APPLICATION OF REACTIVE AGENTS CONCEPTS

IN AUTOMATIC BINDINGS OF LONWORKS NETWORKS

DEVICES

Miguel dos Santos Alves Filho, Rafael de Aquino Cunha and Carlos Eduardo Cugnasca

Escola Politécnica, University of São Paulo, Av. Prof. Luciano Gualberto, travessa 3 nº 158, São Paulo, SP, Brazil

Keywords: LonWorks

, Multi-agents, Artificial Intelligence, Distributed systems, PBX, Telephony, Control networks.

Abstract: This paper aims to propose and test a new method to implement dynamic bindings in LonWorks

technology, allowing a new Distributed System of Private Telephone Switching (DSPTS), also developed

with LonWorks

technology, to make their telephone links. In order to do this, a method for developing

embedded systems and the reactive agent view was applied for each different device in this new system,

offering a unique, practical and innovative solution for both, LonWorks

and PBX systems. This view

allowed the implementation of intelligent and autonomous devices, specially in their internal process,

providing satisfactory and more efficient results based on the DSPTS requirements. This work is the kick-

off and the basis for developing new functions for telephone systems and control networks.

1 INTRODUCTION

In recent years, the paradigm of control systems

design has been changed, moving from the

traditional centralized architecture and proprietary

technology to distributed and open architectures

(Hur, Kim, Park, 2005)(Pu, Moor, 1998).

This change in paradigm takes the concept of

network control, and brings some benefits to

automation systems, such as reducing and

simplifying cabling, increasing reliability and

providing more options of manufacturers and

integrators for the user to chose from (Echelon

Corporation, 2007) (Hur, Kim, Park, 2005).

Aiming to exploit these advantages, a research

project is presented that proposes to develop a

Private Automatic Branch Exchange (PABX) (now

also called Distributed System of Private Telephone

Switching - DSPTS), with distributed architecture

and implemented with the control networking

technology LonWorks

. In this architecture, all

telephone extensions and devices for interfacing

with the telephone line are implemented as devices

(also called nodes) of control network that

communicate with each other, with the audio signals

(voice) digitized and transferred via messages.

The standard method for using such a technology

and the available tools to work with it provide only a

static way for the realization of logical connections

between devices (called bindings). It means that

these connections are defined during the setup

project and then will remain unchanged (Echelon

Corporation, 1999).

Considering that a PABX system must open and

close communication channels every time, it means

that the logical connections between nodes are not

pre-defined, but defined according to the need to

establish a telephone connection. Thus the use of

this control networks technology could not be

exerted in its original form.

The purpose is to propose and analyze a solution

based on the concepts of multi-agent systems to

develop devices of this distributed PABX, using the

LonWorks

technology, with self-management in a

previously configured environment, enabling the

creation of dynamic telephone links for the DSPTS.

The DSPTS characteristics of a distributed

application stimulated the use of the multi-agent

system concepts in its implementation (Durfee;

Rosenchein, 1994).

1.1 LonWorks

Technology

LonWorks

emerged in the 1990s, developed by

Echelon. This is a technology for automation

networks, involving not only a communication

258

dos Santos Alves Filho M., de Aquino Cunha R. and Eduardo Cugnasca C. (2009).

APPLICATION OF REACTIVE AGENTS CONCEPTS IN AUTOMATIC BINDINGS OF LONWORKS NETWORKS DEVICES.

In Proceedings of the 6th International Conference on Informatics in Control, Automation and Robotics - Intelligent Control Systems and Optimization,

pages 258-264

DOI: 10.5220/0002215402580264

 SciTePress

protocol, but also all the necessary infrastructure for

the system development such as management tools,

product development tools and network

configuration tools (Echelon Corporation, 1999).

Its protocol was developed based on the OSI

reference model, from the link layer to the session

layer. It is an open and standardized protocol, ANSI /

EIA 709.1, also known as LonTalk (Echelon

Corporation, 2002).

An emphasis point of this protocol is its

messages structure, which allows direct messaging

between devices, broadcast messages and multicast

messages. Furthermore, it provides system

management messages, allowing a greater control of

devices and even loads and develops programs by

the network itself (Echelon Corporation, 2002).

In the implementation of control networks

various media types can be used, as well as twisted

pair, power network, optic fiber and wireless

communication, which makes the technology very

flexible (Echelon Corporation, 1999) (Chermont,

2007).

The LonWorks

control network nodes may be

treated as objects that communicate through

interfaces, also called functional profiles, which

have network variables of the device in question.

Through these variables, information may be shared

with other devices (Xie, Pu, Moore, 1998).

