Research On Method Of TE Process Monitoring Based On Principal Component Analysis And Partial Least Squares

To meet the productivity needs, As a modern industrial system whose process has been getting complicated, scale has been getting large and production system has been getting intelligent. An advanced process monitoring system is unquestionable playing a important role on improving efficiency and reliability of a production system. However,nonlinear,small sample and identification of complex faults are the major problems in current research on process monitoring. In this paper, multivariate statistical analysis as the core method for the process industry process monitoring launched following research:This paper has analyzed the origins and development of process monitoring techniques, and introduced the classification of process monitoring methods. Has proposed the problems in current process monitoring methods. Principal component analysis(PCA) and partial least squares(PLS) method have been analyzed and deduced in this paper, and has gave a detailed description of implementation steps of fault detection based on PCA and PLS. Has compared the test results of the two methods and identified the advantages of PLS.To solve the nonlinear problems, this paper has introduced the kernel function method. Has generated kernel principal component analysis(KPCA) and kernel partial least squares(KPLS), and has gave a detailed description of implementation steps of fault detection based on KPCA and KPLS. Has compared the test results of the two methods with PCA and PLS,and this has illustrated the advantages of the kernel function in solving nonlinear problems。This paper has introduced the production process, process variables and process faults of Tennessee-Eastman process. Has collected a large number of sample data by the simulation model of TE process and has solved the small sample. In this paper has detected the known faults of TE process by KPLS method, and analyzed the test results. For the identification of complex fault, this paper has introduced contribution chart, and has combined it with KPLS method, solved the fault detection and location of system multi-fault.