Abstract

In this thesis, robust nonblocking supervisory control problem (RNSCP) is studied in which the plant model belongs to a family of models. A drawback of the existing solutions in the literature is that the entire supervisor which is designed offline needs to be stored in computer memory for implementation. In this thesis, we develop an online solution in the form of a variable lookahead policy (VLP). In this approach, the supervisory commands are computed online when the plant is in operation, based on the behavior of the plant models in the near future (lookahead window). In this way, only the plant models need to be stored in computer memory (which can be usually done more efficiently than that of a supervisor). To derive our VLP algorithm, we develop a direct offline solution to RNSCP which is more transparent than the indirect method in the literature. As an illustrative example, we have used our direct method to solve supervisory control problem with multiple markings in which the plant under supervision is required to be able to reach any of the multiple sets of marked states, each corresponding to the completion of a different task. In some problems, the VLP lookahead window may become unbounded and the VLP algorithm cannot be applied. To relieve the limitation, state-based RNSCP (RNSCP-S) is formulated and solved using state-based VLP (VLP-S). If a RNSCP cannot be solved using VLP, it can be solved using VLP-S as a corresponding RNSCP-S after automaton refinement