next up previous
Next: Modeling the Three-Machine Example Up: Comparing Models Previous: Specification Models

Controller Agent and Synthesis Procedure

The underlying difference between the controlling strategies obtained by the different models is related to the methods' capabilities for handling sequential specifications. Because we were only able to obtain the safety controller with TTM's and Petri Nets, these solutions still allow the pushers to move back and forth without accomplishing the desired task (bringing the workpiece from position 1 to position 3). Notice that the NCES sequential controller consists of a single sequence of interim places to follow. This same path is also allowed by the R&W supervisor as well as several others. Depending on our objectives for the system, the restriction of the NCES controller may be undesirable. However, for purposes of implementing controller code (in a PLC, for example), a separate entity would be needed to ``prune'' the R&W supervisor to get a specific sequence of events to be realized by the system. The policies used by this entity in pruning the supervisor would depend,, again, on the goals for system behavior.

The R&W theory guarantees that the supervisor is maximally permissive; that is, it only restricts those events which are not legal with respect to the specifications and no others. The safety controller based on Petri Nets is also maximally permissive. With TTM's and NCES's, however, it is not generally the case that the safety controllers are maximally permissive. The reason is that these algorithms use a backward search in the structures that generate the global states, instead of the ``reachability graph'' itself which is equivalent to R&W's finite state machine.

In terms of the synthesis procedures, it is well known that the R&W algorithm is difficult to implement due to the state explosion problem. (The shuffled plant can grow exponentially in the number of states.) Although explicit representation of all states is what makes the R&W theory complete in terms of its capabilities, it is also what makes the procedure impractical for large, complex systems. With Petri Nets and their compact matrix representation, monitor controllers are efficiently obtained because they require only several matrix operations. Both TTM's and NCES avoid a full search of the state space, thus providing more efficient solutions as well. Furthermore, with NCES it is possible to automatically generate control code for PLC's.


next up previous
Next: Modeling the Three-Machine Example Up: Comparing Models Previous: Specification Models

Luz E. Pinzon
Wed Oct 15 18:10:49 EDT 1997