Regression-based sensitivity analysis and robust design

Abstract

This paper presents the Regression-Based global Sensitivity Analysis method (RBSA). It is an approach for quantitative, variance-based, sensitivity analysis of mathematical models used for design purposes. The method is based on the subdivision of the global variance into its components, due to the design-factor contributions, using general polynomial regression models. The performance of the RBSA is compared to other methods commonly used in engineering for computing sensitivity, namely, the method of Sobol’, the Fourier amplitude sensitivity test, the method of Morris, and the standardized regression coefficients. It was found that RBSA, under certain circumstances, provides very accurate results with a significant reduction of the number of required model evaluations. A test case, using the mathematical models of two subsystems of a spacecraft, demonstrates how RBSA facilitates the discovery and understanding of the effects of the design choices on the performance of the system.