半導(dǎo)體生產(chǎn)線動(dòng)態(tài)實(shí)時(shí)智能調(diào)度方法研究.pdf

1、內(nèi)容摘要生產(chǎn)調(diào)度是在不增加或少增加投入的情況下，通過充分組合和利用現(xiàn)有資源，提高企業(yè)競(jìng)爭(zhēng)力的最有效手段之一。半導(dǎo)體生產(chǎn)線的結(jié)構(gòu)復(fù)雜、設(shè)備多且加工特性各異，具有嚴(yán)重的可重入性、高度不確定性和多目標(biāo)優(yōu)化特征，所有這些給半導(dǎo)體生產(chǎn)線的調(diào)度帶來了極大的困難。在綜述半導(dǎo)體生產(chǎn)線調(diào)度特點(diǎn)的基礎(chǔ)上，對(duì)半導(dǎo)體生產(chǎn)線的動(dòng)態(tài)調(diào)度進(jìn)行了深入而細(xì)致的研究，給出了半導(dǎo)體生產(chǎn)線多目標(biāo)優(yōu)化動(dòng)態(tài)調(diào)度規(guī)則〔MODDMultiObjectiveOptimizationDy

2、namicDispatchingRule).MODD包括五種類型的調(diào)度規(guī)則:正常生產(chǎn)狀態(tài)調(diào)度規(guī)則、瓶頸設(shè)備低在制品水平調(diào)度規(guī)則、非瓶頸設(shè)備高在制品水平調(diào)度規(guī)則、多批加工設(shè)備調(diào)度規(guī)則與緊急工件調(diào)度規(guī)則。該算法考慮了半導(dǎo)體生產(chǎn)線的本質(zhì)特點(diǎn)，如可重入流、多批加工、緊急工件、次序相關(guān)的準(zhǔn)備時(shí)間等等能夠同時(shí)優(yōu)化半導(dǎo)體生產(chǎn)線多個(gè)性能指標(biāo)，如MOVEMENT、加工周期、生產(chǎn)率與準(zhǔn)時(shí)交貨率。但是該算法的局限性在于:在半導(dǎo)體生產(chǎn)線瓶頸變化頻繁的情況下，可

3、能會(huì)影響調(diào)度決策的快速性。群體智能理論為半導(dǎo)體生產(chǎn)線動(dòng)態(tài)調(diào)度提供了新的解決思路。在充分理解群體智能理論的思想的基礎(chǔ)上，提出了半導(dǎo)體生產(chǎn)線動(dòng)態(tài)實(shí)時(shí)智能調(diào)度方法。研究分三個(gè)階段進(jìn)行:第一階段:基于信息素的間接交互方式，提出了基于信息素的半導(dǎo)體生產(chǎn)線動(dòng)態(tài)實(shí)時(shí)智能調(diào)度算法(PBDRPheromoneBasedDynamicRealTimeSchedulingAlgorithm)。首先，模擬蟻群生態(tài)系統(tǒng)，構(gòu)建了實(shí)現(xiàn)半導(dǎo)體生產(chǎn)線動(dòng)態(tài)調(diào)度的MAS系統(tǒng)

4、(SMAS)。在該系統(tǒng)中，使用每個(gè)螞蟻agent分別控制相應(yīng)的工件、設(shè)備、運(yùn)輸工具與人員，將與調(diào)度相關(guān)的信息表示成相應(yīng)的螞蟻agent的信息素。螞蟻agent通過感知其他螞蟻agent的信息素來確定自己下一步行為，即選擇合適的設(shè)備等待加工或選擇合適的工件進(jìn)行加工，從而實(shí)現(xiàn)動(dòng)態(tài)調(diào)度。該算法有兩方面的優(yōu)勢(shì):一是將調(diào)度相關(guān)信息表示成螞蟻agent的信息素后，可以根據(jù)要優(yōu)化的性能指標(biāo)，來相應(yīng)地改變信息素的表示方式，對(duì)調(diào)度的結(jié)構(gòu)卻不發(fā)生影響，可以

