Scholarlink

<p>A deep learning-based method for predicting the optimized mesh of metallic foams is established to enhance the efficiency of the preprocessing for finite element (FE) modeling of metallic foams and reduce the reliance on expert experience. A framework for predicting the non-uniform mesh density of aluminum foam based on a conditioned 3D U-Net (C-3D-U-Net) is proposed. A voxel dataset of size 256³ was constructed using a foam aluminum generation plugin based on a Voronoi-based aluminum foam generation plugin and voxelization processing, with the mesh density information encoded as voxel values. Meanwhile, the stress-strain curves obtained via Abaqus were employed as conditional controls for the network. A C-3D-U-Net model was developed, and the stress-strain curves were injected into each stage of the decoder through the adaptive instance normalization (AdaIN) module as conditions to guide the network to generate non-uniform mesh density distributions that conform to mechanical properties. The results show that the non-uniform mesh partitioning scheme predicted by the model saves an average of 10.4% of the computing time while maintaining a high consistency with the simulation results of the original high-density mesh. This study provides a new data-driven solution for the automation of preprocessing in finite element simulation of complex structures and offers an approach to address the data volume challenge in applying deep learning methods to materials science.</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

Prediction of non-uniform meshes for accelerated simulation of aluminum foam based on conditioned 3D U-Net

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Weimin Zhuang,
  • Bu Yang,
  • Fan Zhang,
  • Xin Deng

摘要

A deep learning-based method for predicting the optimized mesh of metallic foams is established to enhance the efficiency of the preprocessing for finite element (FE) modeling of metallic foams and reduce the reliance on expert experience. A framework for predicting the non-uniform mesh density of aluminum foam based on a conditioned 3D U-Net (C-3D-U-Net) is proposed. A voxel dataset of size 256³ was constructed using a foam aluminum generation plugin based on a Voronoi-based aluminum foam generation plugin and voxelization processing, with the mesh density information encoded as voxel values. Meanwhile, the stress-strain curves obtained via Abaqus were employed as conditional controls for the network. A C-3D-U-Net model was developed, and the stress-strain curves were injected into each stage of the decoder through the adaptive instance normalization (AdaIN) module as conditions to guide the network to generate non-uniform mesh density distributions that conform to mechanical properties. The results show that the non-uniform mesh partitioning scheme predicted by the model saves an average of 10.4% of the computing time while maintaining a high consistency with the simulation results of the original high-density mesh. This study provides a new data-driven solution for the automation of preprocessing in finite element simulation of complex structures and offers an approach to address the data volume challenge in applying deep learning methods to materials science.