Adaptive Supervisory Control

Abstract

This paper discusses how the adaptive control paradigm can be used for reducing the computational complexity of the synthesis of supervisory controllers for discrete event systems (timed or untimed) with partial observations. The adaptive control paradigm simplifies the design by assuming a priori that the control design is decomposed in two parts. The state space of the plant model can be partitioned in different “modes of operation”. The first component of the control loop is a “mode detector”, an observer that estimates the set of all “modes of operation” that are compatible with the past observations. The second component of the supervisory controller activates feedback controllers corresponding to each of these possible current “modes of operation”. This leads to a safe, but not maximally permissive control law, that is computationally much easier to design.

Research supported in part by the ESPRIT project VHS — Verification of Hybrid Systems, and by a BOF project at Ghent University. Part of the results in this paper were based on research carried out within the framework of the Belgian Program on Interuniversity Attraction Poles, Prime Minister’s Office, Science Policy programming. The scientific responsibility rests with its author.