CPNT-Jade Framework

Developing Agent-based Control of Discrete Event Systems

Andrzej Bożek and Marian Wysocki

Department of Computer and Control Engineering, Rzeszow University of Technology,

al. Powstańców Warszawy 12, Rzeszów, Poland

Keywords: Multi-Agent Systems, Jade, CPN Tools, Colored Petri Nets, HTCPN.

Abstract: A framework built as a combination of CPN Tools software and JADE (Java Agent DEvelopment Frame-

work) platform has been proposed. CPN Tools makes it possible to model a controlled system with the use

of hierarchical timed colored Petri nets (HTCPN), a powerful formalism for modeling complex discrete

event systems. The framework supports real-time mode of simulation of the target system model. Issues of

communication and time synchronization between the two software components are transparent to a user.

An example of the framework use has been presented. It refers to a dynamic flexible job shop manufactur-

ing system for which an agent-based control has been developed.

1 INTRODUCTION

Multi-agent systems represent important kind of

intelligent software architecture characterized by

distributed implementation, mobility and scalability.

A multi-agent control system consists of controlled

(sub)system, e.g. a business process, a factory plant,

road traffic, etc., and an agent-based software

(sub)system (Candido and Barata, 2007). A devel-

opment phase is needed for every software system.

In this stage new features are added and tested.

Therefore, a connection between the developed con-

trol system and the controlled system, real or simu-

lated, is required.

The presented solution refers to the scenario

where a computational model of the controlled sys-

tem is used. There exist many formalisms and soft-

ware tools for simulation of continuous or discrete

dynamic systems. In this work the discrete event

systems are the focus of interest. Petri net formalism

with high-level extensions has been selected for the

framework implementation as the powerful tool for

modeling and simulation of discrete event systems.

The framework has been composed of two parts.

The first part simulates discrete system with the use

of a Petri net executable model. The second part is a

multi-agent system that performs a control algo-

rithm. The framework provides transparent bidirec-

tional communication between the two subsystems.

The discrete system model and the multi-agent con-

trol system are built during the development phase.

After that, the developed agent-based control system

is ready to be reconnected to a real-world target

environment.

An exemplary solution based on the framework

has also been presented in this work. It refers to a

control of production process in a screw factory that

has a form of a dynamic flexible job shop problem.

2 RELATED WORK

Many frameworks and generalized architectures

related to multi-agent systems have been proposed.

There exist multi-agent frameworks dedicated for

specific domains. Albashiri and Coenen (2009) have

proposed EMADS framework (Extensible Multi-

Agent Data Mining System). Vokřínek, Komenda

and Pěchouček (2011) have developed abstract ar-

chitecture of a task-oriented multi-agent problem

solver. MABC (Multi-Agent Based Clustering)

framework has been developed by Chaimontree,

Atkinson and Coenen (2011). Hu, Du and Spencer

(2011) have developed Aframe framework for Am-

bient Systems. An agent-based framework that sup-

ports implementation of IPPS system (Integrated

Process Planning and Scheduling) has been pro-

posed by Li, Zhang, Gao, Li and Shao (2010). FAST

(Flexible and Adaptive Scheduling Tool) is a toolkit

184

zek A. and Wysocki M..

CPNT-Jade Framework - Developing Agent-based Control of Discrete Event Systems.

DOI: 10.5220/0004218001840189

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 184-189

ISBN: 978-989-8565-38-9

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

that provides a powerful agent-based framework to

develop flexible, fault tolerant and scalable schedul-

ing systems (Garcia, Valero, Argente, Giret and Julian,

2008). Liu, Abdelrahman and Ramaswamy (2007)

have proposed a multi-agent framework for robust

and adaptable dynamic job shop scheduling. Using

of multi-agent solution in manufacturing process

planning, scheduling and control has become more

and more common recently. The work of Shen,

Wang and Hao (2006) is the good state-of-the-art

survey on this subject.

The Petri net formalism is also used for devel-

opment of multi-agent systems. Miranda and

Perkusich (1999) have presented the modeling, anal-

ysis and verification of MATHEMA multi-agent

system by means of colored Petri nets. Weyns and

Holvoet (2002) have proposed using colored Petri

net for modeling and analysis of an entire multi-

agent system. They have chosen Design/CPN (pre-

decessor of CPN Tools) software and Packet-World

as a case study. The same authors (Weyns and Hol-

voet, 2004) have developed a colored Petri net dedi-

cated for regional synchronization in a multi-agent

system.

