Fault diagnosis based on the enclosure of parameters estimated with an adaptive observer
Abstract
The proposed fault diagnosis approach associates an adaptive observer for residual generation with set-membership computations based on zonotopes for residual evaluation. The main advantage of this approach is its rigorous propagation of pre-specified modeling uncertainty bounds to the computed residuals. Within the assumed modeling uncertainty bounds, fault detection is guaranteed to be free of false alarm, while the efforts made with set-membership computation minimize the conservativeness of fault detection decisions. The novelty compared to earlier works mainly resides in a guaranteed robustness to bounded parameter variations and in a method for dealing with occasional lack of input excitation.