Prediction of injured body regions and injury severity from available accident data can lead to more appropriate and hastier emergency care in automotive accidents. The existing prediction method was based on statistical analysis of a massive amount of real-world accident data. However, numerical crash simulations were also considered to provide a virtual injury database in a relatively short time. Therefore, the purpose of this study was to develop and to evaluate a new method to obtain injury prediction algorithm by utilizing virtual database of numerous computer simulation results.

Occupant and cabin were modeled as multi-bodies. The occupant models have geometries of typical Japanese adult males. The cabin model consists of safety restraint systems and interior panels. Acceleration and intrusion of the door panels during side impact were delivered to the occupant in the simulations. Hundreds of crash simulations were performed where crash parameters were changed systematically. The injury prediction algorithms were developed by logistic regression analysis of the database constructed from the results of the simulations.

The algorithms correctly predicted more than half of the head, thorax, and thigh injuries in 48 accidents. However, this study neglected cabin deformation in frontal crash, break of door-window, as well as occupant’s age and gender, which may affect on the occupant responses and injury severities. These limitations might be the cause of the miss predictions of injury severity in the simulations.

In this study, possibility of developing injury prediction algorithms by using numerical crash reconstructions was presented as a different approach from existing method that used real-world accident data. For more accurate predictions, improvement of the simulation models and consideration of occupant characteristics are required.