The communication between the nodes is made

through links between their network variables, a

process called binding. From these, logical

connections are used that allow messages to be sent

whenever there is a change in the network variables

values (Echelon Corporation, 1999).

Besides the LonWorks

technology design,

Echelon has developed a chip called NeuronChip

that is a full implementation of the LonTalk and

provides the basis for any manufacturer to develop

its products.

The NeuronChip consists of three processors

working in parallel, two of them for the treatment of

the communication protocol and the last one for the

application program execution (Motorola, 1997).

The use of NeuronChip allows developers to

treat the network variables as program variables,

aside from the whole network communication,

developing the programs associated with the

application (Motorola, 1997). A specific

programming language was developed for that

technology, called NeuronC, which follows the

ANSI C and is completely based on events (Echelon

Corporation, 1995a).

1.2 Reactive Agents

Agents are autonomous entities that work together to

achieve the same goal, being able to interact and to

organize efficiently (Liviu, Sean, 2004).

They can also be perceived as real or virtual

automatons that have knowledge of the environment

in which they are inserted, and are able to perceive

changes in that environment. They have knowledge

about other agents, being able to communicate, to

learn, to infer, to form groups or even to reject.

Finally, they are able to decide, through its own

observations, goals, knowledge and interactions

(Wooldridge, Jennings, 2002).

The most common way to classify these agents

divides them into two groups, considering the

deliberation capacity, the environment perception

and communication complexity: one is called

cognitive agent and the other, reactive agent.

The cognitive agents are characterized by having

very complex functions and by having models that

require high processing capacity (Wooldridge,

Jennings, 2002).

There are different internal architectures that can

be used for the development of such agents, being

one of the most used architecture known as “Beliefs,

Desires, Intentions” (BDI) (Rao, 1996) (Rao;

Georgeff, 1991). Its use is considered in systems that

require the ability to exchange complex information,

to own and build complete models of the world

where they belong, their own and other agents, thus

acting spontaneously, creating an organization that

serves a purpose common to all of them (Steels,

1990).

A good example of the use of cognitive agents is

presented by Bigham (2003), which consists of an

antennas chain for mobile phones, each one with a

manageable coverage area, so that intersections

between them can be done and undone.

Within the group of reactive agents, there are

those who are able to understand (though in a fairly

limited way) and to react by acting on the

environment in which they are inserted, through a

pre-defined logic and always with a final goal that

was set in project phase.

A striking feature of this model type is its simple

communication way, which often occurs indirectly,

through the environment itself (Wooldridge;

Jennings, 2002). One advantage of these agents is

the ease of implementation, which may be based on

devices with less processing capacity and great

limitation of energy.

As examples of its use, the study of the behavior

of insects and their development processes may be

APPLICATION OF REACTIVE AGENTS CONCEPTS IN AUTOMATIC BINDINGS OF LONWORKS NETWORKS

DEVICES

259

mentioned, as shown in Liviu and Sean (2004), or

control network devices with low processing

capacity. Finally, this class of agent is used in

systems in which intelligence is expected to arise

from the society overall behavior and not from each

individual (Steels, 1990) (Castelfranchi, 1998)

(Boissier; Demazeau, 1994).

It is worth mentioning that the software agents

are used in systems known as multi-agents systems

or distributed problem resolutions, which have as

one of their characteristics the lack of centralized

control, which fall within one of the DSPTS

requirements as well their simplicity, allowing their

implementation in the NeuronChips (Durfee;

Rosenchein, 1994).

1.3 The DSPTS

One of the most important systems for any

enterprise is the voice communication system.

Currently, these systems tend to be implemented

with equipment called PABX, which are nothing

more than core private telephony, which enables

internal communication between all employees who

have access to branches, and communication

between them and the outside world (external links).

One important feature of this equipment is its

centralized architecture, in which all telephone

extensions and external lines must be connected to a

central device, as illustrated in Figure 1. Among the

main limitations of this architecture there is the limit

of the expansion in the extensions number and

external lines, and the need for a large amount of

cabling, which makes it not very flexible to install

and to change the branches positions.

Figure 1: Typical PBX architecture.

The research presented in this article proposes a

new architecture for the implementation of such

systems, using the technology for control networks

LonWorks

, allowing to remove the mentioned

disadvantages of typical PABXs, as well the

reduction and simplification of cabling, flexibility

for changes, incremental growth (and associated

investments), and greater system reliability (gaps in

some nodes do not prevent the functioning of the

rest of the system).

Moreover, this project aims to add greater

intelligence to these systems, solving problems such

as call direction, to decide which carrier should be

used according to the call type, extensions calls

prioritization to access the outside line, etc.

