A robust parity space approach to fault diagnosis based on optimal eigenstructure assignment

By assigning the left eigenvectors of the observer orthogonal to the disturbance directions, robust fault detection is achievable. Eigenstructure assignment together with a dead-beat design yields the same structure as the 1st order parity relation, when applied to the observer's estimation error dynamics. The fault diagnosis scheme can be implemented in the parity space format. The dead-beat structure also gives a robust and time-optimal monitor performance which is capable of discriminating between the separate effects of plant uncertainties and faults. The directional properties of the residual are used to isolate faults in either sensors or actuators. The paper also shows that the observer-based fault detection approach is equivalent to the generalized parity space approach in a particular case. The proposed method is applied to a tutorial example. Simulation results demonstrate the robustness of the method. >