Nonlinear observer design and fault diagnostics for automated longitudinal vehicle control

An integral part of automated vehicle control systems is their reliance on a host of sensors and actuators that provide the controller with knowledge about the vehicles operating environment and the ability to affect its behavior. While this dependence results in improved driving performance and enhanced safety, it also represents a potential liability when failures in these components are considered. One means of addressing this issue is the incorporation of fault tolerance into the controller design through the use of a fault diagnostic system. The purpose of this thesis is to describe the design and experimental verification a model-based fault diagnostic system for the automated longitudinal control of a passenger vehicle within an automated highway system (AHS) setting.; A classical approach to the fault diagnostic system design is taken by dividing the diagnostic tasks into two modules; the residual generator and the residual processor. The residual generator relies on a simplified nonlinear vehicle model to construct a set of residual signals which are sensitive to faults in the control components. The residuals are formed using a combination of parity equations, linear observers, and nonlinear observers. The residual processor detects and identifies faults by thresholding a least squares estimate of the fault magnitude based on the steady state gain from faults to residuals. Experimental testing of the complete system verifies that all single faults can be detected, and only faults in the braking system are not uniquely identifiable.; A key contribution of this thesis is the development of an observer design technique for incrementally sector-bounded nonlinear systems. Sufficient conditions for the existence of a stabilizing gain matrix are derived as the solution to a convex feasibility problem using Lyapunov's second method and standard convex analysis tools. These results are then extended via convex optimization to guarantee desired performance of the observer with regards to the decay rate of the estimation error dynamics and the $L2$ gain from disturbances to estimation errors. Finally, a method for computing gain matrices which are Pareto optimal in the sense of both performance objectives is also described.