本研究鑑於傳統之EWMA批次製程控制法在對於單變量無趨勢項製程的控制上,當製程前後之自我相關程度較高時會出現控制品質不佳的問題。另外,傳統的數值控制法在為了求得合適的製程控制變數所需之演算式時,亦需要經過複雜的推導過程與長時間的推導。而模糊類神經網路(FNN)具備能夠從所給予的訓練資料中,藉由訓練學習的過程,自動找出輸入與輸出變數間的關係,並可將其運用於推估、預測、決策與診斷上的特性。故本研究之目的為希望以模糊類神經網路作為預測基礎,發展出一個單變量製程控制系統,並將其對於具前後自我相關性之單變量無趨勢項製程的控制結果分別與傳統的EWMA以及Adaptive EWMA二種數值控制法之控制結果進行比較分析,並期望能使原本對於製程前後相關程度較高時,採用EWMA製程控制法控制時所可能發生的控制品質不良問題獲得改善。研究結果顯示,於單變量無趨勢項製程中,當製程之自我迴歸項係數 時,使用FNN控制法控制後之平均MSE(mean square error)值較使用EWMA控制法增加約5.19%,較使用Adaptive EWMA控制法降低約70.18%;當 時,使用FNN控制法控制後之平均MSE值較使用EWMA控制法降低約75.54 %,較使用Adaptive EWMA控制法降低約59.90%;當 時,使用FNN控制法控制後之平均MSE值較使用EWMA控制法降低約97.11%,較使用Adaptive EWMA控制法降低約61.12%;故相較於兩種控制方式,FNN單變量製程控制法使自我相關程度較高時之無趨勢項製程的控制品質獲得改善。
In consideration of when the process with large autoregressive parameter, using traditional EWMA control methods to control SISO process without a deterministic trend couldn’t get a well controlled quality. In addition, before using traditional statistical control method to get the control variable of the process, may need complicated computation. The fuzzy neural network (FNN) could be applied to process control according to training data, and infer the control variable automatically by training and learning process. So this research is trying to establish a SISO process control model which is based on fuzzy neural network, and to use this method to simulate control SISO process without a deterministic trend. The results indicate that, for the SISO process without deterministic trend, when the autoregressive parameter , the mean square error (MSE) after using FNN controller to control is higher than that of using EWMA controller about 5.19%, but lower than using adaptive EWMA controller about 70.18%; when , the MSE after using FNN controller to control is lower than using EWMA controller about 75.54%, and lower than using adaptive EWMA controller about 59.90%; when , the MSE after using FNN controller to control is lower than using EWMA controller about 97.11%, and lower than using adaptive EWMA controller about 61.12%. Consequently, the whole control quality of using FNN SISO control method to control the SISO process without a deterministic trend is better than EWMA control method.