Study Of Multivariable Projection Algorithms Based On ICA On The Process Monitoring

In modern industrial process, detecting, diagnosing and restoring fault timely and efficiently is the precondition of providing products with good performance and consistent quality, which is also the motivation and object of process monitoring. Multivariate statistic process control (MSPC) has developed more than thirty years. Lots of research results have been acquired and applied in this field widely. However, Conventional MSPC is based upon the assumption that the separated latent variables must be subjected to normal probability distribution (or independent and identical distribution), which sometimes can not be satisfied. The aim of this dissertation is to overcome these two assumptions, and to improve the monitoring performance of process and enlarge the range of the application by independent component analysis (ICA).ICA is the primary mathematical tool used in this dissertation. Its principle is to find mutual independent underlying components, to remove the higher-order redundancy between components and to extract the independent original signals by analyzing the higher-order statistical relationships among the multidimensional observations. So ICA is more effective than PCA (Principal Component Analysis) in describing process. The main content of this dissertation is as follows:1. Introductions of the elementary concepts and scope of process monitoring, and simple description of the applications in process monitoring with PCA and ICA.2. Considering process information isn't always subjected to multivariate normal distribution, a new process monitoring method is provided here, which is based on the idea that the process information is driven by a few of components as independent as possible. The simulation results verify its effectiveness.3. In order to reduce the influence of noises, an improved process monitoring method is present, which integrates the multi-scale analysis of wavelet to the unique ability of ICA for non-Gaussian process data. The research results verify that it can improve the monitoring performance efficiently.4. In some real industrial processes, the relations of variables are supposed linear. Sometimes the assumption can lead to some incorrect results. Therefore, a process monitoring method based on nonlinear ICA is applied in the TE process. Some positive results are also given.Finally, some beneficial explorations in the field of process monitoring are made, and some future research areas are highlighted.