Crashworthiness prediction of hexagonal crash box using convolutional neural networks

This study proposes a computationally efficient machine learning approach employing convolutional neural networks (CNNs) to accurately predict the force-displacement responses of hexagonal crash boxes under impact loading. A dataset comprising 53 configurations, modeled using computer-aided engineering
files and finite element analysis (FEA) via LS-DYNA, was utilized for training the CNN model. Upon training, the CNN model demonstrated the capability to predict the force-displacement curves for unseen geometries, with errors in the initial peak crash force and energy absorption being approximately 5% compared to FEA results, while achieving substantial computational efficiency. The proposed CNN-based approach exhibits significant potential for accelerating the design process for crash box structures in the automotive industry.