Fuzzy Automaton for Intelligent Hybrid Control Systems Janos L. Grantner Western Michigan University, Kalamazoo, MI 49008-5329, USA George A. Fodor ABB Automation Technology Products AB, S-721 67 Vasteras, Sweden Abstract - It is a difficult problem to tract the status of an event-driven, large hybrid control system. Those systems often encounter unexpected events in an uncertain environment. Using a fuzzy automaton offers an effective approximation method to model continuous and discrete signals in a single theoretical framework. The concept of the virtual fuzzy automaton will be used to deal with a cluster of relevant states when a decision is made on the next state of a goal path at the supervisory level. The software architecture of an autonomous agent-based industrial control system will be outlined in which the agents utilize a fuzzy automaton that is wrapped in an object. 1. INTRODUCTION Among the problems that characterize industrial process control innovation, and which are not domain-related, the most difficult ones are as follows: (a) how can new knowledge be introduced into a system, (b) how can the system activate stored domain knowledge in an autonomous way, (c) how can the knowledge be validated (or otherwise detected as inappropriate) and (d) how can the system recover if the new, activated knowledge (or the currently active knowledge) is not suitable to handle the situation at hand. The use of agent technology helps to answer question (a): in that paradigm an agent is defined as an architecture- neutral, mobile software entity that can act on behalf of a human and have decision making capabilities similar to a human [1]. The theory of software dynamical architectures describes the dynamics of the environment in which agents can act. The software architecture, using an architecture broker, mediates the information flow among agents in order to achieve overall population-level goals, and also makes various resources (computational power, sensors, and actuators) available to the agent. The activation of the appropriate knowledge is accomplished via identification operations performed by agents that are capable of finding the right model among a set of available models. Thus the answer to problem (b) is defined as the appropriate interaction among the software architecture mechanisms and the agents. Agents with model-based target seeking algorithms utilizing fuzzy logic, interacting in a distributed large system are strong candidates to replace traditional communications channels among units of a distributed system at a higher system level. Thus agents both require more advanced architectures and also reform the architecture. A fuzzy automaton can implement new knowledge by means of the states of the goal path of an event-driven, sequential control algorithm while providing an effective approximation method to model continuous and discrete signals in a single theoretical framework. With respect to problem (c): knowledge validation is achieved by quantifying the degree of deviation from the nominal operating conditions due to unexpected events caused by either abrupt, or gradual changes in the system, or in the environment of the system. With respect to problem (d): these properties can facilitate the development of computationally inexpensive fault detection and identification (FDI) algorithms, and automated recovery from faults (subject to further research). With respect to FDI, the evaluation of the state transitions between states of a large, complex system is done by focusing only on clusters of relevant states along the goal path. A reconfigurable virtual fuzzy automaton is used to model those clusters of states. A qualitative example can illustrate these points. Identification systems, such as those used for early warning systems, operate through a sequence of states: when the target object is first detected the identification is started. When more details are made available by further sensory data, the identification system follows the states of a given classification scheme. Typically, for a geographically distributed system, the scheme implies a complex architecture along with a communication overhead. By contrast, an agent- based system in which agents equipped with fuzzy automaton is more effective: an agent is created and injected into the first identification station. The agent has fuzzy states for each identification step. The agent moves, or copies itself, between the geographically distributed units to follow closely the identification process: each time a new identification step is reached, the state changes and the agent moves to an appropriate place in the control architecture. If the strength of identification for a given class of objects is deemed too low, the identification has failed and a recovery operation is started. What follows can be a attempt to identify a new class of objects, or after all options have been exhausted without achieving a positive identification, a negative result is returned to the supervisory level. In this way, the agent architecture replaces a complex communication and computation architecture, the end result being a system that has a higher level of abstraction. A similar scheme is used in many so-called intelligent systems that are based on classification schemes. 0-7803-7280-8/02/$10.00 ©2002 IEEE