Citation

Sustersic G, Grabec I, Prebil I. Int. J. Impact Eng. 2007; 34(10): 1585-1593.

Copyright

(Copyright © 2007, Elsevier Publishing)

DOI

10.1016/j.ijimpeng.2006.09.093

PMID

unavailable

Abstract

A novel approach to model vehicle collisions for the purpose of road accident investigation is developed. It is based upon statistical description of relations between kinematic variables related to description of collision phenomena. The description starts with an estimate of the probability distribution of these variables, while the relation between them is extracted by non-parametric regression (NPR) using a conditional average (CA) estimator. For the purpose of such a statistical manipulation, information on selected collision phenomena is provided by a set of experiments. If partial data are known in an analysis of a particular collision, then these can be utilized to predict complementary data by NPR . This application is demonstrated using the example of a toy vehicle collision with a massive barrier. The effectiveness of the method is evaluated by comparing the predicted results with corresponding data obtained in an additional set of independent experiments. A high value of predictor quality indicates applicability of the developed method to experimental modeling of vehicle collision phenomena.