DM net: A multiple nonlinear regression net for the inverse design of disordered metamaterials

Abstract

Mechanical metamaterials have attracted great attention because of their lightweight, high strength, and multi-functional performances. They have also been used in aerospace, shipboard equipment, and other fields. The structural efficiency of uniformly distributed lattice structures is not optimal, and some disordered metamaterial design methods have been studied for very limited types of lattice structures, but existing design methods and algorithms have limitations such as difficulty in adapting to three-dimensional disordered metamaterials, lack of overall framework, and very limited mechanical design parameters. Since finite element analysis contour map can be transformed into an algebra matrix form combining spatial position and stress components, a multivariate nonlinear regression algorithm based on point cloud neural network is designed for disordered metamaterials(DM net). The DM net takes nodal stress matrix in finite element analysis as input parameters and disordered metamaterial design parameters as output parameters and realizes inverse operation of finite element analysis. Considering that the contour maps from finite element analysis can be transformed into matrix form combining position and stress, a disordered metamaterial design method based on the point cloud neural network is proposed. The method takes the nodal stress matrix obtained from finite element analysis as input and outputs the disordered metamaterial design parameter matrix, realizing the inverse operation of finite element analysis. In this method, edge convolution and maximum pooling modules are used to solve interactivity and disorder problems of stress information represented by an algebra matrix. The proposed unified framework can be applied to a wide range of lattice materials and different material properties.