Scholarlink

<p>This study proposes a data-driven computational mechanics framework for predicting the damage behavior of hyperelastic materials under complex cyclic loading. Leveraging prior knowledge of the underlying physical mechanisms and an established modeling framework, the 3D stress-strain data required for constitutive modeling are reduced to 1D datasets. A recurrent neural network (RNN) is trained on uniaxial cyclic loading data to capture the stress-strain response, and the trained model is subsequently embedded into a finite element solver. This approach allows 3D structural simulations under complex cyclic loads to be driven by 400 uniaxial test data samples. The predictive accuracy of the proposed approach is validated against the classical Mullins damage model, demonstrating its effectiveness. Finally, the limitations of the present method and potential directions for future improvement are discussed.</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

Mechanistic Data-Driven Modeling on the Damage of Hyperelastic Materials Under Cyclic Loading

  • 期刊论文

  • 发表日期：

    2025年10月14日

查看全文
  • Min Tang,
  • Gang Zhang,
  • Zherui Liu,
  • Lijie Liu,
  • Zefeng Yu,
  • Shan Tang

摘要

This study proposes a data-driven computational mechanics framework for predicting the damage behavior of hyperelastic materials under complex cyclic loading. Leveraging prior knowledge of the underlying physical mechanisms and an established modeling framework, the 3D stress-strain data required for constitutive modeling are reduced to 1D datasets. A recurrent neural network (RNN) is trained on uniaxial cyclic loading data to capture the stress-strain response, and the trained model is subsequently embedded into a finite element solver. This approach allows 3D structural simulations under complex cyclic loads to be driven by 400 uniaxial test data samples. The predictive accuracy of the proposed approach is validated against the classical Mullins damage model, demonstrating its effectiveness. Finally, the limitations of the present method and potential directions for future improvement are discussed.