Optimization of grid composite configuration to maximize toughness using integrated hierarchical deep neural network and genetic algorithm

Abstract

Recently, grid composites have drawn considerable attention in various engineering applications due to their customizable mechanical properties, stemming from a wide spectrum of available configurations. Advanced additive manufacturing techniques have made the production of these configurations more feasible. While much research has focused on optimizing for stiffness and strength, there has been a relative lack of studies targeting toughness optimization, mainly due to the intricate unpredictability arising from the complexity of crack propagation mechanisms. This study seeks to fill that gap by leveraging a hierarchical deep neural network (DNN) model, integrated with finite element method (FEM) simulations, to maximize grid composite toughness. DNN model incorporates both structural configuration and stress field data, improving prediction accuracy for unseen domains. Using a genetic algorithm informed by the DNN model, a substantial 60% increase in toughness was achieved while investigating only 3.5% of the original dataset. Subsequent analyses of the stress field and crack phase field elucidated the mechanisms contributing to this increased toughness. Finally, the optimized configuration was experimentally validated by fabricating specimens using a polyjet 3D printer. The results represent a significant advancement in maximizing energy absorption, offering a substantial contribution to the field of composite material optimization.