assoubuntu9.04
ubuntu9.04 时间:2021-03-28 阅读:()
InnovationsSystSoftwEng(2010)6:219–231DOI10.
1007/s11334-010-0129-9ORIGINALPAPERFormallyverifyinghuman–automationinteractionaspartofasystemmodel:limitationsandtradeoffsMatthewL.
Bolton·EllenJ.
BassReceived:8March2010/Accepted:25March2010/Publishedonline:9April2010TheAuthor(s)2010.
ThisarticleispublishedwithopenaccessatSpringerlink.
comAbstractBoththehumanfactorsengineering(HFE)andformalmethodscommunitiesareconcernedwithimprovingthedesignofsafety-criticalsystems.
Thisworkdiscussesamodelingeffortthatleveragedmethodsfrombotheldstoperformformalvericationofhuman–automationinter-actionwithaprogrammabledevice.
Thiseffortutilizesasystemarchitecturecomposedofindependentmodelsofthehumanmission,humantaskbehavior,human-deviceinter-face,deviceautomation,andoperationalenvironment.
ThegoalsofthisarchitectureweretoallowHFEpractitionerstoperformformalvericationsofrealisticsystemsthatdependonhuman–automationinteractioninareasonableamountoftimeusingrepresentativemodels,intuitivemodelingcon-structs,anddecoupledmodelsofsystemcomponentsthatcouldbeeasilychangedtosupportmultipleanalyses.
Thisframeworkwasinstantiatedusingapatientcontrolledanal-gesiapumpinatwophasedprocesswheremodelsineachphasewereveriedusingacommonsetofspecications.
Therstphasefocusedonthemission,human-deviceinterface,anddeviceautomation;andincludedasimple,unconstrainedhumantaskbehaviormodel.
Thesecondphasereplacedtheunconstrainedtaskmodelwithonerepresentingnormativepumpprogrammingbehavior.
Becausemodelsproducedintherstphaseweretoolargeforthemodelcheckertoverify,anumberofmodelrevisionswereundertakenthataffectedthegoalsoftheeffort.
Whiletheuseofhumantaskbehaviormodelsinthesecondphasehelpedmitigatemodelcomplex-ity,vericationtimeincreased.
AdditionalmodelingtoolsM.
L.
Bolton(B)·E.
J.
BassDepartmentofSystemsandInformationEngineering,UniversityofVirginia,151Engineer'sWay,Charlottesville,VA,USAe-mail:mlb4b@virginia.
eduE.
J.
Basse-mail:ejb4n@virginia.
eduandtechnologicaldevelopmentsarenecessaryformodelcheckingtobecomeamoreusabletechniqueforHFE.
KeywordsHuman–automationinteraction·Taskanalysis·Formalmethods·Modelchecking·Safetycriticalsystems·PCApump1IntroductionBothhumanfactorsengineering(HFE)andformalmethodsareconcernedwiththeengineeringofrobustsystemsthatwillnotfailunderrealisticoperatingconditions.
Thetraditionaluseofformalmethodshasbeentoevaluateasystem'sauto-mationunderdifferentoperatingand/orenvironmentalcon-ditions.
However,humanoperatorscontrolanumberofsafetycriticalsystemsandcontributetounforeseenproblems.
Forexample,humanbehaviorhascontributedtobetween44,000and98,000deathsnationwideeveryyearinmedicalpractice[18],74%ofallgeneralaviationaccidents[19],atleasttwo-thirdsofcommercialaviationaccidents[28],andanumberofhighproledisasterssuchastheincidentsatThreeMileIslandandChernobyl[22].
HFEfocusesonunderstandinghumanbehaviorandapplyingthisknowledgetothedesignofhuman–automationinteraction:makingsystemseasiertousewhilereducingerrorsand/orallowingrecoveryfromthem[25,29].
ByleveragingtheknowledgeofbothHFEandformalmethods,researchershaveidentiedthecognitiveprecon-ditionsformodeconfusionandautomationsurprise[7,10,16,23];automaticallygenerateduserinterfacespecications,emergencyprocedures,andrecoverysequences[13,14];andidentiedhumanbehaviorsequences(normativeorerrone-ous)thatcontributetosystemfailures[8,12].