Figure 2 illustrates the architecture of the

proposed system, in which the following DSPTS

components are present:

• TLM – Trunk Line Module;

• E1M – E1 Module;

• ARM – Audible Response Module;

• DTM – Digital Telephony Module.

Figure 2: DSPTS distributed architecture.

Other features that motivated the choice of the

LonWorks

technology was the possibility of the

object oriented programming and the low jitter in

message exchange, due to the robust features of the

communication protocol.

For the voice message exchange, the project was

specified for the use of voice compression

algorithms, aimed at telephony. In this case, the

G.729 ADPCM algorithm was used, which provides

a good quality for telephony voice communication

and uses a low bandwidth (16 kbps per

communication channel).

Just as occurs in conventional PBX, it is

necessary for the DSPTS to close communication

channels, with the difference that in the first case the

link is physical and in the second it is logical.

In the LonWorks

technology, this logical link is

performed by means of bindings between the

variables of two devices that, as reviewed above, are

built with the tools in the market, during the system

configuration. This makes these links static,

preventing the intended application.

Thus, the possibility to perform bindings

dynamically is fundamental for the DSPTS because

that way two devices could exchange messages

during a phone call, after which they would be free

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

260

to communicate with other devices and participate in

other connections.

2 MATERIALS AND METHODS

The nodes developed have low processing capacity

and low memory amount, and because it is a

research project, in which new requirements can

always be discovered during the evaluation of

prototypes, an incremental development

methodology has been adopted, but devised for

smaller equipment.

Sequence diagrams were also used, quite

common in program development and data flow

diagrams, more used in projects with

microcontrollers.

Figure 3 shows a representation of the equipment

used in the system development, as well as the

physical interconnection between them. In this

assembly, an outside line Public Switched Telephone

Network (PSTN) was used, connected to the TLM to

make external calls.

Figure 3: Experimental assembly.

NodeBuilder and LonMaker softwares were used

for the program development. Both are provided by

Echelon. LonTalk Protocol Analyzer (LPA)

software, developed by Loytec, was also used

(Loytec, 2008).

The TLM and DTM devices were developed

with NeuronChip microcontroller, and meet the

following project requirements:

• low overhead communication: to reduce the

data traffic amount on the network, leaving

a greater free bandwidth for the voice

messages traffic.

• to use of the smallest number of entries in

the tables used for device configuration: the

device aims to keep compatible to be

applied to other control network features.

• to use a telephony compatible time to the

bindings configuration (or closing the

phone link): avoid discomfort to the user,

and maintain compatibility with already

installed PABX systems.

• to dispense human operation and maintain

the device full independence: ensures that

the system does not require constant

maintenance and keeps their integrity for

long periods.

As a way to evaluate the solution proposed, the

LonWorks

network analyzer was used to measure

the total time consumed by the dynamic binding

process proposed, comparing with the standard

binding process, and checking whether the result

meets the time requirements of the project, that is

one second (to meet the standards of Brazilian

telephony).

Several tests were also conducted, involving

external calls request and calls receiving, forcing the

process to be performed many times.

3 PROPOSED MODEL

As a way of meeting the specifications mentioned in

the previous item, an extra software layer was added

to each device, including the functionality of the

device, and representing the reactive agent, as

illustrated in Figure 4:

Figure 4: Reactive agent layer.

APPLICATION OF REACTIVE AGENTS CONCEPTS IN AUTOMATIC BINDINGS OF LONWORKS NETWORKS

DEVICES

261

To facilitate the solution development, the

problem was divided into three stages: the first one

includes the handshake and the decision parameters

to be used in the binding configuration, the second is

to achieve the proper bindings, and the third stage

involves the process for undoing the bindings,

leaving the devices available for new connections.

3.1 Stage 1

To perform this stage, the device must be in an

initial waiting state, in which the agent is active and

attentive to any requests for connections from other

agents.

When the device receives a call or request for

assistance, the agents will exchange direct messages

between themselves, adjusting the parameters

needed to achieve the bindings.

This stage is completed when both have all the

settings necessary for the binding to be achieved.

3.2 Stage 2

With the parameters adjusted, they should begin the

configuration process. At this point, it is worth

mentioning that each agent is responsible for

conducting its own configuration process, not

interfering with the process of another agent.

If there is any failure in the execution of this

stage, the device sends a message of failure to the

other agent, indicating that the process should be

canceled.

After the setup process, they perform a check to

ensure it was successful and that the communication

can be initiated.

With a positive response of the final assessment,

the exchange messages of telephony are initiated,

including the whole call progress process and voice

message exchange. At this time, the agent gets away

from control, passing it to the basic operating

software of the device.

