Architecture of fully connected networksAme layers cannot correspond directly, but have to be finished by the upper layers Thegents of uake charge of dtrolsEach agent in this structure neednt save all the agents' information in the system, just savethe under layers agents related information and knowledge, but it is inferior to thestructure of fully connected networks in correspondenceFig 12 Architecture of fully layered networks(3)Archilereof every ally, it is in charge of different allies correspondence
Different allies are in theopposite relation, similar to the relations of every agent in fully connected networksFig, 13 Architecture of allied networks3 correspondence in the multi-agent systemdifferent agents and agents and the circumstance, thenThere are four kinds oftarget agent directlyby certain protocol, sucTCP/TIP, In this wtion to the target
Coordination Control and Fault Diagnosis of Production System Using Multi-agent Technology 183working time is expressed by the formula 3), the relationship between the numbers ofof faults Coefficients k], k] and c can be given by practical systking time and the dependability of normalachine is not linear, we simplified it to linear as the Fig 2 4 The calculation of im is givenbring tme TFig 2 4 The relationship between reliability and working hoursLet Tsm be the life-span of agents; let Tw be the attrite life-span of agent; let Tas be theoriginal life-span of agents; let Tw be the whole working time of agents
The computation ofMint is giv(6)≤TT(x-Tm)…T
3 3 The factor of influence to the eeans the smallest influent upon t the environment when s/ 1 and is the biggest when Si 0be decided bythe agent based on the practica2 4 Definitions of some interrelated matrix and its calculation methodsf agent A TheThe set of types of tasks that the agentsperform is denoted by T Every agent a e asksable to perform only tasks thatsubset of the overall set of types of tasks in thePi-I if agent al can carry out a task of t; and Pi 0 if agent a; can not carry out a task of t 2rstem Wethat there is a binaan be described as a matrix 12tasks T=(,, t, t3, t4 Is)
The firstcarry out the tasks of type fr and ts thethe tasks of both ty+s ta and ts, the third agentf3 and ts The relation p can be describes as the Fig 25aFig 2 5 An exmple of the relation between agents and tasksThe relation p alsocan be defined a matrix as00105nl, and Si=(i, l Fidonisitrix shows thestates of each agent in the system and the meanings of the symbols in the formula are thatpresents the tab of the agent in the systemI represents the residual tasks for the agentagentA, represents the integrative reliability of the agent8, represents the average energy consumption of agentsi represents the influence level to the eny243Let the dissipated matrix be Chc|1≤isnSthat thne thea agent to accomplishto be the basic dissipatetrix,and define the generalized
Coordination Control and Fault Diagnosis of Production System Using Multi-agent Technology 185consumption which consider the reliability of agents, the energyfactors affect to thebe the generalized dissipatrix The basicThe generalized dissipated matrix can be calculated as(9)1×f(5)×52When the inofλ f(oi) and s equalhe generalizedrix do not change, If one of the factors equal to O when agewe will consider ai has not the ability to fulfill the task and supposed it didrmal situation, the range of Ai, f(oi) and s are [0,1], and Ci will be biggerthe influence of those factors4 4 Task allocation matrix xmThe value of xu is 0 or 1 when xu=1does not to allocate t, to agent a: A task allocation matrix shows a way which the task can be2
5 Thes and allocation arithmetic of the task allocationCertain conditions should be satisfied when in the task allocation here we call thoseonditions as the limiting conditions, when the tasks have been allocated each of the agentther the tasks allocated to them satisfy the limiting condition or not, if thtask satisfy the limitondition, the agent performed it next, if not, the tasks need to bedecomposed again (the workte the tasks again In the paper, the limiting condition is gias follows(here suppose the task t, have been allocated to the agent airsEr= ELet Eimer be the maximal task the agent ai can receive and let Eimwin be the minimum task thethe tasks the agent ai is dealing witfigeneralized dissipated matrix as coefficient matrix in Hungary arithmetic to allocate tasks,he task allocation process is as folAllocating the tasks to the task allocation agent by person
