Arquitetura para a implementação de controle supervisório comunicante

Abstract

Modern industry contemplates a profile of systems that detect and react, at runtime, to changes in the physical context, starting to operate in different ways. Although this characteristic exposes the potential to add improvements to production processes, it also poses challenges to the Automation and Control Engineering community, such as, for example, the limited capacity to process, in parallel, multiple and complex control loops. An alternative arises from the combination of formal synthesis methods, such as the Supervisory Control Theory (TCS), with architectures that decentralize the implementation of the control system. This benefits both controller programming and implementation. However, current models of decentralized implementation do not consider emerging aspects of the industry, such as communication via the Internet of decentralized control modules, something that has potential for customization of production and cooperative production. This work proposes an architecture that decentralizes the control system resulting from TCS and implements the communication between different control agents. A key piece in this proposal comes from the concept of a central agent and a local control agent. The central agent performs an abstraction of the behavior of local agents and is responsible for coordinating, at a high level, the logic of the control system that results in local physical actions. In this abstraction, global control commands actions to local agents after processing certain events reported by local agents. These events, in general, represent operational contexts that abstract chains of events that are locally observable, but not globally observable, thus making global control simpler and more focused on key system actions, operationalized only locally. In terms of results, a significant reduction of the state machine implemented in the central agent (with abstraction) compared to an equivalent state machine (without abstraction) was observed, both using a monolithic approach to the supervisor synthesis, which resulted in 207 states and 464 transitions of a machine with 251 states and 526 transitions, respectively. In addition, other implementation aspects indicate a high potential for scalability, which implies possible gains to TCS, with the possibility of expansion for several researches and use in productive environments.