衛(wèi)星姿態(tài)控制系統(tǒng)的故障診斷研究.pdf

1、南京航空航天大學(xué)碩士學(xué)位論文衛(wèi)星姿態(tài)控制系統(tǒng)的故障診斷研究姓名：王劍非申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：控制理論與控制工程指導(dǎo)教師：姜斌20080101衛(wèi)星姿態(tài)控制系統(tǒng)的故障診斷研究 iiAbstract Fault diagnosis and fault accommodation problems for satellite attitude control system are studied in this dissertation. T

2、he proposed methods are simulated by Matlab software. At First, the history and the development of fault diagnosis, specially, fault diagnosis for the spacecraft is surveyed. Then, the model of satellite attitude control

3、 systems is introduced, and the characteristics and the difficulty of the fault diagnosis are analyzed based on this model, and the fault detection method is also given. Under Lipschitz condition on the nonlinear part of

4、 such systems, a nonlinear adaptive observer is proposed for the actuator fault diagnosis. Moreover, the convergence and its speed of the developed algorithm are also investigated. Then, based on the information of the f

5、ault estimation, a fault accommodation technique is developed to recover the system performance. Finally, simulation results have been included to demonstrate the feasibility and effectiveness of the proposed techniques.

6、 Then, the Unscented Kalman filter (UKF) based fault diagnosis for the sensor fault is proposed. First, the model is discretized, and the faults of sensors are added to the model as augmented state. Then, the UKF is used

7、 to estimate both the state and the fault. Simulation results demonstrate that the UKF is strong tracking filter and can track the abrupt and incipient fault effectively. Finally, least squares support vector machine (LS

8、SVM) is used as a novel fault diagnosis method for satellite attitude control systems. Compared with standard support vector machines, least squares support vector machine LSSVM regressions can always converge rapidly. A

9、ccording to this character, a LSSVM based nonlinear observer is designed to estimate the satellite attitude control system faults, in which LSSVM regressions are on-line trained. The simulation results show the applicabi