123220M.
L.
Bolton,E.
J.
BassWhileallofthisworkhasproducedusefulresults,themodelshavenotincludedallofthecomponentsnecessarytoanalyzehuman–automationinteraction.
ForHFEanalysesofhuman–automationinteraction,theminimalsetofcom-ponentsarethegoalsandproceduresofthehumanopera-tor;theautomatedsystemanditshumaninterface;andtheconstraintsimposedbytheoperationalenvironment.
Cogni-tiveworkanalysisisconcernedwithidentifyingconstraintsintheoperationalenvironmentthatshapethemissiongoalsofthehumanoperator[27];cognitivetaskanalysisiscon-cernedwithdescribinghowhumanoperatorsnormativelyanddescriptivelyperformgoalorientedtaskswheninteract-ingwithanautomatedsystem[17,24];andmodelingframe-workssuchas[11]seektonddiscrepanciesbetweenhumanmentalmodels,human-deviceinterfaces(HDIs),anddeviceautomation.
Inthiscontext,problemsrelatedtohuman–automationinteractionmaybeinuencedbythehumanoperator'smission,thehumanoperator'staskbehavior,theoperationalenvironment,theHDI,thedevice'sautomation,andtheirinterrelationships.
Wearedevelopingmethodsandtoolstoallowhumanfactorsengineerstoexploittheirexistinghumantaskmodelingconstructswiththepowerfulvericationcapabil-itiesofformalmethodsinordertoidentifypotentialprob-lemswithhuman–automationinteractioninsafetycriticalsystemsthatmayberelatedtohumantaskbehavior,theautomateddevice,theoperationalenvironment,ortheirinter-action.
Tothisend,wearedevelopingacomputationalframe-work(Fig.
1)fortheformalmodelingofhuman-automationinteraction.
Thisframeworkutilizesconcurrentmodelsofhumanoperatortaskbehavior,humanmission(thegoalstheoperatorwishestoachieveusingthesystem),deviceauto-mation,andtheoperationalenvironmentwhicharecom-posedtogethertoformalargersystemmodel.
Inter-modelinteractionisrepresentedbyvariablessharedbetweenmod-els.
Environmentvariablescommunicateinformationaboutthestateoftheenvironmenttothedeviceautomation,mis-sion,andhumantaskmodels.
Missionvariablescommuni-catethemissiongoalstothehumantaskmodel.
InterfacevariablesconveyinformationaboutthestateoftheHDI(dis-playedinformation,thestateofinputwidgets,etc.
)tothehumantaskmodel.
ThehumantaskmodelindicateswhenandwhatactionsahumanoperatorwouldperformontheHDI.
TheHDIcommunicatesitscurrentstatetothedeviceauto-mationviatheinterfacevariables.
TheHDIreceivesinfor-mationaboutthestateofthedeviceautomationmodelviatheautomationstatevariables.
Forbroaderapplicability,theanalysisframeworkmustsupportmodelingconstructsintuitivetothehumanfac-torsengineerinordertoallowhimtoeffectivelymodelhumanmissions,humantasks,andHDIs.
Becauseanengi-neermaywishtorerunvericationsusingdifferentmissions,taskmodels,HDIs,environments,orautomationbehaviors,thesecomponentsshouldremaindecoupled(asisthecaseinFig.
1).
Finally,themodelingtechniquemustbecapableofrepresentingthetargetsystemswithenoughdelitytoallowtheengineertoperformthedesiredverication,anddosoinareasonableamountoftime(thiscouldmeanseveralhoursforasmallproject,orseveraldaysforamorecomplicatedone).
ThispaperdescribesaninstantiationofthisframeworkusingamodelofaBaxterIpumpPainManagementSystem[2],apatientcontrolledanalgesia(PCA)pumpthatadminis-terspainmedicationinaccordancewithconstraintsdenedbyahealthcaretechnician(describedinSect.
