人工智能在过程工业绿色制造中的机遇与挑战

Engineering5(2019)995–1002

Contents lists available at ScienceDirect

Engineering

Research

SmartProcessManufacturing:DeepIntegrationofAIandProcessManufacturing—Perspective

OpportunitiesandChallengesofArtificialIntelligenceforGreen

ManufacturingintheProcessIndustry

ShuaiMao,BingWang

⇑

,YangTang

⇑

,FengQian

KeyLaboratoryofAdvancedControlandOptimizationforChemicalProcesses,MinistryofEducation,EastChinaUniversityofScienceandTechnology,Shanghai200237,China

articleinfo

abstract

Smartmtmanufacturing,

thereisatrendtoincorporatedifferentkindsofnew-generationinformationtechnologiesintoprocess-

ent,greenmanufacturingisfacingmajorobstaclesrelatedtosafetymanagement,

duetotheusageoflargeamountsofhazardouschemicals,resultinginspatialinhomogeneityofchemical

industrialpronginforma-

tiontechnologiessuchasartificialintelligence(AI)arequitepromisingasameansofovercomingthese

diffinstate-of-the-artAImethodsandthecomplexsafetyrelationsintheprocessindustry,

weidentifyanddiscussseveraltechnicalchallengesassociatedwithprocesssafety:①knowledge

acquisitionwithscarcelabelsforprocesssafety;②knowledge-basedreasoningforprocesssafety;

③accuratefusionofheterogeneousdatafromvarioussources;and④effectivelearningfordynamicrisk

tandfutureworksarealsodiscussedinthiscontext.

ÓhedbyElsevierLTDonbehalfofChineseAcademyofEngineeringand

anopenaccessarticleundertheCCBY-NC-NDlicense

(/licenses/by-nc-nd/4.0/).

Articlehistory:

Received3January2019

Revised6August2019

Accepted22August2019

Availableonline2November2019

Keywords:

Processindustry

Smartmanufacturing

Greenmanufacturing

Artificialintelligence

uction

Theprocessindustry,whichisasubclassoftherawmaterial

industry,ecadesof

development,theprocessindustryhasmadesignificantprogress

inChina,makingChinaoneofthelargestmanufacturingcountries

[1,2].However,comparedwithdevelopedcountries,China’spro-

cessindustrynowurgentlyrequiresintelligentmanagementand

marketingtechnologiestoincreasetheutilizationrateofraw

materials,whileestablishingmorefunctionalenvironmentaland

suehasdrawnstrongacademic

attention,andmuchprogresshasbeenmadeinrelatedresearch

fields.

Recently,thedramaticgrowthofnew-generationinformation

technologieshaspromptedseveralcountriestoseeknewstrate-

giesforindustrialrevolution[3](Fig.1).TheUnitedStateshas

launchedSmartProcessManufacturing[4],whichaimsatindus-

yhasproposedthe

strategicconceptofIndustry4.0[3],whichfocusesontheintegra-

ted

⇑

Correspondingauthors.

E-mailaddresses:wangb07@(),yangtang@

().

Kingdom(UK),France,andJapanhaverespectivelyannounced

theUK2050strategy,theNewIndustrialFranceprogram,and

context,China’sgovernmenthas

putforwardNewGenerationofArtificialIntelligenceDevelopment

Planfortherealizationofthe‘‘newindustrialrevolution”[5].The

NewGenerationofArtificialIntelligenceDevelopmentPlancon-

centrateson‘‘innovative,coordinative,green,open,andshared

development,”whichiscapableofpromotingintelligent

manufacturing.

Inthecontextofthenationalstrategiesaddressingthenew

industrialrevolution,smartmanufacturingisthecurrenttrendin

theprocessindustry,withgreenmanufacturingasoneofitsindis-

pensablecomponents[1,6–8].Greenmanufacturingfocuseson

high-levelefficiencyandsafety,whichreflectstheneedforstricter

manufacturingcanberealizedwiththreemainobjectivesinmind:

①thereductionofenergyconsumptionandpollutantemissions;

②life-cycleprocess-safetymonitoringandriskcontrol;and

③r,

thereisnowell-understoodwaytoachievetheseobjectives.