hen the task allocation agent receives tasks, it sends the status request to each agent;eives the request sends its status to the task allocation agentThe task allocation agent figure out the generalized dissipated matrix as the coefficientnatrix in Hungary arithmetic based on Equ (8)and Equ (91 Judging the relationship between rows(m) and columns(n) in the matrix, if m=n, usingthe standard Hungaryto allocate tasks (take the generalized dissipatedm*n, then we can translate the coefficient matrix(thegeneralized dissipatedallocate tasks throughHungary arithmetic takes the square matrix as theThe task allocation agent send the task allocation information to agents, then agentsho received the message check whether the task satisfy the limiting condition or not, ifdecompose the tasks and assign to the task allocation agent, thenthe step 2Definition ofulti-agent cooperation refers to a behavior of multi-agent's assorting of themselves togoal Most document regard cooperation as a kind ofnt, On thethat both thened, there is no conflict of targets, an agentplashes theDefinition 3-2omous agent uses another agent's goal, on the assumption that theappends betweenpacity of accepting or refusingDefinition 3-3: between two autonomous agents, if either meets one of the acquirement,say that the two agents cooperate
n agents there is conflict, butn still reach a balanceagents finish their own targets due to their exchangee can see that cooperations goal is to make two autonomous agents get a common targetperation is a interaction to make two orchange information and finish a task togethergents cooperate to finish32Don of multDefinition 3-6: coordination means each agent infers and disposes its behavior in order toDefinitioncoordination is the interaction among group of agents taking theaction, is the adaption of the environment The agent chaits willing to get coordDefinition 3-8[ 17]: coordinatiaction desires and makes decision to let all the member get into harmony and consistency
」Fautsis of Production System Using Multi-agent TechnologyTypical assorting includes timely delivering messages between agents, guarantying theMintzberg considers the three basic coordination proceduresllatual adjustment is the most easy form of coordination It happens when twogents agree to share the sources to get a certaion target Agents always need toexchange information, and continuously modify their behavior according to otheragent's behavipacity of controlling others, this kind of priority relation usually erects by mutual3 Standardization is an usual way to assortertain circumstance, the manager assortin a standard way, namely to erect a standard procedure for its subordinate to follow indThe basic idea of blackboardblackboard isestion and original datarded on the blackboard All experts see the blackboardnd find opportunity to solve question by others' experience knowledge, A solution isexpert finds enough information to makeuntil thealways use some information on the blackboard to modify the information of anotherblackboard layer There are two parts in KS: precondition and action, Precondition isblackboard Actionescribes operation which KS effects blackboard, it is a process
When the informationchange accords with the precondition of ks, the ks is activated and cait adds, deleteupdates solution elements Each Ksndependent and cannot direct call mutually, then only communicate with blackboardontrol mechanism is in charge of monitoring information change on blackboard andecks ks precondition continuously Once some Ks precondition is tenable, the Ksaction is carried out the information on blackboard is modified whichtill find the final solutionBlackboard: solvingrganized in hiberarchy, it storesinformation and state data, such as initial data, part solution, substitution solution andfinal solutiontrol data is storedseekingand the information in blackboard only be added deleted and modified through KMechanism of supervise and control according to the problems on the blackboard andsolution skills of Ks, adaptive Ks is activated which made Ks fit for the blackboardhange The design of control mechanism is the most complex task Its object is to exertK