2.
1).
Modelsweredevelopedintwophases.
TherstphaseinvolvedtheconstructionanddebuggingoftheHDI,deviceautomation,andhumanmissionmodels(anenvironmentalmodelwasnotSystemModelHumanTaskModelDeviceAutomationModelHuman-DeviceInterfaceModelMissionModelEnvironmentModelMissionVariablesHumanActionVariablesInterfaceVariablesInterfaceVariablesAutomationStateVariablesEnvironmentVariablesFig.
1Frameworkfortheformalmodelingofhuman–automationinteraction.
Arrowsbetweenmodelsrepresentvariablesthataresharedbetweenmodels.
Thedirectionofthearrowindicateswhethertherep-resentedvariablesaretreatedasinputsoroutput.
Ifthearrowissourcedfromamodel,therepresentedvariablesareoutputsofthatmodel.
Ifthearrowterminatesatamodel,therepresentedvariablesareinputstothatmodel123Formallyverifyinghuman–automationinteraction221includedbecauseofthegeneralstabilityoftheenvironmentinwhichanactualpumpoperates)withanunconstrainedhumantaskmodelservingasaplaceholderforamorereal-istichumantaskmodel.
Thesecondextendedthemodelpro-ducedinPhase1witharealisticmodelofthehumantask,completingtheframework.
Eventhoughthetargetdeviceinthismodelingeffortwasseeminglysimple,thesystemmodelthatwasinitiallydevel-opedinPhase1(Phase1a)wastoodifcultforthemodelcheckertoprocessquicklyandtoocomplexforittoverify.
Thusanumberofrevisionswereundertaken[3].
InPhase1bareducedandabstractedmodeloftheBaxterIpumpwasproducedwhich,whilecapableofbeingusedinsomever-ications,didsoattheexpenseoflimitingthenumberofgoalsrepresentedinthehumanmissionmodel.
ThisPhase1bmodellimitedtheusefulnessofincorporatinghumantaskbehaviorinPhase2.
Thus,inPhase1c,thesystemmodelwasreducedtoencompasstheprogrammingprocedureforamuchsimplerPCApump.
InPhase2,theincorporationofthemorerealistichumantaskbehavioractuallyresultedinareductionofthetotalsystemmodel'scomplexity,butdidsoattheexpenseofanincreaseinvericationtime.
Thispaperdiscussesthesemodelingphases,thevericationresultspro-ducedinthem,andtheirassociatedcompromisesinrelationtothegoalsofthemodelingarchitecture.
Theseareusedtodrawconclusionsaboutthefeasibilityofusingformalmeth-odstoinformhuman–automationinteraction.
2Methods2.
1ThetargetsystemTheBaxterIpumpisanautomatedmachinethatcontrolsdeliveryofsedative,analgesic,andanestheticmedicationsolutions[2].
Solutiondeliveryviaintravenous,subcutane-ous,andepiduralroutesissupported.
Medicationsolutionsaretypicallystoredinbagslockedinacompartmentonthebackofthepump.
Pumpbehaviorisdictatedbyinternalautomation,whichcandependonhowthepumpisprogrammedbyahumanoperator.
PumpprogrammingisaccomplishedviaitsHDI(Fig.
2)whichcontainsadynamicLCDdisplay,asecu-ritykeylock,andeightbuttons.
Whenprogrammingthepump,theoperatorisabletospecifyallofthefollowing:whethertouseperiodicorcontinuousdosesofmedications(i.
e.
,themodewhichcanbePCA,Basal+PCA,orContinu-ous),whethertouseprescriptioninformationpreviouslypro-grammedintothepump,theuidvolumecontainedinthemedicationbag,theunitsofmeasureusedfordosage(ml,mg,org),whetherornottoadministerabolus(aninitialdoseofmedication),dosageamounts,dosageowrates(foreitherbasalorcontinuousratesasdeterminedbythemode),StartStopEnterOn/OffClearCLCDDisplayInterfaceMessageDisplayedValueCursorSecurityKeyLeftButtonUp(Scroll)ButtonRightButtonClearButtonStartButtonStopButtonEnterButtonOn/OffButtonFig.
