Residual Chart with Hidden Markov Model to Monitoring the Auto-Correlated Processes

doi:10.1007/s12204-018-2029-3

摘要/Abstract

摘要： Autocorrelations exist in real production extensively, and special statistical tools are needed for process monitoring. Residual charts based on autoregressive integrated moving average (ARIMA) models are typically used. However, ARIMA models need a quite amount of experience, which sometimes causes inconveniences in the implementation. With a good performance under less experience or even none, hidden Markov models (HMMs) were proposed. Since ARIMA models have many different performances in positive and negative autocorrelations, it is interesting and essential to study how HMMs affect the performances of residual charts in opposite autocorrelations, which has not been studied yet. Therefore, we extend HMMs to negatively auto-correlated observations. The cross-validation method is used to select the relatively optimal state number. The experiment results show that HMMs are more stable than Auto-Regressive of order one (AR(1) models) in both cases of positive and negative autocorrelations. For detecting abnormalities, the performance of HMMs approach is much better than AR(1) models under positive autocorrelations while under negative autocorrelations both methods have similar performances.

关键词: auto-correlations , hidden Markov models, statistical process control

Abstract: Autocorrelations exist in real production extensively, and special statistical tools are needed for process monitoring. Residual charts based on autoregressive integrated moving average (ARIMA) models are typically used. However, ARIMA models need a quite amount of experience, which sometimes causes inconveniences in the implementation. With a good performance under less experience or even none, hidden Markov models (HMMs) were proposed. Since ARIMA models have many different performances in positive and negative autocorrelations, it is interesting and essential to study how HMMs affect the performances of residual charts in opposite autocorrelations, which has not been studied yet. Therefore, we extend HMMs to negatively auto-correlated observations. The cross-validation method is used to select the relatively optimal state number. The experiment results show that HMMs are more stable than Auto-Regressive of order one (AR(1) models) in both cases of positive and negative autocorrelations. For detecting abnormalities, the performance of HMMs approach is much better than AR(1) models under positive autocorrelations while under negative autocorrelations both methods have similar performances.

Key words: auto-correlations , hidden Markov models, statistical process control

中图分类号:

F273

LI Yaping (李亚平), HUANG Mengdie (黄梦蝶), PAN Ershun (潘尔顺). Residual Chart with Hidden Markov Model to Monitoring the Auto-Correlated Processes[J]. Journal of Shanghai Jiao Tong University (Science), 2018, 23(Sup. 1): 103-108.

