Достижения в области автомобилестроения
Открытый доступ

Абстрактный

Анализ дорожно-транспортного происшествия в Турции

Yilmaz AC, Cigdem AC and Aydin K

No.2918 Turkish Highway Traffic Act has been the reference legislation for traffic accidents in Turkey since 1983. Although this act consists of several explanations and definitions, it has still deficiencies especially in defining fault rates which are vital for traffic accident analyses. Accident experts determine fault rates mostly according to their initiatives without conducting scientific analyses on accidents due to inadequate quantitative instructions on fault rates in the act. Speed analyses of accident involvements play an important role in accident investigations. A more comprehensive parameter, Energy Equivalent Speed, may be defined to explain dissipation and severity of deformation energy and crush amounts formed on vehicles which also give hint about fault rates. In this study, accessible data were collected from a sample accident scene (police reports, skid marks, deformation situations, crush depths etc.) and used as inputs for an accident reconstruction software called “vCrash” which is able to simulate the accident scene in 2D and 3D. Energy equivalent speed calculations were achieved using 784 parameters with a prediction error. Multi-layer Feed Forward Neural Network and Generalized Regression Neural Network models were utilized for estimation of energy equivalent speeds (speeds just before the collision, i.e., in case of absence of skid marks) based on using these parameters as teaching data for the models. It was aimed that, by benefiting from these neural network methods, necessity of using expensive simulation softwares for probable accidents in future may be avoided. In order to observe performance of the neural network models, standard error of estimates (mean square error) and multiple correlation coefficients were also analyzed using 5-fold cross validation on the dataset. It was observed that, in general, Multi-layer Feed Forward Neural Network model yielded better results for both energy equivalent speed and fault rate analyses. Based on simulation results (energy equivalent speeds and deformations) and assumption of a fault rate scale, fault rates were estimated on prediction models by assuming correspondence of every predetermined increment in energy equivalent speed of specific involvement to a specific increment in fault rate of the same involvement to put forward a scientific and systematic approach and compensate deficiencies in the act.