Investigation of the Prediction Model and Assessment Parameters of Head Injury of Children Occupants Based on Machine Learning

The head injury mechanisms of occupants in traffic accidents will be more complicated due to the diversified seating postures in autonomous driving environments. The injury risks and assessment parameters in complex collision conditions need to be investigated thoroughly. Mining the simulation data by the support vector machine (SVM) and the random forest algorithms, some head injury predictive models for a 6-year-old child occupant under a frontal 100% overlap rigid barrier crash scenario were developed. In these head injury predictive models, the impact speed and sitting posture of the occupant were considered as the input variables. All of these head injury predictive models were validated to have good regression and reliability (R20.93) by the ten-fold cross-validation. When the collision speed is less than 60km/h, rotational load is the primary factor leading to head injury, and the trends of BrIC, von Mise stress, Maxshear stress, and MPS are similar. However, when the speed exceeds 60km/h, brain injuries are primarily affected by linear load. The head 3ms acceleration, HIC15, von Mise stress, Maxshear, and MPS have a consistent trend. The causes of head injury are mainly affected by the collision speed and sitting angle. Therefore, in autonomous driving scenarios, the design of child restraint systems should fully consider the influence of collision speed and sitting posture on the risk and mechanism of injury, improving the phenomenon of occupant submarine and head restraint insufficiency under the large angle sitting posture. This research will establish a theoretical foundation for investigating head injury mechanisms, injury thresholds, and the consistency of injury indices, and will provide data support for enhancing the restraint system and virtual testing.