Modeling of manufacturing execution indisperse productive systems using service

oriented technique

Caio C. Fattori ∗ Fabrıcio Junqueira ∗∗

Diolino J. Santos Filho ∗∗∗ Paulo E. Miyagi ∗∗∗∗

∗University of Sao Paulo, Brazil (e-mail: paicaio@usp.br)∗∗University of Sao Paulo, Brazil (e-mail: fabri@usp.br)

∗∗∗University of Sao Paulo, Brazil (e-mail: diolino.santos@poli.usp.br)∗∗∗∗University of Sao Paulo, Brazil (e-mail: pemiyagi@usp.br)

Abstract: Markets are becoming independent of geographic barriers and industries are seekingnew productive systems (PS) configurations, from centralized to distributed structures, movingtheir productive plants to countries with energy reserves and low operating costs. To allow thetasks coordination and management in this new distributed structure, it takes advantage ofthe advances in mechatronics and information technologies, which should assure cooperationbetween the system parties and between the involved users (clients, operators, managers, etc.).Each part of the disperse PS, which is also a PS, has a high level of operational autonomy.Thus, this kind of system presents new integration and coordination problems which must beovercome to achieve effective implementation. The users of the system require communicationframeworks for negotiating the processes involved in production. In this context, this workinitially shows a control architecture for users negotiation in a disperse PS. To implementthe control architecture, computational models exploring the potential of the Petri net wasdeveloped using the production flow schema to systematize the construction of models.

Keywords: industry automation; production control; industrial production system;manufacturing processes; Petri-nets; interacting service stations.

1. INTRODUCTION

According to Garcia Melo et al. [2008], productive systems(PSs) are defined as systems which perform process usingmaterial resources, equipment, human resources, and otherphysical entities and information for the production ofgoods and services.

Since the 1990s, production and supply chain systemshave changed from the traditional mass production ledby products to mass customization in order to face theincrease in global market competition [Thomas and Art-iba, 2009]. Markets demand for products with high qualityat lower costs, highly customized and short life cycles,imposing new requirements on PS, namely in terms ofquality, response agility and flexibility, crucial for a PSto remain in the business. In this kind of market, PSs canno longer be seen as a standalone actor, being forced toreconsider the way they are organized to increase theircompetitiveness [Leitao, 2009]. Consequently, the researchon PS control has moved from traditional centralized andprocess-oriented approaches to distributed structures inwhich the PSs or parts of the PS are spread throughoutthe world, forming a disperse system. One of the reasonsthat contributed to the production decentralization is thedisplacement of the plants to countries with low energyand operation cost [Sousa et al., 2008].

Some studies into distributed systems have focused on theuse of a service-oriented structure, in which the functionsand the information that users (composing the system)provide are seen as services. The growth in the numberof internet services has given rise to communities aroundcommon interests. This growth brings increases in the scaleof the internet and boosts the expectation of informa-tion retrieval [Wang et al., 2009]. These services, on theother hand, can be seen as computational resources ofthe Internet and can be shared among its users to meetcommon interests. However, according Han et al. [2008],conventional computing cannot meet resources sharing,which motivated the development of applications basedon Web services (WS), which provide advantages to shareresources and to develop applications of greater complex-ity.

This work is organized as follows. Section 2 presentsthe methodology for modeling services using productionflow schema and Petri net. Section 3 shows the modelingof an architecture applied to negotiation between users(clients, operators and PS) to answer the demand. Section4 provides the results of models analysis. Section 5 presentssome discussions about this work.

Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011

Copyright by theInternational Federation of Automatic Control (IFAC)

6413

2. PETRI NET AND WEB SERVICES

2.1 Petri net

Petri nets (PNs), as graphical and mathematical tools,provide a uniform notation for modeling, formal analysisand design of discrete event systems (DES) and it iseffective as a process description and specification tech-nique [Yoo et al., 2010]. It is a representation that canbe used both at conceptual and at functional level, inwhich the system can be analyzed and validated beforeproceeding with detailed design and implementation. Ithas the advantage that the same model can be used foranalysis of behavioral properties and performance evalua-tion, as well as for systematic construction of simulatorsand controllers. Mathematically, PN can be described asa set of algebraic equations, and can be used for formalverification of precedence relations between events, con-current operations, appropriate synchronization, freedomfrom deadlock, repetitive activities and mutual exclusionof shared resources [Zurawski and Zhou, 1994].

