Scholarlink

The design of facility layouts is critical in determining the operating costs, efficiency, and productivity of production systems. In advanced manufacturing industries, such as semiconductor or display panel production, the growing scale and complexity of factories, along with the intricate production flows between facilities, pose significant challenges to facility layout design and optimization. While recent studies have applied machine learning methods to address this problem, fully addressing the diverse and complex constraints of real-world scenarios remains a major challenge. This study proposes an approach for designing and optimizing facility layouts using 3D models and a reinforcement learning algorithm, specifically the Double Deep Q-Network (DDQN). The framework supports constraint-aware layout optimization and integrates with the Omniverse platform to provide intuitive 3D visualization. It also incorporates hyperparameter optimization to enhance training efficiency. The platform enables users to review and refine the generated 3D layout models. It is expected that the proposed approach will support both optimization and visualization of facility layouts while effectively accounting for complex real-world constraints.

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

Design and Optimization of Facility Layouts Using 3D Models and Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jueun Yoo,
  • Hyewon Cho,
  • Donghyun Lee,
  • Goo-Young Kim,
  • Gunyeon Kim,
  • Onyu Yu,
  • Youngjun Jung,
  • Sang Do Noh

摘要

The design of facility layouts is critical in determining the operating costs, efficiency, and productivity of production systems. In advanced manufacturing industries, such as semiconductor or display panel production, the growing scale and complexity of factories, along with the intricate production flows between facilities, pose significant challenges to facility layout design and optimization. While recent studies have applied machine learning methods to address this problem, fully addressing the diverse and complex constraints of real-world scenarios remains a major challenge. This study proposes an approach for designing and optimizing facility layouts using 3D models and a reinforcement learning algorithm, specifically the Double Deep Q-Network (DDQN). The framework supports constraint-aware layout optimization and integrates with the Omniverse platform to provide intuitive 3D visualization. It also incorporates hyperparameter optimization to enhance training efficiency. The platform enables users to review and refine the generated 3D layout models. It is expected that the proposed approach will support both optimization and visualization of facility layouts while effectively accounting for complex real-world constraints.