Parametric modeling and deep learning-based forward and inverse design for acoustic metamaterial plates

Abstract

Acoustic metamaterials (AMMs), with extraordinary physical properties, could be used to suppress the specific frequencies elastic waves by altering the structure artificially. However, as the complexity of AMMs structures continues to rise, classical design methods are time-consuming and high-computational. Therefore, there is a pressing need to explore more efficient and accurate design methods for AMMs. In this work, a deep learning algorithm-based forward and inverse design method for acoustic metamaterial plates (AMPs) is proposed. First, the initial samples of AMPs are created with parametric model and the bandgaps properties of the AMPs are generated by the finite element method. The dataset consists of different structure parameters and corresponding bandgap characteristics. Then, A neural network model is constructed by concatenating a pretraining network and an inverse design network. Through inputting the dataset to the concatenated network, the mapping relationship between the structural parameters and the bandgap characteristics of the AMPs can be explored. Ultimately, the trained network enables both forward designs, yielding bandgap characteristics for given structural parameters, and inverse design, deducing structural parameters for specific bandgap characteristics. The accuracy of the proposed design methodology is verified through illustrative examples. The results demonstrate that the trained neuron networks can effectively replace the complex physical mechanisms between the structural parameters and bandgap characteristics.