Scholarlink

<p>Finite element method (FEM) is pivotal in mechanical engineering, particularly for sensor design and reliability analysis, where high-quality meshing critically impacts computational accuracy and efficiency. However, generating adaptive meshes for complex geometries in traditional FEM relies heavily on computationally expensive posterior error estimation and iterative refinement. To address this, we propose the predictive adaptive meshing framework (PAMF) leveraging two fine-tuned LLM agents for a priori mesh generation and error prediction. By feeding predicted errors back to the mesh generator, PAMF actively refines meshes, significantly improving success rates. For validation, we developed an APDL code library with 100 mechanical models, featuring representative sensor geometries, under fixed constraints and force loads. Experiments show PAMF achieves 88.4 % mesh success rate, with 66.8 % of meshes yielding below 10 % error (Zienkiewicz-Zhu estimation), accelerating generation by 3.69 times versus traditional methods. This framework demonstrates LLMs’ potential to revolutionize efficiency in mechanical simulation. To ensure full transparency and reproducibility, the complete dataset, all prompts, and related code for this study are publicly available at: <a href="https://github.com/2002-Pan/PAMF">https://github.com/2002-Pan/PAMF</a>.</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

PAMF: An LLM-driven framework for automated mesh generation in mechanical simulation and CAE workflows

  • 期刊论文

  • 发表日期：

    2026年6月3日

查看全文
  • Feiyang Pan,
  • Chunyan Yin,
  • Guangbin Dou

摘要

Finite element method (FEM) is pivotal in mechanical engineering, particularly for sensor design and reliability analysis, where high-quality meshing critically impacts computational accuracy and efficiency. However, generating adaptive meshes for complex geometries in traditional FEM relies heavily on computationally expensive posterior error estimation and iterative refinement. To address this, we propose the predictive adaptive meshing framework (PAMF) leveraging two fine-tuned LLM agents for a priori mesh generation and error prediction. By feeding predicted errors back to the mesh generator, PAMF actively refines meshes, significantly improving success rates. For validation, we developed an APDL code library with 100 mechanical models, featuring representative sensor geometries, under fixed constraints and force loads. Experiments show PAMF achieves 88.4 % mesh success rate, with 66.8 % of meshes yielding below 10 % error (Zienkiewicz-Zhu estimation), accelerating generation by 3.69 times versus traditional methods. This framework demonstrates LLMs’ potential to revolutionize efficiency in mechanical simulation. To ensure full transparency and reproducibility, the complete dataset, all prompts, and related code for this study are publicly available at: https://github.com/2002-Pan/PAMF.