Sensor fault detection and fault tolerant control with application to vehicle lateral control systems.

摘要

In this dissertation, we design a robust and fault tolerant vehicle lateral control system. The lateral control system performs lane-keeping and lane-changing tasks in the automated highway system (AHS). The lateral controller must be robust with respect to model uncertainties and various road conditions under the constraint of limited actuator power. We propose a controller design method which consists of two steps: (i) designing a robust state-feedback controller of an "augmented plant" by assuming all state variables are available and (ii) converting the state-feedback controller to an output-feedback controller with preserved closed-loop characteristics. Experiments on a passenger car show that small tracking errors are achieved with smooth steering commands.; The lateral control system in this dissertation uses a magnetic reference system which consists of magnetic markers buried along the road centerline and on-board magnetometers. Due to the vulnerabilities of the magnetometers, it is of crucial importance to accommodate magnetometer failures for the safety of the AHS. We propose an observer-based FDI (fault detection and identification) scheme which can detect and identify magnetometer failures right after failures take place. Then the output of the faulty sensor is reconstructed from the output of the healthy sensor. We prove that stability of the lateral control system is assured in the event of magnetometer failures.; The observer-based method is then cast in a stochastic framework. In addition to magnetometers, other on-board sensors can be incorporated into this framework. The switching Kalman filtering and the EM-algorithm are applied to detect faulty sensors and achieve good state estimation. With prior knowledge about the sensor failure modes, concurrent failures can be detected. Stability and performance are observed to be satisfactory in the event of sensor failures.; The inevitable time delay in fault detection, i.e. the time between the occurrence and the detection of a fault, can cause stability problems in control systems. We conduct a theoretical analysis of timely fault detection problem which concerns the detection of faults before the closed-loop system's performance deteriorates to an unacceptable extent. A formal definition of the "timely" fault detection problem will be given. The upper and lower bounds of thresholds of fault detectors will be derived. If the threshold is selected within the bounds, faults belonging to a pre-defined detectable set are guaranteed to be detected in a timely manner and an acceptable level of performance is assured during the detection delay.