Sdudies On A Multi-Mode PCA Based Process Monitoring Method

With the development of multivariate statistical process control theory and technology, principal component analysis (PCA) method has been widely applied in practice. However, when traditional PCA method is used to the process with shifted modes, the statistical model will consider the normal mode change as the failure, which may cause error alarm. Multi-mode PCA establishes monitoring model of process data for each phase according to the characteristics of process patterns.Based on the researches of related technology and methods, a type of multi-mode PCA statistical monitoring method incorporated with dynamic trend analysis is proposed in this paper. Two problems followed are solved. Firstly, the distinction between normal mode change and failure is identified to prevent false alarm, and secondly, the transition phase is monitored to detect abnormal trend and amplitude of each variable. Main contributions are listed as follows: 1.Process control method based on multi-mode PCA is investigated. A multi-mode PCA control method with transition phase is proposed. Methods of division for multi-mode process, modeling and on-line monitor are described in detail.2.A process control method to monitor the dynamic trend of multi-variables in transition phase is proposed based on research of the dynamic extraction and match method. Through the extraction of process variables' standard trend in transition phase, characteristic model describing trend shape and amplitude is achieved for on-line trend matching.A new monitoring indicator for trend amplitude that can reflect more comprehensive trend amplitude characteristics is proposed.3.A type of multi-mode PCA statistical monitoring method incorporated with dynamic trend analysis is proposed. A dynamic trend analysis technique is used to identify the trends of the variables shifted obviously during the transitions phase, while the multi-model PCA is employed to monitor the variables invulnerable to the changes of the operating modes.With the method proposed, not only the difference between the operating mode change and process fault is distinguished, but also the transition phase of the process is monitored more accurately.4.An application on the Tennessee-Eastman process simulation platform is carried out to demonstrate the feasibility and effectiveness of this contribution.A multi-mode PCA monitoring system is developed based on VB software.The system can be applied in real industrial process to establish the statistical model, and implement the single-variable, multi-variable and multi-mode process monitoring.