A Study Of Nonlinear Process Monitoring Method Based On Data

Process monitoring is an important measure which can guarantee the safety and stability of the production. Nonlinear process monitoring method based on data is a vitally important topic in the field of automation in recent years. This thesis mainly focuses on several methods which are based on nonlinear process monitoring and studies the latest kernel method, namely kernel entropy component analysis (KECA), which is on the basis of kernel principal component analysis (KPCA). What is more, this thesis studies a new method of nonlinear process monitoring based on the angular structure of KECA after exploring its angle structure characteristics. This thesis is on the background of Tennessee Eastman (TE) process simulation. The main contents of this thesis are as follows:(1) This thesis studies the development and research status of the nonlinear process monitoring method based on data. And then, it analyzes of the advantages and disadvantages of all kinds of nonlinear methods in process monitoring. Kernel principal component analysis (KPCA) is taken as a typical nonlinear method in process monitoring. By analyzing the theory of principal component analysis (PCA) as well as comparisons and simulations, the feasibility and practicability of KPCA method for nonlinear process monitoring is verified.(2) This thesis also studies the method of kernel entropy component analysis (KECA) for nonlinear process monitoring. After exploring the characteristics of the KECA method and the differences from the KPCA method, and doing some contrastive simulations, KECA method is verified that it has advantages compared with KPCA method when applied to nonlinear process monitoring. The simulation shows that KECA method is better than KPCA method in the monitoring indexes such as the false alarm rate, missing alarm rate and fault sensitivity.(3) The thesis finally studies the angular structure of KECA, and explores the method for nonlinear process monitoring based on the angular structure of KECA. Compared with the previous methods which use the SPE and T2 statistics in process monitoring, the thesis proposes the process monitoring based on the angular structure of KECA would effectively improve the effectiveness of non-gaussian process monitoring, to some extent, it could avoid the limitation of the data being gaussian distribution. Thus it embodies the universality of better in application. By simulation analysis, the angular structure method gets the simulation result with 12.15% missing alarm rate and 8.63% false alarm rate, which is better than the result of previous SPE and T2 method. What is more, the KECA angle structure reflects the obvious superiority in fault monitoring indexes such as false alarm rate, missing alarm rate and fault sensitivity. The method based on the angular structure of KECA is verified to be feasible and practical.