Scholarlink

<p>In high-precision manufacturing environments such as semiconductor fabrication, scheduling decisions must not only minimize makespan but also ensure reliable process execution. This study presents the first extension of the Flexible Job Shop Scheduling Problem (FJSP) to incorporate both station-level and overall system reliability. To address the complexity of large-scale scheduling, a Contextual Snapshot Learning-based Deep Reinforcement Learning (CSL-DRL) is presented. The method constructs contextual states that allow the learning agent to understand how much work each job has remaining and when each machine will become available, while maintaining linear scalability with the number of jobs and machines. Large-scale experiments conducted on semiconductor production data demonstrate the model’s ability to balance scheduling efficiency and process reliability. Furthermore, an Analytic Hierarchy Process (AHP) framework is applied to evaluate scheduling outcomes, integrating practitioner-derived indicator weights for a comprehensive and practical performance assessment. The proposed framework effectively bridges data-driven scheduling optimization with reliability-oriented decision-making, providing insights for smart and sustainable manufacturing systems.</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

A contextual snapshot learning-based deep reinforcement learning for reliability-aware flexible job shop scheduling in semiconductor manufacturing

  • 期刊论文

  • 发表日期：

    2026年3月15日

查看全文
  • Tsung-Jung Hsieh

摘要

In high-precision manufacturing environments such as semiconductor fabrication, scheduling decisions must not only minimize makespan but also ensure reliable process execution. This study presents the first extension of the Flexible Job Shop Scheduling Problem (FJSP) to incorporate both station-level and overall system reliability. To address the complexity of large-scale scheduling, a Contextual Snapshot Learning-based Deep Reinforcement Learning (CSL-DRL) is presented. The method constructs contextual states that allow the learning agent to understand how much work each job has remaining and when each machine will become available, while maintaining linear scalability with the number of jobs and machines. Large-scale experiments conducted on semiconductor production data demonstrate the model’s ability to balance scheduling efficiency and process reliability. Furthermore, an Analytic Hierarchy Process (AHP) framework is applied to evaluate scheduling outcomes, integrating practitioner-derived indicator weights for a comprehensive and practical performance assessment. The proposed framework effectively bridges data-driven scheduling optimization with reliability-oriented decision-making, providing insights for smart and sustainable manufacturing systems.