Fault detection in nonlinear systems: An application to automated highway systems

The aim of this work was to bolster control system reliability for Automated Highway Systems. The means is increase in level of tolerance to faults in sensors and actuators. An efficient fault detection scheme in combination with sufficient system back-up capability suffices in this function.;An approximate feedback linearized transfer function analysis was done to study effects of faults in various components on the platooning objective of constant longitudinal inter-vehicle spacing.;An integral part of the dissertation is design of detection filters for systems with Lipschitz bounded nonlinearities. It was shown that nonlinear observers can be designed to ensure both nominal observer stability and directional characteristics under faults to enable fault identification.;Another important contribution of this work is the design of a fault decision logic for detection of change in mean of Gaussian signals. Normally if the signal exceeds a predetermined threshold value a fault is indicated by the detector. This value is computed based on a desired false alarm rate for the detector. It was shown that if detection delay is minimized too then the threshold to ensure the desired false alarm rate depends on the fault time estimate and is therefore variable. An online algorithm was developed to compute the threshold. This algorithm is easily implementable and demands minor computing requirements.;Sufficient conditions were derived for guaranteeing stability of nonlinear observers for Lipschitz nonlinear systems without an explicit linear part. Such an observer was used for sensor information reconfiguration in the vehicle model. Detection filters were also extended for this class of systems and used for actuator fault detection in the vehicle longitudinal system.;System redundancy available in the vehicle longitudinal model was exploited in the design of detection logic in the sensors and actuator system. Two triplets of double redundancy systems among the sensors were identified and voting schemes were used for detection of faults. Reconfiguration of sensor information was accomplished either by redundancy based or observer based methods. Decision logic to differentiate between an actuator fault and a sensor fault was developed based on a feedback linearized filter for actuator fault detection.