The common feature of the three aforementioned

works is that the Petri net formalism has been used

for modeling and analysis of multi-agent systems. In

this work the Petri net formalism has been used for

modeling and simulation of an external environment

of the multi-agent system that substitutes a target

environment during the development phase and does

not need to be implemented as a multi-agent struc-

ture.

3 FRAMEWORK

ARCHITECTURE

3.1 Components of the Framework

The framework is based on two software packages.

The first one is CPN Tools (CPN Tools Home Page,

2012), the environment for modeling, simulation,

analysis and verification of discrete event systems

with the use of hierarchical timed colored Petri nets

(HTCPN) formalism (Jensen and Kristensen, 2009).

The most important reasons of this selection are:

 Petri nets are convenient formalism for struc-

tural representation of typical features and be-

haviors of discrete event systems, e.g. concur-

rency, synchronization, resources sharing, mu-

tual exclusion.

 The high-level extensions are very useful. The

hierarchy allows to decompose a structure of a

modeled system into nested modules. Colored

tokens carry information that can be repre-

sented by complex data types and can be pro-

cessed by arc expressions. The time extension

allows to model time relations between dis-

crete events.

The second package is JADE (JADE Home

Page, 2012). The most important reasons of JADE

platform selection are:

 Java is classified as one of the most portable

programming languages. A multi-agent sys-

tem can be developed on one platform and

then can be moved to another target platform.

The paradigm of agents mobility is also well

supported by the Java language portability.

 JADE is compliant with FIPA (The Foundation

for Intelligent Physical Agents) (FIPA Home

Page, 2012) specifications and supplies im-

plementations of standards defined by this

specifications.

Both of the software packages are free and open

source. Because of the names CPN Tools and JADE,

the proposed framework has been named as CPNT-

JADE framework.

3.2 Framework Structure

The framework structure is presented in Figure 1.

Both Petri net structure components and multi-agent

system components are divided into two sets, name-

ly framework components and solution components.

Framework components constitute the base part of

the framework that cannot be changed. The solution

components are project-related elements.

CPN Tools and JADE are connected by TCP/IP.

Information frames exchanged between applications

have simple but usable form of list of messages.

Each message is a pair (address, data). On

JADE side, there are five agents that have been pre-

defined for CPNT-JADE framework. The agents are

described in Table 1. A developer that uses the

framework has to implement class of Configurator

agent and class of DispatchingPolicy object. Dis-

patchingPolicy object defines how to translate each

message received from CPN Tools into an instance

of FIPA compliant ACLMessage routed to other

agents and vice versa.

The framework supports the real-time mode of

simulation of a target system model. This mode is

provided by Real-time Mode Subsystem. In the real-

CPNT-JadeFramework-DevelopingAgent-basedControlofDiscreteEventSystems

185

Figure 1: The structure of the CPNT-JADE framework.

time mode timestamps of HTCPN model are

mapped into real time intervals. Thus, model param-

eters include data that defines locations and relations

of events in real time axis.

Table 1: The predefined agents of the CPNT-JADE

framework.

Agent class

name

Description

Initiator

Created by JADE. It is responsible for

creation other predefined agents.

Cpnt-

Connector

Performs low-level TCP/IP communica-

tion with CPN Tools and translates data

between low-level frames and high-

level messages.

Message-

Dispatcher

Redirects messages between CpntCon-

nector and other agents. Routes are

defined on the basis of messages ad-

dresses and rules defined by Dispatch-

ingPolicy object.

Time-

Synchronizer

Optional, needed if the framework runs

in the real-time mode. Details are given

in subsection 3.3.

Configurator

Created by the framework but its behav-

ior is solution-specific and has to be

defined by a user. This agent creates

and administers all solution-related

parts of the multi-agent system.

4 EXAMPLE – DEVELOPING

A PRODUCTION CONTROL

SYSTEM

4.1 Problem Formulation

As an example solution, a prototypical production

control system is presented. Organization of shop-

floor level production in the factory can be classified

as a kind of dynamic flexible job shop manufactur-

ing system (Rajabinasab and Mansour, 2011), with

detailed characteristics as follows:

 Jobs arrive continuously in time. No parameters

of a job are known until it arrives.

