Based On Principal Component Analysis, Fault Detection And Diagnosis

Along with the development of modern production and of science and technology, industrial process becomes more and more complex, amounts of investment becomes larger and larger, automatic level has been higher and higher. Therefore, the security and reliability of Industrial process is more important. Multivariate statistical fault detection and diagnosis technology based on the principal component analysis is one of hot research of process automation.In this dissertation, the fault diagnosis methods, multivariate statistical methods and TE process model were introduced firstly, the fault detection approach based on principal component analysis (PCA) and the fault diagnosis approach based on contribution plot are deeply considered. For some existing shortages of the fault detection and diagnosis approach based on the principal component, PCA is applied with wavelet transform, robust and neural network. Respectively, the fault detection and diagnosis approach of PCA based on wavelet transform, PCA based on robust, and PCA based on neural network is proposed. Appling dynamic principal component analysis (DPCA) with the fault detection and diagnosis approach, the fault detection and diagnosis based on DPCA is developed. And the simulation researches and the result analysis have been done and discussed by Tennessee-Eastman (TE) model. In this dissertation, the main achievement of research can be summarized as follows:1,The basic principle of PCA and the Tennessee-Eastman (TE) model are studied, and the process algorithm of fault detection based on PCA is presented. By the simulating researches by TE model, it has noted that the changes of statistics SPE and T2 have been used to determine whether fault happened and the contributions to statistics of each variable are used to identify the fault source for fault detection and diagnosis.2,For process data containing noises, the wavelet denoising algorithm is presented by studying the basic theory of wavelets transform. Integrated with the fault detection approach based on PCA, the fault detection method of PCA based on the wavelet denoising is investigated. The simulation results by TE model show that the proposed method can effectively reduce the numbers of the principal component, reduce the rate of false positives, and improve the fault detection and diagnosis effect.3,For conventional PCA demands the noise of the modeling data is normal distribution, the fault detection and diagnosis method of PCA based on the robust is investigated. This method uses simple weighted variance-covariance estimate to instead of traditional covariance to build the principal component model, the SPE and T2 statistics is used to detect default, and the contributions to statistics of each variable are used to identify the fault source for fault detection and diagnosis. The simulation results by TE model show that the proposed method is more effective and feasible than the conventional PCA.4,Conventional PCA can not monitor the dynamic multiple process effectively, so the fault detection based on the dynamic principal component analysis (DPCA) is proposed. The dynamic principal component model based on the measuring data is built, and the SPE and T2 statistics is used to detect default. The simulation results by TE model show that considering the timing correlation is good at the fault detection based on principal component analysis.5,For the nonlinear limitations of conventional PCA, the fault detection method of nonlinear PCA based on Radial Basis Function (RBF) neural networks is investigated. This approach combines RBF neural networks with nonlinear PCA, gains the nonlinear PCA formula as the PCA formula, and builds the principal component model by using RBF neural network to gain the nonlinear principal loadings, and the SPE and T2 statistics is used to detect default. At last, this method is applied to fault detection by a third-order nonlinear system and has a good effect which proves the method is more effective and feasible than conventional PCA.