Browse by author
Lookup NU author(s): Dr Xinwei LiORCiD
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
© 2024Lattice metamaterials have received extensive research interest for their superior mechanical properties. However, because of the nonlinear relationship between structure and mechanical responses, designing lattice metamaterials with optimal plastic response for energy absorption still remains a challenge. This study proposes a novel convolutional neural network (CNN) accelerated design approach for lattice metamaterials with customized mechanical properties. The CNN input datasets consist of pseudo-randomized strut geometries and their respective strain-stress curves that are simulated. The trained CNN shows the capabilities of accurate prediction and exploration of new lattice metamaterials that surpass the performance of the original input datasets. From the extended datasets, a new lattice metamaterial with the highest specific energy absorption is obtained, which also surpasses other lattice metamaterials reported in the literature. The new lattice metamaterial exhibits a stretch-dominated failure, with stresses distributed equally in the vertically aligned regions. These regions are supported by the adjacent unit cells, which avoids large expansion and further strengthens load-bearing capability during compression. Overall, the proposed design approach is based on a data-driven model without relying on the complex theory of solid mechanics, which is extendable for numerous new engineering applications.
Author(s): Zhao M, Li X, Yan X, Zhou N, Pang B, Peng B, Zeng Z
Publication type: Article
Publication status: Published
Journal: Thin-Walled Structures
Year: 2025
Volume: 208
Print publication date: 01/03/2025
Online publication date: 15/12/2024
Acceptance date: 14/12/2024
ISSN (print): 0263-8231
ISSN (electronic): 1879-3223
Publisher: Elsevier Ltd
URL: https://doi.org/10.1016/j.tws.2024.112845
DOI: 10.1016/j.tws.2024.112845
Altmetrics provided by Altmetric
