`
`Using multi-agent architecture in FMS
`for dynamic scheduling
`
`K H A L I D K O U I S S , 1 * H E N R I P I E R R E V A L 1 a n d N A S S E R
`M E B A R K I 2
`
`1Laboratoire de Recherche en Syste` mes de Production, Institut Franc¸ ais de Me´canique
`Avance´e, Campus des Ce´ zeaux, BP 265, 63175 Aubie`re Cedex, France
`2Laboratoire PRISMa, Universite´ Claude Bernard, Baˆ t 710, 43 Boulevard du 11 Novembre,
`69622 Villeurbanne Cedex, France
`
`Received July 1995 and accepted June 1996
`
`The proposed scheduling strategy is based on a multi-agent architecture. Each agent of this
`architecture is dedicated to a work centre (i.e. a set of resources of the manufacturing system);
`it selects locally and dynamically the most suitable dispatching rules. Depending on local and
`global considerations, a new selection is carried out each time a predefined event occurs (for
`example, a machine becomes available, or a machine breaks down). The selection depends on:
`(1) primary and secondary performance objectives, (2) the operating conditions, and (3) an
`analysis of the system state, which aims to detect particular symptoms from the values of
`certain system variables. We explain how the scheduling strategy is shared out between agents,
`how each agent performs a local dynamic scheduling by selecting an adequate dispatching
`rule, and how agents can coordinate their actions to perform a global dynamic scheduling of
`the manufacturing system. Each agent can be implemented through object-oriented formal-
`isms. The selection method is improved through the optimization of the numerical thresholds
`used in the detection of symptoms. This approach is compared with the use of SPT, SIX,
`MOD, CEXSPT and CR/SPT on a jobshop problem, already used in other research works.
`The results indicate significant improvements.
`
`Keywords: Dynamic scheduling, dispatching rules, flexible manufacturing systems, multi-agent
`system, simulation-optimization, object-oriented models
`
`1. Introduction
`
`The dynamic scheduling of manufacturing systems is con-
`cerned with the allocation of jobs to the resources in real
`time. This allocation is made according to the state of the
`shopfloor (e.g. breakdown of a machine, availability of a
`resource, or existence of bottlenecks) and the production
`objectives (e.g. reduce the number of jobs in progress, or
`reduce the tardiness). One of the most common approaches
`to dynamic scheduling of the jobs to process is to use dis-
`patching rules (DRs). Dispatching rules can be very simple
`or extremely complex. Examples of simple dispatching
`rules are: ‘select a job at random’ or ‘select the job with the
`longest waiting time’. A more complex example might be
`‘select the job with the shortest due date whose customer’s
`inventory is less than a specific amount’.
`
`*Author to whom all correspondence should be addressed.
`0956-5515 Ó 1997 Chapman & Hall
`
`Numerous DRs exist, but research in recent decades has
`demonstrated that there is no one DR that is globally
`better than the others (Blackstone et al., 1982; Kiran and
`Smith, 1982; Montazeri and van Wassenhove, 1990). Their
`e(cid:129)ciency depends on the performance criteria considered,
`and on the operating conditions (e.g. shop load, tightening
`of due dates, or existence of bottlenecks).
`We propose an approach based on a multi-agent archi-
`tecture. Each agent selects locally and dynamically the DR
`that seems the most suited to the operating conditions, to
`the production objectives, and to the current shop status.
`Because the shop status changes over time, each agent
`analyses the system state each time an event occurs (e.g. a
`machine becomes available, or an urgent job arrives).
`In this paper, we first present the general principles of
`multi-agent systems, and we focus on the benefits of this
`approach in the production management area. Next, we
`present the way in which the dynamic scheduling deci-
`sions can be shared between agents, and the role of these
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1024, IPR2022-00681, Pg. 1
`
`
`
`42
`
`Kouiss et al.
`
`agents is highlighted. Finally, the benefits of this ap-
`proach are demonstrated through the example of a job-
`shop system.
