Abstract

In this thesis, Limited Lookahead Policies (LLP) have been developed for Robust Nonblocking Supervisory Control Problem (RNSCP) of discrete event systems. In the robust control problem considered here, the plant model is assumed to belong to a given finite set of DES models.

The introduced supervisor computes the control action in online fashion and it is named Robust Limited Lookahead (RLL) supervisor. In comparison with offline supervisory control, RLL supervisor can reduce the complexity associated with the computation of control law as it looks at the behavior of system at the current state and of a limited depth in future.

Since a conservative policy is adopted here, the behavior of the system under supervision of the RLL supervisor is generally more restrictive than the optimal offline supervisor. A sufficient condition is presented under which a limited lookahead window can guarantee the optimality (maximal permissiveness) of the RLL supervisor.

In some problems, the required window length for maximally permissive RLL supervisor may become unbounded. To overcome this limitation RNSCP with State information (RNSCP-S) is studied and solved resulting in a state-based RLL (RLL-S) supervisor.

The results of this thesis can be regarded as an extension of previous work in the literature on limited lookahead policies for (non-robust) supervisory control to the case of
nonblocking robust supervisory control.

The robust limited lookahead design procedures are implemented in MATLAB environment and applied to two examples involving spacecraft propulsion systems.