Model-based Actuator Fault Diagnosis And Its Application To Satellite Attitude Control Systems

Since a satellite works in a very harsh and complicated space environment, it isunavoidable to appear all kinds of faults in satellite systems. In order to enhance its re-liability and reduce its dependence on ground systems, a satellite must have the abilityof autonomous fault diagnosis and fault-tolerant recovery. The safety and reliability of asatellite attitude control system, as one of the most critical subsystems of a satellite, is apremise to guarantee the normal operation of the satellite. With the purpose of solve faultdiagnosis problem for satellite attitude control systems, in this dissertation a mathematicalmodel of satellite attitude control systems is established based on the analysis of dynamicand kinematic characteristics of satellite attitude systems. Starting with the point of inte-grating three tasks of fault diagnosis, an integrated actuator fault detection, isolation andidentification problem for satellite attitude control systems has been investigated.An adaptive Luenberger unknown input observer-based robust fault diagnosis schemeis proposed for linearized satellite attitude control systems with actuator faults. The pro-posed fault diagnosis scheme is performed with a Luenberger unknown input detectionobserver and a bank of adaptive Luenberger unknown input isolation observers. Eachisolation observer corresponds to a special possible actuator fault. One isolation observermatching with real fault information will generate zero residual; while, the other isola-tion observers mismatching with real fault information will generate nonzero residuals.The size of actuator fault will be obtained by online adaptive tuning. Besides, the coef-ficient matrices in the designed fault detection observer and fault isolation observers aredetermined based on parametric approaches.A robust adaptive unknown input observer-based fault diagnosis scheme is proposedfor a natural vector second-order linearized satellite attitude control system with multipleactuator faults. The proposed fault diagnosis scheme is performed with a bank of faultdetection and isolation sets which are composed by reorganizing the controllers. Designunknown input observers for each set, and the residual generated by each observer isrobust to the actuator fault in the matching set, and sensitive to the actuator fault out ofthe matching set. The fault estimation observer is activated by the fault detection andisolation result to identify the fault size on-line. Besides, this fault diagnosis schemecould diagnosis multiple faults, and overcome some disadvantages which is produced by augmenting the natural vector second-order linearized satellite attitude control system intoa first-order system, such as, the algebraic structure of vector second-order systems is lost;the state estimation generated by the observer cannot continue maintaining the originalphysical information; the unknown disturbances decoupling condition is not satisfied.An adaptive nonlinear unknown input observer-based fault diagnosis scheme is pro-posed for nonlinear satellite attitude control systems with actuator faults. A nonlinear un-known input detection observer is designed to make a part of the unknown inputs decou-ple with the detection residual; while, make the efect of the remaining part of unknowninputs, model uncertainties and output noises to the fault diagnosis system as small aspossible. Fault isolation is still performed with a bank of adaptive nonlinear unknowninput observers. Similarly, each isolation observer corresponds to a special possible ac-tuator fault. Besides, each isolation observer corresponds to an adaptive threshold. Theresidual signal generated by the fault isolation observer matching with the real actuatorfault will be under the given adaptive threshold; While, the residual signals generated bythe fault isolation observers mismatching with the real actuator fault will be above thegiven adaptive threshold. With this diference, actuator fault can be isolated. A switchingball method is applied to get the online adaptive fault estimation.A hierarchical interactive and a variable structure multiple model fault diagnosisschemes are proposed to solve the multiple faults problems. The proposed fault diagnosisschemes are performed with a bank of unscented Kalman filters and statistical methods.When dealing with lock-in-place actuator faults or loss-of-efectiveness actuator fault, inorder to avoid the performance degradation of fault diagnosis systems caused by the largenumbers of elements in the multiple model set, the nonlinear satellite attitude system willbe augmented by a fault parameter which not only estimate both the system sates andfault parameter, but also reduce the size of the multiple model set. For the multiple faults,interactive hierarchical structure goes into the lower level immediately after the diagnosisof one fault, and then is ready to diagnosis another fault or return back to the higher level.While the variable structure adjust the multiple model set adaptively after the next faultoccurring.