Scholarlink

<p>The efficient design of complex engineering components in data-constrained environments presents significant challenges to traditional methodologies. Existing deep learning-based generative design approaches often depend on large datasets, which limits their applicability in data-scarce contexts. Additionally, conventional image generation techniques often produce impractical designs, requiring extensive manual validation by engineers. This paper presents a novel method that integrates stable diffusion-based Generative AI with computer-aided design automation to minimize data requirements while maintaining high design accuracy. Through the implementation of low-rank adaptation fine-tuning, the proposed method reduces the required training data from over 16,600 to approximately 200 samples. This significant reduction in data ensures efficiency in data-scarce environments while ensuring compliance with stringent mechanical and aesthetic design requirements. Experimental results demonstrate a consistent 90% accuracy in generating feasible designs that meet these constraints. This paper also explores the relevant background and technological developments that support these experimental results, offering context for the challenges and solutions addressed. Furthermore, the automated validation system further enhances efficiency by filtering out all infeasible designs, thereby eliminating the need for manual expert validation. Experimental results demonstrate a 30% reduction in the overall design process, from initial concept to prototyping preparation, compared to traditional workflows, confirming the method’s effectiveness in real-world applications. This research provides a scalable solution to the challenges of generative design in data-limited settings and contributes to advancing intelligent design systems across various engineering sectors.</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

Generative AI and CAD automation for diverse and novel mechanical component designs under data constraints

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年4月9日

查看全文
  • Kun-Ying Li,
  • Cheng-Kai Huang,
  • Qing-Wei Chen,
  • Hsuan-Cheng Zhang,
  • Tsann-Tay Tang

摘要

The efficient design of complex engineering components in data-constrained environments presents significant challenges to traditional methodologies. Existing deep learning-based generative design approaches often depend on large datasets, which limits their applicability in data-scarce contexts. Additionally, conventional image generation techniques often produce impractical designs, requiring extensive manual validation by engineers. This paper presents a novel method that integrates stable diffusion-based Generative AI with computer-aided design automation to minimize data requirements while maintaining high design accuracy. Through the implementation of low-rank adaptation fine-tuning, the proposed method reduces the required training data from over 16,600 to approximately 200 samples. This significant reduction in data ensures efficiency in data-scarce environments while ensuring compliance with stringent mechanical and aesthetic design requirements. Experimental results demonstrate a consistent 90% accuracy in generating feasible designs that meet these constraints. This paper also explores the relevant background and technological developments that support these experimental results, offering context for the challenges and solutions addressed. Furthermore, the automated validation system further enhances efficiency by filtering out all infeasible designs, thereby eliminating the need for manual expert validation. Experimental results demonstrate a 30% reduction in the overall design process, from initial concept to prototyping preparation, compared to traditional workflows, confirming the method’s effectiveness in real-world applications. This research provides a scalable solution to the challenges of generative design in data-limited settings and contributes to advancing intelligent design systems across various engineering sectors.