航空学报 > 2025, Vol. 46 Issue (24): 231877-231877   doi: 10.7527/S1000-6893.2025.31877

固体力学与飞行器总体设计

基于深度学习的复合材料结构性能参数反演

项子健, 麻震宇(), 杨希祥   

  1. 国防科技大学 空天科学学院,长沙 410073
  • 收稿日期:2025-02-19 修回日期:2025-03-06 接受日期:2025-04-25 出版日期:2025-07-03 发布日期:2025-05-13
  • 通讯作者: 麻震宇 E-mail:yuyu1031@163.com
  • 基金资助:
    国家自然科学基金(52372438);国家自然科学基金(51605484)

Inversion of structural performance parameters of composite materials based on deep learning

Zijian XIANG, Zhenyu MA(), Xixiang YANG   

  1. College of Aerospace Science and Engineering,National University of Defense Technology,Changsha 410073,China
  • Received:2025-02-19 Revised:2025-03-06 Accepted:2025-04-25 Online:2025-07-03 Published:2025-05-13
  • Contact: Zhenyu MA E-mail:yuyu1031@163.com
  • Supported by:
    National Natural Science Foundation of China(52372438)

摘要:

提出了一种基于深度学习的复合材料结构性能参数反演方法。传统的复合材料性能测试方法通常需要大量的实验资源,而本研究将有限元仿真与神经网络模型相结合,实现了复合材料层合板参数的高效准确识别。研究探索了2种深度学习模型在参数反演中的性能,并分析了模型超参数、特征类型、数量对反演精度的影响。结果表明,所提方法能够准确识别复合材料的弹性模量、泊松比、强度等参数,误差均小于10%,验证了基于深度学习的复合材料参数反演方法的有效性。此外,分析了数据集构成、铺层信息对参数反演精度的影响,为进一步优化复合材料结构设计提供了理论依据。

关键词: 复合材料, 参数反演, 深度学习, 特征融合, 有限元方法

Abstract:

This paper proposes a deep-learning-based inversion method for determining the structural performance parameters of composite materials. Traditional testing methods for composite material performance typically require substantial experimental resources. In contrast, this study integrates finite element simulation with neural network models to achieve efficient and accurate identification of composite laminate parameters. This study evaluates the performance of deep learning models in parameter inversion and analyzes the effects of model hyperparameters, feature types, and quantities on inversion accuracy. The results demonstrate that the proposed method can accurately identify the elastic modulus, Poisson’s ratio, strength, and other parameters of composite materials, with errors of less than 10%, except for Poisson’s ratio, thus validating the effectiveness of the deep-learning-based composite material parameter inversion method. Additionally, this study examines the impact of dataset composition and layer information on the accuracy of parameter inversion, providing a theoretical foundation for further optimization of composite material structure design.

Key words: composite material, parameter inversion, deep learning, feature fusion, finite element method

中图分类号: