Stochastic Approach for Vehicle Crash Models 2004-01-0460

This paper describes the development of a mathematical model capable of providing realistic simulations of vehicle crashes by accounting for uncertainty in the model input parameters. Advanced and efficient probabilistic and reliability analysis methods are coupled with well-established, high fidelity finite element and occupant modeling software to predict the reliability of vehicle impact scenarios.

The NESSUS probabilistic analysis software was used as the framework for a stochastic crashworthiness FE model. The LS-DYNA finite element model of vehicle frontal offset impact and the MADYMO model of a 50th percentile male Hybrid III dummy were integrated with NESSUS to comprise the crashworthiness characteristics. Response quantities from the models were used to define four occupant injury acceptance criteria and six compartment intrusion criteria. These ten acceptance criteria were used as events in a probabilistic fault tree to compute the overall system reliability of the impact scenario. A response surface model was developed for each acceptance criteria to facilitate the probabilistic analysis and vehicle design tradeoff studies.

NESSUS was used to compute the reliability of each acceptance criteria and the system reliability by combining all acceptance criteria events into a probabilistic fault tree. A redesign analysis was performed using the computed probabilistic sensitivity factors to direct design changes. These sensitivities were used to identify the most effective changes in model parameters to improve the reliability. A redesign using 11 design modifications was performed that increased the original reliability from 23% to 86%. Several of the design changes include increasing the rail material yield strength and reducing its variation, reducing the variation of the bumper and rail installation tolerances, and increasing the rail weld stiffness and reducing its variation. The NCAP star rating was also computed for the original and final designs as another measure of vehicle performance.

Finally, the response surface models were compared to the actual numerical models to verify their accuracy. Accuracy, benefits and limitations of the response surface approach for crashworthiness models is also discussed. The results show that major reliability improvements for occupant injury and compartment intrusion can be realized by certain specific modifications to the model input parameters. A traditional (deterministic) method of analysis would not have suggested several of these modifications.