Fault recovery in discrete-event systems using observer-based supervisors.

摘要

We solve the supervisory design problem using a state-based approach. It is assumed that design specifications are given for normal, transient and recovery modes in terms of legal (safe) states. The system under supervision is also required to be nonblocking in normal and recovery modes.; Following a modular switching approach, we propose supervisory schemes in which separate supervisor modules are designed for normal, transient and recovery modes. We consider failure accommodation in cases where recovery to normal operation is not possible and also recovery in cases in which it is possible to resume normal operation. For each case, we provide two solutions, one in which the recovery supervisor is in the feedback loop when the system is started in its normal mode, and another solution in which the recovery supervisor is engaged only when a fault is detected and isolated. The latter approach is less computationally complex to implement. We investigate supervisor admissibility and nonblocking property of the system under supervision.; All of the supervisor modules are observer-based. In our opinion, the use of observer-based supervisors results in a more transparent solution and simplifies the analysis in our switching scheme when one supervisor replaces another in the feedback loop. In this thesis, in the process of our study of fault recovery, we also propose a systematic method for designing observer-based supervisors using normal languages. (Abstract shortened by UMI.)