`
`2. Agents and multi-agent systems
`
`2.1. Definitions
`
`The use of a multi-agent architecture allows decisions to be
`taken in a decentralized way. In artificial intelligence, this
`approach appears to be well suited to complex problems,
`especially those with a great number of interactions be-
`tween components, and for which classical
`incremental
`methods cannot provide good results. The multi-agent
`method allows one to solve subproblems locally with an
`agent, and to propose a global solution as a result of in-
`teractions between the dierent agents.
`Several researchers have proposed formal definitions for
`agents and multi-agent systems. We retain those proposed
`by Ferber (1993):
`
`(1) An agent is a real or a virtual entity able to act on
`itself and on the surrounding world, generally populated by
`other agents. To perform its actions, this entity contains a
`partial representation of its environment, and can com-
`municate with other agents of this environment. Its be-
`haviour is a result of its observations, its knowledge and its
`interactions with the world and other agents. An agent has
`several interesting features:
`
`(a) it has capabilities of perception and a partial
`representation of the environment;
`(b) it can communicate with other agents;
`(c) it can reproduce son agents;
`(d) it has its own objectives and an autonomous
`behaviour;
`
`(2) A multi-agent system (MAS) is an artificial system
`composed of a population of autonomous agents, which co-
`operate with each other to reach common objectives, while
`simultaneously each agent pursues individual objectives.
`
`2.2. Agent structure
`
`We can split an agent into three layers, as depicted in Fig. 1:
`
`(1) The static knowledge layer: contains knowledge on
`itself and on the other agents. This is an agent’s specific
`memory, used to memorize its observations and its
`knowledge concerning its environment (social knowledge);
`(2) The expertise layer: contains knowledge that repre-
`sents treatments and actions that an agent is able to carry
`out and which can be described in various forms (e.g. al-
`gorithms, production rules, frames, or logical expressions).
`This layer constitutes the agent know-how;
`(3) The communication layer: includes the communica-
`tion tools. They describe the communication protocols
`
`Fig. 1. Agent structure.
`
`between the agent on one side and some other agents and
`resources of the environment on the other side. This layer
`characterizes the way that the agent takes into account the
`messages it receives. This layer also allows the agent to act
`and to apprehend the environment changes.
`
`2.3. Multi-agent structure and production management
`
`For the production management of a manufacturing sys-
`tem, many decisions have to be taken to reach the pro-
`duction objectives (e.g. planning decisions, scheduling
`decisions, and control decisions). Indeed, these decisions
`must be periodically updated in order to take into account
`changes in the production system (e.g. a machine break-
`down, worker absences, or the arrival of an urgent job).
`The use of a multi-agent architecture allows one to share
`out all these decisions between several agents in a hierar-
`chical manner. Each agent is in charge of specific decisions
`(Chandra and Talavage, 1991; Inecker et al., 1991; Bap-
`tiste and Manier, 1993; Kwok and Norrie, 1993; Parunak,
`1993; Barbuceanu and Fox, 1994; Lefranc¸ois and Mon-
`treuil, 1994; Tacquard et al., 1994; Trentesaux and Tahon,
`1995; Ouzrout, 1996). Unfortunately this structure presents
`some disadvantages, due mainly to possible contradictory
`decisions of agents that can lead to a global lock of the
`system (Ayel, 1994; Attoui et al., 1995).
`The allocation of decisions to agents can be made ac-
`cording to several criteria, listed below:
`
`(1) Technological criterion: for example, an agent may
`be dedicated to resources using the same communication
`protocol;
`(2) Topological criterion: for example, an agent may be
`dedicated to resources close (in distance) to each other;
`(3) Functional criterion: for example, an agent may be
`dedicated to a particular function (e.g. quality function,
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1024, IPR2022-00681, Pg. 2
`
`
`
`Using multi-agent architecture in FMS for dynamic scheduling
`
`43
`
`monitoring function, or scheduling function). This paper
`emphasizes the dynamic scheduling function;
`(4) Organizational criterion: the manner in which the
`works are allocated to each agent.
