3D MNetworks 3 and Y. Adachi 1 K. Sugiura 1, T. T. Chen 1, F. Sun 1, T. Ogawa 1, 2, I. Watanabe1 Department of Materials Design Innovation Engineering, Nagoya University 2 Department of Mechanical Engineering, Faculty of Engineering, Aichi Institute of Technology 3 Center for Basic Research on Materials, National Institute for Materials Science (NIMS) 78PP44--1199 icrostructure Reconstruction of MFor a more precise development of metallic materials, including steel, it is essential to analyze not just 2D microstructures, but also their 3D microstructures. Traditionally, experiments and simulations have been used to obtain 3D microstructures, but they require a significant amount of time and effort. To address this challenge, we attempted to reconstruct 3D microstructures using Generative Adversarial Networks (GANs). Specifically, we developed a program based on the GANs algorithm "SliceGAN" proposed by S. Kench et al.1), which generates 3D microstructures from only three orthogonal cross-sectional images. GANs are a type of deep learning technique proficient in image generation. GANs consist of two types of AI models: the Generator, which creates fake images, and the Discriminator, which determines if images are real or fake. GANs autonomously learn the features of the given data and can generate synthesized data that closely resembles the original. With GANs, we can quickly obtain large quantities of 3D microstructures of any size, regardless of the material type or image scale. This offers the potential for obtaining 3D microstructures that were technically challenging with traditional methods and realizing 3D microstructure reconstruction in a shorter and simpler manner. Reference: 1) S. Kench et al., Nat. Mach. Intell. 2021, 3, 299. Poster Presentation ｜NIMS Award Symposium 2023etallic Materials Using Generative Adversarial P4 ｜ Modeling
元のページ ../index.html#78