Model-based Diagnostics for the Supervisory Control of Manufacturing Systems

Because of their nondeterministic nature of behaviour, the supervisory control of manufacturing systems must be carried out in closed loop. This trait greatly increases the size and complexity of the Discrete-Event System (DES)-based supervisorycontrollers of manufacturing systems. Attempting to cope with this complexity, a Hybrid Supervisory Controller (HSC), that distributes operations between a DES-based supervisory controller, an alternate mechanism, and a diagnostic system, was developed. The implemented diagnostic system had to be able to detect multiple faults, be robust, as well as integrate well with the other elements of the HSC. A model-based method derived from the General Diagnostic Engine [de Kleer and Williams, 1989] utilizing a set of models was developed and found to be suitable. Models are chosen and expectations of their predictions are adjusted according to the state of the equipment, the operations they are performing, and the location of parts. Introduction Background and Motivation The design of a supervisory controller entails the formulation of control laws, and the synthesis of supervisors. The laws specify how the supervisor is to react to the behaviour of the manufacturing system, the goal being to have some production specifications satisfied within the standing control enforcement constraints. Petri nets [Long et al., 1992b], [David, 1991], knowledge engineering [Benhabib et al., 1989, Camarinha-Matos and Steiger-Garcao, 1986a], timed transition models, real-time temporal logic [Ostroff, 1990, Ostroff and Wonham, 1990], and controlled automata [Brandin et al., 1991] have been used a.s supervisory controllers for manufacturing systems. Supervisory controllers designed around Discrete-Event Systems (DES) theory have the desirable feature that their behaviour may be proved and verified correct before implementation using the tools developed within the theory. However, the control of even moderately complex systems can easily require an immensely large DES strategy [Ho, 1987]. This necessitated the development of a hybrid approach that uses some alternate mechanism in addition to a DES supervisory controller, which would relieve the latter of the need for so many states by (i) taking on the responsibility for some of the control objectives, and (ii) asserting control whenever events diverge from the significantly reduced number of states of the DES supervisory controller. The tasks of the supervisory controller include the detection of failures and recording of repairs to equipment, which are part of the job of (i) monitoring the workcell, (ii) processing and analyzing large quantities of sensory input, and (iii) confirming the supervisor’s hypothesis about the state of the workcell. These are the responsibilities of an on-line diagnostic system which forms part of the supervisory controller. This paper will focus on the application and development of a diagnostic system, based on the General Diagnostic Engine (GDE) of de Kleer and Williams [de Kleer and Williams, 1987], that meets the needs of a DES-theory based supervisory controller. The environment in which the diagnostic system will operate is described briefly as well 1. 1For a more complete description of the rest of the supervisory controller, see [Williams et al., 1994]. 134 From: AAAI Technical Report SS-94-04. Compilation copyright © 1994, AAAI (www.aaai.org). All rights reserved.