Design of experiment in analyzing uncertainty of simulation results in accident reconstruction

Abstract

In order to obtain a more accurate interval of an accident reconstruction result, two modified experiment design methods were proposed. The first is based on the orthogonal design (OD) and it is named as the orthogonal-based design (OBD), while the other is based on the uniform design (UD) and it is named as the multi-response surface-uniform design (MUD). Furthermore, some conclusions are given according to seven numerical cases: results obtained from the Latin hypercube sampling design (LHS) are the worst, while those obtained from the UD are the last but one, and all are worse than the results obtained from the OBD and the MUD. Furthermore, some conclusions about the number of simulation runs of each design of experiment (DOE) can be given: if the second order response surface model is selected, the number of simulation runs of the LHS is the same as the UD, and all are lesser than the number of simulation runs of the OBD and the MUD in most cases. As for the OBD and the MUD, the number of simulation runs of the OBD is lesser than the number of simulation runs of the MUD, but the MUD can be applied more widely than the OBD. Finally, a true vehicle-pedestrian accident case is given, Pc-Crash was used to reconstruct the accident based on all traces provided by the police, and the interval of the impact velocity of the vehicle was calculated using the OBD, results showed that the research is meaningful in practice.