3.3 Stage 3

When the call is finished, the agent takes back the

control and performs all the configuration of the

device, so that the bindings are undone and the

device is ready to perform a new call.

At this stage, the communication between agents

is not necessary, because it involves only the

configuration of the internal tables of each device.

4 RESULTS ANALYSIS

The stage 1 obtained average time was

approximately 23 ms. This time is due to the need of

several message exchanges between the devices for

the parameters to get adjusted.

Furthermore, depending on the case, it is

necessary for the settings table to get swept several

times, until the agents enter into a consensus about

which values can be used.

The traditional method of conducting bindings,

made from the LonMaker tool is naturally much

faster, since it does not require the exchange of

messages between devices and the configuration tool

already knows the free values that can be used.

Various failures simulations were also

performed, such as loss of communication during

the process, and all tests showed that the solution is

effective in failure recovering and left the devices

ready for another attempt.

In stage 2, the average time needed to perform

the required configurations was approximately 602

ms.

This time was measured from the receipt of the

message containing the configuration values

adjusted between the devices and the first call

progress message.

This time is justified by the fact that the device

must make all the configuration parameters in these

tables that are stored in the EEPROM memory area,

a memory type that demands a longer time for

completing records.

When performing the same procedure with the

LonMaker, the average time obtained was 1,983 ms.

This long time is due to the fact that besides having

to write data into EEPROM memory, all the

configuration and verification is conducted through

the control network, with device management

messages.

Finally, the results analysis for stage 3 did not

take into consideration the completion time, but if

the devices were in a waiting state, as specified

before.

After this step, one can verify that the bindings

were completely undone and the devices were free

to receive new connections. Figure 5 shows the

devices tables at the end of stage 2.

The tables below, containing binding information

between the device TLM (MLT) and DTM (MTD),

show that the telephone link between them was

made.

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

262

Figure 5: Table setup with bindings between devices.

Figure 6 shows the same table at the end of stage 3.

Figure 6: Configuration tables at the end of stage 3.

As seen, all the configuration entries have been

deleted and, therefore, the devices are free for new

connections.

5 CONCLUSIONS

The total mean obtained in completing the telephone

link process, from the beginning of the stage 1 to the

end of the stage 2, was approximately 625 ms, which

is considered a good result for telephone systems.

This result means that after the user enters a phone

number or make a request for external line, he will

hear the calling signal or external line tone signal

after about 625 ms, time barely noticeable by the

user.

At the end of stage 3, the devices had been

effective in returning to the original state and ready

to receive new calls.

From the point of view of multi-agents systems,

a reactive agent was applied with reservations to the

communication means used.

A characteristic of reactive agents is to have a

primitive method of communication, often using

only the very environment in which they are inserted

(Wooldridge, Jennings, 2002). In this solution, the

agents exchanged messages by a complex network

protocol.

However, these agents did not have any model

type: of the environment model, of other agents, and

of their own model. They were not able to make

inferences, to perform learning or to seek

knowledge.

They only reacted to external stimuli as a

previously implemented logic that could not be

changed at runtime, which are features of a reactive

agent (Wooldridge; Jennings, 2002).

Through the application agents’ concepts, the

presented solution meets the requirement of

maintaining the devices autonomy, since they do not

depend on external commands to configure

themselves.

Also, no human intervention was required during

the process execution, no form of communication

used during the exchange of voice messages is

LonWorks

standard and no overhead was added to

the protocol. Only the minimum necessary

parameters were used for the bindings configuration.

Thus, the conclusion is that all project

requirements were met.

As an improvement to consider, there is the

NeuronChip memory implementation. As

LonWorks

technology considers that bindings are

permanent, their settings are stored in an EEPROM

area, which may support approximately 1,000,000

recordings (Motorola, 1997). For example, assuming

that under heavy use 50 calls are made per day from

a certain branch, which corresponds to 100

recording procedures in the EEPROM memory, the

durability of the equipment would be approximately

27 years.

6 FUTURE WORK

In the context of the DSPTS project, one can

imagine the possibility of applying cognitive agents

APPLICATION OF REACTIVE AGENTS CONCEPTS IN AUTOMATIC BINDINGS OF LONWORKS NETWORKS

DEVICES

263

to solve greater complexity problems and that may

require a system with greater integrity and decision

capability.

As example of such problems, there is the

decision of which attendant branch will take an

incoming call for an installation where there are

many attendants’ branches. This could be solved

through the use of auctions with pre-established

metrics, method widely used in cognitive agents

societies (Benisch et al., 2004).