roduction lines system based on multi-blackboard341 Multi-blackboard mechanism model of multi-production lines systemBi nt sends messages to and" peration mechanism applies a public blackboard, which eachsages from hlis model, blackboard is classified into there different levels: central blackboard, middled Different hierarchic agent of the system ist has itsknowledand each agent is given a blackboardThe division among the three-level blackboard and corresponding knowledgeThe main responsibility of central blackboard is to managehigh-level system administration and coordination, for example, the beginning and endingthe overall allocation of resources, and the fault diagnosis The responsibilitymiddle-level blackboard lies in the resources allocation and administration of subsystem,the cooperation and fault diagnosis of subsystem, reporting the movement of subsystem toh-level blackboard and resources request The responsibility of rock-bottom blackboaris to coordinate operation of related agent combination in subsystem, report staggerederyblackboard
The data produced from each agent is classifiedresult date" and "middle data" Besides added into itsbackboard, resultdata need to be added into the blackboard of upper agent at the same time in orderMiddle data need not be addblackboard and only need be added into its own blackboard Thus the quantity of datared and processed by middle-level blackboard and high-level blackboard will reducesignificantly, and in particular when the amount of data in system is large the efficiency oftem wilestablished in fEach agentin the system and differdifferent levels blackboards Each agent in the systemindependently, so theremust be haverdination mechanism toplete the coordination among differenlevels agent and different agent at the same level, in order toerall system The qualitative analysis and quantitative analysis of the system and dynamicalysis to agent system can be accomplished by applying Petri Net to describe the complexlulti-agent system, which time factor is added into Petri Net and Petri Net coordination1 Coordination model of multiFor different levels blackboard, there are agents in three different levels, which are centerent decision agent, preagent and stand-alone agent They work at theme clock signal which called in synchronous forms In conveniency, there are three standalone agents and two production lines agents, The Petri Net coordination model of mu
Coordination Control and Fault Diagnosis of Production System Using Multi-agent Technology 189KKScenter blackboardcenter KKsmiddle-levelddle-level KKsblack boardbottom Kksnter controlbottom controlFig 3
1 Multilevel blackboard model of packing line system+"Fig 32 Petri Net coordination model of multi-agent system
Coordination Control and Fault Diagnosis of Production System Using Mul
ti-agent Technology 175(2)Combine into allies freely, then correspond by Correspondence serverWhen there are too many agents, the cost is expensive by wholly direct correspondenorrespondence server as the media Structure of allied system is as Fig l emong agen &,ocombine mally haspendence function, that is there's no direct correspondencAgentAgent(3)Broadcast corIf an agent need send information to all the agents in the circumstance, or it doesnt knowdirectly to everyby correspondenceWhen the messagwill burden theshares of the message and send them separately, therefor orrespondence needn't copy manyavoid the big burden of the(4)Blackboard Correspondencethe traditional correspondence in the field of artificial agents All the agenaway and read information inor blackboard) In order to realize the secrecyof part of information, the prerally divided into different abstract layd rightsHarmony and Cooperatiethe basic conception of multi-agentnd makes decisions to realize a harmonious work procefor fulfilling theto fulfill their cooperation and solutionl, The ways of cooperation among multi-agent amainly contract net, blackboardnd their relations in the system it describes theependent relations in the accidents,
and so on It is fit for describing the system which has rules and isfeaturedsuch as substance flows information flows and so onThe structure factors of a Petri Net mainly include place, traandPlace mainlydescribes the possible localsystem,fault symptom andhenomenon in the fault diagnose or buffer in the computeron is useddescribe the incident of modifying the system status, suchprocess andansition,describing the direction of the system status changIn Petri Net model, signs are included in the place, their dynamic changes in therepresent the different status of the system If one placethe signs represents the amountIf one place describes15, the circles represent the place, the thin sticks represent the transition, and thelines represent the arc, and the black dots in the place represent the signsP22 The definition of petri NetDefinition 1-1 A three tuple N=(P, T: F)is called directed net, shortened form Petri, itssufficient and necessary conditions arePnT=Φ2
