半導體高架吊運系統之動態適應性調度決策支援模型

Abstract

本研究針對半導體製造中高架軌道運輸系統（Overhead Hoist Transport）之動態搬運挑戰，提出一套結合「動態反應曲面法（Dynamic Response Surface Methodology, Dynamic RSM）」與模擬最佳化技術之決策支援模型。傳統 RSM 方法雖能有效於固定參數條件下擬合系統輸出，惟在高度時變的製造環境中，易因情境轉變而失效。因此，本研究採用動態 RSM，於各模擬時間區段內建立階段性實驗設計(Design of Experiments)，結合分段資料擬合與持續修正，建構動態搬運時間預測模型。透過與基準線模型比較，在平均搬運時間、任務等待時間、緊急任務反應時間等績效指標均獲顯著改善，尤其在多事件壅塞條件下表現尤為穩定。最終，本研究證實動態 RSM 可於變動製程負載與調度規則下提供高韌性的調度決策基礎，並為未來結合強化學習與數位雙生技術之自動化物料搬運系統提供方法論基礎。

This study proposes a decision-support model combining Dynamic Response Surface Methodology (Dynamic RSM) with simulation-based optimization to address the challenges of dynamic material transport in semiconductor fab Overhead Hoist Transport (OHT) systems. Traditional RSM methods are limited to static parameter spaces and fail to adapt to the high variability in modern fab environments. In contrast, the proposed Dynamic RSM approach constructs segmented Design of Experiments(DOE) designs across simulation time windows, enabling stage-wise model fitting and continuous response updates. The approach was validated through simulations, showing significant improvements in average Front Opening Unified Pod (FOUP) transport time, task waiting time, and emergency response time compared to baseline models. Under event-intensive congestion scenarios, the model maintained robust performance. This study demonstrates that Dynamic RSM offers a resilient foundation for adaptive dispatching decisions and paves the way for future integration with reinforcement learning and digital twin frameworks in Automated Material Handling System.