Abstract

In contrast to the deterministic characterization of wave response fixed by the medium’s mechanical property, the elastic metasurface, theoretically grounded in the generalized Snell’s law, manipulates the wave propagation by controlling the phase of each unit cell. While elastic metasurface has advantages of implementation due to its thinner structures than the metamaterial, it requires large computational burden to design each spatially different unit cell. Additionally, the multi-functional elastic metasurface, enhanced versatility version, needs much larger trail-and-errors to find the optimal design with heavy computational costs based on the traditional optimization approach combined with data-driven methods. Therefore, this study presents deep learning-based end-to-end approach, incorporated forward and inverse model, for the inverse design of multi-functional elastic metasurface to replace computationally inefficient conventional inverse design framework.