Statistical Modeling And Monitoring Approach Research For Industrial Process

Industrial process modeling and monitoring is an important method to instruct the manufacture and insure the process safety. It plays a fundamental role in process optimization and control as well as keeping the process running smoothly. Industrial process usually involves high non-linearity and uncertainty, complex kinetic mechanism, and various plants, which makes it a challenge to study the process though mechanism modeling approaches. On the basis of DCS and database systems, a considerable amount of data about the studied process is available, so it is possible to exploit the data using statistical methods, which doesn't need much knowledge of the studied process. Melt Index (MI) is the most important parameter in determining the polypropylene's grade. Since the lack of proper on-line instruments, its measurement interval and delay are both very long, which make the quality control quite difficult. Especially during grade transition, a lot of off-specification polypropylene are produced and leads to the loss of profits. This thesis is focused on the application of statistical methods in industrial process especially the propylene polymerization process.(1) Firstly, the knowledge about process modeling and monitoring is introduced. Then the basic concepts of principal component analysis and partial least square regression are studied, as well as their advantages and disadvantages.(2) Propylene polymerization process is highly nonlinear and characterized by multi-scale nature with lots of variables that are highly correlated. The PCA-MSA-RBF model based on radial basis function network approach integrated with principal component analysis and multi-scale analysis is proposed to infer the melt index of propylene. The results have confirmed the validity of the model.(3) Based on the introduction of statistical learning theory and support vector machine, least square support vector machines and weighted least square support vector machines are developed to predict the melt index of polypropylene.(4) Based on the analysis of the advantages and disadvantages of traditional MSPC, independent component analysis is presented as well as its combination with support vector machines classifier. Its application in propylene polymerization process has shown the effectiveness of this method.