On-line Batch Process Monitoring And Quality Prediction Based On Partial Least-square Algorithm

Online monitoring and quality prediction for batch process is build statisticalmodel based on process history data to detect the fault, process upsets and otherabnormal events promptly, locating and removing the factors causing such events, andthe safety of production process will be assured and the quality of product will beimproved. The achievements of the thesis can be summarized as follows,(1) Analysis and improvement of traditional multiway PLS (MPLS)An improved multi-way PLS algorithm for statistical batch monitoring、faultdiagnosis and quality prdiction is proposed in order to overcome the drawbacks oftraditional MPLS which use different unfolding wise. the method combines theadvantages of traditional expanding methods. It contains information of differentbatches that remove the nonlinear and dynamic characteristics of the process at acertain extent, as well as resolving the problem of data filled in online applications; Incalculating the monitoring statistic of T~2, the method researched in this paper usestime-varying covariance to replace the fixed one of principal components, which fullyconsiders the dynamic characteristics of the score vector. Meanwhile, aiming at theproblem of characteristic of lagging in process that the ordinary contribution plotmethod is difficult to correctly display the source of the fault of the current moment, afault diagnosis method based on time-varying contribution plot is proposed in thispaper. The improved method was proved to be effective by comparing with traditionalMPLS through experiments.(2) Nonlinear batch process monitoring and quality prediction based on MKPLSModified MPLS has a better performance，but it is always a linear method.Mostbatch processes have strong nonlinear characteristics. Monitoring will appear a largenumber of false positives and false negatives, performance prediction of low.Researched the kernel partial least squares (KPLS) method based on the kernelfunction and discussed some defects and problems of the KPLS method is applied tobatch process monitoring, the AT-MKPLS method for extracting FS features is putforward based on this. Retaining the batch information by adopting the AT method totransform the three-dimensional data; For the kernel mapping the huge volume of datanot operational problems, using the FS feature extraction for simplification of the sample, extract the most simple matrix, and then use the KPLS method to extract thenonlinear information of the process. The method is applied to a numerical nonlinearprocess and batch fermentation process, compared with the traditional MPLS andimproved MKPLS method, to verify the effectiveness of the method in monitoringand prediction.(3) Escherichia coli fermentation field experiment testField experimental study of monitoring and forecasting method, the method is ina Beijing biochemical pharmaceutical factory of practical application, carrying onprocess monitoring and quality prediction of recombinant Escherichia coli producinginterleukin-2fermentation process. Through the comparison of monitoring andforecasting performance, it verifies the effectiveness and practicability of theproposed method.