Robust Design in Occupant Safety Simulation 2013-01-0123

Explicit FE-simulation has become a standard design tool in passive safety development as it can help to reduce the number of vehicle prototype tests, especially in the early project stages. Due to fewer hardware tests available as sources of validation, the expectations towards predictability of dummy injury occurrence have strongly increased. To produce robust results, virtual passive safety analysis has to take real uncertainties of test condition and their probabilistic effects upon the system's response into account. Findings by simulation therefore should not be regarded anymore as single point statements in a pure deterministic approach. They rather need to be extended into statements upon grades of correlation between individual system parameters taking into account their stochastic nature. This paper illustrates a process for complete robustness analysis in occupant safety simulation. It introduces techniques to document statistical confidence intervals of the correlations determined, and uses statistical “bootstrapping” in populations where only a limited number of samplings can be generated. Finally, by regression analysis within the reduced relevant physical parameter space, a risk assessment upon a potential failure within the specific test requirements is carried out.

Identification of initial design space parameters requires experience and proper preparation when one is conducting statistical analysis on occupant safety. This work explains relevant design space parameters for a typical full vehicle test and classifies them in proper ways since obtaining their sampling often requires additional pre-simulation efforts. In this context a number of in-house tools are highlighted here that helps the designer in obtaining work intensive design parameters such as the distribution of dummy, seat & belt positions. The entire robustness design process uses automated work flows and tools to visualize the statistical data observed since the simulation model complexity and its computational costs are extremely high. Finally, the gained correlations are discussed, interpreted and explained in terms of real life phenomenon that can be observed in testing.