Research on Reverse Design of Band Gap for Elastic Metamaterials Based on Deep Learning

Elastic metamaterials are the artificial microstructures with the extraordinary mechanical and acoustic properties, possessing unique bandgap characteristics. By dynamically adjusting their bandgap design, they can meet the specific requirements in the aerospace field for vibration reduction and noise reduction performance. In this article, based on the deep learning methods, research on reverse design of the elastic metamaterial bandgap is conducted . Firstly, the parameterized curves and expansion functions are utilized to generate sample configurations because of the image processing advantages of deep learning. Convolutional neural networks are chosen to predict the sample band gap, thereby establishing a mapping relationship between the elastic metamaterials and their constitutive frequencies. Secondly, cGAN is used for the reverse design of elastic metamaterials. In the training of neural networks, the new error evaluation method is introduced to improve the accuracy of inverse design. And then, the reverse design of the band gap of elastic metamaterials is implemented, including widening, generation, and amplification of the band gap. Finally, the bandgap of the designed elastic metamaterials were calculated through the finite element method. The results show that the proposed reverse design technique is reliable, which can be effectively used for designing the elastic metamaterials for actively adjusting the bandgap.