| The Guangzhou petrochemical plant's monitor efficiency has been improved obviously in recent years, with the application of distributed control system(DCS)and Intemor intelligent accident prevention system on the B district of atmospheric pressure device. The accuracy of fault detection and diagnosis still need to be improved. At the same time, the mass data collected by DCS become valuable resources to be used.In this dissertation, we emphasize on the research of fault detection and diagnosis methods based on principal component analysis (PCA), validate and extend the applications of these methods in the real industrial problems.The major contributions include:1,Analyzed the characteristics of the atmospheric distillation process in Guangzhou petrochemical plant, summed up seven monitoring objects based on priori knowledge.2,Research on PCA method and its application :(1) Proposed to search for exceptional samples based on SPE and build PCA model in a iterative way. (2) Studied the fault detection sensitivity of PCA model when mean-shift took place in a single variable. (3)Verified the feasibility of the method via a real failure case study.3,Aiming at the self-correlation features of industrial processes data, proposed two non-model dynamic PCA methods: Improved the traditional DPCA methods by choosing delay variables and delay length according to the degree of correlation.Combined EWMA with PCA, considering the effect from history data.4,Fault detection in chemical process concerns much about the nonlinear problem,a nonlinear PCA fault detection method based on neural networks was proposed, which combines two radial basis function(RBF) networks with a principal curves algorithm.5,A PCA-SDG fault diagnosis method was proposed. SDG model was used to interpret the residual contributions produced by PCA .The application to a real process illustrated its validity.Finally, summarized the whole dissertation, and presented the further exploration issues. |
PDF Full Text Request