Industrial Automation Architecture: AnIntelligent Agent Based Approach∗

ADDISON RIOS-BOLIVAR‡ JOSE AGUILAR‡,∓

FRANCISCO HIDROBO∓,§ LEANDRO LEON‡

‡CEMISID, Facultad de Ingenierıa∓Laboratorio de Sistemas Inteligentes

§Laboratorio SUMA, Facultad de CienciasUniversidad de Los Andes

5101. Merida, VENEZUELA

Abstract: - The purpose of this paper is to present an automation architecture basedon intelligent agents. Starting from the functional requirements of a process automa-tion system, a hierarchical automation architecture is formulated in order to downintelligence to inferior levels. This advantage is possible grace to multi-agent systems,which allow the implementation of interoperativity, heterogeneity and complexitydue to the requirements of industrial automation. The architecture is composed bya superior layer, and a middleware layer. The superior layer is constituted by twomulti-agent systems, which are based on the control and automation functionalities.The middleware layer is designed in order to support superior layer activities andmanages the communications from/to field devices. This layer satisfies the FIPAspecifications, it also contains two subsystems: one to handle the communicationsamong different sites, and other to give management services of agents. The middle-ware can support any multiagent system, and has great design versatility.

Key-Words: - Agent applications, Automation systems, Process control software,Plant design, Plant engineering.

1 Introduction

The industrial plants demand every day so-phisticated systems that allow them to guar-antee a reliable and highly profitable pro-duction. Thus, process automation systemsare applications that have been character-ized by requirements emphasizing safety, re-liability, efficiency and quality [1]. Therefore,the automation systems are complex, large,

∗This work was supported in part by AgendaPetroleo funds under grant project 97003817FONACIT-Venezuela and by the CDCHT-ULA.

distributed and persistent hardware softwaresystems being stamped by the characteristicsof the technical processes they are designedto control [7].

The rapid development of high-capacityhardware components and the informationand communication technologies leads to anincreasing complexity and a strong need forintegration in industrial automation (AI).Traditionally, the automation systems havehad a hierarchical structure with different au-tomation functions at each hierarchical level.

4th WSEAS Int. Conf. on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS Miami, Florida, USA, November 17-19, 2005 (pp134-139)

This hierarchical structure is called the au-tomation pyramid and represents a canon-ical abstract model of the industrial com-plexes [8]. Thus, automation systems shouldbe able to handle complex interactions be-tween hardware and software entities. Thesoftware capabilities to manage the evolvingcomplexity are referred to intrinsic proper-ties of all automation systems, which can besummarizes: automation systems are com-plex and distributed systems; they requiredifferent views; the software must be flexibleand adaptable.

By the another hand, new paradigms havearisen in the design of computational tools.For example, the agent-oriented paradigmpermits to design complex and sophisticatedsoftware systems. A software agent is aproactive object. The decision about how andwhen to perform an action is controlled bythe agent itself. Beyond this, it is able toexecute an action autonomously without be-ing invoked externally. This quality divergesfrom a passive software entity like a softwarecomponent which waits for a remote inter-action [2]. Their most important proprietiesare: autonomy, communication, sociability,reactability, intelligence, and mobility. Thesecharacteristics allow that agent technologycould be used to fulfill the requirements ofprocess automation systems.

2 Industrial Automation

and Agent Technology

Process automation has not typically been anearly adopter of new information technologieslike software agents. However, some researchhas emerged concerning the application ofagent technology to the implementation ofprocess automation systems [2, 3, 9, 10].This application has been characterized bythe match between the operational principlesof process automation systems and agents,where complex and distributed systems of en-gineering can be obtained.

The automated systems can be represented

in different levels, each one has adequate op-eration characteristics: one level for field de-vices in order to capture process information,one level for supervisory control and opti-mization where the tasks of process controlare executed, and another level for processmanagement where the production strategiesare evaluated and developed. This architec-ture of hierarchical operation allows the dis-tribution of functionalities of the automationactivities through the description of differentstrategical, tactical, and operational tasks,where the intelligence is centralized in the su-perior levels.