PTΦFc(p×n)(TxPthere, dom(F)=kx|彐y:(x,y)∈F,cod(F)=kx|彐y:(y,x)∈ FI are defining regiond value region of Fthe net p and tno-intersectant set, called basic factors set of net N, P is places setof net N t is transitionsnet N, f is the flows relations of net n one net can beby a directed dimidiate figure: generally little dots represent place Ipresents transition T, the arrows from x to y represent the (x, y) in the flowits description is shown in Fig 161-2 Prepositive set and postpositive set Set N=(P, T: F) is a Petri Net, X-PUTlyly, x)E F/ is called Prepositiyofro=ly I(x, y)E F is called postpositive set of x
Coordination Control and Fault Diagnosis of Production System Using Multi-agent Technology 17(x,y)∈Fn(P×T)(x,y)∈FnP×T)If placeT, makes P E t n- T, then when t happwill lose andken This feature in the structure shows that thein p affect t similar to thechemical reaction the net which hasnt thisOn the other hand, if x,yEX, x*y, butxuyAx my", then no matter in structure or int distinguishes from y, the net which hasn't this structure is called simplheretransitions which have the same input andDefinition 1-3 If VΦ, then n is called piBx, then n is called simple net2 3 Petri Net systeets and the original sign relatet In the circumstance withor andtion thePetri Net is Petri Net system In the process from net to net system, the original distribution ofurces has to be demonstrated, the activity rule on the frame has to be regulatedDefinition 1- the conditions whichtuple∑=P,T,F,KMo is a Petri NetN=(P, T, F)is a Petri Net, is called basis net of 22
K: P-)N*Ufo/ is capacity functionW:F→N4M:P→ No is original mark, it satisfied:Vp∈P:MP)≤Kf∈Fw)>1, labels w() on the arc When the capacity of place is limited, generally writes k(p)onhe side of the circle of place P, and when K(p)-1, the sign is generally omitted The black dotsof place represent the original sign which represents a kind of resource distribution in placeDefinition 1-5 The condition of trtion that t has friable in mtMs≥Ws)∧Vs∈tM(s)+Wts)≤K(sLabels that f has friable inand M enables f happen or f enabled happens by M Hererepresents all the inputset of trepresents the amount of input place ofrepresents the amount of output place
fif pPetri Net has descritof various structures, these struother net system of all levels, and also instruments in the basic phrelated theories research here are the most basic structuresPPConflict: One of transition happens in the transitions of fi, t2, ts, other two cant happen,the substance of conflict is the competition of resources, t,3 Subsequence: Transitionhappen at the4 Synchronization: Transition can happen just in the circumstance of having all the5 Union: The happens of transition fr, f2, fs affect the same resource, if one ofappen,6
Mixed: The concomitant status of subsequence and conflictThe task allocatioulti-produce linem of the task allocation is a kind of typical problem in combined optimization, iton, equipment collocation, production arrangement and printed circuit boarddesign and so onIn the system of multi-agent, the task allocation, s mechanism is one of the research hotspotsone aspect whether or not it can make the abilitylete the tasks together through the effective dialogue and the negotiation if one of thegents did not has the ability to complete its task Task allocation mechanism establishmentis the foundation of studying the multi-agent cooperation [7] There asynth agent task allocation: task decomposition, task allocation, task solution and result22Tasentration and distribution In then-made allocation in advance, which the tasko agent in the system The other is one a
Coordination Control and Fault Diagnosis of Production System Using Multi-agent Technology 179task decomposition(which is called Trader), this agenttable ofthe able agent whether accepts this task, if it receives the agreeabltell the promulgator of the task, or tell that no agent can fulfill theble intormation Trader willpresent In thedistribution there are also twoacquaintance and contract net2 The mechanism of task allocationThe mechanism of task allocation mainly ha1 General market balancel8l: general market balance furnishes a structure of ddominated by market with son y mme commodities, and thesees that can deal with the task Every commodity is thought limiuccessive
