Design optimization of metastructures integrating machine learning with energy-based homogenization

This study introduces a multi-objective topology optimization framework for designing metastructures with enhanced effective bulk and shear moduli, employing an energy-based homogenization approach. The method utilizes a modified Solid Isotropic Material with Penalization (SIMP) scheme integrated with finite element (FE) analysis, sensitivity analysis, and optimality criteria. The optimization examines diverse initial layouts, volume fractions, and weighting factors to achieve balanced mechanical properties. Additionally, a convolutional neural network (CNN) model was developed to rapidly and accurately predict mechanical properties from microstructural images, attaining coefficients of determination (R 2 ) of 0.94 for bulk modulus and 0.96 for shear modulus. Experimental validation through 3D-printed polylactic acid (PLA) metastructures demonstrated strong agreement with FE simulations, with errors between 3.76% and 7.31%.