Scholarlink

<p>The design of conventional three-dimensional (3D) orthogonal woven composites often relies on extensive theoretical computations, time-consuming simulations, or costly experimental testing. These methods involve high expenses and extended development cycles, which pose significant challenges to rapid design processes. This study develops a parametric multi-scale finite element (FE) model along with a corresponding artificial neural network (ANN) surrogate model for predicting the elastic properties of three-dimensional orthogonal woven composites. The FE model systematically investigates the influence of various parameters on elastic constants, demonstrating less than 4% deviation in both tensile and shear moduli compared to mechanical tests. An automated workflow bridging TexGen and Abaqus was employed to generate a dataset of 4655 samples, encompassing variations in microstructural composition, yarn fabrication, preform weaving, and curing conditions. Based on this dataset, an artificial neural network-based surrogate model was trained, achieving a mean absolute percentage error of only 2.34% relative to the full FE simulations, while reducing computational time by a factor of 58,000. This integrated framework provides a robust foundation for the rapid design and optimization of 3D orthogonal woven composites, establishing an efficient pathway for the development of advanced fiber-reinforced materials.</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 ANN-Based Prediction Method for Mechanical Properties of Hybrid Reinforced Composites

  • 期刊论文

  • 发表日期：

    2025年12月24日

查看全文
  • Jun Ke,
  • Rongkun Li,
  • Jiaqiang Zhang,
  • Zhenyu Wu,
  • Zhongping Le,
  • Wenqi Lu

摘要

The design of conventional three-dimensional (3D) orthogonal woven composites often relies on extensive theoretical computations, time-consuming simulations, or costly experimental testing. These methods involve high expenses and extended development cycles, which pose significant challenges to rapid design processes. This study develops a parametric multi-scale finite element (FE) model along with a corresponding artificial neural network (ANN) surrogate model for predicting the elastic properties of three-dimensional orthogonal woven composites. The FE model systematically investigates the influence of various parameters on elastic constants, demonstrating less than 4% deviation in both tensile and shear moduli compared to mechanical tests. An automated workflow bridging TexGen and Abaqus was employed to generate a dataset of 4655 samples, encompassing variations in microstructural composition, yarn fabrication, preform weaving, and curing conditions. Based on this dataset, an artificial neural network-based surrogate model was trained, achieving a mean absolute percentage error of only 2.34% relative to the full FE simulations, while reducing computational time by a factor of 58,000. This integrated framework provides a robust foundation for the rapid design and optimization of 3D orthogonal woven composites, establishing an efficient pathway for the development of advanced fiber-reinforced materials.