 Arrival (release) date r

and due date d

are de-

fined for each job J

 A non-empty sequence of operations (O

1,k

, O

2,k

..., O

o(k),k

) is defined for each job J

 A non-empty set of machines {M

1,i,k

, M

2,i,k

, ...,

m(i,k),i,k

} is defined for each operation O

i,k

any machine can alternatively execute the op-

eration.

 A processing time p

m,i,k

is defined for each pair

machine-operation O

m,i,k

 Each machine can perform at most one opera-

tion at a time.

CpntConnector

Message-

Dispatcher

Dispatching-

Policy

Time-

Synchronizer

Initiator

TCP/IP

Configurator

Solution-

agents

Framework ComponentsSolution

Components

Solution

Components

Petri Net Structure (CPN Tools) Multi-agent System (JADE)

Real-time Mode

Subsystem

Legend

X Y

Legend

Petri Net Place

Petri Net

Substitution

Transition

Petri Net Arc

Agent

Agent X

Communicates

with Agent Y

Agent X

Creates Agent Y

Java Object

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

186

Figure 2: The structure of the prototypical dynamic production scheduling system.

The production in the screw factory is partially

automated. Machines are equipped with embedded

PC controllers (Żabiński and Mączka, 2012). These

controllers are connected in a network and each one

can be a base for a container of a multi-agent plat-

form.

4.2 HTCPN Plant Model

The structure of the prototype production control

system is presented in Figure 2. The left part repre-

sents a Petri net plant model based on the HTCPN

formalism. The model consists of four pages (sub-

structures): Adpt (JADE Adapter), DFJS (Dynamic

Flexible Job Shop), Machine, Process. DFJS is the

main page of the model. It models dynamic flexible

job shop system structure on high-level. The Ma-

chine subpage represents features of a real machine

important for the prototyped solution. Its implemen-

tation is presented in Figure 3. The modelled proper-

ties are as follows:

 Progress of operations is simulated (place coun-

ter).

 A machine can be stopped over arbitrarily de-

fined time intervals (place brkdwns).

 Start/stop/breakdown events are registered

(place machEvt).

 The multi-agent subsystem can control a queue

of processed operations (place ctrIn).

In the real production environment PLC control-

lers have a connection with item counters and the

progress is delimited by comparison of the numbers

of produced and planned items. PLC controllers also

4.3 Agent-based Control System

The agent-based control system is presented in the

right part of Figure 2. Configurator agent creates all

other solution-related agents, but this is not shown in

Figure 2 to keep it readable.

There are M MachHdw (Machine Hardware) and

Machine agents in the system, one pair of the agents

for each machine. There are N Job agents in the

system, where N is the number of currently pro-

cessed jobs. The control is performed by choosing

one operation in advance for each machine.

Selecting of operations for processing is main

behavior of the Machine agent. When a discrete state

of the production system changes, i.e. a new job

Yellow Pages

CPNT-JadeFramework-DevelopingAgent-basedControlofDiscreteEventSystems

187

Figure 3: The machine subpage of the HTCPN model of the production system.

arrives or some operation finishes, a special message

is broadcasted to all Machine agents. After receiving

this message, each Machine agent checks whether it

needs to reserve a next operation for processing. If

so, the agent starts the JobSelection interaction pro-

tocol.

The sequence diagram of this protocol is pre-

sented in Figure 4. The JobSelection protocol has

been patterned after FIPA-CNIP (contract net inter-

action protocol) (FIPA Home Page, 2012). A sched-

uling decision is made by Machine agent after re-

ceiving operationParameters messages from all

asked Job agents. The parameters represent expected

start and stop times of waiting operations. The Ma-

chine agent compares parameters of waiting opera-

tions and selects at most one operation. The

RuleProvider agent stores selection criteria. For

tests, several typical scheduling dispatching rules

have been implemented: first-come first-served

(FCFS), earliest arrival date (EAD), earliest due date

(EDD) and critical ratio (CR - ratio of the remaining

processing time to the allowance time of a job).

When a machine breaks down, an operation re-

served for this machine is removed from the pro-

cessing queue and marked as unreserved by related

Job agent. Thus, it is possible that the operation will

be reserved and processed by other machine. It is a

form of automatic rescheduling.

5 CONCLUSIONS

The main goal of the presented work is to propose a

solution that can be helpful in developing agent-

based control systems designed for real-world appli-

cations. A user of the framework should at first build

a model of a controlled system using HTCPN for-

malism. Then, the user can develop a multi-agent

control system with the use of JADE. When the

control system is fully ready or safe enough it can be

excluded from the framework and moved to a target

environment.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

188

Figure 4: The sequence diagram of the JobSelection inter-

action protocol.

