Scholarlink

<p>Exploring the relationship between the required properties and internal structures has been historically challenging in the development of composite materials. This paper presents an integrated bi-directional computing framework, combining self-consistent clustering analysis (SCA) for forward homogenization and conditional generative adversarial network (CGAN) for inverse structure design. The original SCA method discretizes the original geometry into discrete clusters and solves the Lippman-Schwinger equation for mechanical problems. We have extended the SCA method to solve thermo-elastic, steady thermal conduction and seepage flow problems, enabling rapid, high-fidelity prediction of effective properties from microstructural features. For inverse material design, CGAN directly utilizes the geometric images and homogenized properties generated by the forward homogenization based on SCA as training dataset. Then, the CGAN facilitates inverse design by generating candidate microstructures that meet target Young’s modulus, thermal expansion, thermal conductivity and permeability requirements. This synergy leverages the solving precision of microscopic mechanical models and the generative power of deep learning, enabling bi-directional material design. Glass-fiber-reinforced polypropylene composite materials are taken as the research objects, and the results of inclusion volume fractions and shapes are validated with experiments and the finite element method. The presented framework offers an integrated and robust platform for exploring composites with customized mechanical, thermal and permeability behaviors.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Integrated Bi-directional Computing Framework for Composite Design Based on Self-Consistent Clustering Analysis and Conditional Generative Adversarial Network (SCA-CGAN)

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • Yixin Feng,
  • Wei Luo

摘要

Exploring the relationship between the required properties and internal structures has been historically challenging in the development of composite materials. This paper presents an integrated bi-directional computing framework, combining self-consistent clustering analysis (SCA) for forward homogenization and conditional generative adversarial network (CGAN) for inverse structure design. The original SCA method discretizes the original geometry into discrete clusters and solves the Lippman-Schwinger equation for mechanical problems. We have extended the SCA method to solve thermo-elastic, steady thermal conduction and seepage flow problems, enabling rapid, high-fidelity prediction of effective properties from microstructural features. For inverse material design, CGAN directly utilizes the geometric images and homogenized properties generated by the forward homogenization based on SCA as training dataset. Then, the CGAN facilitates inverse design by generating candidate microstructures that meet target Young’s modulus, thermal expansion, thermal conductivity and permeability requirements. This synergy leverages the solving precision of microscopic mechanical models and the generative power of deep learning, enabling bi-directional material design. Glass-fiber-reinforced polypropylene composite materials are taken as the research objects, and the results of inclusion volume fractions and shapes are validated with experiments and the finite element method. The presented framework offers an integrated and robust platform for exploring composites with customized mechanical, thermal and permeability behaviors.