在監控高產出製程時,p 管制圖乃假設二項分配滿足逼近常態之前提下所建立,因此當製程不合格率很低時,會因二項分配無法滿足常態性假設之條件,而造成 p 管制圖錯誤警告增加。且因不合格率小,管制圖上容易出現許多為零的點,而管制下限亦容易產生等於零之情形,因此無法作為判斷製程是否有顯著改善之依據。故 p 管制圖並不適合用於監控高產出之製程。近年來發展出使用 CCC 管制圖與 CCC-r 管制圖進行高產出製程監控,其目的是為了要改善傳統管制計數值管制圖不適合用於監控不合格率很低之高產出製程。CCC-r 管制圖比 CCC 管制圖多考慮 r 組樣本統計量,以藉此提升 CCC 管制圖偵測製程不合格率偏移之效率。 本研究之主要貢獻為應用類神經網路來發展一套監控高產出製程的累積合格品個數之管制程序,即是利用類神經網路特別適合用於監控微量偏移之數據與非對稱數據之特性,應用模擬之方式產生高產出製程之數據進行監控,並探討當高產出製程在不同微量偏移與延伸管制變數或是加入不同統計特徵值之績效。研究結果顯示透過類神經網路監控系統使其監控績效優於傳統統計方法。
Shewhart p chart is usually used to monitor the fraction nonconforming of a high-yield process. It has been shown to possess some practical difficulties, such as too many false alarms, meaningless control limits and failure in detecting process improvement when fraction nonconforming or defect rate is quite low. When dealing with high-yield processes, the cumulative count of conforming (CCC) chart has been shown to be more suitable than traditional p chart. In order to improve the sensitivity of the CCC chart, the CCC-r charts are the extension of the CCC chart used to monitor the cumulative count of items inspected until observing r nonconforming ones. In this paper, we proposed a control technique based on the Artificial Neural Networks (ANN) for monitoring the cumulative count of conforming data. Two statistics follow geometric distribution (i.e. charting statistic of the CCC chart) and negative binominal distribution (i.e. charting statistic of the CCC-r chart) are studied in this research. Some statistical features are also considered as an input of the ANN system to increase the sensitivity in detecting the parameter shifts. Some performance comparisons between the ANN and the statistical control schemes, such as CCC, CCC-r and Geometric CUSUM charts, are evaluated by the average run length (ARL). The results show the ANN outperforms than the CCC and CCC-r charts, and slightly outperforms than the Geometric CUSUM chart in monitoring a process when the fraction nonconforming is extremely low.