`
`3. A dynamic scheduling approach based on a multi-agent
`structure
`
`3.1. Principles of the proposed approach
`
`The dynamic scheduling is supported by a multi-agent ar-
`chitecture. Each agent of the system is in charge of a work
`centre of the manufacturing system. An agent solves the
`scheduling problem by selecting dynamically the most ad-
`equate DR to apply locally. Examples of DRs that can be
`applied are: shortest processing time (SPT), smallest critical
`ratio (SCR), earliest due date (EDD), conditional exp-
`editive shortest processing time (CEXSPT), and critical
`ratio shortest processing time (CR/SPT) (Mebarki, 1995).
`To select a DR, the agents take into account primary and
`secondary objectives (e.g. reduce the mean flow time and
`reduce the percentage of tardy jobs), the state of the work
`centre (e.g. length of the waiting job queues, or availability
`of resources), and information received from other agents.
`In order to take into account the changes of the system
`state, the application of new DRs is envisaged by agents
`each time a triggering event occurs; that is, each time a
`resource becomes available, a new job arrives, or a job
`leaves the system.
`It has already been shown that combinations of dierent
`DRs could perform better than applying the same DR to
`all the work centres (Barrett and Barman, 1986). In a
`multi-agent architecture, each agent takes its decisions in
`an independent way, so in a given time dierent DRs may
`be applied to dierent work centres.
`The selection of the DR applied by each agent is carried
`out
`through two steps using the following strategy
`(Pierreval and Mebarki, 1997).
`
`3.1.1. Step 1: Detection of an active symptom
`
`The system status is analysed to try to detect predefined
`symptoms. This is done using knowledge of the following
`form:
`
`If [condition about state variables] then [active symptom]
`
`An example of such a rule is:
`If [job due date ) current time )
`remaining processing time < a] then [active ‘job tardy’]
`
`where a is called a threshold. A symptom becomes active
`when an observed variable (e.g. utilization rate of re-
`sources, waiting time of jobs, or length of queues), has
`exceeded a predefined threshold. This means that the sys-
`tem might be deviating from its production objectives.
`Symptoms may concern the behaviour of the whole system
`
`[global symptoms detected by the supervisory agent (see
`section 3.2), e.g. ‘Too many tardy jobs’], or the behaviour
`of a particular work centre [local symptoms detected by a
`simple agent dedicated to a work centre (see section 3.2)
`e.g. ‘Station S becomes bottleneck’].
`Thresholds depend on the particular scheduling prob-
`lem, and cannot be generally predefined. Pierreval (1992)
`has shown that these thresholds can have a great impact on
`the performance of this scheduling method. Thus thresh-
`olds need to be tuned for each agent using an optimization
`procedure (Pierreval and Mebarki, 1997).
`
`3.1.2. Step 2: Choice of DRs
`
`The DRs to apply are chosen from a set of pre-selected
`DRs, using rules of the following forms:
`
`If [conditions about the objectives
`
`and/or conditions about information received
`
`from other agents
`
`and/or conditions about the state of the
`
`local work centre
`
`and/or conditions about the active symptoms]
`then [apply fselected DRg to the considered queue(cid:138)
`
`An example of such a rule is:
`If [the primary objective (cid:136) ‘reduce the mean flow time’
`and no ‘tardy jobs’
`
`and no ‘urgent jobs’]
`
`then [apply SPT]
`
`3.2. Organization of the agents for dynamic scheduling
`
`The manufacturing system is composed of several work
`centres, each one made up of one or several resources. The
`multi-agent architecture is shown in Fig. 2.
`
`Fig. 2. Multi-agent architecture. WC: work centre.
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1024, IPR2022-00681, Pg. 3
`
`
`
`44
`
`Kouiss et al.
