PLS及其扩展方法在过程监控中的应用研究
|
摘要
生产安全与产品质量一直是过程工业的两大重要主题。故障诊断技术自70年代发展以来,一直是过程监控领域的一个重要分支。在故障诊断理论的各种方法中,较有发展前途的是基于数据的多变量统计分析方法,这是由过程工业本身难以建立精确模型、强耦合的多变量关系等特点决定的。
偏最小二乘(PLS,Partial Least Squares)分析方法是一种利用统计原理提取过程数据的有用信息建立过程模型的降维技术。本文从偏最小二乘的算法原理出发,系统深入地探讨了利用PLS、判别式PLS结合多变量统计图和统计量贡献图对多变量生产过程进行故障检测与诊断的整个思路与应用情况。为了提高高噪声情况下的故障检出率,本文提出了小波偏最小二乘(Wavelet PLS,WPLS)的故障诊断方法,并通过数值仿真实例验证了该方法在故障检测与诊断方面的有效性,最后针对TE过程进行仿真实验研究。仿真实验与对比仿真实验研究表明,与PLS方法相比,判别式PLS对TE过程故障的平均检出率提高了11%左右,而WPLS方法提高了34%左右;而且,与判别式PLS相比,WPLS方法具备对未知故障的检测能力。本文提出的WPLS故障诊断方法在降低误报和漏报方面也有了很大改进,同时在检测对过程影响较小的故障方面收效显著,提高了故障检测与诊断的准确性。因此,WPLS故障诊断方法是一种更为有效的过程监控手段。
Production safety and product quality have been two major themes needed to research in process industry.Fault diagnosis technology,developed since 1970s,has been an important branch of Process Monitoring field.Among the fault diagnosis methods, the statistical method based on data is more promising,which is because it is difficult to establish the precise model of the industrial process and the relations between process variates are strong coupling.
Partial Least Squares(PLS) analysis method is a data dimensionality reduction technology,which extracts useful information from data to build process model using statistical theory.This thesis studied the basic principal of PLS and its application in Fault Detection and Diagnosis(FDD).By the means of multivariate statistical charts and contribution histogram plots,PLS and Discriminant PLS methods were applied in fault detection and diagnosis of simulated systems.The simulation results were expressed by statistical charts and contribution plots.In order to improve the fault detection ratio under high noise,wavelet partial least squares (WPLS) fault diagnosis method was proposed,and its effectiveness was illustrated by the application in the numerical example. Finally,the superior performance of WPLS for process monitoring was reflected in Tennessee-Eastman Process(TEP) simulation experiment.Comparison of fault diagnosis results among PLS,Discriminant PLS,WPLS showed that,the results of Discriminant PLS increased by around 11 percent,results of WPLS method raised by 34 percent,compared with results of PLS for the average fault detection ratio.Moreover, WPLS method has the capabilities of detecting unknown fault better than Discriminant PLS.Furtherly,WPLS makes great improvment in reducing faulse report and omitted report,and WPLS than PLS,DPLS better able to detect relatively weak industrial process changes,which enhances the accuracy of the fault detection and diagnosis.Therefore,the WPLS fault diagnosis method is a more effective means of process monitoring.
偏最小二乘(PLS,Partial Least Squares)分析方法是一种利用统计原理提取过程数据的有用信息建立过程模型的降维技术。本文从偏最小二乘的算法原理出发,系统深入地探讨了利用PLS、判别式PLS结合多变量统计图和统计量贡献图对多变量生产过程进行故障检测与诊断的整个思路与应用情况。为了提高高噪声情况下的故障检出率,本文提出了小波偏最小二乘(Wavelet PLS,WPLS)的故障诊断方法,并通过数值仿真实例验证了该方法在故障检测与诊断方面的有效性,最后针对TE过程进行仿真实验研究。仿真实验与对比仿真实验研究表明,与PLS方法相比,判别式PLS对TE过程故障的平均检出率提高了11%左右,而WPLS方法提高了34%左右;而且,与判别式PLS相比,WPLS方法具备对未知故障的检测能力。本文提出的WPLS故障诊断方法在降低误报和漏报方面也有了很大改进,同时在检测对过程影响较小的故障方面收效显著,提高了故障检测与诊断的准确性。因此,WPLS故障诊断方法是一种更为有效的过程监控手段。
Production safety and product quality have been two major themes needed to research in process industry.Fault diagnosis technology,developed since 1970s,has been an important branch of Process Monitoring field.Among the fault diagnosis methods, the statistical method based on data is more promising,which is because it is difficult to establish the precise model of the industrial process and the relations between process variates are strong coupling.
