Abstract

Fault recovery is a challenging task that is crucial in achieving stringent reliability and safety goals. In this thesis, the problem of fault recovery is studied in discrete-event systems (DES), assuming permanent failures. A diagnosis system is assumed to be available to detect and isolate faults with a bounded delay. Thus, the combination of the plant and diagnosis system can be thought of having three modes: normal, transient, and recovery. Initially the plant is in the normal mode. Once a failure occurs, the system enters the transient mode. After the failure is diagnosed by the diagnosis system, the system enters the recovery mode. This framework does not depend on the diagnosis technique used, as long as the diagnosis delay is bounded. As a result, the diagnosis and control problems are almost decoupled. In general, for each mode there is a set of specifications that have to be met. We propose a modular switching supervisory scheme. The proposed framework contains one normal-transient supervisor and multiple recovery supervisors each corresponding to a particular failure mode. Once a fault is detected and isolated by the diagnoser, the normal-transient supervisor is removed from the feedback loop and one of the recovery supervisors will take sole control of the system. The issue of non-blocking is studied and it is shown that essentially if the system under supervision is non-blocking in the normal mode, then it will remain non-blocking during the recovery procedure. Supervisor admissibility is also studied. This approach is developed for untimed DES and then extended to timed DES. In the process, previous results on supervisor design for untimed DES with partial observation are extended to timed DES. Various examples from manufacturing and process control are provided to illustrate the approach.