The prototypical agent-based control system,

presented as an example, is a subject of continuous

development and improvement. The HTCPN model

is intended to be augmented with additional features,

as representation of setup and transport times, limi-

tation of resources, etc. New functionalities are go-

ing to be added to the control system, among other

things, a possibility of interactive changing of opera-

tion sequences by machine operators, scheduling

many operations in advance for each machine, coop-

eration with a long-term planner.

The work has been supported by the program of

the Polish Ministry of Science and Higher Education

for young scientists (U-8603/DS/M).

REFERENCES

Albashiri, K., A., Coenen, F. (2009). A Generic and Ex-

tensible Multi-Agent Data Mining Framework. Hybrid

Artificial Intelligence Systems. LNCS, 5572, 203-210.

Candido, G., Barata, J. (2007). A Multiagent Control

System for Floor Assembly. HoloMAS 2007, 4659(9-

12), 293-302.

Chaimontree, S., Atkinson, K., Coenen, F. (2011). A

framework for Multi-Agent Based Clustering. Auton

Agent Multi-Agent Syst, 25(3), 425-446.

CPN Tools Home Page (2012). Retrieved November 5,

2012, from http://cpntools.org

Miranda, M., C., Perkusich, A. (1999). Modeling and

Analysis of a Multi-Agent System using Colored Petri

Nets, Workshop on Applications of Petri Nets to Intel-

ligent System Development. Williamsburg.

FIPA Home Page (2012). Retrieved November 5, 2012,

from http://www.fipa.org

Garcia, M., E., Valero, S., Argente, E., Giret, A., Julian, V.

(2008). A FAST Method to Achieve Flexible Production

Programming Systems. IEEE Trans. Syst. Man Cybern.

C, Appl. Rev., 38(2), 242-252.

Hu, X., Du, W., Spencer, B. (2011). A Multi-Agent

Framework for Ambient Systems Development. The

2nd International Conference on Ambient Systems,

Networks and Technologies. Procedia Computer Sci-

ence, 5, 82-89.

JADE Home Page (2012). Retrieved November 5, 2012,

from http://jade.tilab.com

Jensen, K., Kristensen, L.M. (2009). Coloured Petri Nets.

Modelling and Validation of Concurrent Systems. Ber-

lin: Springer.

Li, X., Zhang, Ch., Gao, L., Li, W., Shao, X. (2010). An

agent-based approach for integrated process planning

and scheduling. Experts Systems with Applications,

37(2), 1256-1264.

Liu, N., Abdelrahman, M., A., Ramaswamy, S. (2007). A

Complete Multiagent Framework for Robust and

Adaptable Dynamic Job Shop Scheduling. IEEE

Trans. Syst. Man Cybern. C, Appl. Rev., 37(5), 904-

916.

Rajabinasab, A., Mansour S. (2011). Dynamic flexible job

shop scheduling with alternative process plans: an

agent-based approach. Int. J. Adv. Manuf. Technol. 54,

1091-1107.

Shen, W., Wang, L., Hao, Q. (2006). Agent-Based Dis-

tributed Manufacturing Process Planning and Schedul-

ing: A State-of-the-Art Survey. IEEE Trans. Syst. Man

Cybern. C, Appl. Rev., 36(4), 563-577

Vokřínek, J., Komenda, A., Pěchouček, M. (2011). Ab-

stract Architecture of Task-oriented Multi-agent Prob-

lem Solving. IEEE Trans. Syst. Man Cybern. C, Appl.

Rev., 41(1), 31-41.

Weyns, D., Holvoet, T. (2002). A Colored Petri Net for a

Multi-Agent Application. Second Workshop on Mod-

elling of Components, Objects and Agents, MOCA’02.

Aarhus.

Weyns, D., Holvoet, T. (2004). A Colored Petri Net for

Regional Synchronization in Situated Multi-Agent

Systems. Proceedings of First International Workshop

on Petri Nets and Coordination. Bologna.

Żabiński, T., Mączka, T. (2012). Implementation of Hu-

man-System Interface for Manufacturing Organiza-

tions. Advances in Intelligent and Soft Computing, 98,

13-31.

CPNT-JadeFramework-DevelopingAgent-basedControlofDiscreteEventSystems

189