Partial Least Squares(PLS) analysis method is a data dimensionality reduction technology,which extracts useful information from data to build process model using statistical theory.This thesis studied the basic principal of PLS and its application in Fault Detection and Diagnosis(FDD).By the means of multivariate statistical charts and contribution histogram plots,PLS and Discriminant PLS methods were applied in fault detection and diagnosis of simulated systems.The simulation results were expressed by statistical charts and contribution plots.In order to improve the fault detection ratio under high noise,wavelet partial least squares (WPLS) fault diagnosis method was proposed,and its effectiveness was illustrated by the application in the numerical example. Finally,the superior performance of WPLS for process monitoring was reflected in Tennessee-Eastman Process(TEP) simulation experiment.Comparison of fault diagnosis results among PLS,Discriminant PLS,WPLS showed that,the results of Discriminant PLS increased by around 11 percent,results of WPLS method raised by 34 percent,compared with results of PLS for the average fault detection ratio.Moreover, WPLS method has the capabilities of detecting unknown fault better than Discriminant PLS.Furtherly,WPLS makes great improvment in reducing faulse report and omitted report,and WPLS than PLS,DPLS better able to detect relatively weak industrial process changes,which enhances the accuracy of the fault detection and diagnosis.Therefore,the WPLS fault diagnosis method is a more effective means of process monitoring.
引文
[1]胡峰,孙国基过程监控技术及其应用.北京:国防工业出版社,2002.1-100
[2]蒋浩天,E.L.拉塞尔,R.D.布拉茨.工业系统的故障检测与诊断.北京:机械工业出版社,2003.2-137
[3]蒋立英.基于FDA/DPLS方法的流程工业故障诊断研究.清华大学.2005.1-35
[4]唐凯.基于多元统计过程控制的故障诊断技术.浙江大学.2004.2-35
[5]封波.多变量统计过程控制的应用研究.浙江大学.2002.8-53
[6]方伦钢.应用多元统计过程控制技术进行状态监测.南京航空航天大学2004.1-18
[7]项鲁丁.基于多变量统计过程控制方法—PCA/PLS在工业控制中的应用.2004.1-27
[8] Downs.J.and E.Vogel.A plant-wide industrial process control problem. Computers and Chemical Engineering.1993.20:245-255
[9]郭明.基于数据驱动的流程工业性能监控与故障诊断研究.2004.9-67
[10]朱尔一,杨梵原.化学计量学技术及其应用.北京:科学出版社,2001:92-102
[11]李春富.基于数据的软测量建模方法及其应用的研究.清华大学.2005.1-67
[12]颜杰,丁德荣,刘世庆.小波变换—偏最小二乘算法及其在复方甲硝唑注射液分析中的应用.分析测试学报.2000.19(1):71-73
[13]王延勇,陈淑桂,王洪艳等.小波偏最小二乘光度法用于铜矿石相态分析的研究.长春科技大学学报.1999.29(2):202-205
[14]董胜利,王树青,谢磊.鲁棒PLS在间歇生产过程监控中的应用.控制与决策,20(7):823-826
[15] U.Kruger,etc.Extended PLS Approach for Enhanced Condition. Monitoring of Industrial Processes.AIChE Journal,2001,47(9): 2076-2091
[16] MacGregor,J.F.,etc.Process monitoring and diagnosis by multiblockPLS methods, AIChE.1994.40(5):826-838
[17] P.R.Lyman and C.Georgakis.Plant-wide control of the Tennessee Eastmanproblem.Computer & Chemical Engineering,1995,19(3):321-331