`
`In this architecture we distinguish a supervisory agent
`and several simple agents each dedicated to a specific work
`centre. Those agents perform two types of communication:
`
`(1) Communication with other agents (type a in Fig. 2).
`Examples of this type of communication are: a request to
`apply a specific DR, information about active symptoms,
`and a request of the state of another agent;
`(2) Communication with the environment (type b in
`Fig. 2). Examples of this type of communication are: exe-
`cution of a program, reading of the state of a variable,
`orders from a human operator, and exchange with the
`database system.
`
`The role of the supervisory agent is to monitor the global
`state of the manufacturing system. This agent has only an
`external vision of the state of other work centres; dedicated
`agents keep it informed using messages. It can detect global
`symptoms, and can impose particular DRs to agents con-
`trolling the work centres if it considers this necessary to
`satisfy the global objectives. For example, if it notices the
`global symptom. ‘The number of jobs that become tardy is
`too high’, and if the objective is to reduce the mean flow-
`time of jobs, then it imposes SPT to all agents.
`The agent allocated to a work centre is in charge of the
`scheduling of jobs inside the centre. This agent manages its
`own waiting job queues. The selection of DRs is made
`according to the two steps described above, and depends
`on the system state, the orders received from the supervi-
`sory agent, and the global objectives of the manufacturing
`system (e.g. reduce the mean tardiness). This selection can
`be very simple when the supervisory agent imposes a par-
`ticular DR on the agent.
`
`3.3. Implementation of the dynamic scheduling approach
`in an agent
`
`Each agent has the structure presented in Fig. 1, and can be
`implemented using object-oriented formalisms (Kwok and
`Norrie, 1993; Lefranc¸ois and Montreuil, 1994). This stru-
`cture is based on the three layers described below.
`
`3.3.1. Static and social knowledge layer
`
`This layer contains such pieces of knowledge as:
`
`(1) Threshold values used in the rules to activate the
`predefined symptoms. These values may change during the
`life of the agent according to a learning procedure;
`(2) Data about DRs that the agent can select;
`(3) Data about capabilities of other agents (for example,
`the supervisory agent has data about capabilities of all the
`agents dedicated to the work centres of the manufacturing
`system).
`
`3.3.2. Expertise layer
`
`This is the intelligent part of the agent. It is based on object
`methods representing production rules (as previously de-
`
`scribed), and uses data taken from the static and social
`knowledge layer and information received by the commu-
`nication layer. The expertise concerns the application of
`the two steps described in Section 3.2. It checks the con-
`ditions to detect the predefined symptoms and, if necessary,
`selects a new, adequate DR. Then it applies it, using (from
`the static and social knowledge layer) the relevant data
`necessary for its application, and sends orders to the con-
`cerned entities, using the communication layer functional-
`ities. In the case of the supervisory agent, an order can be a
`choice of a specific DR for another agent. In the case of an
`agent dedicated to a work centre, an order can be the start
`of a machining operation or the notification of the name of
`a new active symptom to the supervisory agent.
`
`3.3.3. Communication layer
`
`Agents need to communicate with the physical environ-
`ment. We include in the term ‘physical environment’ each
`entity that is not an agent. This comprises all the resources
`(e.g. programmable controllers, robots, machining cen-
`tres). Communication between an agent and the environ-
`ment uses the machine’s communication protocols (e.g.
`programmable controller protocols such as MODBUS,
`UNI-TELWAY or SINEC L2, or numerical control pro-
`tocols). The communication layer of an agent must contain
`tools (e.g. drivers) to carry out the communication with all
`the resources of the work centre. This communication al-
`lows an agent, for example, to monitor machines (e.g. to
`start, stop, or download a program), to extract information
`in the database system, or to have information about the
`state of machines (e.g. alarm messages, and the state of a
`sensor that indicates the number of parts in a waiting job
`queue).
