Deep Learning-Based Rapid Prediction for Occupant Injury in Vehicle Restraint System Design

Abstract

Accurate prediction of occupant injury responses is essential for the effective design and optimization of vehicle occupant restraint systems (ORS). To address the complexity and time-consuming nature of injury prediction in existing occupant restraint system (ORS) design workflows, this study proposes a deep learning-based method tailored for frontal collision scenarios. The proposed method enables rapid injury prediction by simultaneously processing crash waveform signals and ORS parameters as inputs. It enables accurate prediction of time-dependent biomechanical response curves across multiple occupant body regions, effectively capturing the complex, nonlinear interactions between dynamic impact conditions and restraint system characteristics. The proposed method requires only 2.7 seconds to predict the occupant's head acceleration–time curve and thorax compression–time curve, as well as to compute the corresponding injury assessment indicators (C-NCAP 2024). The predicted curves achieve a similarity score of over 0.86, and the accuracy of the selected injury assessment indicators exceeds 0.80. Compared with multi-rigid body simulations, the computational efficiency is improved by 533 times. These results demonstrate the model’s potential for intelligent, data-driven, and time-efficient ORS design.