ABSTRACT
In result of a global sensitivity analysis, a singleton sensitivity measure for each input dimension is obtained. All information about the nonlinear, complex functional relation of input and result variables is condensed to a single value. This is reasonable for a high number of input variables (e.g. more than 100). But, for low dimensional problems (e.g. up to 30) a more detailed insight is preferable. Therefore, the approach of sectional sensitivity measures is introduced in (Pannier & Graf 2010). The idea is to partition the sensitivity measure per input dimension additionally and determine the
The simulation based design process of engineering structures is a complex task, especially, when multiple input variables have to be handled. Versatile tools are on hand to optimize the structure or assess the reliability by means of black-box programs. But most often, engineers long for an deepened insight into the specific problem to get an idea about the underlying reality. Therefore, data mining tools can be applied, which enable to detect structures in some predetermined point sets. These point sets are generated with an initial design of experiment, a random sampling or can be the results of a prior optimization or reliability runs. However, the aim is to reason dependencies between variables, mostly between input and result variables. An important issue is to determine the influence of individual input variables in view of specific result variables. This is done by determining the sensitivity of input variables (Helton, Johnson, Sallaberry & Storlied 2006).
