摘要 積體電路測試作業為多資源作業,需要多種資源的搭配才可進行,在控制與管理上亦有多項績效指標作為評估依據。除了多資源的特性,訂單特性、系統擾動因素等亦影響現場的派工方法。在資源搭配、訂單特性及績效指標四種條件綜合考量下,本研究以決策樹歸納學習法,決定最適之派工方法。 本研究以積體電路測試作業為背景,以物件導向模擬為實驗平台,使用決策樹學習法,解決多資源派工問題。生產系統內外在擾動因素下透過決策樹萃取知識,針對不同指標(在製品存量、設置時間、總完工時間、總延遲時間),以決策樹決定派工方法。本研究觀察多種形式(經驗法、通用型、智慧型)派工法則,自訂單資訊中萃取一組具代表性之屬性向量,引用ID3演算法建立決策樹,判斷適用的派工法則。建構之決策樹可用於現場派工,並作為生產管制知識庫的基礎。 關鍵字:積體電路測試、多資源派工、歸納學習、派工。
Abstract Semiconductor testing is characterized by multi-resource constraints and it has many performance measures on controlling and managing. We propose a decision tree-base to determine the suitable dispatching rule by considering resources allocation, order character, performance measures and system uncertainty. This research use decision tree-base learning algorithm that extract knowledge to solve multi-resource dispatching problem. By using decision tree to represent the dispatching rule which is adapted in different performance measures environment. This research uses various dispatching rules. We extract a set of vectors of attribute from attributes of orders and use ID3 algorithm to generate a decision tree that we use to determine the suitable dispatching rule. This decision tree is the base of production control knowledge database in the factory. Keywords: Semiconductor testing, Multi-resource dispatching, Incremental learning, Dispatching