Determining control limits of multivariate system based on principal component analysis

• Article •

Determining control limits of multivariate system based on principal component analysis

LUO Xian-xi,, YUAN Ming-zhe,, WANG Hong, HUANG He

1.Department of Information Service & Intelligent Control, Shenyang Institute of Automation, Chinese Academy of Science, Shenyang 110016, China;2.Mechanics and Electronics Engineering School,East China Institute of Technology, Nanchang 330013, China;3.Graduate School of the Chinese Academy of Sciences, Beijing 100039, China;4.Shenyang Institute of Automation, Guangzhou, Chinese Academy of Science, Guangzhou, 511458, China

Online:2012-10-15 Published:2012-10-25

基于主元分析确定多变量系统控制极限的方法

罗先喜，，苑明哲,，王宏，黄河

1.中国科学院沈阳自动化研究所信息服务与智能控制研究室，辽宁沈阳100016；2.东华理工大学机械与电子工程学院，江西南昌330013；3.中国科学院研究生院，北京100039；4.中国科学院广州沈阳自动化研究所分所，广东广州511458

Abstract

Abstract: Aiming at the control limits problem of Multivariate Statistical Process Control(MSPC)which had multiple stable states, Empirical Distribution Function(EDF)was obtained by analyzing historical data of system variables, and different stable states as well as corresponding state-domains of variables were determined. After standardizing data samples in different state groups, Hotelling's T2 and Squared Prediction Error(SPE)(Q)control limits in different states were obtained by using Principal Component Analysis(PCA). Through comparison with traditional PCA in flow data monitoring for a steel corporation's local steam pipe network, the proposed method was more accurate on determining Hotelling's T2 and SPE(Q)control limits. It could reduce the probability of leaking alarm or false alarm effectively.

Key words: principal component analysis, multivariate statistical process control, control limit, empirical distribution, fault diagnosis

摘要： 针对存在多个稳定状态的多变量统计过程控制的控制极限问题，提出先分析系统变量的历史数据得到其经验分布，确定该变量的不同稳定状态及对应的状态域；将数据样本按不同的状态分组标准化后，分别进行主元分析，得到不同稳定状态下的Hotelling　s T2及平方预测误差SPE（Q）控制极限。用本方法和普通主元分析法对某钢铁公司局部蒸汽管网的流量数据模拟监控的效果进行对比，表明本方法由于区分不同状态，确定的Hotelling　s T2及平方预测误差SPE（Q）控制极限更精确，能有效降低漏报警和误报警的概率。

关键词: 主元分析, 多变量过程控制, 控制极限, 经验分布, 故障诊断

CLC Number:

TP273

LUO Xian-xi,,, YUAN Ming-zhe,, WANG Hong, HUANG He. Determining control limits of multivariate system based on principal component analysis[J]. .

罗先喜,,，苑明哲,，王宏，黄河. 基于主元分析确定多变量系统控制极限的方法[J]. .

[1]	. Business intelligence of automobile “process-quality” stamping production [J]. , 2021, 27(4): 1120-1127.
[2]	. Gas turbine fault diagnosis method under small sample based on transfer learning [J]. , 2021, 27(12): 3450-3461.
[3]	. Open circuit fault diagnosis algorithm of high speed railway inverter with strong generalization [J]. , 2021, 27(11): 3236-3246.
[4]	. Intelligent fault diagnosis of metro traction motor bearing based on convolution neural network and information fusion [J]. , 2021, 27(11): 3247-3258.
[5]	. New automated machine learning based imbalanced learning method for fault diagnosis [J]. , 2021, 27(10): 2837-2847.
[6]	. Fault text data classification based on mutual attention mechanism network model [J]. , 2021, 27(1): 73-90.
[7]	. Rolling bearing early fault classification algorithm based on two-dimensional EKNN [J]. , 2021, 27(1): 91-102.
[8]	. A cloud/edge collaborated bearing fault diagnosis method [J]. , 2020, 26(第3): 589-599.
[9]	. Machine tool fault diagnosis knowledge expansion method based on semantic web [J]. , 2020, 26(第3): 609-622.
[10]	. Product form image design based on elliptic Fourier [J]. , 2020, 26(第2): 481-495.
[11]	. Wafer yield prediction method based on improved continuous deep belief network [J]. , 2020, 26(9): 2388-2395.
[12]	. Many-objectives scheduling optimization based on PCA-NSGAⅡ with improved elite strategy [J]. , 2020, 26(9): 2474-2483.
[13]	. Real-time residual life prediction based on kernel density estimation [J]. , 2020, 26(7): 1794-1801.
[14]	. Aero engine fault diagnosis based on improved density peak clustering [J]. , 2020, 26(5期): 1211-1217.
[15]	. Tool wear condition recognition based on kernel principal component and grey wolf optimizer algorithm [J]. , 2020, 26(11): 3031-3039.