Research Based On Available Measurements For Nonlinear Process Monitoring

Process monitoring is an important research topic to ensure system reliability and security. With the increasing complexity and intelligence of industrial systems, it is difficult to obtain mathematical models and advanced knowledge of the systems. Therefore, it is difficult for model-based as well as knowledge-based approaches to effectively promote and use in such systems. It is worth noting that these industrial systems tend to produce massive historical and real-time data. How to effectively take advantage of these data to monitor the system becomes the focus of attention of scholars and also contributes to the research of data-driven process monitoring. However, most of the approaches are aimed at linear systems up to now. Based on this fact, in allusion to nonlinear static system, modified principal component analysis is extended to nonlinear systems, which provides a novel approach to cope with this issue.Firstly, the most effective method is chosen according to the comparison among several modified principal component analysis(PCA) based approaches. Meanwhile, real data from Tennessee Eastman process is used to verify theoretical analysis in simulation study, which provides the necessary conditions for the subsequent proposed algorithm.Then, the application of modified PCA is extended to nonlinear systems in this paper. Modified PCA is applied to the local linear models under LWPR framework and the global statistic is calculated by a weighted average of local statistics. Besides, two threshold calculation methods are proposed based on the different distributions of the variables. Meanwhile, comparisons between the proposed algorithm and other nonlinear algorithm are made, which indicate the superiority of proposed algorithm in terms of computational complexity, learning speed, robustness and so on. Moreover, nonlinear numerical example simulation results show effectiveness of the proposed algorithm.Finally, the necessity of using data-driven process monitoring is briefly explained for vehicle suspension system in this paper. Besides, the proposed algorithm is applied to the actual industrial system. The experimental results show the superiority of the proposed approach.