Assessment of Techniques for Global Sensitivity Analyses in Conceptual Aircraft Design

Abstract

The design of novel air transportation systems typically targets market introduction several years in the future and is therefore generally characterized by various assumptions, for example on the evolution of materials, manufacturing technologies and subsystem performance. These assumptions are inherently subject to uncertainties that need to be considered to make robust design decisions already in early stages of the design process. To manage the complexity of the design problem and the vast design spaces considered, the execution of sensitivity studies aims at identifying the most influential uncertain factors. While observing these, the factors with low impact can be neglected in the considerations. Nowadays, the majority of sensitivity studies conducted in aircraft design are based on local techniques, i.e. input variables are successively varied around a fixed, single point in the design space. Although this method is straightforward and easy to understand, it only provides limited informative value. Global sensitivity analyses (GSA) have the potential to overcome this limitation by quantifying the sensitivity characteristics of the entire design space. However, its application is numerically expensive and it is not always obvious which of the various GSA techniques fits best to the individual use cases. Therefore, the objective of this work is to assess the applicability of four different state-of-the-art GSA techniques, being the methods of Sobol', FAST, Morris, and Polynomial Chaos Expansion, to application cases representing the numerical challenges faced within conceptual aircraft design. These challenges are the high-dimensionality of the design space due to its multidisciplinary nature, the nonlinearity of the model equations involved, and the computational intensity of the design process. Through a thorough investigation of the convergence properties and a comparison of the resulting sensitivity measures, recommendations for the best choice of GSA methods within conceptual aircraft design are derived.