[18] Gang Chen and Thomas J.McAvoy.Predictive on-line monitoring ofcontinuous processes.Journal of Process Control.1998,8(5-6):409-420
[19] Down,J,J.and vogel,E.F.A plant-wide industrial process control Problem.computers & Chemical Engineering,1993,17(3):245-255
[20] Jesús Lozano,etc.Correlating e-nose responses to wine sensorial descriptors and gas chromatography-mass spetrometry profiles using partial least squares regression analysis.Journal of Hazardous Materical,2007(146):421-427
[21]王惠文等.偏最小二乘回归线性与非线性方法.北京:国防工业出版社.2006:1-117
[22] Pierre Druilhet,Alain Mom.PLS regression:A directional signal -to-noise ratio approach.Journal of Multivariate Analysis, 97(2006):1313-1329
[23] P.Geladi and B.R. Kowalski.Partial least-squares reg ression:A Tutorial.Analytica Chimica Acta,1986(185):1-17
[24]陈勇.基于多元统计分析的生产过程故障诊断研究.浙江大学.2003:31-33
[25]李明.基于多元统计分析的故障诊断方法及其应用研究.山东大学,2006:20-41
[26]王俊锋,钱宇,李秀喜等.用主元分析方法完善DCS过程监控性能.化工自动化仪表.2002,29(3):15-18
[27] L.H.Chiang,E.L.Russell,and R.D.Braatz.Fault diagnosis in chemicalprocesses using Fisher discriminant analysis,discriminant partialleast squares, and principal component analysis.Chemometrics and Intelligent Laboratory Systems,2000,50(2):243-252
[28] Shengjing Mu,etc.Online dual updating with recursive PLS model and its applicationin predicting crustal size of purified terepht-halic acid(PTA) process.Journal of Process Control.2006(16):557-566
[29]宋凯,王海清,李平.PLS质量监控及其在TE过程中的应用.浙江大学学报(工学版).2005,39(5):657-662
[30]潘泉等.小波滤波方法及应用.北京:清华大学出版社,2005.1-40
[31]司圣柱,司娲.小波变换-偏最小二乘法用于三种食用色素的可见分光光度法同时测定.分析仪器.2007(2):48-51
[32]胡昌华等.基于MATLAB的系统分析与设计—小波分析.西安:西安电子科技大学出版社,1999.1-64
[33]孙延奎.小波分析及其应用.北京:机械工业出版社,2005.157-170
[34]刘涛等.实用小波分析入门.北京:国防工业出版社,2006.113-151
[35]李艳兰,蔚文杰.小波分析方法应用于故障诊断的研究.机械管理开发.2007,95(2):40-43
[36]叶昊,王桂增,方崇智.小波变换在故障检测中的应用.自动化学报.1997,23(6):736-741
[37]冯伟东等,信号的多分辨率分析在消噪中的应用.长春工业大学学报(自然科学版).2007,28(1):101-104
[38]艾施连,郭德淳,王之国.基于小波变换消除信号噪声的研究.军民两用技术与产品.2007(1):44-45
[39]罗凯华.三种常用信号处理方法比较.计量与测试技术.2007,34(2):15-17
[40] Hai Qiu,etc.Wavelet filter-based weak signature detection methodand its application on rolling element bearing prognostics. Journal of sound and vibration.2006(289):1066-1090
[41] R.Kalpana,S.Raja Balachandar.Haar wavelet method for the analysis oftransistor circuits.Int.J.Electron.Commun.(AEü).2007(61):589-594
[42] Gand Niu,etc.Decision-level fusion based on wavelet decompositionfor induction motor fault diagnosis using trandient.Expert Systemswith Applications.2007,8(24)
[43]蒋丽英,王树青.基于MPCA-MDPLS的间歇过程的故障诊断.化工学报.2005,56(3):482-486
[44] Skamaledin Setarehdan.Modified evoling window factor analysis forprocess monitoring.Journal of chemometrics.2004(18):414-421
[45] Gibaek Lee.Multiple-Fault Diagnosis of the Tennessee Eastman ProcessBased on System Decomposition and Dynamic PLS.Ind.Eng.Chem.Res.2004(43):8037-8048
