| With the rapid development of the technology,more and more engineers are devoting themselves to improving product's quality,reducing downtime and increasing reliability of facilities.Many effective monitoring methods are applied to industrial process.Fault diagnosis based on data-driven detection method is an important part of process monitoring field.It depends on process data,but not precise mathematical model,therefore,the detection method is paid more and more attention in processing monitoring filed.Because the data is dynamic and the statistical model can not be updated in time,the methods of fault diagnosis based on the Multivariate Statistical Process Control methods in industrial process are studied systematically in this thesis.The main research work is as follows:(1)The basis of fault diagnosis including its performance indexes and classification method are introduced,research results in several aspects of dynamic data fault diagnosis methods are summarized.(2)Researches on the independent component analysis(ICA)algorithm,the dynamic independent component analysis(DICA)algorithm,the canonical variate analysis algorithm and adaptive principle component analysis algorithm(includes the recursive principle component Analysis algorithm and the moving window principle component analysis algorithm).(3)Researches on the SI-DICA method of the canonical variate analysis and dynamic independent component analysis combination,improving the testing performance.The effectiveness of the proposed method is verified by the Tennessee-Eastman Process(TEP)simulation.(4)Based on the moving window principle component analysis algorithm,the on-line adaptive SI-DICA is presented.The method introduces the ideas of RPCA algorithm and gives the simplified recursive formula of the correlation matrix.The method improves the online update rate and saves the hardware storage space.The effectiveness of the proposed method is verified by the Tennessee-Eastman Process(TEP)simulation. |
