1 RECONFIGURATION OF DISCRETE EVENT SYSTEM CONTROLLERS WITH DYNAMIC SENSING SET Jing Liu and Houshang Darabi University of Illinois at Chicago, Department of Mechanical and Industrial Engineering Abstract: This paper presents a finite automaton model to describe the control reconfiguration of discrete event systems (DES) with respect to the dynamic changes of observation means, especially when the changes are not acceptable for the conventional observability theory. The model includes other classes of events besides the regular DES events, such as repair and failure events of the observation means. Given a regular DES controller, through a systematic procedure it is extended to include the effect of those events on the control. The potential application of this work is the optimization of the reconfiguration strategies. Copyright © 2005 IFAC Keywords: supervisory control, observability, sensor failures, formal methods. 1. INTRODUCTION This paper models the control reconfiguration of DES with respect to the change of observation means. Specifically, the effect of the reconfiguration strategies on DES when the change of observation means is not acceptable for the conventional observability theory is addressed. The investigation on the observability theory of DES was triggered by two parallel and original works of (Lin and Wonham, 1988) and (Cieslak, et al., 1988). The DES observability is critical for the DES controller as the controller takes the control actions according to what it observes from the DES. It is assumed that for each event, there is a sensor reporting its occurrence to the controller. If a sensor is not working properly or its communication channel with the controller fails, the controller cannot obtain the information regarding the occurrence of the corresponding event. In this case, from the controller’s point of view, this event is unobservable. Once the sensor and/or communication channel are repaired, that event becomes observable again. In practice, the sensors or communication channels can be broken down or repaired at any time during the control. In a sense, in certain state the controller may “see” a different set of events than it does in the initial state. In other words, the observability of events changes over time. In conventional observability theory of the DES (Lin and Wonham, 1988), it is assumed that the set of observable events remains unchanged. Furthermore, there are some events which are critical for keeping the right control - the “must-be-observed” events. However, if these events are not seen by the controller, the controlled DES will go out of control. But there are some occasions where the controller can survive even if some of the “must-be-observed” events are not observed according to (Darabi, et al., 2003). Here they propose a control switching policy to discover an appropriate control policy and switch to the new policy on the fly. The search for the new control policy and the switching actions are performed by another control agent called the mega-controller. The proposed switching theory in this work is based on the set of observable projections originally introduced in (Haji-Valizadeh and Loparo 1996). Therefore if a new control policy is found when the set of observable events changes, then this policy can survive for infinite time given that the observable events set does not change again after the switch. Such policies, called infinite time policies, have some limitations according to (Liu and Darabi 2004). In fact the infinite time policies cannot provide the controller maximum survival time subject to the dynamic observable event set. In (Liu and Darabi 2004) the authors have developed the class of finite time policies. This class is a superset of the infinite time policies introduced by (Darabi, et al., 2003). This class, in addition to the infinite time policies, consists of all control policies that are feasible for a finite duration of time. A finite time observation policy is good for the current state, but may not be good for other states. Using finite time policies provides a more general solution for the