A Stochastic Approach for Parameter Relevance Estimation in Vehicle Interior Simulations of Frontal Impacts

Abstract

The evaluation of frontal crash performance of vehicles is mainly performed using standardized barrier tests and the safety system is designed against these. Innovation has led to improved technologies such as more accurate simulation models enabling us to improve how the safety system can work in a variety of scenarios. Thus, virtual crash tests using finite elements (FE) have almost replaced the physical tests among vehicle manufactures. The combination of faster computers and more accurate models has led to the possibility to simulate the variability in the system performance using stochastic simulations. If stochastic simulations are performed, which allows for testing of different design options and variability, information can be gathered as how to create a balanced and optimized safety system where all components work together at their best. In this thesis, simulations and physical tests of a currently used car platform are studied in full-frontal crash tests in order to determine which components need to be represented in the stochastic simulations. Based on the discovered behavior in theses tests, a list of relevant input and output parameters for the stochastic simulation is defined. Input parameters are sorted into three different types; production, design and real life, and both risk injury criteria and kinematics of a crash test dummy are used as output. Furthermore, a simplified and parameterized FE model of a vehicle interior for frontal impacts is developed and validated. Passenger airbag and knee airbag are simplified as unfolded airbags and a curve fitting function that measured the root mean square error between the reference airbag pressure and the simplified airbags is used to reproduce the behavior of the studied airbags. Floor and windscreen are modeled as rigid surfaces and the seat is simplified by removing parts of the backrest and prestensioned bolt connections are changed to rigid. The validation show that the simplified model deviates slightly compared to the studied tests, but it is approximately 14 times faster compared to the reference. Further, stochastic simulations are performed with geometrical, impact and airbag parameters varied. The statistical information such as mean, standard deviation and correlation between parameters is evaluated. The results produced from the simplified model follows the existing trends in the reference model making it possible to draw trend-based conclusions from the stochastic simulations. The stochastic simulations indicates that a crash pulse with low Vehicle Pulse Index is to strive for when developing a car to get low values of the risk indicators in full frontal impact. In general, the linear correlation matrix is a valuable tool to see the influence a variation of the input parameters have on the risk injury criteria and dummy kinematics. Lastly, different ways to present the data, i.e. scatter plots and history curves based on time data are visualized. In summary, the methodology developed offers an approach for using stochastic vehicle interior simulations in frontal impact, which can lead to a deeper understanding of the safety system and the opportunity to optimize and make it more robust.