Optimization for crashworthiness is discussed here with focus on the consideration of physical and numerical scatter. Due to parallel computing, crash simulations are not completely repeatable and the numerical noise may render optimization obsolete. In addition, numerical optima are never realized in industrial practice. Several sources - starting from manufacturing, wear to variations in material or loadings alter the ideal numerical design. To assure nevertheless an overall success of optimization, this physical scatter should also be taken into account. If scatter leads to radical losses in performance the optimal solution is not robust. Because optimization drives the design to the limits, this instable reaction is quite natural. Hence robust design techniques should be considered as well. While robustness is concerned with alterations around the mean value, which is hence relevant in the context of crash optimization, reliability focuses more on failure, i.e. on rare events. An estimation of the corresponding probabilities is up to now not realizable for finite element crash simulations. This paper presents a literature survey on the inclusion of numerical and physical scatter considerations into crash optimization and concludes with some examples performed by the author at the research center of BMW. In particular, the question if response surface methods are appropriate is discussed here.     «

Optimization for crashworthiness is discussed here with focus on the consideration of physical and numerical scatter. Due to parallel computing, crash simulations are not completely repeatable and the numerical noise may render optimization obsolete. In addition, numerical optima are never realized in industrial practice. Several sources - starting from manufacturing, wear to variations in material or loadings alter the ideal numerical design. To assure nevertheless an overall success of optimiz...     »