應用類神經網路與支援向量機於多變量自我相關製程變異來源之辨識

Identifying the Source of Multivariate Autocorrelated Process Shiftsby Artificial Neural Networks and Support Vector Machine

邱玉琳(Yu-Lin Chiu)

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(2007年)

https://doi.org/10.6838/YZU.2007.00216

摘要

實際工業製程中，連續生產過程、自動化生產以及檢驗等等，將導致製程數據間存在自我相關性，且現今同時監控數個品質特性已成了必然的趨勢。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，進而發展出有效之監控系統，可及早偵測到製程之變化。然而在多變量自我相關製程中，傳統 T-square 管制法卻無法的偵測到異常狀況。本研究以 Hotelling T-square 管制法判斷多變量自我相關製程平均值是否發生異常，利用類神經網路及支援向量機兩種分類方法對已發生異常之製程進行異常來源之分類，以判斷發生異常之變量為何。
本研究使用多變量自我相關製程中三種模型，並探討不同共變異數矩陣和自我相關參數矩陣之平均值偏移型態。此外，除了以原始數據做為分類法之輸入向量，也考慮以統計距離做為變量之特徵值，比較原始數據與特徵值之績效。以正確分類率作為各分類法之評估指標，觀察各分類法對於異常來源分類之辨識能力。結果顯示，特徵值統計距離對各分類法均提升不少效益，表示統計距離對本研究具有可行性。且支援向量機之正確分類率相對類神經網路具有較佳辨識能力。

關鍵字

多變量製程管制 ； 自我相關性 ； 類神經網路 ； 支援向量機 ； 統計距離

並列摘要

In many industrial processes, a product may have two or more related quality characteristics which should be monitored simultaneously. However, the measurement data from many manufacturing processes are not independent in practice. Thus, the traditional T-square chart was insufficient for detecting mean shifts in multivariate auto-correlated processes. The Hotelling’s T-square control chart has been designed for detecting mean shifts. In this research, we purposed two mean shifts classifiers based on artificial neural network (ANN) and support vector machine (SVM). When an out-of-control signal is appeared, the classifier will determine which variable is responsible for the mean shifts.
In this research, we considered three models of multivariate auto-correlated process. Various shift scenarios expressed in covariance matrices and autocorrelation parameter matrices were investigated. Statistical distance was proposed to be used as the component of the input vectors. The performance of the proposed method was evaluated by computing correct classification accuracy. The results showed that the proposed approach is a successful method in identifying the source of mean change in multivariate auto-correlated process. Results from our experiment also indicated that SVM-based classifier performs better than the neural network-based classifier.

並列關鍵字

multivariate process control ； autocorrelation ； artificial neural network ； support vector machine ； statistical distance

參考文獻

39. 阮冰如，「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」，元智大學工業工程與管理所碩士論文，2006。

40. 楊慧萍，「以類神經網路建立偵測自我相關製程平均值偏移和參數估計之雙邊管制法」，元智大學工業工程與管理所碩士論文，2005。

2. Alwan, L. C., and Radson, D., “ Time-series investigation of subsample mean charts,” IIE Transactions, 24, 66-80 (1992).

4. Charnes, J. M., “Tests for special causes with multivariate autocorrelated data,” Computers and operations research, 22, 4, 443-453 (1995).

5. Chen, L. H. and Wang, T. Y., “Artificial neural networks to classify mean shifts from multivariate x2 chart signals,” Computer & industrial engineering, 47, 195-205 (2004).