Data-based approach in modeling automobile crash

Abstract

State of the art techniques in crash-worthiness engineering include crash simulation using finite element models or lumped parameter models, as well as full car and component tests. Simulations are used to optimize design and reduce the number of tests. The physical tests are used to verify the crash performance of an automobile. However, the development of analytical models is based on the basic knowledge of the system and makes little use of the full car test data. In this paper, we propose to develop analytical models directly from crash test measurements using system identification techniques. The analytical model is made up of two parts: a differential equation part consisting of mass, stiffness and damping characteristics, and a transfer function part, consisting of an autoregressive moving average (ARMA) of white noise. The lumped parameters used in the model are time varying and are estimated recursively by minimizing the quadratic criterion of the one step ahead prediction errors. These parameters are also correlated with the structural characteristics involved during the crash test. This approach is verified by estimating known parameters in the presence of random as well as autocorrelated noise. Furthermore the usefulness of this technique is demonstrated by estimating structural parameters from frontal and side impact test data. Limitations and future work are also highlighted.