Page 22 - IJAMD-2-3

International Journal of AI for
Materials and Design AI applications in composite materials

A

B C

Figure 9. Structure of GAN and research utilizing GAN in the field of composite materials. (A) Structure of GAN; (B) Strain and stress field prediction
results using conditional GAN and comparison with FEM (ground truth). Reprinted with permission from Yang et al. Copyright © 2021 The American
122
122
Association for the Advancement of Science; (C) Stress field results derived from the microstructure of chopped fiber composites using conditional GAN. 110
Abbreviations: FEM: Finite element method; GAN: Generative adversarial network.
as the fiber diameter, fiber volume fraction, fiber spatial potential of GANs for complex multi-variate, multi-
distribution, and resin-rich regions, which traditional objective material design challenges in civil engineering
random microstructure generators struggled to represent. applications.
Comparisons of the generated microstructures with real Another innovative application of GANs is in topology
data highlighted the accuracy of GANs in modeling the optimization. Li et al. integrated GANs with subset
125
detailed characteristics of composite structures. simulation to guide the design of periodic structures
In addition, GANs have been utilized for multi- with desired bandgap properties. This hybrid approach
objective optimization in the design of engineered allows for efficient generation of rare samples in high-
cementitious composites. The tensile stress, strain, dimensional design spaces, facilitating the identification of
124
and cost are optimized simultaneously for varying optimal topologies for composite structures. The method
mixture proportions and fiber types, demonstrating the has proven to be effective in the topology optimization of

Volume 2 Issue 3 (2025) 16 doi: 10.36922/IJAMD025210016