Multiscale topology optimisation with nonparametric microstructures using three-dimensional convolutional neural network (3D-CNN) models

ABSTRACT

Additive manufacturing enables the fabrication of parts with complex geometries, thereby opening up the design space from part scale to microarchitecture scale. By optimising the structure in the expanded design space, structural performance can be improved. Topology optimisation is commonly used as the tool to optimise the structures according to specific application requirements. However, multiscale topology optimisation can be computationally expensive and with limited choices in microscale structures. Therefore, we propose a surrogate model based on three-dimensional convolutional neural networks (3D-CNN) to model the effective elasticity tensor and its gradients for general voxel-based nonparametric microstructures. The proposed 3D-CNN-based surrogate model greatly extends the flexibility over existing surrogate-based methods which can only be applied in relatively simple parametric microstructures. Given the microscale structure, the proposed 3D-CNN-based model can effectively predict its material properties. Furthermore, being able to estimate the gradient of the material properties with respect to microscale structure changes makes the proposed 3D-CNN-based surrogate readily adaptive to existing multiscale topology optimisation frameworks. Through extensive simulations, by comparing with both SIMP and existing surrogate-based methods, we demonstrate the advantages of the proposed 3D-CNN-based surrogate model.

KEYWORDS:

• Topology optimisation
• multiscale analysis
• neural networks
• surrogate model

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Guo Yilin

Guo Yilin received the B.E. degree from the Department of Mechanical Engineering at NUS and M.Sc. degree from the Department of Industrial and System Engineering at NUS. She is currently a Ph.D. candidate in the Department of Mechanical Engineering, NUS, and now works as a research associate in NUS Centre of Additive Manufacturing (AM.NUS). Her research interests include deep learning, generative models, and topology optimization with applications to manufacturability assessment and optimal design in additive manufacturing.

Jerry Fuh Ying Hsi

Dr Jerry Fuh Ying Hsi is a Professor in the Department of Mechanical Engineering, NUS and the Co-Director of NUS Centre for Additive Manufacturing (AM.NUS). He has devoted himself to the research of additive manufacturing or 3D printing processes and materials since 1995. In 2016, he and his colleagues established a 3DP-enable biomedical hub within the NUS premise across five different faculties, i.e. engineering, science, design, dental and medicine, to be one of its few kinds to promote transnational works on biomedical and healthcare research.

Lu Wen Feng

Dr Lu Wen Feng is currently the director of Additive Manufacturing center (AM.NUS) and the Associate Professor of Department of Mechanical Engineering at NUS. He received his Ph.D. degree in Mechanical Engineering from University of Minnesota, USA. He had been the group manager of Singapore Institute of Manufacturing Technology and the director of research and industry for Design Technology Institute at NUS. His research interests include additive manufacturing, bioprinting, design technology, industry informatic, and intelligent manufacturing.