`Agents also need to communicate with each other. This
`communication is performed by exchange of messages, and
`is supported by a communication network installed be-
`tween the agents’ host computers. The protocol can be a
`speech–act type (Trouilhet, 1993) such as KQML (Finin
`et al., 1992), which is an agent communication language
`(ACL). This communication allows, for example, the su-
`pervisory agent to know the state of a waiting queue in a
`work centre, or to request the application of a given DR in
`another agent.
`
`4. Simulation of the distributed dynamic scheduling
`
`At present, this approach has not been implemented on a
`real FMS. In order to evaluate its performance, specific
`object-oriented simulation software was designed. Al-
`though this software is implemented in a simple program, it
`is based on the distributed dynamic scheduling that we
`have presented. To make the comparison as relevant as
`possible, we have chosen a jobshop model that has been
`already used by several researchers to compare DRs. Eilon
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1024, IPR2022-00681, Pg. 4
`
`
`
`Using multi-agent architecture in FMS for dynamic scheduling
`
`45
`
`and Cotteril (1968) have used this model to test the eects
`of the SIX rule, Baker and Kanet (1983) to demonstrate the
`benefits of the MOD rule, Baker (1984) to examine the
`interaction between dispatching rules and due-dates as-
`signment methods, Russel et al. (1987) to analyse the eects
`of the CoverT rule comparatively with several other DRs,
`and Schultz (1989) to demonstrate the benefits of the
`CEXSPT rule.
`The system is a four-machines jobshop. Each machine
`can perform only one operation at a time. The number of
`operations of the jobs processed in the system is uniformly
`distributed between two and six. The routeing of each job is
`random. More precisely, when a job leaves a machine and
`needs another operation, each machine has the same
`probability of being the next, except the one just released,
`which cannot be chosen. The processing times on machines
`are exponentially distributed, with a mean of 1. The arrival
`of jobs in the system is modelled as a Poisson process. The
`mean arrival rate of this process is equal to the shop uti-
`lization, which is defined as follows:
`Shop utilization (cid:136) 1
`m
`
`Xm
`
`k (cid:136) 1
`
`@k
`
`(cid:133)1(cid:134)
`
`where @k is the steady-state utilization rate of the kth re-
`source, and m is the number of workstations in the shop.
`Due dates of jobs are determined using the TWK method
`(Baker, 1984).
`The dispatching rules compared are: SPT, CEXSPT,
`CR/SPT, plus MOD and SIX. These DRs seem to be ac-
`cepted as being among the most e(cid:129)cient (see for example
`Baker, 1984; Russel et al., 1987; Engell and Moser, 1992).
`A simulation model of the system previously described
`was build using our simulation–optimization software. It
`was first run to tune the thresholds, and then to compare
`the dynamic change of DRs, managed by the multi-agent
`system (called SFSR), with regard to the two following
`pairs of objectives:
`
`(1) Reduce the mean tardiness as a primary objective and
`reduce the mean flow time as a secondary objective (noted
`as SFSR1);
`(2) Reduce the conditional mean tardiness as a primary
`objective and reduce the mean tardiness as a secondary
`objective (noted as SFSR2).
`
`The experiments were conducted with a mean arrival
`rate of jobs of 0.9, which corresponds to a utilization rate
`of the resources of 90% (i.e. a high level of utilization). For
`this case, an average flow allowance of 30 time units rep-
`resents tight due dates (i.e. an allowance factor k of 7.5),
`whereas an average flow allowance of 60 time units rep-
`resents loose due dates (i.e. an allowance factor k of 15).
