Deep neural network based head injury criterion estimation for more efficient pedestrian protection performance evaluation in bonnet structure design

Abstract

In the early development stage of a new car model, fast evaluation of pedestrian protection performance is required to accelerate the bonnet structure design process. Current evaluation methods rely on finite element (FE) simulation to calculate the Head Injury Criterion (HIC) values at specified points on the bonnet, before proceeding to the more expensive real-world impact test using headform impactors. However, the FE based approach typically takes several days to complete the meshing, boundary condition setting and HIC calculation to evaluate a single design candidate. To further increase the evaluation efficiency, this paper presents a novel HIC estimation approach based on deep learning. An end-to-end deep neural network model is proposed which can directly generate the HIC value without resorting to FE methods. It uses seven variables pertaining to the panel height, structural difference, thickness and head type as the input based on the definition of HIC. Convolution layers are utilised to aggregate the surrounding structural information for each target point. To demonstrate the effectiveness of the proposed approach, cross validation results are presented based on a dataset of over five thousands target points collected from 28 cars. For the green target regions, the average HIC estimation accuracy is 93.1 %, which outperforms the result of 83 % reported in previous work. A comparison with the traditional support vector regression method demonstrates the advantages of the proposed approach.