参考文献 32

[1]	BERTHOUEX P M, HUNTER W G, PALLESENL. Monitoring sewage treatment plants: Some qualitycontrol aspects [J]. Journal of Quality Technology,1978, 10(4): 139-149.
[2]	MONTGOMERY D C, MASTRANGELO C M. Somestatistical process control methods for autocorrelateddata [J]. Journal of Quality Technology, 1991, 23(3):179-193.
[3]	MARAGAH H D, WOODALL W H. The effect of autocorrelationon the retrospective X-chart [J]. Journalof Statistical Computation & Simulation, 1992,40(1/2): 29-42.
[4]	MONTGOMERY D C. Statistical quality control: Amodern introduction [M]. 6th ed. New York: John Wiley& Sons, 2009.
[5]	VASILOPOULOS A V, STAMBOULIS A P. Modificationof control chart limits in the presence of datacorrelation [J]. Journal of Quality Technology, 1978,10(1): 20-30.
[6]	WARDELL D G, MOSKOWITZ H, PLANTE R D.Control charts in the presence of data correlation [J].Management Science, 1992, 38(8): 1084-1105.
[7]	JIANG W, TSUI K L, WOODALL W H. A new SPCmonitoring method: The ARMA chart [J]. Technometrics,2000, 42(4): 399-410.
[8]	LU C W, REYNOLDS M R Jr. CUSUM charts formonitoring an autocorrelated process [J]. Journal ofQuality Technology, 2001, 33(3): 316-334.
[9]	YASHCHIN E. Performance of CUSUM controlschemes for serially correlated observations [J]. Technometrics,1993, 35(1): 37-52.
[10]	CASTAGLIOLA P, TSUNG F. Autocorrelated SPCfor non-normal situations [J]. Quality and ReliabilityEngineering International, 2005, 21 (2): 131-161.
[11]	SCHMID W. On the run length of a Shewhart chartfor correlated data [J]. Statistical Papers, 1995, 36(2):111-130.
[12]	GARZA-VENEGAS J A, TERCERO-GOMEZ V G,HO LEE L, et al. Effect of autocorrelation estimatorson the performance of the ˉX -control chart [J]. Journalof Statistical Computation and Simulation, 2018,88(13): 2612-2630.
[13]	YANG S F, YANG C M. An approach to controllingtwo dependent process steps with autocorrelated observations[J]. International Journal of Advanced ManufacturingTechnology, 2006, 29(1): 170-177.
[14]	DAVOODI M, NIAKI S T A. Estimating the stepchangetime of the location parameter in multistageprocesses using MLE [J]. Quality and Reliability EngineeringInternational, 2012, 28: 843-855.
[15]	PAN J N, LI C I,WU J J. A new approach to detectingthe process changes for multistage systems [J]. ExpertSystems with Applications, 2016, 62: 293-301.
[16]	MASTRANGELO C M, MONTGOMERY D C. SPCwith correlated observations for the chemical and processindustries [J]. Quality and Reliability EngineeringInternational, 1995, 11 (2): 79-89.
[17]	LU C W, REYNOLDS M R Jr. EWMA control chartsfor monitoring the mean of autocorrelated processes[J]. Journal of Quality Technology, 1999, 31(2): 166-188.
[18]	ALWAN L C, ROBERTS H V. Time-series modelingfor statistical process control [J]. Journal of Business& Economics Statistics, 1988, 6(1): 87-95.
[19]	MONTGOMERY D C, MASTRANGELO C M. Somestatistical process control methods for autocorrelateddata [J]. Journal of Quality Technology, 1991, 23 (3):200-202.
[20]	WARDELL D G, MOSKOWITZ H, PLANTE RD. Run-length distributions of special-cause controlcharts for correlated processes [J]. Technometrics,1994, 36(1): 3-17.
[21]	ZHANG N F. Detection capability of residual controlchart for stationary process data [J]. Journal of AppliedStatistics, 1997, 24(4): 475-492.
[22]	VANHATALO E, KULAHCI M. The effect of autocorrelationon the Hotelling T2 control chart [J]. Qualityand Reliability Engineering International, 2015, 31(8):1779-1796.
[23]	CHIU C C, CHEN M K, LEE K M. Shifts recognitionin correlated process data using a neural network [J].International Journal of Systems Science, 2001, 32(2):137-143.
[24]	ARKAT J, NIAKI S T A, ABBASI B. Artificial neuralnetworks in applying MCUSUM residuals charts forAR(1) processes [J]. Applied Mathematics and Computation,2007, 189(2): 1889-1901.
[25]	PACELLA M, SEMERARO Q. Using recurrent neuralnetworks to detect changes in autocorrelated processesfor quality monitoring [J]. Computers & Industrial Engineering,2007, 52(4): 502-520.
[26]	CAMARGO M E, FILHO W P, RUSSO S L, et al.Control charts for monitoring autocorrelated processesbased on neural networks model [C]//InternationalConference on Computers & Industrial Engineering.Troyes, France: IEEE, 2009: 1881-1884.
[27]	YU J L, HAN Y, MIAO M X, et al. Self-related processresidual control chart based on neural network[C]//International Symposium on Knowledge Acquisitionand Modeling. London, UK: Atlantis Press, 2015:41-43.
[28]	YANG H H, HUANG M L, YANG S W. Integratingauto-associative neural networks with Hotelling T 2control charts for wind turbine fault detection [J]. Energies,2015, 8: 12100-12115.
[29]	RAI A, UPADHYAY S H. The use of MD-CUMSUMand NARX neural network for anticipating the remaininguseful life of bearings [J]. Measurement, 2017, 111:397-410.
[30]	ALSHRAIDEH H, RUNGER G. Process monitoringusing hidden Markov models [J]. Quality and ReliabilityEngineering International, 2015, 30(8): 1379-1387.
[31]	ROSS S M. Introduction to probability models [M].11th ed. Beijing: Posts & Telecom Press, 2015.
[32]	CHEN Z, XIA T B, LI Y P, et al. Degradation modelingand classification of mixed populations using segmentalcontinuous hidden Markov models [J]. Qualityand Reliability Engineering International, 2018,34(5): 807-823.