[46] LEO HAO-TIEN CHIANG.Fault Detection and Diagnosis for Large-scaleSystems.B.S.,University of Wisconsin,1997:10-73
[47] Q.Peter He and S.Joe Qin.A New Fault Diagnosis Method Using Fault Directions in Fisher Discriminant Analysis.AICHE Journal.2005,51(2):555-571
[48] Bahram Hemmateenejad,etc.Partial least squares-based on multivariate spectral calibration method for simultaneous determination of beta-carboline derivatives in Peganum hatmala seed extracts. Analytica Chimica Acta.2006(575):290-299
[49] A.Simoglou,E.B.Martin,A.J.Morris.Multivariate statistical process control of an industrial fluidised-bed reactor.Control Engineering Practice.2000(8):893-909
[50] Q.Chen,U.Kruger,M.Meronk,etc.Synthesis of and statistics for processmonitoring.Control Engineering Practice.2004(12):745-755
[51] Nomikos P.MacGregor J F.Monitoring batch processes using multiway principal component analysis.AIChE Journal.1994,40(8):1361-1375
[52] MacGregor J F,etc.Process monitoring and diagnosis by multiblock PLSmethods.AIChE Journal.1994,40(5):826-838
[53] Tiina Komulainen,etc.An online application of dynamic PLS to a dearomatization process.Computer & Chemical Engineering 2004(28):2611-2619
[54] Dae Sung Lee,etc.Nonlinear dynamic partial least squares modelingof a full-scale biological wastewater treatment plant.Process Biochemistry 2006(41):2050-2057
[55]黄启明,钱宇等.化工过程故障诊断研究进展.化工自动化及仪表,2000,27(3):1-5
[56] Svante Wold,Nouna Kettaneh-Wold and Bert Skagerberg.Nonlinear PLSmodeling.1989,7(1-2):53-65
[57]时瑞研,潘立登.一种新型非线性偏最小二乘方法研究与应用—基于Chebyshev多项式改进的PLS方法.控制工程.2003,10(6):506-509
[58] Yan-Ping Zhou,etc.Artificial neural network-based transformation for nolinear partial least-square regression with application to QSAR studies.talanta 8561(2006)6
[59] Lu Xu,etc.Optimized sample-weighted partial least aquares.Talanta8497(2006)6
[60] A.Simoglou,E.B.Martin,A.J.Morris.Multivariate statistical processcontrol of an industrial fluidized-bed reactor.Control Engineering Practice.2000(8):893-909
[61] Q.Chen,U.Kruger,M.Meronk,etc.Synthesis of T 2 and Q statistics forprocess monitoring.Control Engineering Practice.2004(12):745-755
[2]蒋浩天,E.L.拉塞尔,R.D.布拉茨.工业系统的故障检测与诊断.北京:机械工业出版社,2003.2-137
[3]蒋立英.基于FDA/DPLS方法的流程工业故障诊断研究.清华大学.2005.1-35
[4]唐凯.基于多元统计过程控制的故障诊断技术.浙江大学.2004.2-35
[5]封波.多变量统计过程控制的应用研究.浙江大学.2002.8-53
[6]方伦钢.应用多元统计过程控制技术进行状态监测.南京航空航天大学2004.1-18
[7]项鲁丁.基于多变量统计过程控制方法—PCA/PLS在工业控制中的应用.2004.1-27
[8] Downs.J.and E.Vogel.A plant-wide industrial process control problem. Computers and Chemical Engineering.1993.20:245-255
[9]郭明.基于数据驱动的流程工业性能监控与故障诊断研究.2004.9-67
[10]朱尔一,杨梵原.化学计量学技术及其应用.北京:科学出版社,2001:92-102
[11]李春富.基于数据的软测量建模方法及其应用的研究.清华大学.2005.1-67
