91 Fahrzeugkonstruktion
This study aimed at developing an injury estimation algorithm for AACN technologies for Germany and compared them to findings based on Japanese data. The data to build and to verify the algorithm was obtained from the German in-depth Accident Database (GIDAS) and split into a training and a validation dataset. Significant input variables and the generalized linear regression model to predict severe injuries (ISS>15) were selected to maximize area under the receiver operating characteristic curve (AUC). Probit regression with the input parameter multiple impact, delta v, seatbelt use and impact direction gave the largest AUC of 0.91. Sensitivity of the algorithm was validated at 90% and specificity at 76% for an injury risk threshold of 2%. It appears that no major differences between Japan and Germany exist for injury estimation based on delta v and impact direction. However, far side impact and multiple crash events appear to be associated with a larger risk increase in the German data.
Institute for Traffic Accident Research and Data Analysis <Tokyo>rnAbstract: Analyses were conducted to clarify the features of rear-end collisions, using an integrated accident database developed by the Institute for Traffic Accident Research and Data Analysis (ITARDA). Focusing on neck injuries in rear-end collisions, analyses were made of the relation to struck-vehicle properties. Regarding the relation to the initial year of registration, the results did not show that newer vehicles tended to have a lower no-neck-injury rate, which was defined in this study as an index. On the contrary, in some passenger car classes, it was observed that the no-neck-injury rate was higher in newer vehicles. The effect of an active head restraint system, which is one type of anti-whiplash device, was analyzed by using not only the no-neck-injury rate but also a regression analysis. The results showed that the effect of an active head restraint system on suppressing the incidence of neck injuries was statistically significant.rn