Research On Observer-based Robust Fault Detection Approaches

Due to the huger and more complex of modern control systems, there is an increasing demand for higher safety and more reliable standards. Fault diagnosis technology is an important approach to improve the safety and reliability for dynamic systems.Recently, H_and H∞techniques have been introduced to characterize the fault sen-sitivity and the disturbance robustness problems, respectively. Many approaches have been presented, however, there are still some problems to be developed associated with the existing approaches. For example, the existing approaches did not consider the effect of time-delay to fault detection (FD) filter for linear time-delay systems. On the other hand, there are also some problems, which arise from the robust fault diagnosis methods for linear systems with affine uncertainties, to be solved. For instance, only the FD filter with fixed gains is considered in the existing literatures. However, if the range of the un-certainties are large, the existing techniques which are concerned with designing FD filter with fixed gains will be very conservative. In addition, it is difficult to design a nonlinear observer for FD of a class of nonlinear system which contains parameter uncertainties, stochastic disturbance, time delays, as well as Lipschitz-like nonlinearities, especially the model cannot be linearized exactly. The problem of FD for networked control systems has received more and more attentions. For in the same position with controller to FD unit, it is impossible to receive messages of packet dropout between sensor-to-controller and controller-to-actuator. Saturation phenomenon also makes the problem more complex.Corresponding to the aforementioned problems, this thesis, based on previous works of others, presents some new fault diagnosis methods. First, the existing finite frequency H_/H∞FD approaches are extended to linear time-delay systems, in addition, the delay dependent fault detection filter design conditions are given to reduce the conservatism of the existing FD approaches. For the linear systems with time-varying affine uncertainties, this thesis designs the FD filter with varying gains through introducing the switching mechanism, and such a scheme can reduce the conservativeness inherent in the traditional FD filter with fixed gains. A high-gain nonlinear observer is presented based on the approach of robust L2filtering for nonlinear stochastic systems with time delays. For networked control systems with packet dropout subject to sensor saturation, a FD filter is designed to make sure the robust FD system is uniformly bounded mean-square stable and satisfies the prescribed H∞performance constraint.In conclusion, the main innovations of the paper are listed as follows:1. For linear time-delay systems, a new finite frequency FD approach is developed to deal with the inaccurateness of the frequency weights through directly characterizing the frequency ranges of disturbances and faults. Moreover, the delay dependent FD filter design conditions in finite frequency domain are given to reduce the conservatism inher-ent in the existing FD approaches with delay independent design conditions. Finally, an example is given to illustrate the effectiveness of the proposed FD approach.2. For the linear stochastic systems with time-varying affine uncertainties, FD filter design conditions are given based on the switching techniques. Compared with the exist-ing approaches, one of the innovations of this thesis is to design the FD filter with varying gains based on switching techniques and the lower and upper bounds of time-varying un-certainties. In fact, such a scheme can reduce the conservatism inherent in the traditional FD filter with fixed gains.3. For nonlinear stochastic systems with time delays, a high-gain nonlinear observer is designed based on the approach of robust L2FD filtering. The system under consid-eration contains parameter uncertainties, stochastic disturbances, time delays, as well as Lipschitz-like nonlinearities. The problem addressed is to design a nonlinear filter such that, for all admissible uncertainties, nonlinearities and time delays, the dynamics of the filtering error is to be exponentially mean square stable and a prescribed L2robustness. By using the Lyapunov stability theory and stochastic analysis techniques, the design of the proposed filter does not necessitate the resolution of any dynamics systems and its expression is explicitly given, i.e., its calibration is achieved through choosing a single parameter, and does not rely on solving any matrixes inequalities by numerical computa-tion. A simulation example is given to illustrate the performance of the proposed filter.4. Fault detection for networked control systems with packet dropout subject to sen-sor saturation is discussed with bounded disturbances. The packet dropout is assumed to be existed in both sensor-to-controller and controller-to-actuator links. Assume two inde-pendent Bernoulli distributed white sequences are used to describe the packet dropouts, a fault detection filter is designed to decrease the effects of disturbances and fault to eval-uate error. Based on linear matrix inequalities(LMIs) techniques, a sufficient condition is derived such that the robust fault detection system is uniformly bounded mean-square stable and satisfies the prescribed H∞fault detection performance constraint.Finally, the results of the dissertation are summarized and further research topics are pointed out.