2AsimpliedrepresentationoftheBaxterIpump'shuman-deviceinterface.
Notethattheactualpumpcontainsadditionalcontrolsandinformationconveyancesthedelaytimebetweendosages,and1hlimitsontheamountofdeliveredmedication.
Duringprogramming,thesecuritykeyisusedtolockandunlockthecompartmentcontainingthemedicationsolution.
Theunlockingandlockingprocessisalsousedasasecuritymeasuretoensurethatanauthorizedpersonisprogrammingthepump.
Thestartandstopbuttonsareusedtostartandstopthedeliveryofmedicationatspecictimesduringpro-gramming.
Theon–offbuttonisusedtoturnthedeviceonandoff.
TheLCDdisplaysupportspumpoperationoptions.
Whentheoperatorchoosesbetweentwoormoreoptions,theinter-facemessageindicateswhatisbeingchosen,andtheinitialordefaultoptionisdisplayed.
Pressingtheupbuttonallowstheprogrammertoscrollthroughtheavailableoptions.
Whenanumericalvalueisrequired,theinterfacemessageconveysitsnameandthedisplayedvalueispresentedwiththecursorunderoneofthevalue'sdigits.
Theprogrammercanmovethepositionofthecursorbypressingtheleftandrightbuttons.
Heorshecanpresstheupbuttontoscrollthroughthedifferentdigitvaluesavailableatthatcursorposition.
Theclearbuttonsetsthedisplayedvaluetozero.
Theenterbuttonisusedtoconrmvaluesandtreatmentoptions.
Asidefromtheadministrationoftreatment,thepump'sautomationsupportsdynamiccheckingandrestrictionofoperatorenteredvalues.
Thus,inadditiontohavinghardlim-itsonvalueranges,theextremacanchangedynamicallyinresponsetootheruserspeciedvalues.
2.
2ApparatusAllformalmodelswereconstructedusingtheSymbolicAnalysisLaboratory(SAL)language[9]becauseofitsasso-ciatedanalysisanddebuggingtools,anditssupportforboththeasynchronousandsynchronouscompositionofdifferent123222M.
L.
Bolton,E.
J.
Bassmodels(modulesusingSAL'sinternalsemantics).
ThetaskmodelrepresentationsdescribednextweretranslatedintotheSALlanguageasasinglemoduleusingacustom-builtjavaprogram[5].
AllvericationsweredoneusingSAL-SMC3.
0,theSALsymbolicmodelchecker.
1Vericationswereconductedona3.
0GHzdual-coreIntelXeonprocessorwith16GBofRAMrunningtheUbuntu9.
04desktop.
Humantaskmodelswerecreatedusinganintermediarylanguagecalledenhancedoperatorfunctionmodel(EOFM)[4],anXML-based,generichumantaskmodelinglan-guagebasedontheoperatorfunctionmodel(OFM)[21,26].
EOFMsarehierarchicalandheterarchicalrepresentationsofgoalormissiondrivenactivitiesthatdecomposeintolowerlevelactivities,andnally,atomicactions—whereactionscanrepresentanyobservable,cognitive,orperceptualhumanbehavior.
EOFMsexpresstaskknowledgebyexplicitlyspec-ifyingtheconditionsunderwhichhumanoperatoractivi-tiescanbeundertaken:whatmustbetruebeforetheycanexecute(preconditions),whentheycanrepeat(repeatcon-ditions),andwhentheyhavecompleted(completioncondi-tions).
Anyactivitycandecomposeintooneormoreotheractivitiesoroneormoreactions.
Adecompositionoperatorspeciesthetemporalrelationshipsbetweenandthecardi-nalityofthedecomposedactivitiesoractions(whentheycanexecuterelativetoeachotherandhowmanycanexecute).
EOFMscanberepresentedvisuallyasatree-likegraphstructure(examplescanbeseeninFigs.
3,4,5,6,7).
