Page 24 - IJAMD-2-3

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

A

B C

Figure 10. Structures of diffusion models and normalizing flows, as well as research utilizing diffusion models in the field of composite materials. (A) Structure
of diffusion models and NFs; (B) The training process of the conditional diffusion model and comparison between the generated microstructures and the
original microstructures; (C) Metamaterial synthesis for four stress-strain responses. 130
128
thermal conductivities was conducted through Bayesian original training set. NFs are also implemented by Mirzaee
inference, combining principal component analysis to and Kamrava to generate microstructures of porous
133
reduce dimensionality, a VAE for data compression and to materials for a given target property from CNN-encoded
learn the prior distribution, and an NF model to generate 3D images via stochastic inverse design. NF models enable
new samples. The framework was able to identify a range the incorporation of the stochastic nature of composite
132
of novel microstructures with a target property, rather than materials, providing a more realistic approach to design
a single optimal sample, including samples beyond the and optimization.

Volume 2 Issue 3 (2025) 18 doi: 10.36922/IJAMD025210016