Model-based Fault Detection And Fault-tolerant Control Method

With complexity of modern control systems increasing and the requirement of higher performance of the system, reliability, maintainability and fault-tolerant capacity have attracted more and more attention of the world. For two decade, fault detection and tolerant control have made great improvement, and many remarkable results have been obtained.Based on the system model, different fault detection methods are presented for different characteristic of the linear and non-linear systems. Several fault tolerant control method are proposed for linear and nonlinear systems respectively.Based on the analysis of the system architecture and working principle of the linear system, a novel parity space approach is given for fault detection. This method is proposed for linear time-varying systems by using the fault residual, which is obtained from the measured state and output of the system. The filter is not only used to monitor the system for abnormal behavior, but also can track the fault and predict development trend of the fault. Based on the neural network observer, we expend the parity space method to nonlinear Wiener systems, with the combination of the neural network and Kalman filter, the fault detection and the fault magnitude estimation can be obtained. Considering the consideration of multiple faults occurrence simultaneously, Parity space approach is introduced to design the filter matrix which used to filter the nuisance faults and detect the target fault.Nowadays, many researchers have paid much attention to fault tolerant control. Considering the parameters disturbance fault, and the periods of sampling and the uncertain matrix are not satisfied the matrix decomposing condition, the partly guaranteed cost target is presented by confirming the state space, so the whole performance is obtained target for time-delay linear systems.Finally, we propose the fault tolerant control methods for nonlinear systems, fuzzy multitude reference model control (FMRM), RBF neural network control (RBFNN) and fuzzy-neural network control (FNN) was studied. The method (RMRM) uses the fuzzy logic rules to choose the suitable reference model, and it is applied to control the speed servo system of dynamic model of BLDCM, the simulation results show it works well with high dynamic performance under the condition of great change in reference speed and variation of parameters. RBFNN method combines model reference adaptive control (MRAC) and RBF adaptive neural network which is proposed for permanent magnet synchronous motor (PMSM). The effect of the unknown uncertainties and disturbances is overcome by online adaptive tuning network parameters of RBFNN which can implement the mapping between speed error and the control current. It is proved that the presented scheme can guarantee the stability of the designed system with parameters faults happened. A fault detection and accommodation methods based on fuzzy neural networks for non-linear systems is presented. The fault parameters adaptive updating method is using to detect malfunction, fuzzy neural networks is suggested to adjust the fault parameters for tracking the faults. Fault compensation control input is introduced for fault accommodation. The theory analysis and simulation results show that when the parameters faults happen, the systems can still preserve certain performance by fault accommodation control, and the applicability and validity of RBF fuzzy neural networks in fault time-varying system accommodation control is proved.