasingly difficult, which, in turn, makes the decision-making process cumbersome for practicing designers. To address this problem, proposed is a parameterization of the input probability distribution functions, to account for several statistical moments. This, combined with efficient uncertainty propagation and inverse computation techniques, results in a computational system that performs order(s) of magnitude faster than a state-of-the-art optimization technique. The approach is demonstrated by means of an illustrative example and a representative aircraft thermal system integration example.