A priori prediction of the probability of survival in vehicle crashes using anthropomorphic test devices and human body models.

Objective

In the development of restraint systems, anthropomorphic test devices (ATDs) and human body models (HBMs) are used to estimate occupant injury risks. Due to conflicting objectives, this approach limits an injury severity risk tradeoff between the different body regions. Therefore, we present and validate a protocol for the aggregation of injury risks of body regions to a probability of survival (PoS).

Methods

Injuries were clustered in regions similar to ATD or HBM investigations and the most severe injury as rated by the Maximum Abbreviated Injury Scale (MAIS) per body region was determined. Each injury was transformed into a dichotomous variable with regard to the injury severity level (e.g., MAIS 3+) whose injury risk was computed using the German In-Depth Accident Study (GIDAS) and NASS-CDS databases. Without loss of generality, we focus on 2 body regions-Head/face/neck (HFN) and chest (C)-at the MAIS 3+ level. The PoS was calculated using injury outcomes from the databases. The method of predicting PoS was validated by stratifying the database by crash type and technical crash severity.

Results

The PoS of occupants injured in both HFN and C at the AIS 3+ level was found to be lower, at a statistically significant level, than that of occupants with AIS 3+ injuries to just one of the body regions. Focusing on occupants with only one body region injured at the AIS 3+ level, HFN injuries tended to decrease PoS more than chest injuries. For the validation cases, observed PoS could be reproduced in the majority of cases. When comparing predicted to observed values, a correlation of R2 = 0.92 was observed when not taking the restraint system into account. Focusing on frontal crashes, the correlation was R2 = 0.89. Considering only belted occupants, R2 increased to 0.93, whereas for cases with deployed airbag systems the R2 decreased to 0.68. The PoS for side crashes is reproduced with R2= 0.97 independent of the restraint system; it was 0.95 with belted occupants and 0.55 when also factoring in airbag deployment.

Conclusions

The method showed an excellent predictive capability when disregarding the restraint system, or restraint-specific subgroups, for the considered validation cases.