Inverse design of topological metaplates for flexural waves with machine learning

Abstract

The mechanical analog to the topological insulators brings anomalous elastic wave properties which diversifies classic wave functions for potential broad applications. To obtain topological mechanical wave states with good quality at desired frequency ranges, it needs repetitive trials of different geometric parameters with traditional forward designs. In this work, we develop an inverse design of topological edge states for flexural wave using machine learning method which is promising for instantaneous design. Nonlinear mapping function from input targets to output desired parameters are adopted in artificial neural networks where the data sets for training are generated by the plane wave expansion method. Topological edge states are then realized and compared for different bandgap width conditions with such inverse designs, proving that wide bandgap can promote the confinement of the topological edge states. Finally, direction selective propagations with sharp turns are further demonstrated as anomalous wave behaviors. The machine learning inverse design of topological states for flexural wave provides an efficient way to design practical devices with targeted needs for potential applications such as signal processing, sensing and energy harvesting.