Research On Learning Observer-based Fault Reconstruction For Satellite Attitude Control Systems

An Attitude Control System(ACS), as an indispensable subsystem in a satellite, is prone to potential faults especially in actuators and sensors during its long operational life in the extremely harsh outer space environment. Faults in ACSs of on-orbit satellites may dramatically degrade control performance and cause closed-loop ACSs unstable, and even shorten the mission lifetime or result in tumbling out of control. To guarantee high reliability and survivability of satellites, hardware redundancy and simple Fault Detection and Isolation(FDI) functions are provided. However, to achieve autonomous fault diagnosis as well as fast and accurate fault-tolerant control for satellites, it is necessary to obtain fault magnitude and trends for satellite ACSs. Currently, only a few research achievements have been reported on fault reconstruction for satellite ACSs. On the other hand, although much progress has been made on Learning Observer(LO)-based fault reconstruction approaches since 2001, there are still many problems on this research topic to be solved, such as robustness issue, sensor fault reconstruction and systematic LO design.In view of these, this dissertation investigates LO-based fault reconstruction approaches with their application to microsatellite ACSs. The main contributions of this dissertation are outlined in the following aspects.P-type LO-based fault reconstruction issue for microsatellite ACSs subjects to loss of reaction wheel e?ectiveness is investigated. The existing P-type LO-based multiplicative fault reconstruction approaches have not provided a systematic design for P-type LOs, and must satisfy the condition that control inputs are bounded. In view of these, an improved P-type LO-based approach for reconstructing actuator e?ectiveness factors is proposed, and it not only provides a systematic P-type LO design based on Linear Matrix Inequality(LMI) technique, but also requires no actuator constraint condition. Further,based on H∞ technique, robust improved P-type LO is designed for attenuating the influence of disturbance inputs on fault reconstruction performance. Simulation studies on microsatellite ACSs with loss of reaction wheel e?ectiveness are carried out, and simulation results clearly demonstrate the e?ectiveness of the proposed fault reconstruction approach.For gyroscope fault reconstruction in microsatellite ACSs, we propose two kinds of augmented LO: augmented P-type LO and augmented PD-type LO. By using a suitable output filter, dynamic system with sensor faults can be transformed into an augmentedsystem with an actuator fault form. For this augmented model, an augmented P-type LO is designed to reconstruct sensor faults. In the existing P-type LOs, the assumption on time-varying fault constrains has no specific physical meaning such that it is di?cult to determine time-varying characteristics of sensor faults. Further, to achieve fast sensor fault reconstruction, by introducing di?erential term of measurement output estimation errors into P-type learning algorithm, a novel augmented PD-type LO is proposed to accurately reconstruct sensor faults. Based on LMI technique, a systematic PD-type LO design method is explicitly provided. Utilizing H∞ technique, robust augmented PD-type LO is designed to attenuate the influence of disturbance inputs on sensor fault reconstruction.The above two augmented LO-based approaches are applied for reconstructing gyroscope faults in microsatellite ACSs, simulation studies are provided to verify the e?ectiveness of the proposed two sensor fault reconstruction approaches, and simulation comparison between them are also given to demonstrate their advantages and disadvantages.Continuous-time Learning Unknown Input Observer(LUIO)-based robust thruster fault reconstruction in microsatellite ACSs is studied. First, UIO-based robust actuator FDI approach is proposed, and robust UIO design and robust FDI conditions are investigated. Based on the existing P-type LO designs, we propose a P-type LO-based additive actuator fault reconstruction approach. Further, based on the UIO and the P-type LO, a novel LUIO is proposed to robustly reconstruct actuator faults. Considering the decoupling between actuator faults and disturbance inputs, stability conditions and existence conditions of the proposed LUIO are explicitly discussed. Considering the coupling between them and existence of measurement noises in the system, the LUIO gain matrices are designed such that the LUIO is partially decoupled from disturbance inputs and has robustness against the remaining part of disturbance inputs and measurement noises.Simulation studies on microsatellite ACSs with thruster faults are performed to verify the e?ectiveness of the proposed two fault reconstruction approaches, and simulation comparison between the P-type LO and the LUIO is also provided in detail.Discrete-time LUIO-based robust thruster fault reconstruction in microsatellite ACSs is investigated. For the Euler-approximate model of continuous-time dynamic system,a Euler-approximate UIO is designed and its robust stability analysis is provided. Based on the discrete-time UIO and the continuous-time LUIO, a Euler-approximate LUIO is proposed for robust actuator fault reconstruction. Considering the coupling between actuator faults and disturbance inputs, as well as existence of measurement noises in the system, based on the disturbance-decoupling ability of the UIO, some LUIO matrices aredesigned such that the LUIO is partially decoupled from disturbance inputs; further, H∞technique is used to design other LUIO matrices such that the LUIO has robustness against the remaining part of disturbance inputs and measurement noises. To guarantee the implementation of the Euler-approximate LUIO on the exact model, semiglobal practical convergence property of the proposed LUIO is discussed in detail. The proposed discrete-time LUIO-based approach is applied for robust fault reconstruction of thrusters in microsatellite ACSs, and simulation results are provided to demonstrate the e?ectiveness of the proposed approach.