Deep neural network-based method for designing tunable band gaps metamaterials based on shape memory alloy embedded perforated elastic metamaterials

Abstract This paper proposes a deep neural network-based optimization strategy for elastic metamaterials. It aims to attain excellent tunable elastic wave attenuation performance of shape memory alloy (SMA) embedded perforated plates by optimizing the shapes of the SMA inclusions. Firstly, the design of the SMA-embedded perforated plate is presented. By utilizing the property of SMA to phase change between martensite and austenite, a new design for achieving tunable bandgaps is introduced. The finite element method based on the Bloch-Floquet theorem is used to solve the tunable energy band structure, and the effects of geometrical variations of the SMA on the width and position of the tunable bandgap are explored. Next, a deep neural network is employed to establish the relationship between the geometrical parameters of the SMA and the tunable bandgap. The accuracy of the agent model is verified by performance evaluation. Finally, a strategy combining genetic algorithms and deep neural networks is proposed for inverse design optimization to obtain metamaterials with superior tunable bandgap performance. The results of five optimization cases demonstrate that the proposed strategy performs well in terms of computational efficiency and real-time design of multiple sets of targets. This study provides an important reference for the development and application of advanced elastic metamaterials with tunable functions.