Loading...

Multi-Material Topology Optimization of Metamaterials Using Convolutional Neural Networks

Babaei, Hossein | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57896 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Asghari, Mohsen; Mohammadi, Kaivan
  7. Abstract:
  8. Metamaterials, due to their unique structure-dependent properties, exhibit characteristics beyond those of natural materials. One of the design methods for these materials is topology optimization. This research presents a novel topology optimization approach that directly employs convolutional neural networks (CNNs) to determine material distribution within the design domain. The input features of the neural network are physically meaningful attributes that have a direct correlation with the objective function, applied constraints, and optimal structure. This characteristic accelerates the optimization process and enhances the accuracy of the results. The network architecture is inspired by the well-known U-Net, which has demonstrated a high capability in solving data-driven topology optimization problems. Unlike conventional methods, the proposed approach eliminates the need for additional filtering of the results, significantly reducing the number of gray elements. Moreover, the implementation of manufacturability and geometric constraints is straightforward in this method. Two commonly used constraint symmetry and overhang are incorporated in this study, with the symmetry constraint implemented in a way that reduces computational cost by up to 40%. To evaluate the proposed method, it was implemented in Python using the PyTorch framework, and a set of benchmark topology optimization problems in both single-material and multi-material cases were analyzed. The results indicate that the proposed approach reduces the optimization time compared to conventional methods while producing structures with lower objective function values and minimal gray elements. Furthermore, optimized structures designed using this method were fabricated through additive manufacturing and subjected to three-point bending tests. The experimental results demonstrated that, in optimized components, the ratio of maximum load-bearing capacity to volume was approximately 2.5 times higher than that of solid components, highlighting the effectiveness of this approach in designing optimized structures
  9. Keywords:
  10. Topology Optimization ; Convolutional Neural Network ; Three Dimentional Printing ; Mechanical Metamaterials ; Overhang Constraint

 Digital Object List

 Bookmark

No TOC