High-resolution topology optimization method of multi-morphology lattice structures based on three-dimensional convolutional neural networks (3D-CNN)

The application of lattice structures provides significant benefits for lightweight structural design. To further strengthen structural stiffness, multi-morphology lattice structures are integrated into topology optimization. Considering the high costs associated with microstructural mechanical calculations and modeling, a novel three-dimensional Convolutional Neural Network (3D-CNN) with Transfer Learning (TL) is proposed to rapidly predict the performance of lattice structures with any morphology. The optimization framework is reconstructed to accommodate multi-morphology lattice structure design, combining density updates with cell topology iteration using a modified sensitivity formula. Furthermore, a cutting-edge post-processing method based on 3D-CNN is employed to achieve a substantial improvement in structural resolution levels within acceptable costs. Through comprehensive simulations comparing with both single-morphology and existing multi-morphology optimizations of lattice structures, we demonstrate the superiority of our proposed approach. Lastly, the effectiveness of the result through post-processing is validated by the Finite Element Method (FEM).