Deep learning-based inverse design of lattice metamaterials for tuning bandgap

Abstract

In this paper, deep learning neural networks is used to predict the band structure of metamaterial lattices, and proactive inverse design is employed in bandgap modulation. A parametric design of the metamaterial lattice is proposed to achieve a rich design space. The corresponding band structure data is calculated by finite element method (FEM) to construct the data set. We successfully bypass complex theoretical or numerical methods to establish the mapping relationship between the lattice geometry parameters of metamaterials and the band structure data by constructing and training fully connected neural networks and convolutional neural networks (CNN). By combining the trained neural network model into an inverse design method of bandgap tuning, the geometric parameters of the metamaterial lattice can be obtained directly by inputting the target band structure. Finally, three object band structures are designed and verified by finite element simulation and experiment, which verifies the effectiveness of the inverse design method. This design approach can be extended to design other metamaterial properties.