A data-driven inverse design framework for tunable phononic crystals

Abstract

Soft phononic crystals have significant advantages for tuning bandgaps and undergoing reversible large deformations. Despite the superior tunability performance of soft phononic crystals, most existing research methods typically rely on iterative searching aided by researchers’ experience and optimization algorithms to obtain a structure with desired properties. In this paper, we develop an inverse design framework based on deep learning methods. The framework combines Residual Network (ResNet) and Conditional Generative Adversarial Network (CGAN) to establish a bidirectional relationship between tunable phononic crystal structures and their dispersion relations. The results show that the framework can accurately predict the dispersion relations for given structures and design near-optimal structures for the given dispersion relations through a statistical optimization strategy. In addition, the framework can be used to design structures with specific bandgaps. The developed framework can accelerate the design process of tunable phononic crystals and provide new solutions for material design research.