[12]颜杰,丁德荣,刘世庆.小波变换—偏最小二乘算法及其在复方甲硝唑注射液分析中的应用.分析测试学报.2000.19(1):71-73
[13]王延勇,陈淑桂,王洪艳等.小波偏最小二乘光度法用于铜矿石相态分析的研究.长春科技大学学报.1999.29(2):202-205
[14]董胜利,王树青,谢磊.鲁棒PLS在间歇生产过程监控中的应用.控制与决策,20(7):823-826
[15] U.Kruger,etc.Extended PLS Approach for Enhanced Condition. Monitoring of Industrial Processes.AIChE Journal,2001,47(9): 2076-2091
[16] MacGregor,J.F.,etc.Process monitoring and diagnosis by multiblockPLS methods, AIChE.1994.40(5):826-838
[17] P.R.Lyman and C.Georgakis.Plant-wide control of the Tennessee Eastmanproblem.Computer & Chemical Engineering,1995,19(3):321-331
[18] Gang Chen and Thomas J.McAvoy.Predictive on-line monitoring ofcontinuous processes.Journal of Process Control.1998,8(5-6):409-420
[19] Down,J,J.and vogel,E.F.A plant-wide industrial process control Problem.computers & Chemical Engineering,1993,17(3):245-255
[20] Jesús Lozano,etc.Correlating e-nose responses to wine sensorial descriptors and gas chromatography-mass spetrometry profiles using partial least squares regression analysis.Journal of Hazardous Materical,2007(146):421-427
[21]王惠文等.偏最小二乘回归线性与非线性方法.北京:国防工业出版社.2006:1-117
[22] Pierre Druilhet,Alain Mom.PLS regression:A directional signal -to-noise ratio approach.Journal of Multivariate Analysis, 97(2006):1313-1329
[23] P.Geladi and B.R. Kowalski.Partial least-squares reg ression:A Tutorial.Analytica Chimica Acta,1986(185):1-17
[24]陈勇.基于多元统计分析的生产过程故障诊断研究.浙江大学.2003:31-33
[25]李明.基于多元统计分析的故障诊断方法及其应用研究.山东大学,2006:20-41
[26]王俊锋,钱宇,李秀喜等.用主元分析方法完善DCS过程监控性能.化工自动化仪表.2002,29(3):15-18
[27] L.H.Chiang,E.L.Russell,and R.D.Braatz.Fault diagnosis in chemicalprocesses using Fisher discriminant analysis,discriminant partialleast squares, and principal component analysis.Chemometrics and Intelligent Laboratory Systems,2000,50(2):243-252
[28] Shengjing Mu,etc.Online dual updating with recursive PLS model and its applicationin predicting crustal size of purified terepht-halic acid(PTA) process.Journal of Process Control.2006(16):557-566
[29]宋凯,王海清,李平.PLS质量监控及其在TE过程中的应用.浙江大学学报(工学版).2005,39(5):657-662
[30]潘泉等.小波滤波方法及应用.北京:清华大学出版社,2005.1-40
[31]司圣柱,司娲.小波变换-偏最小二乘法用于三种食用色素的可见分光光度法同时测定.分析仪器.2007(2):48-51
[32]胡昌华等.基于MATLAB的系统分析与设计—小波分析.西安:西安电子科技大学出版社,1999.1-64
[33]孙延奎.小波分析及其应用.北京:机械工业出版社,2005.157-170
[34]刘涛等.实用小波分析入门.北京:国防工业出版社,2006.113-151
[35]李艳兰,蔚文杰.小波分析方法应用于故障诊断的研究.机械管理开发.2007,95(2):40-43
[36]叶昊,王桂增,方崇智.小波变换在故障检测中的应用.自动化学报.1997,23(6):736-741
[37]冯伟东等,信号的多分辨率分析在消噪中的应用.长春工业大学学报(自然科学版).2007,28(1):101-104
[38]艾施连,郭德淳,王之国.基于小波变换消除信号噪声的研究.军民两用技术与产品.2007(1):44-45
[39]罗凯华.三种常用信号处理方法比较.计量与测试技术.2007,34(2):15-17
[40] Hai Qiu,etc.Wavelet filter-based weak signature detection methodand its application on rolling element bearing prognostics. Journal of sound and vibration.2006(289):1066-1090
