Machine-learning based design of digital materials for elastic wave control

Materials for wave control need to be both anisotropic and spatially distributed. Traditional method is to first design a microstructure with anisotropic property, and then change geometric parameters of the microstructure according to analytical theory or numerical calculation. Unlike the traditional method, mechanical properties of digital materials can be easily tuned by changing the 0/1 ordering without changing the geometry of digital materials. However, determining suitable orderings of digital materials according to target properties remains a key challenge. In this paper, we establish a digital structural genome to solve this problem. By combining the developed machine learning method with finite element method, we can quickly calculate elastic wave properties of digital materials with all orderings and finally establish a digital structural genome quickly and accurately. The complete digital structural genome provides us with a fast approach to design anisotropy and spatial distribution of materials. Our research unequivocally shows that the establishment of a complete structural genome database of digital materials is of great significance for inverse design multifunctional structures, and can opens an avenue to achieve wave control on demand, such as corner cloak and acoustic carpet cloak.