應用殘差修正灰預測模型與自組性演算法建構工程製程管制系統

Abstract

隨著科技的進步，各個產業的競爭也越來越激烈，要能夠保持競爭優 勢及領先地位，廠商必須持續提供消費者高品質的產品，因此如何強化製 程管制的能力以製造出穩定且高品質的產品，為各廠商維持競爭優勢的關 鍵因素。在生產的過程中，工程製程管制(Engineering Process Control, EPC) 是提升產品品質的重要工具之一，EPC 主要是透過調整製程參數，使製程 之輸出值能迅速修回目標值，以穩定及提昇製程品質。過去EPC 執行之方 法通常是根據工程人員的經驗來調整製程的參數，但人為的調整方式沒有 一定的準則且無法正確的補償製程的偏移量，因此，本研究之主要目的即 是建立一套有效之工程製程管制系統，當製程之輸出偏離目標值時，能迅 速調整關鍵製程參數，使製程輸出值能回到目標值。本研究首先利用殘差 修正灰預測模型(Modified GreyModel)來預測下一時間點之製程輸出值； 接著利用自組性演算法(Group Method of Data Handling, GMDH)建立一個 參數調整系統，當估計之下一時間點的製程輸出值偏離目標值時，可應用 此系統調整適當之製程參數值，使製程輸出值能迅速修回至目標值。本研 究最後分別以台灣某半導體公司的一個實例及模擬的製程數據來說明本 研究所提出之回饋系統確實有效可行。

The growth of technologies deserves more attention in competitive international markets. To succeed in these markets, Manufacturers must place premium consideration on quality of their products. Therefore, the key factor for keeping the competitive advantage is how to strengthen the process control capabilities and create stable and high quality products. An engineering process control (EPC) is one of the most important tools to promote product quality in the manufacturing process. EPC is used containing controllable variables that can be adjusted to keep the process output on target. This study aims to construct a feedback system to adjust the process parameters, when process output is beyond the adjustment threshold. So, the process output can be revised quickly to the target. This study firstly uses the modified grey model to forecast the output of the succeeding lot and then group method of data handling (GMDH) is utilized to establish a parameter adjustment system. A real case from a Taiwanese semiconductor manufacturer and a set of simulated data are used to demonstrate the effectiveness of the proposed approach.