Prediction of Head Injury Criteria in Pedestrian Crashes Using Frequency Response Function-Based Deep Neural Networks

This study aims to develop a deep neural network model capable of predicting the Head Injury Criterion (HIC), which is traditionally obtained through time-consuming and costly headform impact tests using dynamic stiffness measurements. The correlation between dynamic stiffness and HIC was initially analyzed in a simple steel plate to explore feasibility. Subsequently, this correlation was evaluated using the hoods of two actual vehicles. The results revealed significant differences compared to the steel plate analysis, primarily due to the influence of contact with the lower components beneath the hood. To overcome these physical differences, a deep neural network model incorporating various features was designed to improve HIC prediction accuracy. Additionally, an algorithm was developed to estimate the penetration of the hood into the lower components, and this feature was included as an input to enhance the performance of the model. This study demonstrates the potential of the deep neural network model as an auxiliary tool for predicting changes in HIC resulting from modifications to the hood design.