應用支援向量機於相關性製程監控階梯式干擾

Monitoring step-change disturbance in an autocorrelated manufacturing process using support vector machine

李天行(Tian-Shyug Lee) ； 邱群穎(Chiun-Ying Chiou) ； 呂奇傑(Chi-Jie Lu)

《Journal of Data Analysis》 11卷4期 (2016/08) Pp. 1-20

https://doi.org/10.6338/JDA.201608_11(4).0001

摘要

統計製程管制(statistic process control, SPC)工具大都假設製程觀測值間互為統計獨立。但在實務上，製程多屬相關性製程，亦即製程資料間存在自我相關，如此限制了傳統SPC 管制圖於實務應用上之有效性。有鑒於傳統SPC 管制圖常無法有效監控相關性製程，本研究提出以新興人工智慧工具─支援向量機(support vector machine, SVM)做為監控相關性製程之管制工具，並以實務上常出現之一階自我迴歸時間序列模式(first order autoregressive, AR(1)) 及階梯式干擾(step-change disturbance)為研究對象。本研究建構兩種製程監控模式，一為判別製程正常與否之二類模式，另一為辨識製程平均數偏移量之四類模式，分別以平均連串長度（average run length, ARL）及辨識正確率做為指標來驗證所提方法之有效性，並與傳統的SPC 管制圖的結果進行比較。研究結果顯示，在二類模式中，所提方法於各情況下，皆能對相關性製程資料進行有效監控，能較傳統蕭華特管制圖得到更令人滿意之結果；而在四類模式中，本研究所提之SVM 監控模式，在製程存在階梯式干擾時，能有優於傳統時間序列管制圖之監控正確率，僅有在製程之自我相關係數小於0.5 且不具階梯式干擾情況下，辨識正確率明顯低於時間序列管制圖。根據研究成果，說明了支援向量機應用於監控相關性製程之可行性及發展潛力。

關鍵字

統計製程管制 ； 支援向量機 ； 相關性製程 ； 階梯式干擾

並列摘要

The traditional statistical process control (SPC) control charts assume that process data are identically and independently distributed. However, the real process data are actually serially correlated. Thus, traditional SPC techniques are not applicable in many process industries due to the assumption of the independence between observations. In this research, a process monitoring scheme using support vector machine (SVM) is proposed to identify step-change disturbance in an autocorrelated process. There are two kinds of monitoring models are built in this study. The first one model is used to identify whether the process is in-control or out-of-control. The other one model is applied to recognize the magnitudes of step-change disturbance of a process. The traditional Shewhart chart and residual Shewhart chart are used in this study as comparison techniques. For evaluating the effectiveness of the proposed SVM monitoring scheme, simulated autocorrelated process datasets with step-change disturbance are evaluated. Experiments reveal that the proposed SVM monitoring scheme outperforms the traditional control charts in most instances and thus is effective for monitoring the step-change disturbance in an autocorrelated process.

並列關鍵字

statistics process control ； support vector machine ； autocorrelated process ； step-change disturbance

參考文獻

Adams, B. M.,Tseng, L. T.(1998).Robustness of forecast-based monitoring schemes.Journal of Quality Technology.30,328-329.

Alwan, L. C.,Roberts, H. V.(1988).Time series modeling for statistical process control.Journal of Business and Economic Statistics.6,87-95.

Box, G. E. P.,Jenkins, G. M.,Reinsel, G. C.(1994).Time series analysis: Forecasting and control.Englewood Cliffs, NJ:Prentice-Hall.

Burbidge, R.,Trotter, M.,Buxton, B.,Holden, S.(2001).Drug design by machines learning: Support vector machines for pharmaceutical data analysis.Computer & Chemistry.26,5-14.

Cheng, C. S.,Tzeng, C. A.(1994).A neural network approach for detecting shifts in the process mean and variability.Journal of the Chinese Institute of Industrial Engineers.11(2),67-75.