Scholarlink

<p>This paper presents a method for predicting the mechanical behavior of patch laminate composites using deep neural networks (DNN) to enhance prediction efficiency. Progressive damage analysis was conducted on patch-laminated materials, simulating fiber and matrix damage using the Hashin damage criterion and delamination failure of interfacial layers through cohesive contact properties. A database of stress–strain curves was constructed based on various patch structural parameters and stacking methods, serving as training inputs for the DNN, with the generated stress–strain curves as the training outputs. The Hyperband optimization algorithm was applied to determine the neural network’s hyperparameters, optimizing its performance. The mechanical behaviors predicted by the numerical simulations and DNN were validated through mechanical testing. The experimental and numerical results show a strong correlation, with the DNN method significantly reducing prediction time while maintaining high accuracy. The results demonstrate that the DNN can quickly and accurately predict the stress–strain curves of arbitrary patch-laminated composites under tensile loading. This approach offers an innovative solution for the optimization and design of composite materials, providing broad application prospects and expected to play a significant role in future material research and engineering design.</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

Study on the Tensile Properties and Damage Analysis of Patch Laminated Composite Based on Deep Neural Networks

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • Xiaowei Zhang,
  • Chengchang Ji,
  • Xinfu Chi,
  • Zhijun Sun,
  • Ma Lei

摘要

This paper presents a method for predicting the mechanical behavior of patch laminate composites using deep neural networks (DNN) to enhance prediction efficiency. Progressive damage analysis was conducted on patch-laminated materials, simulating fiber and matrix damage using the Hashin damage criterion and delamination failure of interfacial layers through cohesive contact properties. A database of stress–strain curves was constructed based on various patch structural parameters and stacking methods, serving as training inputs for the DNN, with the generated stress–strain curves as the training outputs. The Hyperband optimization algorithm was applied to determine the neural network’s hyperparameters, optimizing its performance. The mechanical behaviors predicted by the numerical simulations and DNN were validated through mechanical testing. The experimental and numerical results show a strong correlation, with the DNN method significantly reducing prediction time while maintaining high accuracy. The results demonstrate that the DNN can quickly and accurately predict the stress–strain curves of arbitrary patch-laminated composites under tensile loading. This approach offers an innovative solution for the optimization and design of composite materials, providing broad application prospects and expected to play a significant role in future material research and engineering design.