Achieving fault-tolerance and safety of discrete-event systems through learning

Abstract

A system is said to be fault-tolerant if it remains functional even after a fault occurs. By describing faults as unpredicted events, we study the active fault-tolerance of discrete-event systems (DES) while ensuring safety requirements. Starting from a finite automaton model of the uncontrolled plant, our proposed control framework consists of nominal supervision, fault diagnosis and active post-fault control reconfiguration. First a nominal supervisor is designed with respect to the nominal mode to ensure the control specification prior to the occurrence of faults. Second, a learning-based algorithm is proposed to compute a diagnoser that can detect the occurrence of a fault. Necessary and sufficient conditions under which a post-fault safety-enforcing control reconfiguration is feasible are explored, and a second learning-based design algorithm for the post-fault supervisor is presented by using the limited lookahead policies. Effectiveness the proposed framework is examined through an example.