For a broader case that involves the entire

LonWorks

technology, the implementation and

evaluation of new hardware for the network nodes

can be proposed, enabling the storage of bindings

information in RAM. This would remove the

restrictions associated with the EEPROM memory,

allowing faster bindings and a longer system life

time.

A natural DSPTS continuity suggestion is related

to the design of building automation systems based

on LonWorks

technology that integrate the PABX

functions.

ACKNOWLEDGEMENTS

The authors wish to thank the Fundação de Amparo

à Pesquisa do Estado de São Paulo (FAPESP),

project number 05/56045-6, and the company

Conceito Tecnologia, for the support granted to this

research.

REFERENCES

Benisch, M., Greenwald, A., Grypari, i., Lederman, R.,

Naroditskiy, v., Tschantz, M., Botticelli. A Supply

Chain Management Agent. Proc. 3

. Int. Conf. On

Autonomous Agents and Multi Agent Systems

(AAMAS´04), ACM Press (New York), p. 1174-1181,

2004.

Bigham, J., Du, L. Cooperative Negotiation in a Multi-

Agent System for Real-Time Load Balancing of a

Mobile Cellular Network. In: Proceedings of the

Second International Joint Conference on Autonomous

Agents and Multiagent Systems, 2., 2003, Melbourne

(Australia). International Conference on Autonomous

Agents. Nova York: ACM, 2003. p.568-575.

Boissier, O., Demazeau, Y. Asic: An Architecture for

Social and Individual Control and its Application to

Computer Vision. Modeling Autonomous Agents in a

Multi-Agent World, v. 1069, 1994.

Castelfranchi, C. Modeling Social Action for AI Agents.

Artificial Intelligence, Elsevier Science Publishers

Ltd, v. 103, n. 1, p. 157-182, 1998.

Conceito Tecnologia. The company's website containing

information on the LonWorks

technology and some

examples of application. Available at:

<http://www.conceitotecnologia.com.br>. Access at:

12 feb. de 2008.

Durfee, E. H., Rosenchein, J. S. Distributed Problem

Solving and Multi-Agent Systems. Proceedings of the

International Workshop on Distributed Artificial

Intelligence, 1994.

Echelon Corporation. The company's website containing

information on the LonWorks

technology. Available

at: <http://www.echelon.com>. Access at: 17 de jun.

de 2007.

Estados Unidos. Echelon Corporation. Neuron C

programmer’s guide. Palo Alto, 1995, 1 v.

Estados Unidos. Echelon Corporation. Introduction to the

LonWorks

System, Palo Alto, 1999. 1 v. (078-0183-

01A).

Estados Unidos. Echelon Corporation. LonMark Layer 1-6

Interoperability Guidelines, Palo Alto, 2002. 1 v.

Estados Unidos. Echelon Corporation. LonMaker User’s

Guide, Palo Alto, 2003. 3.1 v. (078-0168-02E). (a)

Estados Unidos. Echelon Corporation. NodeBuilder

User’s Guide. Palo Alto, Echelon, 2003. 1 v. (b)

Estados Unidos. Motorola. LonWorks

Technology

Device Data, rev. 4, 1997.

Liviu, P., Sean, L. A Pheromone-Based Utility Model for

Collaborative Foraging. Proc. 3

. Int. Conf. On

Autonomous Agents and Multi Agent Systems

(AAMAS), p. 36-43, 2004.

Lonmark International. Site of standardized organ for

LonWorks technology and products certification.

Available at: <http://www.lonmark.org/>. Access at:

19 jan. 2008.

Loytec. The company's website. Available at:

<http://www.loytec.com/english/products/lcore.htm>.

Access at: 19 jan. 2009.

Rao, A. S. AgentSpeak(L): BDI Agents speak out in a

logical computable language. Australian Artificial

Intelligence Institute, 1996.

Rao, A. S., Georgeff, M. P. Modeling Rational Agents

within a BDI-Architecture. Proceedings of the 2nd

International Conference on Principles of Knowledge

Representation and Reasoning (KR'91), USA, p. 473-

484, 1991.

Shoham, Y. Agent-Oriented Programming. Journal of

Artificial Intelligence, n. 60, p. 51-92, 1991.

Steels, L. Cooperation between distributed agents through

self-organization. Intelligent Robots and Systems '90.

'Towards a New Frontier of Applications',

Proceedings. IROS '90. IEEE International Workshop,

Japan, p. 8-14, 1990.

Wooldridge, M., Jennings, N. Introduction to Multi-Agent

Systems. USA: JASSS, 2002.

Xie, C., Pu, J.-S., Moore, P. R. A case study on the

development of intelligent actuator components for

distributed control systems using LONWORK neuron

chips. Mechatronics v. 8, n. 2, p. 103-119, 1998.

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

264