[41] R.Kalpana,S.Raja Balachandar.Haar wavelet method for the analysis oftransistor circuits.Int.J.Electron.Commun.(AEü).2007(61):589-594
[42] Gand Niu,etc.Decision-level fusion based on wavelet decompositionfor induction motor fault diagnosis using trandient.Expert Systemswith Applications.2007,8(24)
[43]蒋丽英,王树青.基于MPCA-MDPLS的间歇过程的故障诊断.化工学报.2005,56(3):482-486
[44] Skamaledin Setarehdan.Modified evoling window factor analysis forprocess monitoring.Journal of chemometrics.2004(18):414-421
[45] Gibaek Lee.Multiple-Fault Diagnosis of the Tennessee Eastman ProcessBased on System Decomposition and Dynamic PLS.Ind.Eng.Chem.Res.2004(43):8037-8048
[46] LEO HAO-TIEN CHIANG.Fault Detection and Diagnosis for Large-scaleSystems.B.S.,University of Wisconsin,1997:10-73
[47] Q.Peter He and S.Joe Qin.A New Fault Diagnosis Method Using Fault Directions in Fisher Discriminant Analysis.AICHE Journal.2005,51(2):555-571
[48] Bahram Hemmateenejad,etc.Partial least squares-based on multivariate spectral calibration method for simultaneous determination of beta-carboline derivatives in Peganum hatmala seed extracts. Analytica Chimica Acta.2006(575):290-299
[49] A.Simoglou,E.B.Martin,A.J.Morris.Multivariate statistical process control of an industrial fluidised-bed reactor.Control Engineering Practice.2000(8):893-909
[50] Q.Chen,U.Kruger,M.Meronk,etc.Synthesis of and statistics for processmonitoring.Control Engineering Practice.2004(12):745-755
[51] Nomikos P.MacGregor J F.Monitoring batch processes using multiway principal component analysis.AIChE Journal.1994,40(8):1361-1375
[52] MacGregor J F,etc.Process monitoring and diagnosis by multiblock PLSmethods.AIChE Journal.1994,40(5):826-838
[53] Tiina Komulainen,etc.An online application of dynamic PLS to a dearomatization process.Computer & Chemical Engineering 2004(28):2611-2619
[54] Dae Sung Lee,etc.Nonlinear dynamic partial least squares modelingof a full-scale biological wastewater treatment plant.Process Biochemistry 2006(41):2050-2057
[55]黄启明,钱宇等.化工过程故障诊断研究进展.化工自动化及仪表,2000,27(3):1-5
[56] Svante Wold,Nouna Kettaneh-Wold and Bert Skagerberg.Nonlinear PLSmodeling.1989,7(1-2):53-65
[57]时瑞研,潘立登.一种新型非线性偏最小二乘方法研究与应用—基于Chebyshev多项式改进的PLS方法.控制工程.2003,10(6):506-509
[58] Yan-Ping Zhou,etc.Artificial neural network-based transformation for nolinear partial least-square regression with application to QSAR studies.talanta 8561(2006)6
[59] Lu Xu,etc.Optimized sample-weighted partial least aquares.Talanta8497(2006)6
[60] A.Simoglou,E.B.Martin,A.J.Morris.Multivariate statistical processcontrol of an industrial fluidized-bed reactor.Control Engineering Practice.2000(8):893-909
[61] Q.Chen,U.Kruger,M.Meronk,etc.Synthesis of T 2 and Q statistics forprocess monitoring.Control Engineering Practice.2004(12):745-755