`The simulation experiments have been designed in the
`same way as those of Schultz (1989): that is, the transient
`phase is estimated at 500 jobs, ten replications are carried
`out, and each run yields estimates of the performance
`
`Table 1. Comparison of the dynamic selection with various
`dispatching rules, with tight due dates
`
`Rule
`
`MFT
`
`SPT
`SIx
`MOD
`CEXSPT
`CR/SPT
`SFSR1
`SFSR2
`
`17.4
`22.6
`23.0
`21.9
`22.2
`21.1
`23.4
`
`MT
`
`3.38
`1.97
`1.87
`1.78
`1.42
`1.5
`1.86
`
`CMT
`
`45.45
`9.16
`12.11
`5.37
`8.73
`9.18
`4.5
`
`PT
`
`0.07
`0.2
`0.14
`0.31
`0.15
`0.15
`0.39
`
`Table 2. Comparison of the dynamic selection with various
`dispatching rules, with loose due dates
`
`Rule
`
`MFT
`
`SPT
`SIx
`MOD
`CEXSPT
`CR/SPT
`SFSR1
`SFSR2
`
`17.4
`19.9
`25.9
`20.3
`23.8
`21.1
`26.3
`
`MT
`
`1.75
`0.04
`0.13
`0.1
`0.03
`0.03
`0.21
`
`CMT
`
`PT
`
`85.17
`4.55
`6.48
`1.89
`2.81
`3.33
`1.29
`
`0.02
`0.007
`0.01
`0.04
`0.007
`0.007
`0.16
`
`measures collected on 5000 jobs. The performance measures
`collected are the mean flow time (MFT), the mean tardiness
`(MT), the conditional mean tardiness (CMT), and the
`proportion of tardy jobs (PT). These measures are averaged
`over the ten replications. The results are given in Tables 1
`and 2.
`In order to find out the significant dierences between
`the strategies, we used a statistical test, based on 0.95
`confidence intervals of the means of the dierences between
`the results of each couple of strategies. This test is known
`as the paired-t confidence interval method (Law and Kel-
`ton, 1982).
`Table 3 lists the best three strategies for each criterion,
`with loose or tight due dates (i.e. mean flow allowances of
`60 and 30 time units), and the average of the percentage of
`the dierence between the best and second-best strategies
`and the second and the third-best strategies. Note that all
`mean percentage dierences are significant at a = 0.05 ex-
`cept those indicated with an asterisk.
`From these tables we can see that SFSR can overcome
`the dispatching rules on primary objectives. In the jobshop
`example, the SFSR2 strategy was found to perform the
`best on the conditional mean tardiness as the primary ob-
`jective. The results of SFSR1 were the best on the mean
`tardiness, except in the case of tight due dates, where
`CR/SPT gives slightly better results.
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1024, IPR2022-00681, Pg. 5
`
`
`
`46
`
`Kouiss et al.
`
`Table 3. Best three performing rules under each criterion
`
`Criterion
`
`Mean
`tardiness
`
`Mean cond.
`tardiness
`
`Mean
`flowtime
`
`Proportion
`tardy
`
`Due date
`tightness
`
`Loose
`Tight
`
`Loose
`Tight
`
`Loose
`Tight
`
`Loose
`Tight
`
`Rank
`
`1
`
`SFSR1
`CR\SPT
`
`SFSR2
`SFSR2
`
`SPT
`SPT
`
`SFSR1
`SPT
`
`2
`
`CR/SPT
`SFSR1
`
`CEXSPT
`CEXSPT
`
`SIx
`SFSR1
`
`CR/SPT
`MOD
`
`3
`
`SIx
`CEXSPT
`
`CR/SPT
`CR/SPT
`
`CEXSPT
`CEXSPT
`
`SIx
`SFSR1
`
`Average percentage
`of the dierence between
`2 and 1
`3 and 2
`
`1.0*
`5.6
`
`46.0
`19.3
`
`14.4
`21.8
`
`1.0*
`100.0
`
`33.0
`18.7
`
`48.7
`62.6
`
`2.0
`3.3
`
`1.0*
`7.0
`
`*Indicates that the dierence is not statistically significant at a = 0.05
`
`5. Conclusions and perspectives
`
`In this paper we have presented a distributed dynamic
`scheduling approach based on a multi-agent paradigm.
