Inverse design and mechanistic analysis of randomly graded re-entrant honeycombs under dynamic compression

Auxetic cellular structures exhibit strong potential for impact protection due to their superior energy absorption capability. However, conventional re-entrant honeycombs with uniform or simple gradient designs often suffer from limited tunability in mechanical response, and the underlying relationship between gradient distribution and deformation mechanisms remains insufficiently understood. In this study, a randomly graded re-entrant honeycomb (RGRH) architecture is proposed, introducing heterogeneous geometric gradients and inherent interlayer misalignment to enrich the accessible design space and mechanical response diversity. To efficiently explore this complex design space, a surrogate-assisted inverse design framework is developed by integrating a hybrid neural network (HybridCurveNet) with a genetic algorithm, using AlSi10Mg as the base material and validated through finite element simulations under dynamic compression. The framework enables the identification of RGRH configurations that match five prescribed stress-strain responses, including plateau, strain-hardening, strain-softening, double-plateau, and multi-plateau behaviors. Dynamic compression simulations are performed to investigate the underlying deformation mechanisms. The results reveal that gradient-induced interlayer misalignment serves as the dominant mechanism governing stress localization, deformation sequencing, and the transition between collapse modes. Stable misalignment promotes homogeneous crushing with a relatively low peak stress and plateau response, whereas suppressed or delayed misalignment leads to localized or staged deformation. Meanwhile, the cell-size gradient plays a complementary role by regulating the spatial evolution of densification and the emergence of multi-stage responses. Overall, this study establishes a direct structure-mechanism-response linkage for RGRH lattices and demonstrates an effective pathway for performance-driven inverse design of advanced auxetic metamaterials.