利用決策樹模式評估TFT-LCD偏光板最佳檢驗方案

A Decision Tree Approach to Construct an Inspection Sampling Economic Optimization Model for TFT-LCD Polarizer products.

周夏暐(Hsia-Wei Chou)

指導教授 ： 徐旭昇

元智大學/工程學院/工業工程與管理學系/碩士(2008年)

摘要

本論文主旨在提出一個以最佳檢查方式作為經濟成本考量的決策模式，研究中以TFT-LCD某一零組件為例，其產品具有不可修復之特性，發生缺點後，無法加以重工的方式來降低或使缺點消失，因此，產品於生產線上製作到包裝完成後，最終再出貨至客戶手中，這一段的過程中如何能使產品支出的期望總成本為最佳。模式考量產品檢驗成本、抽樣資訊以及產品失效成本，可選擇行動方案分兩階段：(一)決定抽樣數，(二)根據抽樣結果對批量剩餘組件之最佳處置。研究方法以決策樹為工具，來建立問題決策模式，並以決策樹理論推導出整體決策目標函數，並據此分析模式性質，再以C++語言撰寫模式程式。
報告中以某公司為研究案例，提供以實際之檢驗數據，並透過自行撰寫之程式來計算期望最小之支出成本。因此，本研究期望在建立一個適用於TFT-LCD偏光板缺點具有不可修復特性的最佳經濟型抽樣檢驗計畫，提供相關業界在進行檢驗時的參考決策使用，用以避免人力及檢驗成本的浪費。

關鍵字

經濟成本 ； TFT-LCD ； 決策樹 ； 抽樣檢驗計畫

並列摘要

This thesis presents an inspection sampling economic optimization model for TFT-LCD polarizer products from the supplier’s view point using decision tree approach. Inspections are taken at both the supplier’s and its manufacturer’s phases. At the supplier’s phase, the inspection is destructive and after observing the sampling outcome, the supplier has to decide whether or not to send the remainder of the polarizer products to its downstream manufacturer. If the decision is rejecting, the lot will be discarded or sold at a cheap price. The inspection at the manufacturer’s phase is nondestructive and imperfect, that is, inspection errors of types 1 and 2 may occur. All polarizer products judged to be nonconforming by the manufacturer will be sent back to the supplier. The inspection cost and additional cost, such as repair and shipment, are also charged to the supplier. The model assumes that the nonconforming products will be complemented by the supplier until all of the complmented products are accepted by the manufacturer. In the model, only the sampling information obtained at the destructive inspection phase is used in the original lot. No sampling information is used for the complemented products. The cost to be considered includes the destructive and nondestructive inspection cost and product failure cost.
The study utilizes the decision tree theory to construct the economic inspection sampling model with the objective of minimizing the expected total cost. Some properties of the model using Beta prior for the product quality are derived, and such findings are used to facilitate the computation for finding an optimal solution. A computer program coded in Visual C++ is written for solving the model. A real life case is studied in the thesis, and the effects of the model parameters, such as prior parameters, inspection cost, and lot size, are also analyzed and discussed. The results provide a good reference for the management in making operational decisions.

並列關鍵字

TFT-LCD industry ； Decision tree ； Inspection sampling plan ； Beta distributions

參考文獻

10.沈政諭，「精實生產與六標準差整合模式之研究-以光電零組件偏光板為例」，中原大學工業工程研究所碩士論文，2006年6月。

11.周雅君，「以資料探勘為基建構偏光板品質異常診斷系統」，元智大學工業工程與管理研究所碩士論文，2006年7月。

14.賴政昌，「利用決策樹模式評估TFT-LCD產品出貨之最佳處置方案」，元智大學工業工程與管理研究所碩士論文，2007年7月。

8.黃珮菁，「模糊理論在單次抽樣檢驗計畫之應用」，國立成功大學統計學系碩士論文，2005年6月。

9.郭權興，「品質監控系統應用於偏光板裁斷製程之改善研究」，元智大學工業工程與管理研究所碩士論文，2005年7月。