Fault-tolerant supervisory control of discrete event systems: Methods and examples

Faults can be considered as uncontrollable events that suddenly happen in a system and change the behaviour of the system in a negative way. In particular, in case a fault happens in a discrete event system (DES), certain actions or operations might no longer be possible. This thesis studies the supervisory control of DES that are subject to faults. Hereby, two concepts are employed: Fault-tolerant and fault-recovery control. Regarding fault-tolerant control, it is desired to use a controller that works both for the system with and without a fault. Hence, we first identify necessary and sufficient conditions for the existence of a supervisor that realizes a given behavioral specification both in the non-faulty and in the faulty case. We further show that it is possible to determine a supremal fault-tolerant sublanguage in case the existence condition is violated. Finally, we propose an algorithm for the computation of this sublanguage and prove its correctness. Different from existing work, our fault-tolerant supervisor allows fault occurrences and system repairs at any time. Regarding fault-recovery control, we study both the case of operating a DES after a fault and after repair. We first develop a new method for the fault-recovery of DES. In particular, we compute a fault-recovery supervisor that follows the specified nominal system behavior until a fault occurrence, that continues its operation according to a degraded specification after a fault and that finally converges to a desired behavior after fault. We next show that our method is also applicable to system repair and we propose an iterative procedure that determines a supervisor for an arbitrary number of fault occurrences and system repairs. Finally, we extend our fault-recovery and repair method with multiple and different faults and repairs. As a result, we obtain a supervisor that follows the specified nominal system behavior in the fault-free case, converges to a desired degraded behavior for each fault type and recovers the nominal behavior after corresponding repair. All developed methods are demonstrated with a small manufacturing system.