Global Sensitivity Analysis in Structural Optimization

The main purpose of global sensitivity analysis is to identify the most significant model parameters affect- ing a specific model response. This helps engineers to improve the model understanding and provides valueable information to reduce the computational effort in structural optimization. A structural optimiza- tion is characterized by a set of design parameters, constraints and objective functions formulated on basis of model responses. The computational effort of a structural optimization depends besides the complexity of the computational model heavily on the number of design parameters. However in many cases an objective function is dominated only by a few design parameters. Results of global sensitivity analysis may be used to select the most significant design parameters from a number of potential can- didates and thereby reduce the optimization problem by the insignificant ones. In this paper different sensitivity measures and associated algorithms are evaluated with respect to their capabilities and com- putational costs. In particular the variance-based approach after Sobol is compared with the correlation analysis, the linear and quadratic ANOVA approaches, and the FAST approach. This is done using a comprehensible academic example as well as an optimization problem from engineering practice. Keywords:

application/pdf F-I-03.pdf — 1021.2 KB