Techniques for Estimating Uncertainty Propagation in Probabilistic Design of Multilevel Systems

In probabilistic design of multilevel systems, the challenge is to estimate uncertainty propagation since outputs of subsystems at lower levels constitute inputs of subsystems at higher levels. Three uncertainty propagation estimation techniques are compared in this paper in terms of numerical efficiency and accuracy: root sum square (linearization), distribution-based moment approximation, and Taguchi-based integration. When applied to simulation-based, multilevel system design optimization under uncertainty, it is investigated which type of applications each method is best suitable for. The probabilistic formulation of the analytical target cascading methodology is used to solve the multilevel problem. A hierarchical bi-level engine design problem is employed to investigate unique features of the presented techniques for uncertainty propagation. This study aims at helping potential users to identify appropriate techniques for their applications.