Fault Diagnosis Of Networked Control System Based On Support Vector Machine

The thesis mainly focuses on Model-based fault diagnosis with the application to networked control system, and support vector machine (SVM) is used to consummate the fault diagnosis method.With the improvement in stability and safety of the networked control system as well as its large scale, it is imperative to have a credible fault diagnosis method for the system. Though much progress has been made, problems still exist in current fault diagnosis of networked control system, chiefly represent in two facets: (1) Fault detection: Networked control systems can be considered as the synthesis of complex system and remote control system. The influence of environment, disturbances, nonlinear and uncertain dynamics on it is much more serious than the traditional ones, thus a new challenge raises up for researchers. (2) Fault identification: Networked control system exploits artificial neural network to identify the fault, just as the traditional system does, yet even some theoretical deficiencies still exist in the neural network. So how to solve this problem or adopt other identification methods as substitute is on edge for the researchers.As regards fault detection, SVM is examined in detail at first. Due to the dependence of SVM on parameter selection, and the deficiency in intelligence, a parameter selection approach based on swarm particles optimizer is proposed. Then, SVM is used to identify the uncertain term mentioned above, and the result, as the compensation, is added to the state observer, which can reduce the influence of environment, disturbances and/or uncertain dynamics on system residual, and the residual is only related to fault in ideal condition. The simulation result represents that SVM has a perfect performance in identification while too many SVs decrease real-time application. So at last, relevant vector machine is considered and used in identifying the uncertain term, and the comparison between RVM's result and SVM's tells that RVM is more suitable in off-line identificaion than SVM.As for fault identification, SVM is used to identify the fault instead of neural netework. Active set method used to solve the problem of quadratic optimization is introduced firstly, which is applied in incremental SVM in further step. So when new...