現行主流的製造執行系統架構多是針對標準化量產流程所優化，而精密鑄造產業具 備高度設備異質性，且製程高度依賴隱性知識，其數位化過程具有獨特的實務複雜度。 特別是在試作階段則具有頻繁的動態製程調整與參數演化特徵。因此，本研究探討的核 心議題是提升系統架構的彈性，使其能更貼合此類非線性製程的需求，並減少現場作業 與數據收集之間的落差。 為應對上述產業特性，本研究提出一套模組化製造執行系統架構與導入方法論。透 過行動研究取向，深入台灣一家上市精密鑄造企業進行實證，並於不干擾既有生產運作 之前提下，透過系統與既有作業流程之並行運作進行同步驗證，歸納出三項關鍵實作策 略：（一）異質通訊解耦：利用通訊閘道整合底層多樣化的通訊協定；（二）追加式版本 控管：建立支援參數迭代履歷的資料模型，完整保存試作階段的參數演化過程；（三）配 置化邏輯引擎：將業務規則與核心程式碼分離，賦予系統在不重新編譯的情況下，即能 靈活對應製程變更的彈性。 在技術實作與驗證層面，本研究建構一套支援資料整合與製程履歷追溯之多層式系 統架構。導入成效顯示，關鍵績效指標獲得正向改善趨勢：平均報表產出延遲縮短 98.9%， 品質異常追溯時間縮短 98.7%。此外，跨國廠區之複製部署結果顯示本架構於不同場域 中具備一定程度之可攜性。本研究成果表明，透過彈性的模組化架構設計，能有助協助 傳統產業將寶貴的試作經驗轉化為標準化的數位知識資產，為面臨類似高變異製程挑戰 的製造業者，提供了一條具備實務參考價值的轉型路徑。;Most Manufacturing Execution System (MES) architectures are developed for stable， high-volume production environments. However， the precision casting industry presents challenges in system implementation due to equipment heterogeneity and reliance on tacit knowledge. During trial stages， frequent process adjustments and parameter changes are required， which creates a gap between shop-floor operations and data collection. To address this issue， this study proposes a configuration-driven modular MES architecture and an implementation approach. The study was conducted in a listed precision casting company in Taiwan using an action-oriented approach. The system was introduced through parallel operation with existing processes， without interfering with ongoing production. This allowed data to be validated and compared during the early stage of implementation. Based on the implementation process， three strategies were identified: (1) decoupling heterogeneous communication through a gateway; (2) append-capable versioning to retain parameter changes during trials; and (3) configuration-driven logic to separate business rules from core code， enabling system adjustment without program modification. A multi-layer MES architecture focusing on data integration and process traceability was developed and implemented. The results indicate improvements in key performance indicators. Average report latency was reduced by 98.9%， and quality traceability time was reduced by 98.7%. In addition， implementation in a second site suggests that the proposed architecture can be reused with limited modification. The findings indicate that a modular MES design can support the transformation of trial-and-error practices into structured digital records， providing a practical reference for manufacturing environments with high-mix， low-volume characteristics.
