Fault-tolerant control and its application to lane-keeping control of automated vehicles

In this dissertation, we develop procedures for the design of real-time control systems that are insensitive to failures i.e., fault tolerant control systems. These procedures are then applied to lane-keeping control of automated vehicles on Automated Highway Systems (AHS). Our focus is on accommodating failures which have an immediate effect on the stability of the system under consideration. We consider two situations related to the design of failure tolerant controllers.; First, we formulate the design of failure tolerant controllers as a simultaneous stability problem which reads as follows: Given linear time-invariant (LTI) systems P₀, P₁,,P_n, design a controller C (also LTI) such that each of the feedback inter connections (P _i,C), i = 0, 1, 2,,n is internally stable. We deal with two problems related to the design of simultaneously stabilizing controllers. The first problem deals with the design of a simultaneously stabilizing controller that stabilizes two LTI systems. A new procedure is developed which uses a sufficient condition for simultaneous stability to reduce this problem to a standard $H\infty$ control problem. Results indicate that for certain plants this procedure results in a controller of significantly lesser order than the classical interpolation based methodology. The second problem deals with the simultaneous stability problem with an added performance criterion. In this case, we express a sufficient condition for simultaneous stability as a Linear Matrix Inequality (LMI). This condition is then used to obtain, once again, to a standard $H\infty$ control problem. The highlight of this procedure is that it applies to the case of simultaneous stability of three or more plants. However, both procedures introduced above suffer from the limitation of being conservative. This is because they are both based on only sufficient conditions for simultaneous stability.; Next, we consider the problem of accommodating a specific type of failure in one of two sensors used for controlling a two-output system. The failure model considered here is that in the event of a failure in a sensor, the associated output goes to a constant value. A heuristic control architecture utilizing dedicated observers is proposed to solve this problem. It is argued that the limitation of the dedicated observer based scheme is its heavy dependence on the model used to describe the system.; The strategies developed above are tested on a safety-critical control system namely, the lane-keeping control system of fully-automated test vehicles used by the Partners for Advanced Transit on Highways (PATH), California in the development of an AHS. The lane-keeping control system is responsible for ensuring that vehicles maintain their lanes. This dissertation addresses the accommodation of hard failures in two lateral error measuring sensors which act as the primary sensory components of the lane-keeping control system. High and low-speed experimental results demonstrating failure tolerant control action are documented.