利用选择性DAG-SVM集成在线检测与诊断多变量过程均值异常

中国机械工程 ›› 2013, Vol. 24 ›› Issue (21): 2895-2902.

利用选择性DAG-SVM集成在线检测与诊断多变量过程均值异常

朱波;刘飞

重庆大学机械传动国家重点实验室，重庆，400030

出版日期:2013-11-10 发布日期:2013-11-15
基金资助:
国家自然科学基金资助重点项目（51035001）；国家科技重大专项（2011ZX04001-041-06）
National Natural Science Foundation of China（No. 51035001）;
National Science and Technology Major Project ( No. 2011ZX04001-041-06)

On-line Detecting and Diagnosing of Mean Shifts in Multivariate Processes Using Selective DAG-SVM Ensemble

Zhu Bo;Liu Fei

The State Key Laboratory of Mechanical Transmission,Chongqing University,Chongqing,400030

Online:2013-11-10 Published:2013-11-15
Supported by:

National Natural Science Foundation of China（No. 51035001）;
National Science and Technology Major Project ( No. 2011ZX04001-041-06)

摘要/Abstract

摘要：

针对多变量过程均值异常，提出了选择性有向无环图支持向量机(DAG-SVM)集成，以之为模式识别工具对过程状态进行识别，以探测异常和判别异常源。集成结合Bagging方法的重复采样技术和对DAG-SVM结构的调整，对数据和模型进行双重扰动，以获得差异的候选个体;再通过二进制粒子群优化(BPSO)算法得到最优集成方案。离线仿真测试证明所提选择性DAG-SVM集成具有分类正确率和效率的双重优势;在线仿真测试表明基于选择性DAG-SVM集成的模型探测过程均值阶跃异常优于χ2图和BPN（误差反传神经网络）模型，且判别异常源比BPN模型更准确。针对实际齿轮加工过程数据的应用验证进一步证实了模型的有效性和实用性。

关键词: 统计过程控制, 多变量过程, 均值阶跃, 有向无环图支持向量机

Abstract:

Aiming at mean shifts in multivariate processes, a selective DAG-SVM ensemble was proposed, which was used as pattern recognition tool to construct an on-line detecting and diagnosing model for the shifts. To produce diverse candidate components of the ensemble, a double disturbance mechanism of data and model was implemented, i.e. , through the Bootstrap sampling approach from Bagging method to create different training data sets, and to randomly adjust the structure of DAG-SVM. Finally, the BPSO(binary particle swarm optimization) algorithm was adopted to obtain the optimal ensemble. The simulative off-line test verify that the selective DAG-SVM ensemble have better performance on both classification rate and efficiency. The simulative on-line test shows that the model based on the selective DAG-SVM ensemble outperform χ2 chart and BPN(back propagation neural network) model on detecting mean shifts, and identify the sources of shifts more accurately than the BPN model. An application examination on actual gear machining process data further verified the validity and practicability of this model.

Key words: statistic process control, multivariate process, mean shift, directed acyclic graph support vector machine(DAG-SVM)

中图分类号:

朱波;刘飞. 利用选择性DAG-SVM集成在线检测与诊断多变量过程均值异常[J]. 中国机械工程, 2013, 24(21): 2895-2902.

Zhu Bo;Liu Fei. On-line Detecting and Diagnosing of Mean Shifts in Multivariate Processes Using Selective DAG-SVM Ensemble[J]. China Mechanical Engineering, 2013, 24(21): 2895-2902.

参考文献

[1]Hachicha W, Ghorbel A. A Survey of Control- chart Pattern-- recognition Literature (1991-2010) Based on a New Conceptual Classification Scheme[J]. Computers & Industrial Engineering, 2012,63 (1) :204-222.
[2]Wang T Y, Chen L H. Mean Shifts Detection and Classification in Multivariate Process: a Neural-- fuzzy Approach[J].Journal of Intelligent Manufac- turing,2002,13(3) :211-221.
[3]Chen I H,Wang T Y. Artificial Neural Networks to Classify Mean Shifts from Multivariate Chart Sig- nals[J].Computers and Industrial Engineering, 2004,47(2/3) : 195-205.
[4]Guh R S. On--line Identification and Quantification of Mean Shifts in Bivariate Processes Using a Neu- ral Network--based Approach[J].Quality and Re- liability Engineering International, 2007, 23 ( 3 ) : 367-385.
[5]Niaki S T A, Nafar M. An Artifieial Neural Network Approach to Monitor and Diagnose Multi--attrib- ute Quality Control Process[J]. Journal of Industri- al Engineering International, 2008,4 (7) : 10-24.
[6]Yu Jianbo, Xi Lifeng, Zhou Xiaojun. Identifying Source(s) of Out--of--control Signals in Multiva- riate Manufacturing Processes Using Selective Neu- ral Network Ensemble [J]. Engineering Applica- tions of Artificial Intelligence, 2009,22 ( 1 ) : 141 - 152.
[7]Vapnik V N. The Nature of Statistical Learning Theory[M]. New York: Springer Verlag, 2000.
[8]马超[1],陈西宏[1],徐宇亮[1],姚懿玲[2].基于支持向量机属性约简集成的模拟电路故障诊断[J].仪器仪表学报,2011,32(3):660-666.
[9]李烨[1],蔡云泽[1],尹汝泼[1],许晓鸣[2].基于证据理论的多类分类支持向量机集成[J].计算机研究与发展,2008,45(4):571-578.
[10]Platt J C, Cristianini N, Shawe--Taylor J. Large Margin DAGs for Multi- class Classification [C]//Sara A Solla, Todd K Leen, Klaus-- Robert Muller. Advances in Neural Information Process- ing Systems 12. Cambridge, MA: MIT Press, 2000:547-553.
[11]Dietterich T G. An Experimental Comparison of Three Methods for Constructing Ensembles of De- cision Trees Bagging Boosting and Randomization [J]. Machine Learning, 2000,40 (2) : 139-158.
[12]Zhou Zhihua, Wu Jianxin, Tang Wei, et al. Combi- ning Regression Estimators: GA--based Selective Neural Network Ensemble[J]. International Jour- nal of Computational Intelligence and Applica- tions, 2001,1 (4) : 341-356.
[13]Breiman L. Bagging Predictors[J]. Machine Learn- ing, 1996,24(2) : 123-140.
[14]Khanesar M A,Teshnehlab M,Shoorehdeli M A, et ah Novel Binary Particle Swarm Optimization [C]//Mediterranean Conference on Control & Au- tomation. Athens, 2007 : 1-6.
[15]史锋,王辉,胡斐,等.MATLAB智能算法30个案例分析[M].北京:北京航空航天大学出版社,2011:69-72.
[16]Hassan A, SharifI N B M, Shaharoun A M, et al. Improved SPC Chart Pattern Recognition Using Statistical Features [J]. International Journal of Production Research,2003,41(7) : 1587-1603.
[17]Alt F B. Multivariate Quality Control [M]//S. Kotz, N. L. Johnson, C. R. Read(Eds. ), The En- cyclopedia of Statistical Science. New York: Wiley, 3 1984.