`Each agent makes local decisions, under the control of the
`supervisory agent, by selecting the dispatching rules that
`are the most suited to meet the global production objec-
`tives. The knowledge used by each agent has been gathered
`from our literature review on dispatching rules, and from
`multiple simulation experiments. This knowledge is generic
`enough to be implemented in agents responsible for work
`centres of a large variety of FMS, but it needs to be
`completed to take into account such problems as assembly
`and machine selection.
`This approach turns out to be capable of yielding quite
`good performance, as was demonstrated on the jobshop
`example. Moreover, the use of agents oers more modu-
`larity, so that it becomes easier to change the decisions
`related to a work centre, and to reuse parts of the software
`for another work centre (creation of a new agent) or an-
`other factory. It may also be noted that, because of the
`distributed implementation and the ‘autonomous’ behavi-
`our of an agent, the global system is less disturbed in the
`case of a computer malfunction.
`Nevertheless, this approach has only been tested using
`simulation. Our perspectives are to implement it on a real
`FMS, so as to improve it. The addition of other functions,
`such as monitoring, is also contemplated.
`
`References
`
`Attoui, A., Hasbani, A. and Maouche, A. (1995) Specification
`environment for multi-agent systems based on anonymous
`communications in the CIM context, in Proceedings of IMSE,
`European Workshop on Integrated Manufacturing Systems
`Engineering, 12–14 December, Grenoble, pp. 243–252.
`
`Ayel, J. (1994) Concurrent decisions in production management.
`Integrated Computer-Aided Engineering Journal, special issue
`on ‘AI in manufacturing and robotics’.
`Baker, K.R. (1984) Sequencing rules and due-date assignments in
`a job-shop. Management Science, 30, 1093–1104.
`Baker, K.R. and Kanet, J.J. (1983) Job shop scheduling with
`modified due dates. Journal of Operations Management, 4,
`11–22.
`Baptiste, P. and Manier, H. (1993) Pilotage d’un anneau flexible
`de production: une approche de´finissant un comportement
`autonome par station, in Proceedings of GSI4: Quatrie` me
`Congre`s International de Ge´nie Industriel, Marseilles, 15–17
`December, pp. 117–126.
`Barbuceanu, M. and Fox, M.S. (1994) The information agent: an
`infrastructure agent supporting collaborative enterprise ar-
`chitectures,
`in Proceedings of 3rd Workshop on Enabling
`Technologies: Infrastructure for Collaborative Enterprises,
`Morgantown, WV, USA, pp. 112–116.
`Barrett, R.T. and Barman, S. (1986) A SLAM II simulation study
`of a simplified flow shop. Simulation, 47, (November), 181–
`189.
`Blackstone, J.H., Phillips, D.T. and Hogg, G.L. (1982) A state-of-
`the-art survey of dispatching rules for manufacturing job
`shop operations. International Journal of Production Re-
`search, 20, 27–45.
`Chandra, J. and Talavage, J. (1991) Intelligent dispatching for
`flexible manufacturing. International Journal of Production
`Research, 29, 2259–2278.
`Eilon, S. and Cotteril, D.J. (1968) A modified SI rule in job shop
`sequencing. International Journal of Production Research, 7,
`135–145.
`Engell, S. and Moser, M. (1992) A survey of priority rules
`for FMS scheduling and their performance for the bench-
`mark problem, in Proceedings of the 31st IEEE Conference
`on Decision and Control, Tucson, AZ, December, pp. 392–
`396.
`Ferber, J. (1993) Mode`le de syste`mes multi-agents–Du re´actif au
`cognitif,
`in Proceedings of INFAUTOM’93 ‘du traitement
`re´ parti aux syste`mes multi-agents et a`
`l’autonomie des sys-
`te`mes’, Toulouse, 18–19 February, pp. 26–56.
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1024, IPR2022-00681, Pg. 6
`
`
`
`Using multi-agent architecture in FMS for dynamic scheduling
`
`47
`
`Finin, T. et al. (1992) Specification of the KQML Agent Com-
`munication Language, DARPA Knowledge Sharing Initia-
`tive, External Interfaces Working Group.