In that same sense, the agent-oriented ap-proach is a natural way of system decompo-sition and a reasonable alternative to con-temporary approaches in software engineer-ing [9]. The levels of a automatized sys-tem can be represented by subsystem com-ponents, which are mapped to agents andagent communities (MAS); the interactionsbetween subsystem components are mappedto cooperation, coordination and negotiationmechanisms; this same way the organiza-tional relationships are represented. Thus,intelligence can be distributed through thedifferent levels. See Figure 1. Therefore, theagent-oriented paradigm is suitable to meetthe requirements of modern automation sys-tems, where the reconfigurability and flexi-bility are important aspects [5, 6, 8, 10].

Figure 1: Agents Topology in Automation

4th WSEAS Int. Conf. on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS Miami, Florida, USA, November 17-19, 2005 (pp134-139)

3 IA Architecture based

on Agents

Just as it is shown in the Figure 1, the con-trol automated systems can be characterized,by their great heterogeneity, by two aspects:the first is the hardware; which is constitutedby components of very low level, i.e. remote,PLCs, etc, up to very complex platforms ofhigh level, i.e. industrial computers, servers,clusters, etc. The second level is representedby application software, which is constitutedfrom configurations of field communicationprotocols, viewers, including pertinent appli-cations to the business. Then, such as theMAS, control automated systems are com-plexes and heterogenic.

Based on the functional relationship be-tween the industrial automation process andthe agent technology, we propose an archi-tecture for automation that consider agentsin order to lower and distribute the intelli-gence to different levels. The proposed hier-archical automation architecture has the fol-lowing different layers and components (seeFigure 2):

Figure 2: Hierarchic Architecture

• A superior layer. In this layer exists twoagents communities: the process and theapplication agents communities.

• An interface layer. It offers services tothe agents communities and its archi-tecture and functionalities are based onFIPA specifications [4].

• The base layer. This layer contains twosubsystems: one to handle the commu-nications among different sites, and an-other one to give agent management ser-vices.

• An operating system layer. This layeris represented by the real time operat-ing system installed in the industrial PC.This can be any real time operating sys-tem. In our case, an open source oper-ating system based on LINUX is used.

The interface layer and base layer can betreated as an unique layer. This unified layer(middleware) manages resources and servicesfor superior layer. The process agent mod-els the elements of the real world, the ap-plication agents execute specific automationtasks, and the middleware agents provide thebasic functionalities to manage a MultiagentSystem.

3.1 Functions in superior layer

Our automation system is a complex informa-tion system with automation tasks. In thissense, the automation system have to make,at least, these functions:

1. Monitoring a set of operation variablesin a dynamic way. This allows to detectand manage emergencies and productionproblems in the plant.

2. Management of the complex interactionsamong the operation variables to trans-form them into control commands.

3. Transmission of the control commandsto a set of actuators located in the plant.

4th WSEAS Int. Conf. on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS Miami, Florida, USA, November 17-19, 2005 (pp134-139)

4. Implementation of economic criteriasthat can be applied as control commandsor as a part of the plant programmingfunctions to improve productivity.

5. Reconfiguration of the control and pro-duction systems to fulfill the require-ments in one specific moment.

6. Giving information to management andoperations personnel about the plant sit-uation and the fabricated products.

7. Simplification of the production and op-eration data, as well as of the qualitydata of products, with the intention toform historical data bases that could beused like references to make engineeringor marketing decisions.

8. Scheduling of the production to reachthe user necessities, maintaining thehighest productivity to the smallest cost.In addition, the system must allow theplanning of the appropriate maintenancefunctions (preventive and corrective).

9. Determination of the inventory levelsand suitable use, as much for the materi-als, energy, spare parts, goods in processand products to reach wished economicand productions levels.