The structure of a PN is a 4-tuple N = (P, T, F,W )[Zurawski and Zhou, 1994], where P and T are finite,nonempty, and disjoints sets. P is the set of places, and T isthe set of transitions. F is the set of relationships betweenplaces and transitions and is also called set of directedarc. W is a set of nonnegative and nonempty integers thatrepresents the weight of each directed arcs. A PN is calledordinary net if ∀f ∈ F,W (f) = 1 and its structure isdenoted as N = (P, T, F ) [Li and Zhou, 2008].

A PN is a 5-tuple (P, T, F,W,M0), where M0 is a set ofnonnegative integers that represents the initial tokens ofplaces, which is also known as initial marking of places[Yoo et al., 2010].

The main properties of PN used in the analysis of PSs ispresented as follows.

Reachability In order to find out whether the modeledsystem can reach a specific state as a result of a requiredfunctional behavior, it is necessary to find such a sequenceof firings of transitions [Li and Zhou, 2008] which wouldresult in transforming a marking M into Mi, where Mi

represents the specific state and the sequence of firingsrepresents the required functional behavior. The existenceof an additional sequence of transitions firings whichtransform M0 into Mi indicates that the PN model hasalternative ways in which certain states are achieved. Itcan also indicate the presence of unexpected functionalbehavior of the real system [Zurawski and Zhou, 1994].

Marking Mi is called reachable from marking M if thereis a sequence of transitions firings which transform M intoMi, Mi ∈ R(N,M) [Han et al., 2008].

Boundedness and safeness A PN is said to be k-boundedif the number of tokens in any place p, where p ∈ P ,is always smaller than or equal to k (k is a nonnegativenonempty integer number) for every marking M reachablefrom the initial marking M0. With p ∈ P , if ∃B > 0integer, so that ∀M ∈ R(N,M0) and M(p) ≤ B, p isbounded. The boundary B of a PN is called max{B},∀p ∈ P [Han et al., 2008].

A PN is safe if it is 1-bounded; in other words, B = 1 forall places of the PN [Han et al., 2008].

Conservativeness A PN is conservative if the numberof the tokens is constant, independently of the transitionsfirings. From the net structural point of view, this can onlyhappen if the number of the input arcs for each transitionis equal to the number of output arcs. However, in realPSs, resources are frequently combined together so thatcertain tasks can be executed, then separated after thetask is completed [Zurawski and Zhou, 1994].

A Petri net is also said to be conservative if there is a vectorm, m = [m1,m2, . . . ,mk], where k is the number of places,and the weighted sum of mi tokens remains the same foreach marking M reachable from the initial marking M0

[Zurawski and Zhou, 1994].

Liveness The concept of liveness is closely related to thedeadlock situation (state at which the PN model cannotfire any of its transitions). Zurawski and Zhou [1994]showed that four conditions must hold for a deadlock tooccur. These four conditions are:

• Mutual exclusion: a resource either available orallocated to a process which has an exclusive accessto this resource;

• Hold and wait: a process is allowed to hold aresource while requesting more resources;

• No preemption: a resource allocated to a processcannot be removed from the process, until it is re-leased by the process itself;

• Circular wait: two or more processes are arrangedin a chain in which each process waits for resourcesheld by the next process in the chain.

A PN is considered live if ∀M , M0[> M , i.e. it can progressthrough some firing sequences. From a marking M0, it canbe deadlock-free immediately, yet it can have a sequence offirings which can lead the PN to a deadlock situation [Hanet al., 2008]. For this reason, different levels of liveness fortransition t, and marking M0, can be considered.

Reversibility An important issue in the operation of realPS is the ability of these systems for error recovery (returnto a normal state) after failure or error occurrence. I.e.,these systems are required to return from failure statesto the preceding correct states. This requirement is closelyrelated to the reversibility property of a PN [Zurawski andZhou, 1994].

A PN is said to be reversible if M is reachable from M0

and M0 is reachble from M [Zurawski and Zhou, 1994].

2.2 Production flow schema

The production flow schema (PFS) is derived from thePN of the channel-agent type and used to systematize themodel building in PN. The PFS shows, at conceptual level,how items involved in the production act in the executionof functions necessary to obtain the desired products orservices. An important concept presented by the PFSis the flow of items (materials or information), which isassociated with allocation and deallocation of resources.The PFS indicates that the activities or production steps

Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011

6414

involve the interaction between items, their flows and otheractivities [Hasegawa et al., 1999].

