Research On Several Methods Of Fault Diagnosis For Nonlinear Systems And Their Applications

Modern systems are becoming more and more complex and sophisticated in their demand for performance, reliability and increasing autonomy. It is inevitable for sensors, actuators, and impotents inside the system that the fault occurs. Fault diagnosis and tolerant control technology is an important approach to improve the safety and reliability for dynamic systems. Research on fault diagnosis and tolerant control strategy has both theoretical and practical importance. However, the existence of the nonlinearity in the practical plant and uncertainty and noise of the plant models make it more and more difficult. At present, it has drawn wide attention, and has been one of the main topics in the control domain. In the thesis, according to the problems existed in the field of nonlinear fault diagnosis and the development trend of this subject, the fault diagnosis approaches for nonlinear systems and its applications are studied, and the main research results are given as follows:Robust fault detection approaches for a class of time-delay systems with nonlinear perturbations are studied. The design procedure of the fault detection filter is based on two methods: the H∞/ H? filter based approach and the reference model approach. By analyzing the characteristics of the system, design a fault detection filter, and then based on the robust control theory, the problem of designing the gain matrix of the fault detection filter can be solved by using the system's robust stability analysis method. The existence and calculating methods of the gain matrix of the fault detection filter are also given in terms of LMI equality. The fault detection methods mentioned in this chapter,takes into account the sensitivity to system faults and robustness against system uncertainty simultaneously.Fault detection approach for a class of nonlinear time-delay systems with uncertainty is studied. First of all, according to the system equation, a Backstepping observer is constructed. For the purpose of fault detection, a general residual system is constructed by the system equation and residual system. Then, based on the Backstepping methods, a Lyapunov function is used to prove the stability of the Backstepping observer, and the solvable conditions of the gain matrix of the Backstepping observer are given at the same time.A novel fault tracking approximator (FTA) is proposed for fault diagnosis based on the predictive control and iterative learning control theory. Based on the predictive control theory, choose an optimization time span and adjust the virtual fault by using iterative learning algorithm according to the errors of the outputs of fault tracking approximator and the actual system outputs, until the errors meet the requirements. At this time, the virtual faults can approach the real system faults to diagnose the system faults. The convergence of this algorithm and the analysis of the tracking characteristics of the FTA are given in this paper. The results are given as follows: the initial conditions of the iterative learning algorithm do not have an effect on the accuracy of the fault tracking in the time axis. The system uncertainty, including the modelling errors and the noises, will bring fault tracking error. If the system uncertainty can be erased, the FTA can track the system faults。The fault tracking approximator can not only be used in linear systems, but in nonlinear systems; not only in general systems, but in uncertainty systems, and can be widely applied in real systems.The extended states observer (ESO) of the active disturbance rejection controller is used for fault diagnosis. According to the theory of ESO, the system faults and uncertainty are viewed as an extended system state. An fault diagnosis observer is constructed for the purpose of fault diagnosis and the estimation of the system states. Under the assumption that the uncertainty is bounded in terms of norm, the system faults can be detected by selecting appropriate threshold. Compared with the neural-network based fault diagnosis approach, the approach proposed in this chapter can detect and estimate the system fault in real time, which improve the efficiency of the fault diagnosis. Moreover, the approach is applied to Vander Pol oscillator system and robot arm system.