應用類神經網路與支援向量機於多變量製程 變異來源之辨識

Identifying the Source of Multivariate Process Shifts by Artificial Neural Networks and Support Vector Machine

阮冰如(Ping-Ru Yuan)

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(2006年)

摘要

在過去研究中，多變量管制法主要在偵測多變量製程異常之發生，並進而發展出有效的監控系統，可及早偵測到製程之變化。在多變量管制法中，至今遇到最大的問題在於管制外信號之解釋，尤其當製程已發生變異時，是由哪個變量引起管制外信號。本研究採用類神經網路與支援向量機兩種分類方法，對已發生異常之多變量製程進行製程異常來源之分類，以判斷發生異常之變量為何。

研究中分別以傳統 管制法、 管制法判斷製程平均值或變異性是否發生異常，並利用不同分類法對發生異常之資料作分類，以判別出發生異常之來源。在研究中，我們以正確分類率作為評估指標。結果顯示，在辨識平均值異常來源時，類神經網路與支援向量機之效益差異不大；但在辨識變異性改變之來源時，支援向量機之結果表現遠高於類神經網路。另外，從研究結果可得知，正確分類率會受到相關性大小、偏移量大小之影響。

關鍵字

多變量製程管制 ； 類神經網路 ； 支援向量機 ； 正確分類率

並列摘要

Multivariate quality-control chart is used in the situations which simultaneous with two or more related quality characteristics. The challenge in any multivariate control chart is practical interpretation of an out-of-control signal. Particularly, which variable causes the out-of-control signal? We adopted two classifications: artificial neural networks (ANN) and support vector machine (SVM), to identify the sources of multivariate process which has had unnatural.

We employ traditional chart and chart to indicate mean shifts or variability changes. And apply two classifications to classify unnatural data, identify the source of process change. In this paper, the performance is evaluated by computing its correct classifying rate. Results from simulation researches indicate that ANN and SVM performance are similar in classifying the source of mean shifts. But in classifying the source of variability changes, SVM shows better results. Furthermore, the results indicate the performance is dependent on the correlation and the magnitude of shift.

並列關鍵字

multivariate process control ； artificial neural networks ； support vector machine ； correct classifying rate

參考文獻

2. Chen, L. H. and Wang, T. Y., “Artificial neural networks to classify mean shifts from multivariate chart signals,” Computer & industrial engineering, 47, 195-205 (2004).

3. Cheng, C. S. and Tzeng, C. A., “A neural network approach for detecting shifts in the process mean and variability,” Journal of the Chinese Institute of Industrial Engineers, 11, 67-75 (1994).

4. Cheng, C. S., “A multi-layer neural network model for detecting changes in the process mean,” Computers and Industrial Engineering, 28, 51-61 (1995).

5. Chinnam, R. B., “Support vector machines for recognizing shifts in correlated and other manufacturing processes,” International Journal of Production Research, 40,4449-4466 (2002).

7. Guo, Y., and Dooley, K. J., “Identification of change structure in statistical process control,” International Journal of Production Research, 30, 1655-1669 (1992).