Agent and Multi-Agent Systems: Technologies and Applications

Agent–Based Approach for LabVIEW Developed Distributed Control Systems (p. 21-22)

Grzegorz Polaków and Mieczys³aw Metzger
Faculty of Automatic Control, Electronics and Computer Science, Silesian University of Technology, Akademicka 16, 44–100 Gliwice, Poland

Abstract. Idea of networked software agents is particularly popular in the field of information sciences dealing with distributed content, whereas in industrial automation its use is usually limited to manufacturing systems. This work presents a concept of multi–agent networked system for automation of continuous processes. Some properties of typical software agent (i.e. advanced high level languages and social skills) had to be dropped in exchange for determinism and satisfying time performance in negotiations between network– connected control components. The proposed environment is National Instruments LabVIEW, very popular solution for automation and measurement. LabVIEW is equipped with advanced data acquisition tools and is capable of artificial intelligence methods, although it lacks agentification mechanisms. A framework is presented, providing LabVIEW the required functionality. Keywords: multiagent–based networking, producer–distributor–consumer, distributed control systems, process control, systems integration.

1 Introduction

There is strong ongoing progress in the field of agent–based technologies. The theory proved its usefulness and commercial interest in its applications boosted rapidly, resulting in countless implementations, especially in the most popular environment of distributed content i.e. World Wide Web [1], [2]. However, process of adaptation and implementation of the original idea in automation and control science is not so successful and varies greatly depending on subject. On the one hand, natural entity based structure of software agencies reflects perfectly processes occurring in manufacturing systems and already proved its usability in that field, overtaking classical approaches [3]. On the other hand, in the automation of continuous processes it is hard to tie agents to non–discrete physical phenomena like flow of liquids, gases or heat. In this case agents have rather to be bound to logical structure of control system. Great implementation of such agent–based frameworks in continuous plants was done by van Breemen and de Vries [4]. Their works are a significant contribution inttroducing agents theory in the field of process automation, however they overlook important issue, that is implementation of theory in existing physical plants and systems. Solving this issue is significant and non–trivial, since while there is wide choice of frameworks for developing intelligent agents in abstract knowledge environments, there is no ready solution for agentification of spatially distributed control systems. Specific requirement, needed to be fulfilled by software to be usable in the field of data acquisition and control, is easy and unproblematic integration of logical software layer with physical hardware of control instrumentation. One of applications fulfilling this requirement, therefore suitable for the task, is National Instruments’ LabVIEW.

LabVIEW is a very popular software development solution for engineers. A programming language which LabVIEW implements is simple yet powerful graphical language based on functional block diagrams, often called a "G language". Capabilities of the NI LV are the same as capabilities of any other classical text– based programming language. It also includes a wide set of predefined function blocks providing functionalities useful in advanced software, i.e.:

- artificial intelligence tools (neural networks and fuzzy logic),
- basic functions of a desktop computer’s operational system’s network protocol stack,
- control instrumentation connectibility (using standards like OPC, and proprietary protocols like Logos),
- rich Human–Machine Interfaces and visualisation tools.

As it can be concluded from the above list, internal agent’s intelligence can easily be programmed by an end user using library of components supplied by default. However, it was necessary to design a protocol for agents communication and build a required framework supporting it. A new potential of continuous control systems gained by agentification at control loop layer turned out to be unprecedented, which was motivation to describe it in detail. Concept of networked multiagent system for continuous process control (including structure of the framework, communication protocols and language of agents) is an original contribution.