類神經網路應用在品質管制中相關性製程數據之管制

A Neural Network Approach to Process Quality Control in the Presence of Serial Correlation

吳聰宏(Tsung-Hung Wu)

指導教授 ： 鄭春生

元智大學/工程學院/工業工程與管理學系/碩士(1993年)

摘要

傳統之統計製程管制法是利用滿足獨立、常態的合理樣本組資料來判斷生
產製程是否有異常的狀況發生，由於個別值之間常具有相關性，在此種情
況下沿用傳統之製程管制法並不適當。目前對於具有相關性之數據的管制
方法，大多是先用時間序列法將其數據予以模式化，然後再採用傳統之管
制法對其具有獨立性之殘差進行管制，這種方式雖然可以符合傳統管制法
的要求，但卻是一種非常沒有效率的管制方式。本研究是使用類神經網路
中的倒傳遞網路模式，發展一偵測製程平均值跳動之管制法。　　本研究
是以平均連串長度做為評估類神經網路成效之基準。由模擬之結果來看，
本研究所發展之管制法，對於微量至中量的製程平均值跳動之偵測，較
Shewhart-CUSUM管制法為優。

關鍵字

相關性製程數據 ； 類神經網路 ； 統計製程管制

並列摘要

Control charts are an important tool in statistical process
control. Control chart analysis is based on the assumption that
process data are normally and independently distributed. But in
the real world, the data measured from industrial process are
often serial correlated. It will lead to many false alarms if
the users neglect the correlation structure of process data. In
the past, traditional control methods are applied to the
uncorrelated residuals of a time series model. It has been
shown that this approach may be feasible but not effective. In
this research, a control method based on artificial neural
networks techniques has been developed as an alternative to
monitor serial-correlated data. Simulation results show that
the proposed approach is more effective comparing to Shewhart-
CUSUM control schemes in detecting small to medium process
changes.

並列關鍵字

Correlated Data ； Neural Networks ； Statistical Process Control