There are two kinds of agents in market: manufacturers and consumers: thederedfactumhile the agents whichbalance the market by bargainThe way of auction is widely used in the task allocation of muAuction is a market mechanism, it decides resources allocation and price by a series ofclear rules Price decision is based on the huckster of market anticipantslsContract net: Contract net is an important mechanism of task allocation; it is widelysed in the arithmetic of task allocation the basic ideatasks to be solved by other agents, it broadcasts thetasks to other agents,d thehe ability tosolve the problem; and then send out its value of bidding and become the bidderthose values and elect the most appropriate bidder toaward the task, that is tt fulfill the negotiation process according to themechanism of tender -bidding-selected as itthe market9 The task allocatenegotiation process is given as the Fig 21Manager agentM Task notifyREceivedReceived tenderAccept the task
21 The goal of task allocatThe goal of task allocation is to find a feasiblmake the system achieving the set goal and minimize the sumthe tasks The dissipation needs to synthesize factors of the tinlity of the system, the energy consumption of the system and the effect tothe environment of the system222 Principles of task allocation&s presents the time agent i finish the task ) It contains two parts of the time to fulfill the taskeach agentSuppose the load of agent i is Li, that is, it need to satisfy L1-L2 Ln, thisk withload, avoiding the phenomena thatlongtime but some of agents work with relatively little load
The load of each agent needs toThe reliability needs to bed in the task allocation The reliability needs synthesizethose influence factors: rate of average failure of system, the working time of system and theorking environment of system The system which a person has joined must consider merfactor Here, we did not consider the incredible because of cheating of agents; the reliabilityd by the failure theking time and the environmentTask allocated assign the task for the equipment which has thehighest reliabilityDifferent quantity of energy ption4 Minimize the energy coronsumed when accomplish a particular taskifferent agents, task should be allocated to the agent who with minimal energyonsumption when accomplish the task5 Minimize the influence to thewill cause the different influence to the environment, we should allocatere tasks to the system which with the minimal influence to the23 Assumptions and simplion to the practical systemthat is, a particular task may take a differentment if executedfor an agent to deal with different type of taskthe difference at ability, that is, if an agent can solve differentypes of tasks, theg those tasksDeeming theate costsO Speaking strictlyarts of the time are cost tofinish the task the time is used forting and the timnication(101accord with the practical system andmake the discussion
Coordination Control and Fault Diagnosis of Production System Using Multi-agent Technology 18The tasks needed to be allocated are independent, that is, there are not dependenif wefinish t(irn)The agents are absolutely honest That is, deeming the values of state which are returned bynts are absolutely believable in the article There is no cheating when send the values ofstate for all the agents(include personinttion, but we will not take the task decomposein the article and think the task have been decomposedbe done by the former agent23 Factors influence on task allocation and its compution nThere are three factors influumption of agent; the factor influence on the environment; let A behe integrative reliability of agent i; let d be the average energy consumption of agent i andlet f(0i)be the factor of energy consumption; let si be the influence level to thenvironment of agent i The computational methods are as follow2
31 The integrative reliability computattegrative reliability of agents and the dimension is equal to the number ofcompletely credibd unbelievable when 1Auf be the reliability related to faults: let wt be the reliability related to the working timend let Mine be the reliability related to workingl the values of that variableare fall into [0,1The computation of integrative reliability Ai is given by (1)Assuming agent i has gone wrong N times and its maximal allowed number is M, then AoptedThe reliability is the highest when Agz=1 and is the lowest when lig: =0 The computation ofM is given by (3)and (4)based on the relationship between rate of equipment faults and theworking time, as in fig 22t+bGZLk2(Txet-Tmh)T≤T≤T
(c+b)Ir-=kIT(c+b) -=k+b(The attrite periopesiolinear relationship, as in Fig 2
3 There is not much influence on thepractical model after simplified, bubetter for the follow anaThe effectiveInherent faultsFig 23 Simplified graph of equipment fault ratio(5), Into denotes to acquire integer for the content in the parernd the GrL inrelationship betvfaults