10. Providing of interfaces with the exter-nal actors that interact with the plantproduction system, such as accounting,trade, investigation, development andengineering, external transport, suppli-ers and salesmen, purchases, clients andcontractors, corporative managements,etc.

These functions can be distributed and ex-pressed through a hierarchic logical structure.Figure 1 shows an Automation Topology basedon Agents where the functionalities are dis-tributed with certain hierarchies and eachagent has an specific role.

In the superior layer there are two agentcommunities: Process Agents, Application

Agents. The community of process agents,that represents components and abstractionsof real processes (pump, oil well, machines,etc. ), is composed by agents based on thephysical division of the controlled processand the functional division of agent tasks.This functional representation of the pro-cesses through agents has the advantage ofexecuting intelligent tasks, for example, eval-uation of parameters and physical variables,performance comparison, etc. The processagents are distributed hierarchically so thetotal plant can be represent for an agent.Several sub-process agents represent, then,sub-systems of the plant. This descompo-sitional arrangement allows to support ab-straction of information, allowing to the pro-cess agents of superior level (or higher) toexchange refined knowledge of the process.The main function of the process agents isto maintain the knowledge about the state ofthe area under its responsibility. The agentsproduce a refined notion of the state, basedon the measured variables of their domain.

The community of application agents makespecific functions for industrial automation(visualization, supervision, control, optimiza-tion or any specialized function). Special-ized agents can require of wrappers. Thus,the application agents are conceived as anspecialized multi-agent system for coordina-tion, execution, and evaluation of control,supervision, optimization and administrationtasks, which is necessary in the prosecutionof the process information and the taking ofdecisions. Then, the application agents of-fer services to the process agent community.Both communities interact with the middle-ware through the interface layer.

3.2 Middleware Layer

The middleware is the basic group of softwaremodules that implant the minimum abstrac-tions for the specification, installation andmanipulation of agents and objects.

To assure the operability of the automationsystem, due to the heterogeneity and com-

4th WSEAS Int. Conf. on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS Miami, Florida, USA, November 17-19, 2005 (pp134-139)

plexity, the middleware proposal has a mul-tilayer focus, which is composed of homoge-neous layers for the operating system, com-munications management, and agents man-agement. In general, the middleware isformed by three layers: interface layer, baselayer and operating system layer. The In-terface layer is based on FIPA specification[4]. It is constituted by five agents: Agentfor Administration of Agents (AAA), whichis called Agent Management System in FIPA,Resource Management Agent (RMA), Ap-plication Management Agent (AMA), DataManagement Agent (DMA) and Agent forCommunication Control (ACC). The RMA,AMA and DMA agents are specializationsof the Directory Facilitator (DF) defined byFIPA. This layer manages the operations andcommunications in the MAS.

The base layer contains two subsystems:one to handle the communications amongdifferent sites, and other to give services tothe agents. It allows agent localization andassociation of agents with operating systemprocesses. Also, it gives distributed char-acteristics to the system implementing someproperties of these types of systems: migra-tion, interoperativity, naming, etc. The baselayer is formed by two modules: the AgentManager and the Communication Manager.The Agent Manager implements the differentservices required to create, delete and man-age agents, which are viewed as processes inthe computational platform. At this level,naming and migrate characteristics of agentsis implemented. The Communication Man-ager allows the communication among dif-ferent processes, in the same site or in differ-ent sites. The Communication Manager mustprovide reliable communication with the net-work oriented to invocation, and it is herethat interoperativity characteristic of agentsis implemented.

The Agent Manager is structured in threesub-modules: Dispatcher, Mapper and Loca-tor:

• The dispatcher module allows the invo-

cation among agents.

• The mapper module allows mapping ofagents in a process.

• The locator module allows the locationof the agents in the system.

Figure 3 shows the components of the mid-dleware layers. In order to accomplish alltheir activities, the agents of interface layermust invoke these modules.

