A Crashworthiness Design Framework based on Temporal-Spatial Feature Extraction and Multi-Target Sequential Modeling

Temporal-spatial crashworthiness design remains a challenging issue in engineering applications. Metamodeling techniques have been widely used to improve design efficiency by reducing the need for extensive experiments or simulations. However, these methods often fail to capture the essential information of temporal and spatial during the dynamical procedure. In this study, a novel multi-target modeling and optimization framework is introduced to overcome these limitations. This framework utilizes autocorrelation functions to identify key temporal-spatial segments, ensuring that the most influential factors are captured, and then builds a metamodel using multi-target regression techniques and partial autocorrelation functions, effectively capturing the complex relationships among different time steps. An adaptive sampling strategy is also employed to generate additional training data according to the objective functions, thereby enhancing the accuracy and robustness of the metamodels. These improvements enable a more accurate and interpretable integration of temporal-spatial information compared to popular methods. The effectiveness of the proposed framework is demonstrated through its successful implementation in optimizing crashworthiness across diverse scenarios: a cylindrical tube, a multi-cell energy-absorbing structure, and a B-pillar designed to withstand side impacts. The results show that the proposed method provides reliable predictions for subsequent optimization tasks and has the potential to address complex crashworthiness design challenges by comprehensively considering temporal-spatial information.