The PFS has 3 structural elements [Hasegawa et al., 1999]:

• Active elements or activities: (represents theevents) elements which can be refined by obtaininga new PFS or some PN, shown in Fig. 1 (a);• Passive elements or distributors: (similar to the

places in PN) elements which can represents itemsand are responsible for activity enabling, shown inFig. 1 (b);• Arcs: elements which describe the relations between

the distributors and the activities, shown in Fig. 1(c);

(a) Activity (b) Distributor (c) Arc

Fig. 1. PFS elements

The PFS can represent the main flow of items (Fig.2) or represent relations of asynchronous communication(Fig. 3), which explore the concept of secondary flow ofinformation [?].

Fig. 2. Example of modeling of the PFS main flow ofactivities

Fig. 3. Example of modeling in PFS of asynchronouscommunication with secondary flow of information

The refinement of an activity is conducted by replacingthat for a graph containing new activities (of PFS) ortransitions (of PN) joined by distributors (of PFS) orplaces (of PN), shown in Fig. 4. The refinement of adistributor is conducted by replacing that for a graphcontaining new distributors (of PFS) or places (of PN)joined by activities (of PFS) or transitions (of PN), shownin Fig. 5.

2.3 Web services modeling

According to Hamadi and Benatallah [2003], a WS can bedescribe as a tuple

S = (NameS,Desc, Loc, URL,CS, SN), (1)

Fig. 4. Refinement of activities of PFS

Fig. 5. Refinement of distributors of PFS

where, NameS is the name of the service (used as itsunique identifier), Desc is the description of the serviceprovided (it summarizes what the service offers), Loc isthe server where the service is located in, URL is theinvocation of the WS, CS is a set of its componentservices (if CS = NameS then S is a basic service oratomic, otherwise, S is a composite service of atomicsservices), and SN = (P, T, F,W,M0) is the PN model ofthe dynamic behavior of the service.

The services and their compositions can be modeled byusing PN [Hamadi and Benatallah, 2003] associated withone of its interpretations, such as the PFS:

• ε is an empty service, ie, a service which performs nooperation, Fig. 6 a);• S1 is a constant service, used as an atomic or basic

service in this context, Fig. 6 b);• S1

⊙S2 is a composite service that performs service

S1 followed by service S2, Fig. 7;• S1

⊕S2 is a composite service that behaves as either

service S1 or service S2. Once one of them executesits first operation the second service is discarted, Fig.8;• S13S2 is a composite service that performs either

service S1 followed by service S2, or service S2followed by service S1, Fig. 9;

(a) Empty (b) Atomic

Fig. 6. Services

Fig. 7. Service S1⊙

S2

Fig. 8. Service S1⊕

S2

Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011

6415

Fig. 9. Service S13S2

3. CONTROL ARCHITECTURE

There are different configurations for a disperse PS, so thefocus of this work is to initialy consider a disperse PS withits basic users (client, operator and PS).

Consequently, in this work, some users of the dispersePS architecture have specific functions to perform: oneuser has equipment, resources and other physical entities;one user has the knowledge to operate equipments; andanother user has a demand to be met.

The users above can be called productive system (or PS),operator and client, respectively. Each user must followsome rules to ensure proper system operation. In thisarchitecture, the operator user is not an employee of thePS user and can operate the equipment of different PSs.

The PS user must negotiate the deadline to meet thedemand (this could be an estimate of past demands orother estimate) and must fulfill the negotiated deadline(in other words, it is forbiden to propose a deadline inwhich the demand can not be met). Likewise, the operatoruser must negotiate the deadline to meet the demand andmust fulfill the negotiated deadline. The client user mustnegotiate with the operator and PS users the deadlines tomeet its demand and, based on the negotiated deadlines,it must choose the longest deadline as the deadline to meetits demand (because that is the most conservative time).

The negotiation between users to meet the demand of adisperse PS can be divided into 3 basic activities, each oneperformed by a different user (client, operator and PS).The activities execution involves an asynchronous com-munication, with secondary flow of information, in whichthere is specific data flow between users sub-activities, asin Fig. 10 (a). In this case, after a refinement, activity[A2] is divided into 2 activities, each one responsible forcommunicating with activities [A1] or [A3] (Fig. 10 (b)).

