累積和管制法在偵測製程平均值的微量偏移時,優於蕭華特管制圖的偵測能力。但由於累積和管制法是以簡單對立假設為基礎的逐次機率比檢定法所衍生而得,因此操作累積和管制法時,必須先固定此管制法中的參考值。而參考值的設定方法與製程為管制外時之狀態有關,以實務而言,參考值應隨時變動以反應製程之改變。但在學理上,目前參考值的設定方法,均以事先設定之固定參考值大小於管制圖的操作中。為改善此一缺點,本研究提出以類神經網路為基礎之參考值設定方法,利用製程之現有數據以類神經網路提供及時之參考值大小,再由累積和管制法判斷製程是否異常。此方法之目的在於改善固定參考值的缺點,以擴大累積和管制法的有效偵測範圍。研究結果顯示,本文所提之改善方案將比使用固定參考值之傳統累積和管制法更具偵測效益。
Cumulative sum (CUSUM) charts are known to be very sensitive in detecting small shifts in the process mean, the magnitude of which should be known before using such a chart. However, in the real world, the mean shifts may be changing dynamically in a running process; therefore, it is difficult to obtain an exact value in practical operation. Moreover, the superiority of a CUSUM chart will be lost if improper reference values are chosen. Usually a constant reference value for a CUSUM control chart is selected on the basis of what one is interested in detecting quickly. In this study, a neural network was employed as an alternative approach to the estimation of the reference parameters of a CUSUM. This network provided the reference value for a CUSUM chart on the basis of current process data, but the discriminating criterion was still constructed by a general CUSUM. Average run lengths (ARLs) using simulation were used to evaluate the performance of this procedure. The results show that this charting procedure, based on the hybrid of a statistical method and a neural network, extended not only the utility of the CUSUM chart but also the alternative SPC (statistical process control) solution.