Design optimization of additively manufactured sandwich beams through experimentation, machine learning, and imperialist competitive algorithm

Abstract

The primary aim of this study was to optimize 3D-printed sandwich beams incorporating auxetic cores. Four key design variables were considered: core thickness, width, and two specific geometric parameters related to the auxetic core–the cell's wall thickness-to-length ratio and the cell angle. Initially, sixteen sandwich samples, each varying in geometric parameters, were printed following Taguchi's experimental design. These samples underwent a three-point bending test, and the resulting data were used to train a deep neural network (DNN). The trained DNN proficiently predicted the bending response of the structures. Another network, using a similar architecture, was trained to precisely predict the structure's mass. These trained networks were then integrated into the imperialist competitive algorithm (ICA) to optimize the sandwich beams design for minimal mass. A contour plot was subsequently generated to visually illustrate the minimum required mass distribution for sandwich beams under various conditions and constraints, considering applied bending forces and maximum allowable deflection. The optimization results underscored the importance of fabricating and testing an adequate number of sandwich samples, aligning with the number of design parameters. The results provide a valuable reference for future research on designing advanced 3D-printed lightweight structures, including integrating machine learning techniques and metaheuristic algorithms.