臺灣博碩士論文加值系統

半導體製造業是一種要求高精密度且技術密集的工業，昂貴的生產設備成本佔所有生產成本的百分之四十以上，而生產設備能否正常的運作直接影響產量與品質，為了有效的利用生產設備以達到設備生產的極限，許多製造者便以降低準備時間與故障時間為年度的目標，大部分現存的方法之中都是以技術手法或是線上管理為主，極少以系統控制與管理的發展為本，然而，以技術導向的方法需要較高的發展成本而且並不是適用於所有不同特性的機台，雖然台灣半導體製造業已發展為一項重要的產業，但對於設備維護管理方面卻尚未作足夠的研究。
因此，在探訪許多半導體製造廠商的管理者與工程人員了解目前實際的設備維護管理狀況與問題之後，本研究提出一個半導體製造設備全面維護管理系統的概念性架構，使半導體產業能參考此一系統架構發展出適合自己企業體質的設備維護管理系統。依據此一系統架構，本研究發展設備故障預測模式，經由對設備故障歷史資料的分析與設備故障模式建立的過程，可獲得設備故障間隔之機率模式，將所得之模式運用貝氏方法 (Bayesian Method) 並考慮定期維護(Preventive Maintenance) 的影響計算設備移動期間內 (rolling period) 可靠度，此故障預測模式除了提供設備故障預測的資訊作為施行預知維護的參考也能提供未來研究發展設備維護排程的基礎。因為在發展故障預測模式的過程中，設備歷史資料是非常重要的一個分析來源，本研究考量系統架構發展所需的資料建立設備維護資料庫，因此，此資料庫除了能供給故障預測模式所需的設備歷史資料外，並能支援設備全面維護管理系統發展所需的一切資訊。

In high precision required semiconductor manufacturing, the expensive equipment investment causes high depreciation cost which is more than forty percent of all production cost. And the equipment performance directly affects the production capacity and product quality. In order to effectively and efficiently utilize equipment to reach equipment capacity limits, many manufacturers set the reduction of standby time and down time as a yearly unchanged goal. Most of the existing methods for attaining the above purposes are based on technology approach and in-line management. Very few methods are development based on system control and management concepts. However, the technology-oriented methods need high development cost and are not generic for all kind of machine. Although the semiconductor manufacturing is important in Taiwan''s industry, the management of equipment maintenance is not well studied in current stage.
In this research, after interviewing with managers and engineers of many semiconductors manufacturing companies to understand the practitioners'' concerns and problems, we have developed the concept structure of Total Preventive Maintenance (TPM) Management System for semiconductor manufacturing. The semiconductor industry can apply this system structure to develop a reliable and suitable equipment maintenance management system. According to the system concept structure, the failure prediction model is developed. Followed is the analysis of machine failure history and the development of failure model. The probability model of mean time between failure can be obtained. Using the failure model, we obtained the failure prediction model, which applies the Bayesian method and considers the effect of preventive maintenance to calculate the reliability of equipment. The failure prediction models can be applied to the prediction of machine failure and the development of scheduling methods for equipment maintenance in the further research. Due to the failure history data is important to the failure prediction model, this research also develop a TPM database. The database supports all equipment information of TPM Management System.