Mechanical properties of additively manufactured lattice structures designed by deep learning

Abstract

Lattice structures, characterized by their repetitive lightweight cellular forms, enable more effective load distribution compared to solid bodies. Designing lattice structures with tailored mechanical properties remains challenging due to the numerous design variables and their complex relationship with mechanical performance. This paper presents a novel approach employing a deep learning-based Generative Adversarial Network (GAN) model to address this engineering challenge. With its potential for creativity and innovation, GAN provides design diversity that cannot be achieved with traditional design methods or other generative design models. Distinct from previous studies, the GAN training data set consists of lattice structures with improved mechanical properties obtained using parametric design and simulated annealing method. This data set enables the GAN model to create lattice structures with high strength-to-weight ratio. These lattice designs were fabricated using a commercial Material Jetting Additive Manufacturing (MJ-AM) machine, allowing for the production of complex structures. The mechanical performance of the 3D-printed unit cell samples was evaluated through Finite Element Analysis (FEA), compression, and impact testing. The results reveal that the lattice structures generated using the GAN model demonstrated improved mechanical strength (i.e. up to 108 % and 150 % improved strength and elongation performance, respectively). This study shows AI's potential to widen lattice structure design space and create tailored parts with improved mechanical properties. The research also paves the way for future exploration of deep learning techniques in revolutionizing the design and fabrication of parts with tailored mechanical properties.