Artificialintelligence(AI)[9],whichisacomprehensivefrontier

technology,hasattractedextensiveattentionworldwideforits

extraordinaryperformanceinAlphaGo[10].Today,AIisregarded

asoneoftheworld’sthreemostadvancedtechnologiesandhas

l./Engineering5(2019)995–1002

mentalprogramsproposedbyseveralcountriestoaddressthenewindustrialrevolution.

remarkablyaffectedseveralfields,includingcomputervision,nat-

urallanguageprocessing,er,itiswidely

believedthatAIiscriticalforsmartmanufacturing.

Inthispaper,thechallengesassociatedwithgreenmanufactur-

san

importantroleinimprovingprocess-safetymanagementand

increasingefficiencythroughtheintelligentutilizationofmaterials

ainderofthispaperintroduces

theexistingtechnicalchallengesofintegratingAIintotheprocess

industry[11].Themaincontributionsofthispaperareasfollows:

(1)ThecurrentstatusofprocesssafetyinChina’spetrochemical

industryissummarizedandmajorproblemsforachievinggreen

manufacturingintheprocessindustryareidentified;theseprovide

significantguidelinesforgreenmanufacturing.

(2)InviewofthecurrentstatusandmajorproblemsofChina’s

petrochemicalindustry,ourperspectiveisoffered:AIisthecore

ltechniques

areavailabletoaddressthesemajorproblemsinordertoachieve

greenmanufacturing,includingknowledgegraphs,Bayesiannet-

works,anddeeplearning.

(3)Consideringthecomplexsafetyrelationshipbetweenthe

processindustryandthecharacteristicsoftheprocessindustry,

severaltechnicalchallengesareputforwardregardingtheapplica-

hallenges

mayattracttheinterestoffutureresearchers.

rent

statusandproblemsofgreenmanufacturingintheprocessindus-

ialAItechniqueswithrele-

l

technicalchallengesrelatedtoAIingreenmanufacturingare

progressandfutureperspectives

arediscussedinSection5,followedbyconclusionsandanoutlook

inthelastsection.

tstatusandproblemsofgreenmanufacturing

anufacturing

Itiswidelyacknowledgedthatthemanufacturingindustryhas

experiencedthreeindustrialrevolutionsandisnowundergoinga

fourth(Fig.2).Eachofthefirstthreerevolutionstremendously

contributedtoproductivityandeconomicdevelopment[12].Thus,

itiswidelybelievedthatthefourthrevolution,termed‘‘smart

manufacturing,”willalsomakesuchcontributions.

Petroleumandchemicalsaresignificantcomponentsofthepro-

ingtotheNationalBureauofStatisticsofthe

People’sRepublicofChina,thepetrochemicalindustryhasbecome

oneofthepillarindustriesofChina’snationaleconomy,contribut-

ing12%ithtremen-

douscontributionstothedevelopmentofChina’snational

economy,thepetrochemicalindustryhasbroughtaboutnegative

effectsonbothpublichealthandenvironmentalsafety,dueto

long-termpollutioneffectsandfrequentaccidents[13].Inthe

informationera,industrialaccidentsarereportedandspread

worldwidethroughvariousnewschannelsandsocialmedia,mak-

tivatesthe

governmenttoestablishstricterstandardsandregulationsfor

thecorrespondingindustry,whichincreasesthedemandforgreen

majoraccidentssuchastheXiangshui

dustrialrevolutions.

l./Engineering5(2019)995–1002

997

‘‘3

Á

21”chemicalplantexplosionaccident,theTianjinPort‘‘8

Á

12”

fireandexplosionaccident[14],andtheQingdao‘‘11

Á

22”crude

oilleakingandexplosionaccident[15]notonlycausedserious

casualties,tremendouseconomiclosses,andsevereenvironmental

impacts,butalsohadanegativeeffectonthedevelopmentofthe

ernmentofJiangsu

Provinceplanstoclosemorethanhalfoftheexistingchemical

enterprisesby2022.

Inadditiontoprocesssafety[16],theenvironmentalimpactof

thepetrochemicalindustry—boththeshort-termimpactandthe

long-terminfluence—arisonwithdiscrete

manufacturing,theprocessindustryinChinaischaracterizedby

lowmaterialandenergyefficiency,andseriouspollutionproblems.

