Arbitrary target frequency cloaking for flexural waves using deep learning

Differing from electromagnetic and acoustic waves, the governing equation for elastic waves in flexural plates is not form invariant, hindering straightforward cloak design based on coordinate transformation theory. In this work, we propose a novel idea instead of approximately equivalent transformation method, and employ scattering cancellation techniques to design a multi-layer cylindrical structure for achieving flexural wave cloaking at desired target frequencies. Moreover, we use deep learning to effectively address the time consuming issue dealing with fine-tuning design parameters for the desired response. More importantly, we adopt a method based on a tandem neural network to tackle the one-to-many mapping challenge in inverse design. It not only accurately predicts the scattering spectra of multi-layer structures in advance but also efficiently performs inverse design to obtain the required design parameters for arbitrary target frequency cloaking.