Decision Fusion Based Multi-Classifiers For Fault Detection And Identification And Design Of GUI Platform

In recent years, due to safety and quality requirements of industrial production, process monitoring has become an important research area, and then more effective and practical methods for fault detection and identification are proposed. Whereas, most individual approaches are based on certain assumed condition, which means, one method under some condition with good performance is likely to behave badly in other state. Therefore, we focus on the limitations of a single method, and propose a multiple classifier system integrated by decision fusion strategies which mainly to deal with the situation with linear, non-linear and non-Gaussian process characteristics. This thesis introduces the basic idea, related algorithms, and the corresponding design of GUI simulation platform. These include:(1) In order to improve the diversity of the entire system, by filtering out samples that contain more useful information to improve the diversity of data, the resampling technique is used. Besides, by selecting classifiers which can handle process feature of non-linear, non-Gaussian and linear process, it could obtain good classification performance, as well as to provide a strong diversity for the subsequent decision fusion system.(2) ALL decision fusion system based on Dempster-Shafer theory of evidence for fault detection and identification is proposed; moreover, a new indicator measuring the degree of correlation between classifiers is imported, namely the correlation coefficient index, calculated based on classification performance parameter, which is a measure of the linear correlation of each two classifiers; then on the basis of ALL decision fusion system, a SELECTIVE decision fusion framework is developed, referring to the correlation coefficient index. It is a method to retain classifiers that initially identify fault for integration through filtering of great relevance classifiers. Studies on the Tennessee Eastman Platform demonstrate the superiority of the two new proposed integration frameworks, which reduce the delay time of detection and improve diagnostic accuracy.(3) A Matlab-GUI platform based on the fusion framework in this thesis is developed, which could deal with online monitoring and fault identification, and also has the ability to offer sound or light alarms for fault information. The platform is equal to a man-machine interface integrated on the client host. Besides, it includes model library of fault detection algorithm, monitoring module, pre-processing module, as well as unified management of Access database to historical data and system information. In addition, tests show that the platform not only is a friendly interface but also has strong versatility and scalability.Finally, a summarization of the full text is given and some prospects for further research are discussed.