GiventhescaleoftheprocessindustryinChina’snationalecon-

omy,thereisanurgentneedtoreduceenergyconsumptionand

processemissionsundertheincreasinglyrestrictiverequirements

ofenvironmentalprotection.

Greenmanufacturing[17,18]isregardedasasolutionthatcan

achieveprocesssafety,energyconsumption,andemissionreduc-

totracksafety-relatedaspectsthroughouttheprocess

life-cyclebyintegratingsmartmonitoring,intelligentearlywarn-

ing,intelligentdecision-making,andoptimization-basedpollution

anufacturingcansignificantly

improvethesafetyandefficiencyoftheprocessindustries,andis

likelytobecomeanecessaryrequirementforhigh-leveleconomic

development.

tstatusandmajorproblems

Thecurrentstatusoftheprocesssafetyandenvironmentalpro-

tectioncapabilityofChina’spetrochemicalindustriescanbe

describedbyaccountingforthefollowingthreeaspects(Fig.3).

(1)ticalreportsissuedbytheNational

RegistrationCenterofChemicalsshowthatin2017,morethan

70000typesofcommonchemicalswereproducedandconsumed

inthemainlandofChina,3962ofwhichwerehazardouschemicals.

Ahazardouschemicalisachemicalthatcanpotentiallybeharmful

tothehealthofahumanoranimal,theenvironment,orproperty.

Withindecadesofdevelopment,Chinahasbecomeoneofthe

teAdminis-

trationofWorkSafetyreportsthatthenumberofhazardous

chemical-relatedenterprisesinChinaexceeds300000,withover

gthofthepipelinesexceeds120000

ateconomicvalueofthesechemicals—especially

hazardouschemicals—hasnotonlyresultedintherapiddevelop-

mentofthegrossdomesticproduct,butalsoledtoincreasingly

serioussituationsrelatedtoenvironmentalprotectionandpublic

safety.

(2)

ofthepopulationdistributionandeconomicmismatchinthewest-

to-eastdirectionofthemainlandofChina,thepetroleumand

chemical-relatedenterprisesaremostlylocatedintheeastern

coastalareas(Fig.4).However,foraspecificchemical,thelife-

cycleprocessesofproduction,storage,transportation,usage,and

wasteusuallyoccurindifferentplants,counties,cities,orevenpro-

ult,processsafetyandenvironmentalprotection

shouldbeconsideredonalargerspatialandtemporalscale;how-

ever,thisisnotaneasytaskduetothecomplexitiesassociated

withthedifferentstages,includingmaterialtreatmentandinfor-

mationfletheseissues,moreeffortisneededininfor-

mationintegrationanddataanalysis,whichcanbeachievedusing

AIandcloudcomputing.

(3)ic

developmentisimprovingthegenerallivingstandardinChina,

makingenvironmentaldeteriorationlessacceptableforresidents

nment-related

issuesareattractingincreasingattentionfrombothgovernment

eofpublicconsciousnessisdrivingthegov-

ernmenttodropthecrudemodeofeconomicdevelopmentand

tion,the13th

Five-YearPlanrequiresa20%reductioninthenumberofmajor

accidentsin2020anda20%reductioninthenumberofrelated

deaths,therebyincreasingthedemandforsmartprocess

monitoringandriskmanagementsystemsinthepetrochemical

industry.

Thecurrentstatusesofprocessindustriesotherthanthepetro-

chemicalindustryarequitesimilartowhathasbeendiscussed

roblemsthatprecludegreenmanufacturinginthe

processindustryarelistedhere.

(1)Informationisolationamongmultiplefields.

Inthepro-

cessindustry,thestagesofproduction,storage,transportation,

usage,r,eachstagefocuses

onitsspecificfieldsandhasitsowninformationanddatabasesys-

ity,theprocessesinthelife-cyclearephysicallycon-

nected;however,theyareusuallyisolatedfromtheperspective

opriateinformationexchange

amongdifferentprocesseshindersthecomprehensiveanalysisof

mple,ifinformationaboutmaterials

andthereal-timelocationofhazardouschemicaltransportation

canbeconsideredinthedynamicriskassessmentprocess,thelike-

lihoodofpotentialaccidentscanbeestimatedandriskcanbebet-

termanaged,withdynamicroutinesandemergencypreparedness.

spectsofthecurrentstatusoftheprocesssafetyandenvironmentalprotectioncapabilityofChina’spetrochemicalindustries.

l./Engineering5(2019)995–1002

butionofchemicalenterprisesinthemainlandofChina.

