應用類神經網路與支援向量機以辨認製程多重品質特性出錯之研究

A Study of Identification of the Out-of-Control Multiple Quality Characteristics Using Neural Network and Support Vector Machine Techniques

10.6338/JDA.200810_3(5).0006

邵曰仁(Yueh-Jen E. Shao)；黃于庭(Yu-Ting Huang)；顏慈儀(Tzu-Yi Yen)；鍾緯華(Wei-Hua Chung)；沈凡超(Fan-Chao Shen)；蕭庭岳(Ting-Yue Hsiao)；游雅淇(Ya-Chi Yu)

多變量管制圖 ； 類神經網路 ； 支援向量機 ； Multivariate Control Chart ； Neural Network ； Support Vector Machine

Journal of Data Analysis

3卷5期（2008 / 10 / 01）

69 - 81

繁體中文

科技的發展日新月異，各項產品不斷的創新與進步，同時為了因應現代複雜的產品製程，傳統的單變量管制圖已不敷需求，因此，多維管制圖監控產品成為現代產品製造重要的課題之一。然而，複雜的多重品質特性也意味著監控及改善製程的困難度越來越高，一般而言，多變量管制圖通常可有效的得知失控訊息，但是當多變量管制圖產生製程失控訊號時，品質管制人員並不易了解究竟是那一項品質特性或那一群品質特性出錯？也因此需要花費較多的時間及成本，始能分析出錯誤的品質特性為何。本文將探討Hotelling T2多變量管制圖整合類神經網路與支援向量機之方法，以期有效辨認製程出錯之品質特性，並進而提升判斷出正確失控之品質特性，本文將藉由模擬範例以展示本文方法之優異性。

Because of rapid advances in technology, consumers have higher concern for the quality of products than before. There are many aspects of quality in each product. Typically, the univariate statistical process control (SPC) charts may not be suitable for monitoring a process that has multiple quality characteristics. Instead, the multivariate control charts are able to simultaneously monitor multiple quality characteristics of a process. A process is hypothesized to be out of control when a signal is triggered by a multivariate SPC chart. The problem is that it is difficult to determine the contributors of this signal. That is, which of the monitored quality characteristics is responsible for this out-of-control signal? This determination is not straightforward. This issue is, of course, a very important research topic for industries. If the monitored quality characteristics that are fault can be quickly and correctly determined, the remedial action can be taken in time. As a consequence, the process can be significantly improved. This study aims to integrate the Hotelling T2 multivariate control chart with neural networks (NN) and support vector machine (SVM) to effectively identify the set of quality characteristics that is responsible for the out-of-control signal. The superior results of the proposed approaches are demonstrated with the use of a series of simulations.

1. Chen, Xue-wen,Zeng, Xiangyan,Van Aiphen, Deborah(1996).Multi-class Feature Selection for Texture Classification.Pattern Recognition Letters,27,1685-1691.
2. Cortes, C.,Vapnik, V.(1995).Support-Vector Networks.Machine Learning,20,273-297.
3. Cybenko, G.(1989).Approximation by Superpositions of Sigmoidal Function.Mathematics of Control, Signals and Systems,2,303-314.
4. Kim, Hyun-Chul,Pang, Shaoning,Je, Hong-Mo,Kim, Daijin,Sung, Yang Bang(2003).Constructing Support Vector Machine Ensemble.Pattern Recognition,36,2757-2767.
5. Li, L.,Tang, H.,Wu, Z.,Gong, J.,Gruidl, M.,Zou, J.,Tockman, M.,Clark, R. A.(2004).Data Mining Techniques for Cancer Detection Using Serum Proteomic Profiling.Artificial Intelligence in Medicine,32,71-83.
6. Montgomery, D. C.(2005).Statistical Quality Control.John Wiley and Sons Inc.
7. Rumelnhart, D. E.,Hinton, D. E.,Williams, R. J.(1986).Learning Internal Representations by Error Propagation in Parallel Distributed Process.Cambridge, MA:MIT Press.
8. Runger, G. C.,Alt, F. B.,Montgomery, D. C.(1996).Contributors to a Multivariate Statistical Process Control Chart Signals.Communications in Statistics-Theory and Methods,25(10),2203-2213.
9. Song, Jie,Tang, Huanwen(2005).Support Vector Machines for Classification of Homo-oligomeric Proteins by Incorporating Subsequence Distributions.Journal of Molecular Structure,722,97-101.
10. Tsai, Chih-Fong(2007).Image Mining by Spectral Features: A Case Study of Scenery Image Classification.Expert Systems With Applications,32,135-142.