Scholarlink

<p>Topology optimization is a crucial technique for designing lightweight structures, widely utilized across various engineering disciplines. Despite ongoing research integrating various deep learning techniques into topology optimization, there remains a significant gap in addressing the complexity of diverse shapes within design domains. To bridge this gap, we introduce Disp2Topo, a novel methodology employing the Pix2Pix deep learning framework tailored for topology optimization. Disp2Topo generates training data through finite element analysis and topology optimization processes, subsequently training a neural network to predict structures with relative density values that closely mimic those derived from traditional analysis. Unlike conventional methods that primarily focus on external load information, Disp2Topo leverages deformation across all structure nodes, facilitating a more nuanced differentiation between design and non-design domains and substantially reducing computational costs. Our experimental findings demonstrate that Disp2Topo can achieve results comparable to the conventional SIMP method, while drastically reducing computation time to approximately 0.2% of the traditional approach. This represents a significant advancement in efficiently producing optimized structures tailored to varied design requirements.</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

Disp2Topo Method: A Novel Deep Learning Framework for Efficient Topology Optimization

  • 期刊论文

  • 发表日期：

    2025年2月22日

查看全文
  • Woongbeom Heogh,
  • Yong Son,
  • Hyub Lee

摘要

Topology optimization is a crucial technique for designing lightweight structures, widely utilized across various engineering disciplines. Despite ongoing research integrating various deep learning techniques into topology optimization, there remains a significant gap in addressing the complexity of diverse shapes within design domains. To bridge this gap, we introduce Disp2Topo, a novel methodology employing the Pix2Pix deep learning framework tailored for topology optimization. Disp2Topo generates training data through finite element analysis and topology optimization processes, subsequently training a neural network to predict structures with relative density values that closely mimic those derived from traditional analysis. Unlike conventional methods that primarily focus on external load information, Disp2Topo leverages deformation across all structure nodes, facilitating a more nuanced differentiation between design and non-design domains and substantially reducing computational costs. Our experimental findings demonstrate that Disp2Topo can achieve results comparable to the conventional SIMP method, while drastically reducing computation time to approximately 0.2% of the traditional approach. This represents a significant advancement in efficiently producing optimized structures tailored to varied design requirements.