Machine learning-combined topology optimization for functionary graded composite structure design

Abstract

This study presents new framework in which the representative volume element (RVE) method and machine learning (ML) model are used to construct continuous anisotropic effective material properties for simultaneous design of the overall topology configuration and local fiber material layout in functionally graded composite structures. It is an alternative to the asymptotic homogenization design method (AHDM) to obtain continuous effective material property functions. While the AHDM uses the asymptotic homogenization theory (AHT) and Legendre polynomials, the RVE method calculates anisotropic effective material properties having nonlinear behavior with respect to design variables of microstructures, and it is easier to implement than AHT given the governing equations and appropriate boundary conditions. More efficient and accurate than Legendre polynomials, ML is used to build a continuous model of the RVE results required for simultaneous design of the overall topology configuration and local fiber material layout. To show the convenience and expandability of the proposed method, a 3D RVE model is also proposed through the extension of the 2D model. The proposed method is verified through 2D and 3D numerical examples to minimize structural compliance and obtained results are compared with those from the application of AHDM.