Manufacturability-aware deep generative design of 3D metamaterial units for additive manufacturing

Mechanical metamaterials are artificial structures that possess exceptional mechanical properties that are not naturally occurring. The complex geometrical and topological features of these metamaterials pose significant challenges to both structure design and manufacturing, despite the recent rapid development of additive manufacturing (AM) techniques. Thus, an effective framework for designing 3D metamaterials with desired mechanical properties, while also ensuring AM manufacturability, is urgently needed. In this paper, an AM manufacturability-aware deep generative model-based design framework is proposed for designing 3D metamaterial units for target properties. To accomplish this, we propose using Variational Autoencoder (VAE) as the feature extractor, which maps the 3D metamaterial geometries to a low-dimensional latent feature space. The latent feature space is concurrently linked to discriminators/regressors to predict manufacturability metrics and mechanical properties. We demonstrate that the proposed design framework is capable of designing high-performance metamaterial units with various user-defined manufacturability metrics. To showcase the effectiveness of the proposed design framework, three design cases with different objective functions are presented, and the final optimal designs are validated by comparing them to state-of-the-art designs or the optimal designs obtained by topology optimization methods.