A machine-learning based framework for design and characterization of honeycombs with partial self-similar hierarchical architectures

Abstract

Honeycomb structures with hierarchical topology are widely used in engineering fields for their excellent mechanical properties. Multiphase topology is proved to possess a larger design space than the traditional hierarchical topology. In this study, a partial self-similar hierarchical honeycomb (PSSHH), comprising of zero-order cell and first-order cell, was designed. Numerical simulations, validating against the experimental results, were carried out to investigate the in-plane crushing performance of PSSHHs with different configurations. An artificial neural network model was trained using the results of simulations. The ability and effectiveness of artificial neural networks in predicting and characterizing the crushing responses of PSSHH were evaluated through comparison with simulations. The model was primarily employed for optimizing the design of PSSHH to enhance the crushing performance. The results indicate that the optimized PSSHH outperforms traditional hexagonal honeycombs and fully self-similar hierarchical honeycombs in terms of initial peak force, mean crushing force, and energy absorption. Furthermore, the effect of number and organized patterns of the first-order cells along the loading direction on the mechanical properties of PSSHH is examined. The findings reveal that the formation and development of localized collapse bands are significantly affected by the number and distribution of hierarchical first-order cells. The deformation, which involves a sequential crushing process with the initial collapse of specific zero-order cells followed by the subsequent collapse of first-order cells, contributes to enhancing the crashworthiness performance of PSSHH.