Researches On Performance Monitoring For Process Industry Based On ICA And Multi-parameter

In process industry, such features as large-scale, high-complexity and multi-variable-coupling increase the possibility of fault. In order to achieve stable, persistent and high-quality products, it is necessary to detect and diagnose faults in the equipment at early stage. In view of the characteristics of process industry, the independent components (ICs) extracted through the independent component analysis (ICA) are utilized to study the performance monitoring in Tennessee-Eastman process.(1)In view of the characteristics that the sample data in process industry may not follow normal distribution, the ICA-based method is used to monitor the system performances without assumption that data follow normal distribution. Furthermore, these ICs span the characteristic space of normal operating condition, and the changes of coefficients matrix, which are obtained by projection of process variables onto the space of normal operating condition, are monitored. The method is applied to the Tennessee-Eastman process (TE process). The result shows that the ICA-based method is feasible and can detect the faults early.(2)The ICs are in the same important position. To the industry process, it is generally impossible to get the number of the ICs exactly. At present there isn't a good method to get the number of ICs. The principal component analysis (PCA) is used for dimensionality reduction in this article. Then the ICA is used to analyze the data after dimension reduction. The data reconstruction to Tennessee-Eastman process shows that the ICs obtained through the method reserve most information of original data.(3)The sample data in process industry are auto-correlative. The current data and the previous data are not independent, and they are time series in effect. Considering the dynamic behavior of the data, the dynamic independent component analysis (DICA) is used to monitor the system performance. This method can resolve the static modeling problem of traditional ICA. The DICA-based method is used to Tennessee-Eastman process, and satisfying results are obtained.(4)A fault detection method based on multi-parameter is discussed. The characters are extracted from the time series data, and those characters construct the state matrix. The running status or fault style can be identified through the distances of the state matrixes. The simulation of Tennessee-Eastman shows that the state matrixes have a better differentiation to different running status.In the last chapter, the conclusions are given and some prospective research fields about this paper are proposed.