Learning the dynamics of metamaterials from diffracted waves with convolutional neural networks

Abstract Conventional methods used to identify the dynamical properties of unknown media from scattered mechanical waves rely on analytical or numerical manipulations wave equation. These show their limitations in scenarios where analyzed medium is moderately sized and diffraction material edges influences fields significantly, such as non-destructive diagnostics metamaterial characterization. Here, we that convolutional neural networks can interpret diffracted learn mapping between all effective parameters including mass density stiffness tensors a small set simulations. Furthermore, trained with synthetic data process physical measurements are very robust measurement errors. More importantly, network provides insight into dynamic behavior matter quantitative measures field sensitivity each property how changes depending under test.