Scholarlink

In this study, we introduce a novel algorithm that utilizes graph neural networks (GNNs) to predict the Mises equivalent stress distribution throughout three-dimensional (3D) elastic bodies, based solely on the surface node deformation displacements under tensile static loads. Traditional methods for stress analysis in solid mechanics often involve inputting constraints and loading conditions in the finite element method (FEM), which can be complex and computationally intensive. Our approach represents a significant advancement by directly inferring stress distributions from surface deformation data, thereby eliminating the need for volumetric mesh generation and intensive numerical simulations. By training the GNN on a dataset of structural deformations and corresponding stress fields generated by FEM, the algorithm achieves high predictive accuracy, with a mean error of 0.53%. The proposed methodology holds promising applications in structural health monitoring, non-destructive testing, and predictive maintenance, offering a new paradigm for efficient and accurate stress analysis in engineering and material science.

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

Graph Neural Network-Based Three-Dimensional Stress Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yexing Yang,
  • Xinzhan Hong,
  • Na Liu,
  • Xuefeng Li,
  • Meng Zhao,
  • Jing Ji

摘要

In this study, we introduce a novel algorithm that utilizes graph neural networks (GNNs) to predict the Mises equivalent stress distribution throughout three-dimensional (3D) elastic bodies, based solely on the surface node deformation displacements under tensile static loads. Traditional methods for stress analysis in solid mechanics often involve inputting constraints and loading conditions in the finite element method (FEM), which can be complex and computationally intensive. Our approach represents a significant advancement by directly inferring stress distributions from surface deformation data, thereby eliminating the need for volumetric mesh generation and intensive numerical simulations. By training the GNN on a dataset of structural deformations and corresponding stress fields generated by FEM, the algorithm achieves high predictive accuracy, with a mean error of 0.53%. The proposed methodology holds promising applications in structural health monitoring, non-destructive testing, and predictive maintenance, offering a new paradigm for efficient and accurate stress analysis in engineering and material science.