3D Denoising Diffusion Probabilistic Models for 3D microstructure image generation of fuel cell electrodes
Abdelouahid Bentamou, Stéphane Chrétien, Yann Gavet
Abstract
The generation of realistic 3D microstructure images is crucial for understanding and optimizing materials in various fields, including fuel cell technology. In this article, we present a novel approach for generating synthetic 3D microstructure images using 3D Denoising Diffusion Probabilistic Models (3D DDPM). This approach extends to n-phase materials. Unlike conventional image generation techniques, our method leverages the principles of diffusion in three-dimensional space to simulate the intricate evolution of microstructures. By incorporating stochastic processes and diffusion equations, 3D DDPMs enable a more realistic and controlled representation of the dynamic processes occurring within materials. This approach generates synthetic microstructures that capture the spatial complexities inherent in real-world materials across multiple phases. Through experimental evaluation, we demonstrate that our approach generates realistic 3D microstructure images of O 2 fuel cell electrodes for two or three phases. • A novel approach using 3D DDPMs to generate synthetic 3D microstructure images. • The method supports n-phase materials, extending to complex multiphase systems. • Stochastic processes simulate 3D microstructure evolution with diffusion equations. • Generates realistic 3D microstructures of O 2 fuel cell electrodes. • Experiments show the generated microstructures resemble real-world material structures.