Fault diagnosis for a satellite system using sliding mode observers

This thesis presents the theoretical development along with verification and validation campaign for fault detection and isolation (FDI) schemes applied to a satellite system. Specifically FDI schemes developed in the thesis are applied to the problems of actuator and sensor fault detection based on the nonlinear model of Mars Express Satellite (MEX). Model based FDI schemes are not implemented in the current health monitor schemes on board the MEX satellite, which is an over actuated system with the actuators and sensors having a specific tetrahedral configuration. A simple nonlinear sliding mode observers based on unit vector and super twisting methods developed in the thesis are verified on both rigid body and flexible mode models of the nonlinear MEX satellite. Faults for thrusters are applied during initial strong controller mode phase and also later stages of sun acquisition mode (SAM). Both dedicated observers and a bank of observers are designed to generate residuals which are used to detect and isolate faults. Unlike the reduced order systems typically obtained when using sliding modes in literature, this particular application produces no reduced order dynamics. All the schemes utilize the equivalent injection signal found in the sliding mode literature, which are then filtered using a simple integrator to generate threshold based residual analysis to detect faults or isolate faulty components. Verification and validation campaign consists initially with specific fault simulation and later with Monte Carlo simulations with uncertainties and noise are used to verify the performance of the designed schemes. A new multi-variable version of the super twisting algorithm is proposed along with a Lyapunov approach to analyze its stability. Existing super twisting schemes assumes a single control input structure. The proposed scheme demonstrates that it can be applicable to vector cases, where the decoupling of the control input between states is not possible. Finally a comparison between first order sliding mode observers based on unit vector methods and a second order super twisting observer is made using Monte Carlo simulations in the presence of noise and uncertainty for both the rigid body and flexible mode models of the satellite. The analysis establishes that both methods show good performance and robustness for fault diagnosis, which can be applied for all controller mode changes during several phases of SAM