Tunable anisotropy in lattice structures via deep learning-based optimization

Abstract

The advancements in additive manufacturing have sparked extensive exploration of lattice truss structures, renowned for their exceptional properties suitable for diverse engineering applications. Achieving a high mechanical modulus while maintaining low density poses a challenge, prompting numerous studies aimed at addressing the trade-off relationship between porosity and modulus. However, most efforts have concentrated on enhancing modulus for unidirectional loads, a limitation in addressing the random loads commonly found in industrial settings. To tackle this issue, our study designs lattice structures that achieve significantly reduced anisotropy at low densities, employing a combination of neural networks and genetic optimization. This novel approach allows for the efficient derivation of optimized models. Importantly, our approach solely alters the beam shapes within the basic lattice structure configurations, facilitating manufacturability without reducing average stiffness. Furthermore, our research elucidates the mechanism behind shape-dependent anisotropy and confirms these findings through both numerical and experimental results, offering insights into the design limitations and potential for next-generation lightweight structures.