Modern industrial processes have many characteristics, such as large-scale, continuous, intelligent, over load, high-speed. Therefore, it will result in a great loss of personnel and property once there has faults occur. Obviously, it has great significance and practical value if we take the fault diagnosis method as a research topic. This thesis regard the Tennessee-Eastman Process (TEP) as the background because the characteristics of industrial processes. Then it applies the Kernel independent component analysis (KICA) and Fisher discerns analysis (FDA) to achieve the process fault detection and diagnosis. The main contents of this thesis are as follows:(1) The concepts, tasks and processes of fault diagnosis method are introduced in details in this thesis. It describes the research status of fault diagnosis method, and researches on its classification from the perspective of qualitative and quantitative analysis.(2) It studies of independent component analysis (ICA) and the kernel independent component analysis (KICA) theory and its application of multivariate statistical fault detection method in the TE. After that, using the two detection methods in the TE process, and it obtains that multivariate statistical fault detection method based on the KICA is superior to the ICA.(3) The KICA fault detection method is improved in this thesis. It combined feature vector extraction (FVS) and KICA, and then proposed a new KICA fault detection method which is based on FVS. In this method, at first, looking for a subset by FVS. This subset of the mapping in the feature space F sufficient to demonstrate the mapping of all sample data. When the number of samples is large, it could reduce the calculation amount of KICA and shorten the calculating time. Finally, verified the effectiveness of the proposed method by researching on the typical fault of TE process simulation.(4) The FDA fault diagnosis method is improved in this thesis.Put forward a kind of fault diagnosis methods based on KICA-FDA. At the beginning, the method extracts statistically independent key variables by KICA, and then uses the extracted key variables to establish the FDA fault diagnosis model. Clearly, the simulation results show that fault diagnosis methods based on KICA-FDA has a better performance than the other fault diagnosis methods based on FDA in the application of the method in TE process of typical fault. |