Research On Fault Diagnosis Of Nonlinear Process Based On Data-driven

Modern industrial processes become more and more huge, automatic and complex,which has been requring higher reliability in all aspects. If one part has fault, the wholesystem may collapse. Therefore the fault diagnosis for industrial process becomes moreand more important. Meanwhile, the method based on data-driven is the most widelyused in fault diagnosis field, which uses collected data continuously detect changes andfault information during operation. The method based on data-driven does not requirecomplex mathematical models and accurate prior knowledge. Conventional PCAmethod for fault detection and diagnosis has a higher accuracy due to most industrialprocesses with nonlinear characteristics. As for the problem of fault detection anddiagnosis associated with nonlinear, the main puprose of the theisis is to make study onfault diagnosis method based on data-driven. And they are applied to detect anddiagnose the faults in TE processes and three tank water level control system based onthe actual data respectively. The main contents are as follows:Firstly, this thesis introduces research situation and development status of faultdiagnosis. The TE process and the three tank water level control system is introduced. Adetailed analysis of their data features and composition of the system is given.Secondly, this thesis describes the issues that need to be addressed by datapreprocessing, which are crucial for the establishment of the process monitoring modeland monitoring effect in the actual industrial process. The theoretical knowledges of theprincipal component analysis and kernel principal component analysis have beendescribed in detail, and based on above-mentioned TE process and the three tank waterlevel control system, There is a comparison of fault detection result in PCA method andKPCA method, which verifies fault detection in KPCA for nonlinear processes is moreefficient and accurate.Third, as for the fast-response problem and blindness of kernel parameter selectionin KPCA, the effective improving methods are raised. Meanwhile, they will be appliedto the two above-mentioned processes. Compare to traditional KPCA methods, theseimproved methods are faster, and its false negative rate is lower.Fourth, aim at deficiencies of fault identification in KPCA, this thesis presents twofault identification methods, and based on TE process and three tank water level control system, these two methods were verified. Simulation results illustrate theireffectiveness.Finally, in order to the multivariate statistical method is more easily applied toonline application, this thesis established a process monitoring system applicationplatform. In the environment of VB6.0, process monitoring system is designed for TEprocess and the three tank water level control system.