关键词: 发动机盖；深度学习；行人保护；头部伤害预测

Abstract: In vehicle-pedestrian collisions, the impact response between the pedestrian's head and the engine hood (hereinafter referred to "hood") is a key factor in automotive passive safety design. To address the low computational efficiency and long iteration cycles of traditional finite element simulation methods in new vehicle development, this paper proposes a deep learning–based rapid prediction model for the head injury criterion (HIC). An input vector consisting of six key parameters is constructed, including the ground clearance of the impact point on the outer panel Z1, the ground clearance of the corresponding point on the inner panel Z2, the hard-point gap Z3, the outer hood panel thickness T1, the inner hood panel thickness T2, and impact headform category. Fully connected layers combined with residual activation modules are used for feature encoding, and a convolutional neural network is employed to achieve spatial aggregation of local structural features around the impact point. In this way, an end-to-end mapping relationship from hood structural parameters to HIC values is established, enabling rapid prediction of pedestrian head injury caused by the hood. The results show that, within the hood test area defined by the China-new car assessment program (C-NCAP) safety standard, the proposed prediction model achieves an average accuracy of 81.6%, representing a 39% improvement compared with traditional support vector regression methods, thus providing reliable technical support for rapid safety assessment during the hood development stage.

Key words: hood; deep learning; pedestrian protection; head injury prediction