Font Size: a A A

Research On Robust Adaptive Fault Detection And Isolation Approaches

Posted on:2012-09-11Degree:DoctorType:Dissertation
Country:ChinaCandidate:X J LiFull Text:PDF
GTID:1228330467981126Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Due to the huger and more complex of modem 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, in full and finite frequency domain, H_and H∞techniques have been used to characterize the fault sensitivity and the disturbance robustness, respectively. Although many effective robust fault diagnosis methods have been developed in this field, there are still some problems to be developed associated with the existing approaches. For example, the known results cannot directly apply to the stochastic systems. The finite frequency fault detection approaches combined with generalized KYP lemma are only valid for a class of linear time-invariant (LTI) systems. For some feedback control systems with small stuck faults, the existing fault detection schemes are invalid. On the other hand, there are also some problems to be solved in the adaptive techniques-based on fault diagnosis approaches. For instance, in order to achieve complete fault detection and isolation, the existing approaches require the number of sensors to be larger than or equal to the system order (or the number of faults), and some signals to satisfy the persistency of excitation conditions. Furthermore, the known schemes cannot detect and isolate the small stuck faults.Corresponding to the aforementioned problems, this thesis, based on previous works of others, presents some new fault diagnosis methods. First, the existing finite frequency fault detection approaches are extended to LTI systems with time-delay. By defining the fault sensitivity performance and disturbance robustness performance from the viewpoint of signals, the fault detection problem of stochastic systems is investigated. Designing the residuals to be sensitive to the reference inputs in the faulty cases, this thesis develops a new fault detection approach to detect the small stuck faults. Also, this thesis provides convex filter design conditions to guarantee the fault sensitivity. As for adaptive fault diagnosis methods, for a class of nonlinear systems, the assumption that the number of sensors should be larger than or equal to the system order (or the number of faults) is removed and, no persistency of excitation condition is required. For a class of linear feedback control systems, by modeling the closed-loop as multiple models and effectively using the reference inputs, the small actuator stuck faults can be detected and isolated.In conclusion, the main innovations of the paper are listed as follows:1. For deterministic delay systems, a new fault detection (FD) approach in finite frequency domain is provided to deal with the inaccurateness of the frequency weights through directly characterizing the frequency ranges of disturbances and faults. The com-parisons with the full frequency approaches illustrate the advantages of our method. On the other hand, for stochastic delay systems, this thesis defines the fault sensitivity and disturbance attenuation performances from the viewpoint of signals, and then provides the LMI conditions for guaranteeing these performances. Especially, if the delay and the stochastic disturbance are neglected, the above LMI conditions can improve the existing results, in other words, a fault detection filter design condition whose conservativeness can be precisely captured domain is established to guarantee the fault sensitivity.2. A fault detection approach is developed for linear feedback control systems with finite frequency reference inputs and small actuator stuck faults. Considered all the pos-sible actuator stuck faults, the closed-loop is modeled via multiple models, that is, faulty models and fault-free model. One of the innovations of this thesis is that the residuals are designed to be sensitive to finite frequency reference inputs for the faulty cases and robust against reference inputs for the fault-free case, this strategy guarantees that the designed residuals have large discrepancies between fault-free model and faulty models even if the actuator stuck values are very small. It should be pointed out that,in finite frequency domain, the convex FD filter design conditions for guaranteeing the fault sen-sitivity and disturbance attenuation performances are provided, and the conservatism of these conditions can be precisely captured.3. Due to the closed-loop may affect the detection performance, some literature considered the problem of simultaneous fault detection and control. However, many ex-isting results required that the detector and controller should be a common unit, which may lead to the conflict between the control performance and detection performance. In order to solve this problem, two different units are designed as detector and controller, respectively. In finite frequency domain, based on a new matrix partition technique, this thesis provides a scheme to simultaneously design detector and controller parameters. The simulation example illustrates that the proposed method can achieve better integrated detection and control performances.4. A fault detection and isolation (FDI) scheme based on adaptive technique is de-veloped for a class of linear feedback control systems. By efficiently using reference inputs and modeling the closed-loop as multiple models, the explicit fault detectability and isolability conditions are derived, which show that the proposed adaptive FDI scheme is not only able to detect and isolate the large faults but performs very well for the small actuator stuck faults including outage cases (the stuck values are zero). On the other hand, in the proposed FDI framework, no persistency of excitation condition is required.5. This dissertation studies the fault detection and isolation problem for a class of multiple-input-single-output (MISO) nonlinear systems with q faults, q+2observers are designed, one is the fault detection observer, another is the adaptive state estimation ob-server, and the other q observers are adaptive fault isolation observers, q+1residuals are generated by using the states of the q+2observers, and the FDI problem for the aforementioned nonlinear system is then solved through designing zero as threshold. In practice, in order to achieve FDI, some existing literature required that the number of sen-sors must be larger than or equal to the system order (or the number of faults). However, this requirement cannot be satisfied in some cases, which shows the significance of our work.Finally, the results of the dissertation are summarized and further research topics are pointed out.
Keywords/Search Tags:Fault detection (FD), fault detection and isolation (FDI), fault detectionfilter, linear systems, nonlinear systems, stochastic systems, reference inputs, actuatorfault, sensor fault, finite frequency, H_∞norm, H_index, adaptive
PDF Full Text Request
Related items