Global Sensitivity Analysis in Industrial Application with LS-OPT

The presented paper is related to a global sensitivity analysis in the field of numerical simulations. The objects of investigation are different car side crash scenarios and the influence of model variable variation on the occupant safety. The work is motivated by the enhancement of the understanding for complex system behaviour of occupant safety components. Therefore, the methodology of the investigation should be conditioned in such a way, that it can be used by engineers in the standard development process for car projects. For the systematisation of such studies it is essential to know how many experiments are necessary to approximate the global behaviour in a suitable manner. Therefore, a convergence study is examined. Other addressed questions are related to the comparability of different sensitivity measures and the influence of the distribution of experiments in the design space. To provide a comprehensive view on the results, different possibilities to compress the large amount of information are developed, tested and compared. Furthermore, the variable sampling in the design space and the influence of the type of the global response approximation model are addressed. For the evaluation of the sensitivities between variables and system responses, different sensitivity analysis methods are applied. LS-OPT, the main software component for this analysis, provides different methods for the evaluation of dependencies between system input and output quantities. Namely, the linear ANOVA and the more general formulation of Sobol’s indices are evaluated within this study to classify the individual variable influence. A number of design variables are selected for variation. In the vehicle model, the thickness of constructional elements from the side structure are introduced. Also some values of friction are considered as variables. As a value of model uncertainty, the positioning of the barrier is changed in the range of allowed deviations of the experimental execution. The selected set of variables is simultaneously used for the investigation of four side crash load cases.