Thepossibilityofaccidentscanthenbereducedtoanacceptable

ore,integrating

abundantinformationandestablishingknowledgebasesarechal-

lengingtasksthatshouldbeaddressedfirstintheimplementation

ofgreenmanufacturingintheprocessindustry.

(2)Diverseinformationtypesanddifferentkindsofdata.

Fromthelife-cycleperspective,thedifferentstagesof,forexample,

manufacturing,storage,androadtransportationhavetheirown

ifferencesoccurin

mple,thevalues

ofthetemperature,pressure,level,andsoforthareimportant

becauseofembeddedinformationregardingabnormalsituations

comestotransportation,

theroutes,enterprises,real-timelocations,vehiclestatus,anddri-

vers’r,this

informationbelongstodifferentsystemsthataredifficulttocom-

municateorintegrate,nottomentiondifferencesinthedatasam-

plingrate,dataformats,ssues

posedifficultiestotheintegrationofsafety-relatedinformation

er,therearevariousdisciplines

behindthedatacollectedduringdifferentlife-cycleprocesses,

andtheintegrationofbothfactualdataandknowledgeintoacon-

sistentsystemisanotherchallengingtask.

(3)Lackofaprocess-safety-orienteddecision-makingsys-

ge-scaleproductionintheprocessindustry—especially

inthepetrochemicalindustry—geometricmagnificationisthebest

waytoreducecostsandgainthebenefi-scale

productionthenresultsincomplexsupplychainsandmarkingsys-

ustryspreadsoutofonecity,oneprovince,andfinally

strialchains,the

productionanddemandindifferentspatialandtemporalregions

requiresnationaltransportationandstoragenetworks,covering

bulkstorageofhazardouschemicalscouldposeamajorriskto

comestoproduction,thescale-upof

theprocessrequiresafullyfunctionalcontrolsystem,correct

humanoperation,

handledcorrectlyintime,anysmallmistakesorfailuresmight

rwords,intermsofrisk,too

manyfactorsarecorrelatedandtheirinteractionsareusuallynot

nagementistakenintoaccountduringthedesign

ofmodernpetrochemicalplants,byintegratingmultiplepractical

hazardousidentificationandmanagementtechniquessuchashaz-

ardandoperabilityanalysis(HAZOP),layerofprotectionanalysis

(LOPA),andsystemintegritylevel(SIL).Yettheseanalysesare

somewhatstaticandtherelevantdocumentationisbeyondthe

reachofonsiteoperators,whoneedtobefullyawareofthe

behaviorofoperatorsishighlydependentontrainingandmanage-

ntsrecentlyoccurredattheGermanchemicalcom-

panyBASF[19,20],resultingincasualtiesandeconomiclosses.

Thecauseoftheaccidentswascloselyrelatedtooperationerrors

sereflectsthenecessityofhaving

adecision-makingsystem,whichissignificantforoperationsafety.

Thus,thereisademandforanintelligentsystemthatcantake

advantageofthestaticknowledgeembeddedinexisting

documentssuchasHAZOP,LOPA,andSIL,inordertodynamically

asystemmightimprovesafetymanagementbytakingmore

safety-relatedfactorsintoconsiderationthanahumanworkerin

rkeyaspectofsafetypractice

isinherentlysaferdesign—aconceptthatpermanentlyreducesor

eliminatesthehazardsassociatedwithmaterialsandoperations

usedintheprocess—whichiswidelyappliedthroughthevery

earlystagesofprocessdesign,normaloperation,andchange,until

alprocesseshavestructural

licationofan

inherentlysaferdesigninonechemicalprocessfacilitywouldset

rtosupportsafety-

orienteddecision-makinginabroaderview,thebasicruleandcon-

ceptofinherentlysaferdesignmustbeextractedandintegrated

intointelligentsystems,alongwiththeexperiencesgainedfrom

y,thegeneralchallengeofestablish-

ingadecision-makingsystemforprocesssafetyliesinhowhuman

experienceandknowledgewritteninnaturallanguagecanbe

understoodandusedbymachines.

