The Research Of KPCA And Its External Methods Applied In Processes Monitoring

The object and motivation of industrial process monitoring is to timely and effectively find and restore fault in process, insure the safety of production procedure and consistency of product quality and reduce the productivity loss. Multivariate statistic process control (MSPC) methods are known to be effective for detecting and diagnosing abnormal operating conditions. Among them, PCA is the most popular one, which relates to its conceptual simplicity. However, many industrial processes are fairly more complex with nonlinear characteristic which limit its application. Since kernel principal component analysis (KPCA) first put forward by Scholkopf et al.(1998) is a nonlinear extension of principal component analysis (PCA), and it has already shown better performance than PCA in fault detection and diagnosis, kernel method has been developed rapidly. Especially in recent years, many extended methods were proposed based on KPCA. These methods extend applications of KPCA in the industry process control. This dissertation develops the research based on the predecessor’s work. The main research contents are as follows:1. Although KPCA method has been successful in many practical situations, the measurement data may also have outliers and multi-scale characteristic except nonlinear. When applying KPCA to fault detection, it will lead to false-alarm of the system. To overcome the limitations of KPCA handling the data corrupted with outliers and multi-scale characters. SMF-MSKPCA method is proposed, which is developed by introducing the sliding median filtering and KPCA into the MSPCA method. This method not only utilizes the advantage of sliding median filtering to preprocess the data to eliminate outliers, then reduce false alarm, but also wavelets transform to analysis multi-scale. Then, this method is applied to fault detection, which can not only improve the detecting ability for small but important changes in data, but also resolve the false-alarms problem caused by the outliers, which is difficult of being dealt with by the existing MSPCA, when measured data contain outliers. The simulation result of the process monitoring shows this method can not only reduce the false-alarms, but also improve the effect of the fault detection. 2. Based on the further research on data statistical characteristics of industrial processes, this dissertation focuses on the dynamics of industrial processes, the dynamic kernel technique (DKPCA) is introduced into the SMF-MSKPCA method and a SMF-MSDKPCA-based statistical performance monitoring and controlling algorithm is presented. Applications on a simple dynamic nonlinear process and the Tennessee process demonstrate that the proposed approach can effectively capture dynamics and nonlinearities of industrial process which proves the method is effective and feasible.Finally, there are concluded with a summary and some further research areas in this thesis.