在統計製程管制 (Statistical Process Control, SPC) 中,當製程存在 可歸屬原因時,製程之異常可分為平均值改變和變異性改變。 當管制對 象為製程之變異性時, 通常是以 R, S, S?等管制圖來監視製程變異性 之變化。由過去的研究結果得知上述管制方法並不能有效偵測製程變異性 之微量變化, 因此一些學者提出了 CUSUM 、 EWMA等管制法來改善此 一缺點,但其對於製程之分析仍需仰賴人力來完成。然而為配合現今自動 化生產及 100% 檢驗,並增進製程分析之效率,發展一個電腦化之製程異 常偵測系統乃是未來的趨勢。因此本研究提出一應用類神經網路 (Artificial Neural Networks) 之管制程序,用以偵測製程變異性之變 化。在本研究中,我們採用平均連串長度 (Average Run Length, ARL) 來評估類神經網路之偵測效益, 而評估結果的比較對象是 Crowder 和 Hamilton(1992) 所提出之 EWMA管制法。經由比較後我們發現,本研究所 提出之類神經網路對於製程變異性變化之偵測,較傳統之 EWMA 管制法更 具效益。
In statistical process control the quality characteristics of product may shift in process mean and/or change in process variability when the out-of-control conditions were caused by assignable causes. When detecting the changes of the process variability, people were used to using R, S, S?charts to monitor the process. But previous research has shown that these charts are not effective in detecting a small change of the process variability. In this research, a control procedure based on artificial neural networks for monitoring the process variability was developed. The performance of the proposed artificial neural network has been evaluated by the average run length (ARL), and a compression of the performance of artificial neural network and exponentially weighted moving average( EWMA) chart which using ln(S? to monitor process is presented. Simulation results show that the neural network performance in detecting the changes of the process variability is superior to traditional EWMA control scheme.