- 學位論文
利用資料挖礦找出影響產品品質之關鍵因素:以LED封裝為例
Using Data Mining to Identify Critical Factors of Product Quality: An Empirical Study on LED Packaging
摘要
隨著市場競爭激烈，LED封裝產品價格也持續下探，嚴重影響了毛利率。面對這樣嚴苛的考驗，通常最簡單的方式是將問題轉嫁給供應商，要求原物料的成本下降。除此之外，對於產品的製程表現，也是值得著手去改善的方向，如何有效提升產品良率，才能從被動地要求供應商成本下降外，還可以更主動地提升自己產品的競爭力，並且維持產品毛利率。 工程單位擅長用實驗設計和統計手法去改善製程，提升良率。然而，因為資訊系統的發展迅速，建置也相當普及，每家工廠製造過程都會使用更系統化的方式去輔助生產，當中也儲存大量的交易記錄保留至自家的資料庫內，所以如何有效地利用這些資料，轉換成資訊，甚至成為公司重要的知識，不僅不浪費儲存資料的成本外，更能創造公司產業製造的智慧，提升產品品質。 大數據的時代，人們開始接受從資料中找出隱含的意義，希望能夠帶來價值，往往可以從資料的結果分析出可能的原因。因此本研究希望透過資料挖掘的方式，使用決策樹分類歸納、預估的方式，以及關聯規則找出關聯規則的方法，找出LED封裝製程產品的生產條件因素，是否存在影響產品規格的關鍵因子，讓產品設計和生產製造單位，可以針對關鍵因子進行提升或是改善，甚至可以盡量避免使用造成不良產品的因素，例如：原物料、生產機台。 關鍵詞：資料挖礦，知識發現，決策樹分析，關聯規則，決策支援。
並列摘要
Since stiff global competition, the price of light-emitting diode (LED) package is going down. How to keep the gross profit of product had been a critical issue for the industry. Although we can shift the losses to the vendor for asking more cost down, we also have to promote productive yield with data mining. The better our product is, the higher gross profit is, so we can use data mining to identify critical factors and make our company more profitable and competitive. Data mining is a very powerful and useful tool to find out the root cause and decrease the moment of detecting problems among the flow of data analysis. Knowledge discovery from database (KDD) is the procedure that we can define our problem and find out critical factors by using data mining skills. Owing to the development of information system, all company has their own database to store transaction data. From data collecting to data analyzing, KDD just set up a framework to make manufacturing much better intelligent. In this research, we used two common data mining skills, decision trees and association rules, to identify critical factors of product quality. In the empirical study of LED packaging, we know critical factors in bill of materials (BOMs) base on decision tree, and understand how to choose the operation machine in manufacturing. Using data mining is very different from the rule of thumb that’s used in the past, and it’s more efficient to make decision. Last, we hope to enhance our manufacture by mining historical data, and make intelligent manufacture with rolling discover knowledge from database. Keyword: Data mining, Knowledge discovery from database, Decision tree, Association rules, Decision support.