Font Size: a A A

Process Monitoring Based On Fisher Disciminant Analysis

Posted on:2012-04-29Degree:MasterType:Thesis
Country:ChinaCandidate:H H XinFull Text:PDF
GTID:2178330338993737Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
With the wide application of DCS and computer technology, a large number of processdata can be collected and stored. How to fully utilize the deep-level information to improvethe performance of process monitoring has been gradually becoming one of the hotspots inthe field of process control. As to deal with the nonlinear system and statistical correlationbetween variables, a novel fault detection method using improved kernel Fisher discriminentanalysis is studied in this paper. Moreover, an optimal searching algorithm for the kernelparameter is employed to fault diagnosis based on kernel Fisher feature extraction. Takingcontinuous stirred tank reactor(CSTR) as a background, specific analysis and simulation wascarried according to the diverse disturbance of different variables.First, the thesis analyzed the principle and algorithm of Fisher discriminant analysis andcarried out the fault detection of continuous process based on FDA. Then, according to thenonlinear between process variables, an improved algorithm of kernel methods KFDA isstudied, which can separate nonlinear data effectively through mapping data tohigh-dimensional feature space, so as to improve the performance of fault detection.Independent component analysis can extract the intrinsic factor of the process under themeanings of the statistical independence. A mixture method of fault detection, ICA_KFDA, isproposed to get fewer key variables, which can drive the process and describe the essentialcharacteristics of the production more efficiently. The simulation on CSTR shows that,compared with the method based on FDA and KFDA, fault detection adopting ICA_KFDAalgorithm can obviously improve the performance. Meanwhile, the method of probabilitydensity estimation is adopted to calculate the control limit of distance statistic in monitoringprocess, for its distribution unknown.On the basis of fault detection, Fisher feature extraction method is put forward to reduce the dimension of date and diagnose various faults. And then the improved method based onkernel Fisher feature extraction is exploited to acquire better data separation efficiency. Anunconstrained nonlinear programming algorithm is employed to obtain the optimal kernelparameter. For the two feature extraction methods, distance classifier and similarity classifieris applied respectively in order to compare the diagnosis effect, the result of simulation showsthat the method based on nonlinear Fisher extracting features improve the diagnosisperformance.
Keywords/Search Tags:Fisher discriminant analysis, Kernel function, Independent componentanalysis, Process monitoring, CSTR
PDF Full Text Request
Related items