傳統之統計製程管制（statistical process control, SPC）係利用統計的技巧，迅速地偵測出品質特性的變動，藉此推測產品品質發生變異的原因，使得在更多不良品被製造出來之前能及早發現變異與原因進行矯正，以減少產生報廢、重工、退貨或遲延...等各項成本產生的機會。然而在實際工作中，倘若製程數據明顯產生相互影響的時候，如再使用傳統SPC之技巧對製程異常進行偵測時往往會使得錯誤警訊增加進而導致管制圖的誤判。因此如何發展出一套能處理相關性數據的製程變異偵測系統以協助製程分析，便成當前重要之研究課題。 　　過去學者陸續提出許多統計原理的管制法來對製程中所產生非隨機性模型之異常進行偵測。然而，大多數之研究皆在製程間數據為獨立之基礎下作探討，鮮少涉及製程數據呈相關特性之研究。因此，本研究旨在以類神經網路（artificial neural network）中之倒傳遞網路（back-propagation network）為基礎對製程間具有相關性數據特性的趨勢模型進行辨識與分析。利用神經網路優越之學習、歸納及辨識能力，期能達到迅速且正確地辨識出變異模型的目的。研究中除以平均串連長度（average run length）為指標進行系統效益之評估，並進行各項影響網路偵測效益的參數分析，提供適當之參數組合，以獲得一個較合理的系統辨認效率。

　　　　　The Statistical Process Control technique (SPC) can be used as a control tool to detect the manufacturing process variation early. Therefore, the variation causes can be found out and remedied immediately before more defective products have been produced. However, in practical production environment, it is often resulted in missing judgment from the control chart, one of important control technique in SPC, if manufacturing information substantially exists correlated observations in production process. So, developing a system that can recognize and analyze the variation in the manufacturing process with correlated information has become an important issue. In this research, a system which is accomplished by Back-Propagation Neural Network (BPN) is developed for recognizing the trend pattern in manufacturing process in a very short time. Following, an experimental design is performed to analyze the factors which effect the performance of the developed neural network model significantly. Furthermore, a set of parameters which effect the model performance are provided to increase the system recognition efficiency.