The object of this thesis was to develop a time- and memory-efficient optimization strategy for a discretized structure of a finite element analysis. It should be shown that this can be done by estimating the computational expensive objective function with a more easily optimizable classical linear regression surface. To this end, the theoretical basis for bounded optimization, regressor parameter estimation and diagnosis of linear regression models were explained.
Finally, these principles were implemented in a software environment and the working hypothesis has been tested using a simple drop test case example. Unfortunately it turned out that classical linear regression models are inappropriate for such complex tasks.