In Fig. 10, activity [A1] represents the activity performedby an operator during the negotiation. Activity [A2] rep-resents the activity performed by a client during the nego-tiation. Activity [A3] represents the activity performed bya PS during the negotiation. Activity [A2.1] represents theactivity performed by a client in communication with anoperator. Activity [A2.2] represents the activity performedby a client in communication with a PS.

The communication activities involve 3 distinct sub-activities: one for sending and receiving requests; onefor sending and receiving proposals; and one for sendingand receiving confirmation. These sub-activities are repre-sented in Fig. 11.

In this case, [A1.1] and [A3.1] represent the receiving of arequest from a client by the operator and PS, respectively;[A2.1.1] and [A2.2.1] represent the sending of request by

(a) (b)

Fig. 10. Activities flow: (a) Main flow (b) [A2] refinemet

Fig. 11. Refined activities flow

the client to the operator and PS, respectively; [A1.2] and[A3.2] represent the sending of proposal to the client bythe operator and PS, respectively; [A2.1.2] and [A2.2.2]represent the receiving of proposal by the client from theoperator and PS, respectively; [A1.3] and [A3.3] representthe receiving of confirmation from the client by the oper-ator and PS, respectively; [A2.1.3] and [A2.2.3] representthe sending of confirmation by the client to the operatorand PS, respectively.

In the case of multiple operators, the client has a choice(conflict) which the operator will negotiate. Fig.12 showsa representation of this conflict.

In Fig. 12, the choice (conflict) is in the relation betweenthe first distributor and the first activities of client. Inthis figure, the activities [A2.1.i] represent the activities ofclient in communication with operator 1 and the activities[A′2.1.i] represent the activities of client in communicationwith operator 2. In the same figure, the activities [A1.i]

Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011

6416

Fig. 12. Activities flow for multiple operators

represent the activities of operator 1 and the activities[A′1.i] represent the activities of operator 2. The samestructure can be applied to a larger number of operatorsand to multiple PS.

The proposal sending activities ([A1.2] and [A3.2]) cancontain two kinds of information: one with costs, condi-tions, deadline and others; and another saying unable tomeet the request. The two kinds of information make theclient take different decisions: first, proceed to the nextactivity; and second, return to the beginning of the processto look for new operators or PSs. This conflict is solved bythe operators or by the PS, as shown in Fig. 13.

Fig. 13. Refinement of the proposal sending and receivingactivities

In Fig. 13, the activity [Unable], in the activities flowof operator or PS, tells the client that they are notable to meet the request. This information finishes theactivities flow of the operators and PS and makes the clientactivities flow return to the first distributor, before sendinga request. The activity [Proposes] allows proceeding tothe last activity of the activities flow of operator or PSand communicates with the activity [Receives] of client,which allows proceeding to the last activity of his activitiesflow.

The activity of sending confirmation ([A2.1.3] and [A2.2.3])can also contain two kinds of information: one to accept;

and another to reject the proposal sent. In the two kindsof information, the activities flow of operator or PS isfinished. However the client has two possibilities in his ac-tivities flow: one finishes the negotiation with the operatoror PS; and other returns to the beginning of the processto seek new operators or PS. This conflict is solved by theclient, as shown in Fig. 14.

Fig. 14. Refinement of the activities of sending and receiv-ing confirmation

In Fig. 14, the activity [Accepts], in the client activitiesflow, says to the operator or PS that the client acceptsthe proposal and finishes this negotiation. The activity[Rejects], in the client activities flow, says to the operatoror PS that the client rejects the proposal and returns tothe first distributor of the activities flow, before sending arequest.

The representations above can be refined to PN. Thedistributors are refined as PN places. The activities withsecondary flow of information can be refined as 2 transi-tions and 1 place, where the first arc of secondary flow ofinformation is linked to the first transition and the secondarc is linked to the second transition (Fig. 15 (a)). Thebrackets that include the activities flow can be refined as2 transitions, and the activities flow is between these twotransitions (Fig. 15 (b)).

(a) Activities with secondary flow of informa-tion

(b) Activities without secondary flow of infor-mation

Fig. 15. Ativities refinement

Finally, the models are combined and Fig. 16 presents theresulting PN.

Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011

6417

Fig. 16. Structured PN of the architecture

4. ANALYSIS

One of the analyzed cases consider clients, operators andPS unoccupied and the PN in Fig. 16 with initial marking(or initial condition), as shown in Fig. 17.

The results obtained by simulation were, the PN: reachesonly the desired states; is 1-bounded (and, therefore, issafe); is conservative if considering the weighted sum ofthe tokens; is L3-live because it can be fired infinitely fromthe initial marking; is reversible.

Fig. 17. Simulated PN from architecture

Other analyzed cases consider the client, initially, in com-munication with a PS and with an operator (client, oper-ator and PS busy). The results were the same as for theprevious case, confirming the expected behavior.

Then, the practical implementation of the architecture wascarried out [Fattori, 2010]. In that work, a set of webservices were developed and implemented in a disperse PSemulated by stations of an automated flexible manufac-turing system. The tests made in these systems using thearchitecture confirm the results of the model analysis.

5. CONCLUSION

The use of information systems techniques in PS is not re-cent, yet the change of the paradigm from process-orientedcentralized production to service-oriented disperse produc-tion has been investigated more strongly in recent years.

The negotiation between clients, operators and PS to meetthe demand can also be understood as service and, asthey are well known solutions in service-oriented appli-cations. The web services facilitates the implementationof a structure which allows that negotiation with service

orientation. Consequently, based on that, it was proposedan architecture for disperse PS in this work.

The operation in a PS plant can be understood as asequence of events. Consequently, the behavior of the PScan be modeled as a discrete event system and it can usethe PN. Combining this model to the PN model createdby the systematics, it is possible to create a fuller modelof disperse PSs behavior.

The systematic applied in this work allows the analysis ofdemand meeting using the PN to verify the properties ofthe modeled architecture. The use of PFS in the systematicallows a smooth conversion from a conceptual model to aformal model (as PN).

ACKNOWLEDGEMENTS

The authors would like to thanks CNPq, FAPESP andCAPES for the finantial support to the present project.

REFERENCES

C. C. Fattori. Implementacao de uma arquitetura decontrole distribuıdo para sistemas produtivos. Master’sthesis, Universidade de Sao Paulo, 2010.

J. I. Garcia Melo, F. Junqueira, R. A. Morales, and P. E.Miyagi. A procedure for modeling and analysis ofservice-oriented and distributed productive systems. InIEEE International Conference on Automation Scienceand Engineering, pages 941–946, 2008.

R. Hamadi and B. Benatallah. A Petri net-based modelfor web service composition. In Proceedings of theAustralasian Database Conference, pages 191–200, 2003.

R. Han, K. Liu, Y. Ju, and J. Zhao. A Petri nettheory-based method for modeling web service-basedsystems. In 4th International Conference on WirelessCommunications, Networking and Mobile Computing,pages 1–7, Dalian, China, Oct. 2008.

K. Hasegawa, P. E. Miyagi, D. J. Santos Fo, K. Takahashi,L. Ma, and Sugisawa M. On resource arc for Petri netmodelling of complex resource sharing system. Journalof Intelligent and Robotic Systems, 26:423–437, 1999.

P. Leitao. Agent-based distributed manufacturing control:A state-of-art survey. Engineering Applications of Arti-ficial Intelligence, 22(7):979–991, oct. 2009.

ZW Li and MC Zhou. Control of elementary and depen-dent siphons in Petri nets and their application. IEEETransactions on Systems, Man and Cybernetics, PartA: Systems and Humans, 38(1):133–148, jan. 2008.

R. Sousa, N. Shah, and L. G. Papageorgiou. Supply chaindesign and multilevel planning–an industrial case. Com-puters & Chemical Engineering, 32:2643–2663, 2008.

A. Thomas and AH Artiba. Modelling and control of pro-ductive system: Concepts and application. InternationalJournal of Production Economics, 121:1–3, 2009.

D. Wang, Q. C. K. Tse, and Y. Zhou. A decentralizedsearch engine for dynamic web communities. Knowledgeand Information Systems, dec. 2009.

T. Yoo, B. Jeong, and H. Cho. A Petri nets basedfunctional validation for services composition. ExpertSystems with Applications, 37(5):3768–3776, may. 2010.

R. Zurawski and MC Zhou. Petri nets and industrial ap-plications: A tutorial. IEEE Transactions on IndustrialEletronics, 41(6):567–583, dec. 1994.

Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011

6418