Artificial intelligence and optimization technique-driven unified framework for dynamic response analysis and explainable design optimization of sandwich structures

The increase in complexity arising from topology design and morphology optimization hinders a deep understanding of the dynamic response of sandwich structures (SSs). Accordingly, a tailored framework of Sandwich Structure Performance Lifecycle Engine (SSPLE) is proposed by integrating eXtreme Gradient Boosting (XGBoost), SHapley Additive exPlanations (SHAP), and Nondominated Sorting Genetic Algorithm II (NSGA-II). To demonstrate its superiority, an SS with three-layer aluminum foam cores is selected as the case study. Firstly, specific energy absorption (SEA) and peak deflection (PD) are employed to characterize the SS's blast resistance performance, considering the influence of 11 features encompassing geometry, material, bonding, and loading parameters. Subsequently, 600 instances are collected from validated numerical simulations. Employing the trained XGBoost (R² = 0.9443 on the test set) and SHAP, T_top (thickness of the top sheet) is identified as the most influential feature with an effect range from -0.32 kJ·kg⁻¹ to 0.91 kJ·kg⁻¹, and is confirmed to exhibit a substantial interaction effect with SoD (standoff distance) on SEA. As for PD, the main effect of AM (adhesive material) is restricted to the range of -2.35 mm to 7.27 mm following the exclusion of interaction effects. In addition, two optimization strategies provide explainable schemes for performance enhancement, with an increase of 67.39 % in SEA and a decrease of 32.60 % in PD. These findings confirm that the SSPLE can provide a systematic solution for dynamic response analysis and design optimization of SSs. Lastly, a software tool is developed to facilitate the implementation of SSPLE functionalities in practical applications.