关键词: 多孔材料, 多相, 微观结构, 重构, 高阶统计, 深度卷积生成式对抗网络模型

Abstract: Porous materials exhibit typical heterogeneous characteristics, with their macroscopic properties fundamentally governed by microstructural features. Therefore, establishing an accurate microstructure characterization system and realizing high-fidelity numerical modeling were the core scientific issues to reveal the structure-activity relationship of materials. This study proposes a deep learning-based reconstruction method for multiphase anisotropic porous microstructure. By integrating a deep convolutional generative adversarial network model with a slice sampling strategy, and decomposing the complex multiphase reconstruction task into a series of two-phase reconstruction tasks based on the One-hot encoding principle, the proposed method efficiently reconstruct multiphase anisotropic porous materials. In addition, a three-point statistical correlation function algorithm is introduced as a high-order statistical controller for the reconstruction model, addressing the limitations of traditional second-order statistical descriptors and enabling precise quantification of high-order microstructural features. This work provides valuable theoretical insights and practical guidance for performance prediction and microstructure optimization of porous materials.

Key words: porous material, multiphase, microstructure, reconstruction, high-order statistic, deep convolutional generative adversarial network model