(4)ro-

cessindustry,mostaccidentsbecomeseverebecauseofthelackof

aneffectivewarningmechanism[21].Ingeneral,apetrochemical

processrunsinmultipleconfigurations,eachwithitsownopera-

tionalwindowand/rocessiscontrolledby

automationsystems,smallfluctuationsinprocessparameters

canspreadtothedownstreamprocessesandaffectcriticalunits.

Thecomplexinnercorrelationsoftheprocessparametersusually

makethesechangesandfluctuationsincriticalunitsnon-

ent,processsupervisors

processfunctioningrequirestheabilitytorecognizeabnormal

situationsbasedonsophisticatedprocess-monitoringsystems,in

ordertoidentifyactualprocessconfi

diagnosisfunctionsshouldworkinrealtimetoevaluatethecur-

rentriskandpotentialshiftsofnormaloperationconfigurations

stantdecision-makingsystemisneeded

thatcanprovidepossiblecausesforthecurrentsituationand

u-

nately,nosuchsmartsystemiscurrentlyavailable.

ficialintelligenceforgreenmanufacturing

AI[22],usuallyreferredtoasmachineintelligence,hasbecome

-

binesdomainknowledgefromcomputerscience,automation,

informationengineering,mathematics,psychology,linguistics,

blemsfacedbyAIhavebeenseparatedinto

severalsub-problemsbasedonspecificcharacteristicsorcapabili-

ties,asshowninFig.5.

Thefourmajorproblemsingreenmanufacturingcanbedivided

intothreecategories,accordingtotheircharacteristics:①infor-

mationintegration,②dynamicriskassessmentandaidin

decision-making,and③ltechniquesare

availabletoaddresstheseproblemsandachievegreenmanufac-

turing,includingknowledgegraphs,Bayesiannetworks,anddeep

learning,whichwillbedetailedinthefollowing.

l./Engineering5(2019)995–1002

999

lsub-problemsinartificialintelligence.

ationintegrationviaknowledgegraphs

Aknowledgegraphisafamousandpromisingtechniquefor

organizinglinkeddatainthefitructuredsemantic

er,

knowledgegraphscanprovidereasoningandinferencecapabilities

basedonrulesordeeplearningstrategies,andthesecapabilities

empowertheimplicationofrelationsbetween‘‘entities”within

predefidgegraphsarewidelyusedin

Internet-basedapplicationssuchasencyclopedias,socialnet-

works,onlinefinancialsystems,andsocialsecuritysystems

[23,24].

UnlikethegeneralknowledgeusedinInternet-relatedapplica-

tions,theprocessindustryrequiresmuchgreaterspecializationin

chemicalengineering,processsafety,processcontrol,automation,

lementationofknowledgegraphs

insuchaspecializedindustryrequiresnotonlyfactualinforma-

tion,butalsothespecificknowledgeforthefidge

modelingisusuallydifficult,andadeepunderstandingofthe

specificfihegeneralproceduretobuilda

knowledgegraphofanarbitrarydomain(Fig.6)[25],theestablish-

mentofknowledgegraphforprocesssafetycanberoughlydivided

intothreestages:gatheringprocess-safety-relatedinformation,

knowledgefusion,andknowledgeprocessing,asshownbelow.

(1)structionof

aknowledgegraphstartswithgatheringrelevantinformation[26]

aboutprocesssafety,includingchemicals,reactions,process-

relateddocuments,controlsystems,mechanicalinformation,and

erageofthisinformationanddata

shouldbebroadanddiverse,coveringnotonlystructuraldatabut

alsodiagrams,tables,andtextswritteninnaturallanguage.

Becausethekeypurposeofknowledgegraphsistogeneratelinked

data,themainchallengeistoidentifyindividual‘‘entities”and

rocesses

areusuallyreferredtoasentityextraction,relationshipextraction,

andattributeextraction[27,28].

本文标签： 过程 知识 工业 数据 制造