Study Of Fault Detection Based On Tennessee-eastman Process

Multivariate statistical process control(MSPC) has been widely used in the study of chemical process monitoring, and it has been paid more attention by experts and scholars. Due to the chemical process monitoring method is to verify that the proposed algorithm is effective and feasible through the Tennessee Eastman(TE) data, which is from the United States Eastman Chemical Company of the downs and Vogel according to the company a real United chemical reaction process, the development TE benchmark test platform to generate the data. Therefore, the main research object of this paper is TE data, in view of the actual industrial process, put forward the following several effective process monitoring methods.(1) a new fault detection model based on principal component analysis and support vector data description(PCA-SVDD) is established for nonlinear and non Gauss information problems in complex industrial processes. Because the support vector data description(SVDD) model has the advantage of being free from the limitation of linear and Gauss hypothesis, it overcomes the shortcomings of the traditional principal component analysis(PCA) statistical testing method to satisfy the linear and Gauss distribution. First, the principal component analysis is used to decompose the process data, and the score matrix is extracted. Then, the support vector data description(SVDD) algorithm is used to build the statistic based on distance for the score matrix and the corresponding statistical limit is constructed. Finally, the simulation experiment of TE data shows that the algorithm is effective and feasible, and it can improve the detection rate of the fault.(2) data got from chemical process monitoring may contain sparse noises. A fault detection method based on robust principal component decomposition and support vector data description(RPCA-SVDD) is proposed. Due to the low rank matrix data can be recovered from the sparse noise pollution data by the robust principal component decomposition(RPCA), and the support vector data description(SVDD) algorithm can overcome the shortcomings of linear and Gauss’ s hypothesis. The solution of robust principal component analysis is realized by using the accelerated proximal gradient(APG) algorithm. Finally, based on the Tennessee Eastman(TE) simulation results show that the proposed algorithm is feasible. The results also show that the proposed algorithm can effectively improve the monitoring effect.(3) the data generated by the actual industrial process will be subject to large and sparse noise pollution, so the traditional principal component analysis(PCA) to deal with such data is not good. Low rank matrix and sparse decomposition(LRSD) can decompose the nature of reaction in high dimensional data of low rank matrix data, and have advantages not affected by the sparse noise, this chapter put forward fault detection approach based on a low rank matrix and sparse decomposition a nd principal component analysis(LRSD-PCA). First, low rank matrix and sparse decomposition(LRSD) are used to extract the low rank data in high dimension data. Secondly, low rank data is used to establish the principal component analysis model, which is obtained from the low rank matrix and sparse decomposition(LRSD) algorithm. Finally, through a numerical simulation example and TE simulation data, the results show that the algorithm is effective.