`Inecker, C., Ferber, J. and Zawadzki, D. (1991) Un mode`le
`multi-agents pour l’aide a` la coope´ration de produits e´lect-
`rome´caniques, in Proceedings of Onzie`mes Journe´ es Interna-
`les Syste`mes Experts
`tionales
`et Leurs Applications,
`Confe´rence Ge´ne´rale Syste` mes Experts de Seconde Ge´ne´ra-
`tion, Ed. EC2, Vol 2, Avignon, May, pp. 137–151.
`Kiran, A.S. and Smith, M.S. (1984) Simulation studies in job shop
`scheduling: a survey. Computers and Industrial Engineering,
`8, 46–51.
`Kwok, A. and Norrie, D. (1993) Intelligent agent systems for
`manufacturing applications. Journal of Intelligent Manufac-
`turing, 4, 285–293.
`Law, A.M. and Kelton, W.D. (1982) Simulation Modelling and
`Analysis, 2nd edn, McGraw-Hill, Singapore.
`Lefranc¸ois, P. and Montreuil, B.
`(1994) An object-oriented
`knowledge representation for intelligent control of manu-
`facturing workstation. IIE Transactions, Industrial Engi-
`neering Research and Development, 26, (1), 11–26.
`Mebarki, N. (1995) Une approche d’ordonnancement temps re´el
`base´e sur la se´lection dynamique de re`gles de priorite´, Doc-
`toral dissertation, Universite´ Claude Bernard lyon I, Lyon,
`France.
`Montazeri, M. and Van Wassenhove, L.N. (1990) Analysis of
`scheduling rules for an FMS. International Journal of Pro-
`duction Research, 28, 785–802.
`Ouzrout, Y. (1996) Mode´lisation et simulation d’organisation
`productive re´actives: une approche multi-agents, PhD dis-
`sertation, INSA de Lyon, 96 ISAL 0032, Lyon, France.
`
`Parunak, H.V.D. (1993) Industrial applications of multi-agent
`systems, in Proceedings of INFAUTOM’93: Du Traitement
`Re´parti aux Syste`mes Multi-Agents et a`
`l’Autonomie des
`Syste`mes’, 18–19 February, Toulouse, Session B3 P. 16.
`Pierreval, H. (1992) Expert system for selecting priority rules in
`flexible manufacturing systems. Expert Systems with Appli-
`cations, 5, 51–57.
`Pierreval, H. and Mebarki, N. (1997) Dynamic selection of dis-
`patching rules for manufacturing systems. International
`Journal of Production Research, in press.
`Russel, R., Dar-EL, E.M. and Taylor, B.W. (1987) A compara-
`tive analysis of the COVERT job sequencing rule using
`various shop performance measures. International Journal of
`Production Research, 25, 1523–1540.
`Schultz, C.R. (1989) An expediting heuristic for the shortest
`processing time dispatching rule. International Journal of
`Production Research, 1, 31–41.
`Tacquard, C., Baptiste, P. and Manier, H. (1994) Model of a be-
`havior approach to schedule and control a flexible manufac-
`turing ring: an extension to the pull flow, in Proceedings of
`IMSE, European Workshop on Integrated Manufacturing Sys-
`tems Engineering, 12–14 December, Grenoble, pp. 333–339.
`Trentesaux, D. and Tahon, C. (1995) Dynamic and distributed
`production activity control: a multicriteria approach for task
`allocation problematic, in Proceedings of IEPM’95 Interna-
`tional Conference on Industrial Engineering and Production
`Management, Marrakech, April, pp. 137–155.
`Trouilhet, S. (1993) Pre´sentation et traitement des connaissances
`sociales chez l’agent: Application a` l’environnement multi-
`agents SYSNERGIC, Doctoral dissertation, Universite´ Paul
`Sabatier, Toulouse, France.
`
`
`
`View publication statsView publication stats
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1024, IPR2022-00681, Pg. 7
`
`