5、方便地實(shí)現(xiàn)方法的重用二是決策時(shí)間短、計(jì)算量小、實(shí)時(shí)性好、易于實(shí)現(xiàn)，非常適用于動(dòng)態(tài)調(diào)度。PBDR的局限性在于缺乏對(duì)整體性能的把握和預(yù)見能力，因此其調(diào)度結(jié)果可能會(huì)與系統(tǒng)的全局優(yōu)化有一定的偏差。第二階段:為了盡可能縮小PBDR的調(diào)度結(jié)果與全局優(yōu)化的偏差，提出了集成優(yōu)化的半導(dǎo)體生產(chǎn)線智能調(diào)度模型(ODISMOptimizedDynamicIntelligentSchedulingModel)。該模型由兩個(gè)部分組成，即分布式動(dòng)態(tài)控制人工蟻群MAS

6、系統(tǒng)(DMAS)與集中優(yōu)化控制MAS系統(tǒng)(CMAS)oDMAS是在SMAS的基礎(chǔ)上，引入了蟻后AbstractSchedulingisoneofthemostefectivewaystoefficientlyutilizeexistingresourcestoimprovecompetitivepowerofenterpriseswithnoorlessinvestment.Semiconductorwaferfabhasitsownp

7、roductionstyleandschedulingcharacteristicssuchasreentrancehighuncertaintyandcomplexityandmultipleoptimizedobjectives.Asaresultitisdifficulttodirectlyapplysuccessfulschedulingmethodsinjobshopandflowshoptosemiconductorwafe

8、rfab.Onthebasisofsummarizingschedulingcharacteristicsofsemiconductorwaferfabmultiobjectiveoptimizationdynamicdispatchingrule(MODD)isproposed.TherearefivekindsofdispatchingrulesinMODDi.e.defaultdispatchingruledispatchingr

9、uleforbotleneckmachineswithlowerWIPleveldispatchingrulefornonbottleneckmachineswithhigherWIPleveldispatchingruleforbatchingmachinesanddispatchingruleforhotlots.MODDhasconsideredtheintrinsiccharacteristicsofsemiconductorw

10、aferfabschedulingsuchasreentrancebatchingmachinehotlotandorderrelatedsetuptime.Itcanoptimizemultipleperformancessuchontimedelivery.ThedeficiencyofMODDwithfrequentchangeofbottleneckmachines.asmovementcycletimethroughputan

11、disnotapplicabletosemiconductorwaferfabSwarmintelligenceofersanovelsolutiontodynamicschedulingofsemiconductorwaferfab.Basedondeeplycomprehendingswarmintelligenceideasemiconductorwaferfabdynamicrealtimeduringtheresearch.i

12、ntelligentschedulingapproachisproposed.Therearethree曲asesThefirstphase:pheromonebaseddynamicpresented.Atfirstimitatingantcoloniesthescheduling(SMAS)isconstructed.Eachantrealtimeschedulingalgorithm(PBDR)isonemachineschedu

13、lingisMASforsemiconductorwaferfabdynamicagentinSMAScontrolsoneworkpieceororonetransportationtooloroneperson.Theinformationrelatedtoexpressedaspheromoneofantagents.Theantagentsensesotheranttodetermineitsactioni.e.selectin

14、gagentsselectingproperworkpiecetoprocess.AsTherearetwoadvantagesinPBDR.Onepropermachinetobeingprocessedoraresultthedynamicschedulingisrealized.beonisthattheexpressionofpheromonecouldeasilymodifiedaccordingtorequiredoptim

15、izedperformanceschedulingstructure.SoPBDRcouldmakingtimeofPBDRisshortandthe，withoutimpactingbereusedeasily.TheothercomputationofPBDRisless.berealizedandcouldquicklyresponsetochangedenvironmentisthatthedecisionSoPBDRiseas

16、ytoapplicabletodynamicscheduling.ThedeficiencyofPBDRmaydeviatetotheglobaloptimization..AsaresultPBDRisthatitsschedulingisveryresultsSecondphase:tocomplementthedeficiencyofPBDRschedulingmodel(ODISM)isproposedwhichisoptimi