Figure 3: Middleware Implantation Ap-proach

The operating system composes the low-est layer. Normally must be an operatingsystem that support real time. We proposeLINUX, which is a popular operating system,that is executed in most of the hardware ar-chitectures, and offers support for real timeand embedded devices based on firmware.LINUX can give a common layer to all au-tomation hardware.

LINUX, as well as most of the operat-ing systems, base their reliable remote com-munication on TCP/IP and UDP/IP proto-cols, that difficult to reach the performancerequirements and real time communicationconstraints. For that reason, we propose ad-ditional communication services: rendezvous,asynchronous messages, synchronous mes-sages and RPC messages. In all the cases,the communication is reliable and minimizesthe number of network packages transmitted.

4th WSEAS Int. Conf. on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS Miami, Florida, USA, November 17-19, 2005 (pp134-139)

4 Conclusion

In this paper we have proposed an automa-tion system based on agents. Our system ex-ploits the main features of the multi-agentsystems, like autonomy and intelligence ofagents, among others. In this way, we de-scribe the automation process like and intelli-gent distributed system that support the het-erogeneity and interoperativity of its compo-nents transparently.

Our system has three components, twoagent communities to describe the typical au-tomation processes and tasks, and a middle-ware to support industrial automation activ-ities. Specifically, our middleware give sup-port to automation systems based on agentsfollowing the FIPA standard.

The process agent community describesthe components of a process automation sys-tem, and the application agent communitydescribes the different automation tasks (forexample, optimization, visualization, etc.).This is an innovative approach to representautomation process models with large scala-bility and adaptability capabilities.

Our middleware is composed of layers,where each layer can be defined individually.The only requirement is the set of servicesthat each layer must provide to others. Thatis, each layer has a set of functionalities. Thelowest function is the kernel of the operatingsystem, the second one is the extension ofthe operating system to support distributedsystem, and finally, the last one manages themulti-agent system. This last one has beendesigned using the FIPA standard. In thisway, our middleware can support any multi-agent system, and has a great design versa-tility.

References:

[1] J. Aguilar, A. Rios, F. Rivas, O. Teran,L. Leon, and N. Perez, Definicion de do-minios y paradigmas en una arquitecturade automatizacion industrial, Tech. Re-

port 05-04, Fundacite-Merida, Merida,2004.

[2] Martin Albert, Thomas Langle, HeinzWorn, Michele Capobianco, and AttilioBrighenti, Multi-agent systems for in-dustrial diagnostics, XV IFAC WorldCongress (Barcelona, Spain), 2003.

[3] Alexei Bratoukhine, Yoseba Penya, andThilo Sauter, Intelligent software agentsin plant automation, Tech. report, Vi-enna University Of Technology, WienAustria, 2002.

[4] Foundation for Intelligent Physi-cal Agents, www.fipa.org.

[5] Nicholas Jennings and Stefan Buss-mann, Agent-based control systems,IEEE Control Systems Magazine June(2003), 61–73.

[6] Axel Klostermeyer, Revolutionisingplant automation the pabadis approach- white paper, Tech. Report IST-1999-60016, PABADIS: Information SocietyTechnology, Germany, 2003.

[7] Jim Pinto, Instrumentation & control onthe frontiers of a new millennium, Jour-nal Instruments & Control Systems 1(2000), no. 2, 78–90.

[8] Ilkka Seilonen, Teppo Pirttioja, andPekka Appelqvist, Agent technologyand process automation, Tech. report,Helsinki University of Technology, 2003.

[9] Thomas Wagner, An agent-oriented ap-proach to industrial automation sys-tems, Tech. report, Institute of Indus-trial Automation and Software Engi-neering, University of Stuttgart, 2002.

[10] , Applying agents for engi-neering of industrial automation sys-tems, MATES 2003 (M. Schillo et al.,ed.), Springer-Verlag, Berlin Heidelberg,2003, p. 6273.

4th WSEAS Int. Conf. on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS Miami, Florida, USA, November 17-19, 2005 (pp134-139)