| A fault-tolerant estimator is developed in which state estimates remain untainted in the presence of faults and for which performance degrades gracefully. The essential component for fault-tolerant estimation has been identified as a bank of local fault detection filters computing state estimates and generating residuals with directional properties allowing for fault detection and identification. If no fault is detected, the structure uses a blending method to fuse the estimates from the local fault detection filters into a global minimum variance estimate. Once a fault is detected and identified, the fault detection filters are labeled corrupted or uncorrupted according to their design; the untainted estimate is obtained from the uncorrupted filter(s). In previous fault detection schemes, a voting algorithm relied on statistical properties of the local state estimates, whereas here the directional properties of the detection filter residuals are used to both detect and identify the system faults.; The fault-tolerant estimator is developed in discrete time. The properties of the discrete-time fault detection filter are investigated under the light of combined robust fault detection, decisive fault identification and qualitative estimation in the presence of modeled faults. The discrete-time filters are studied in three different forms: full-order, in the limit and reduced-order. Stability is assessed and blocking properties are proven. The role of transmission zeros is investigated in details and lead to the possible alteration of the nuisance fault direction to ensure closed-loop stability. Similarly, the addition of a disturbance direction, like wind gust, in the nuisance fault direction yields a fault-tolerant estimator that is robust to the presence of that very disturbance.; The fault-tolerant estimator is tested as part of a navigation system applied to Unmanned Aerial Vehicle. The instrumentation includes a single GPS receiver and an IMU. A bank of seventeen filters is designed, of which two thirds are reduced-order filters and one third is full-order limiting filters. Sensor faults are generated in both the GPS and the IMU output signals; actuator faults are created in the measured control surface deflections. In all cases, the faults are detected, identified and the fault-tolerant estimator reconfigures to provide an uncorrupted state estimate. Robustness is tested by modifying the model used to design the filters so that it does not match the “real” aircraft. Finally wind gust is added to demonstrate the robustness of the fault-tolerant estimator to exogenous disturbances. |
