Back to list of 2001/02 seminars

Supervisory control theory provides tools for the synthesis of feedback controllers guaranteeing the safe, and autonomous operation of large, complex plants. Supervisory controls also ensure the maximal freedom for the plant to perform as many useful tasks as possible. The complexity of the problem forces the designer to use finite automata (or Petri net) models of the plant. Even then the complexity of the control design task grows very fast with the size of the plant (much faster than linear in the number of components). This talk will first review the classical theory of supervisory control design for controlled automata and for controlled Petri nets. This theory will be compared to classical feedback control theory, formulated in the behavioural framework.

The talk will next discuss one possible approach to simplifying the supervisory control design complexity, giving up the requirement of maximal freedom. The adaptive control paradigm will be considered, where the feedback control loop is decomposed a priori into a mode detector and a controller. This adaptive approach assumes that the mode of operation of the plant changes rarely as a result of changes in the external operating conditions, or due to internal changes such as uncontrollable failures, or repairs. The mode detector estimates which mode is active at a given time. The controller is designed as if this mode is indeed the true mode of operation of the plant at the current time. The model for one fixed mode of operation is a lot simpler than a global model, leading to a much lower design compelxity. Finally the talk will discuss the requirements for a mode detector, or equivalently for a failure detector.