應用類神經網路偵測製程平均值偏移之研究─訓練樣本之影響

管制圖是監控製造程序或服務流程之重要工具。傳統蕭華特 (Shewhart) 管制圖是一個非常簡單之工具，但對平均值微量的偵測並不靈敏。累積和管制法 (CUSUM) 是偵測製程微量變化之有效工具。在應用累積和管制法時，我們必須事先決定一組參數，以獲得較佳之偵測效益。但在實務上，我們對於製程平均值會發生何種程度之變化，可能一無所知。一個改善方法是使用多重累積和管制法，同時操作數個累積和管制圖，來擴大有效之偵測範圍。此方法雖可改善偵測成效，但在使用上較為費時。在本研究中，我們提出一個以類神經網路為基礎之管制法，透過訓練樣本之規劃，希望能夠對一個較大範圍內之平均值偏移做有效地管制。在此研究中，我們是以平均連串長度 (ARL) 來評估各種管制法之效益。比較結果顯示，以適當樣本訓練所獲得之類神經網路可以比多重累積和管制法更具有偵測效益。

關鍵字

類神經網路；訓練樣本；多重累積和管制法；平均連串長度

並列摘要

Control charts are widely used for both manufacturing and service industries. Although traditional Shewhart chart is a very simple tool, it is insensitive to small shifts in the process mean. Cumulative sum (CUSUM) charts are known to be very sensitive in detecting small shifts in the mean. On the implementation of CUSUM charts, a set of parameters has to be determined in advance. However, it is very difficult to determine these parameters when future shifts are unknown. The multiple CUSUM modeled by applying several standard CUSUM charts simultaneously, is more robust at signaling a wide range of process shifts. The multiple CUSUM will improve the conventional CUSUM while requiring little additional computational effort. In this research, we propose a neural network as an alternative approach to the multiple CUSUM charts when monitoring the shift in the process mean. The focus is on the generation and configuration of the training data set. The performance of neural network was evaluated by estimating the average run lengths (ARL's) using simulation. The results obtained with simulated data suggest that control scheme based on neural network is significantly more sensitive to process shifts than the multiple CUSUM charts.

並列關鍵字

neural network ； training samples ； multiple CUSUM ； ARL

參考文獻

1.王仁達，應用類神經網路偵測製程平均值之變化︰設計策略之研究，

2.吳聰宏，類神經網路應用在品質管制中相關性製程數據之管制，中壢，

3.陳建至，以類神經網路為基礎監視計數值品質特性，中壢，元智工學院

4.賴錦興，類神經網路在品質管制中製程平均值管制之應用︰設計策略及

5.Chang, T. C. and F. F. Gan, “A cumulative sum control chart

國際替代計量

應用類神經網路偵測製程平均值偏移之研究─訓練樣本之影響

主題瀏覽