Intheserepresentations,actionsarerepresentedasrect-anglesandactivitiesarerepresentedasroundedrectan-gles.
Anactivity'sdecompositionispresentedasanarrow,labeledwiththedecompositionoperator,extendingbelowitthatpointstoalargeroundedrectanglecontainingthedecomposedactivitiesoractions.
Inthiswork,threedecompositionoperatorsareused:(1)ord(allactivitiesoractionsinthedecompositionmustexecuteintheordertheyappear);(2)or_seq(oneormoreoftheactivitiesoractionsinthedecompositionmustexecute);and(3)xor(exactlyoneactivityoractioninthedecompositionmustexecute).
Conditionsonactivitiesarerepresentedasshapesorarrows(annotatedwiththeconditionlogic)con-nectedtotheactivitythattheyconstrain.
Theform,posi-tion,andcoloroftheshapearedeterminedbythetypeofcondition.
Apreconditionispresentedasayellow,downward-pointingtriangleconnectedtotheleftsideoftheactivity.
Acompletionconditionisrepresentedbyamagenta,upward-pointingtriangleconnectedtotherightsideoftheactivity.
Arepeatconditionisdepictedasanarrowrecur-sivelypointingtothetopoftheactivity.
Moredetailscanbefoundin[4].
1SomemodeldebuggingwasalsoconductedusingSAL'sboundedmodelchecker.
2.
3VericationspecicationTwospecicationswereemployedineachofthemodelingphases:bothwerewritteninlineartemporallogicandeval-uatedusingSAL–SMC.
Therst(Eq.
1),usedformodeldebugging,veriesthatavalidprescriptioncouldbepro-grammedintothepump:GiInterfacemessage=TreatmentAdministering∧iMode=iPrescribedMode∧lFluidVolume=iPrescribedFluidVolume∧lPCADose=iPrescribedPCADose∧lDelay=iPrescribedDelay∧lBasalRate=iPrescribedBasalRate∧l1HourLimit=iPrescribed1HourLimit∧lBolus=iPrescribedBolus∧lContinuousRate=iPrescribedContinuousRate(1)Here,ifthemodelisabletoenterastateindicatingthattreatmentisadministering(iInterfaceMessage=Treat-mentAdministering)withtheentered(orprogrammed)prescriptionvalues(iMode,lFluidVolume,…,lContinuous-Rate)matchingtheprescriptionvaluesgeneratedbythemissionmodel(variableswiththeiPrescribedprex),acounterexamplewouldbeproducedillustratinghowthatpre-scriptionwasprogrammed.
Variableswithaniprexindicatethatthevariableisaninputtothehumantaskmodel.
Vari-ableswithanlprexindicatethatthevariableislocaltoagivenmodel.
Thesecondspecication(Eq.
2)representedasafetyprop-ertythatwasexpectedtoverifytotrue,thusallowingthemodelcheckertotraversetheentirestatespaceofeachphase'smodel.
BecausesuchavericationallowsSALtoreportthesizeofamodel'sstatespace,vericationsusingthisspecicationwouldprovidesomemeansofcomparingthecomplexityofthemodelsproducedineachphase.
GiInterfaceMessage=TreatmentAdministering∧iMode=Continuous∧lDelay=0(2)Here,thespecicationisassertingthatthemodelshouldneverenterastatewheretreatmentisadministeringinthePCAorBasal+PCAmodes(iMode=Continuous)whenthereisnodelaybetweendoses.
2Thus,ifEq.
(2)veriestotrue,thepumpwillneverallowaprogrammertoenterprescriptionsthatwouldallowpatientstocontinuouslyadministerPCAdosestothemselves[2].
2AdelaycanonlybeensetwhenthePCAorBasal+PCAmodeshavebeenselectedbythehumanoperator.
TherearenodelaysbetweendoseswhenthepumpisintheContinuousmode.
123Formallyverifyinghuman–automationinteraction2233Phase1a:arepresentativemodeloftheIpump3.
1ModeldescriptionAninitialmodelwascreatedtoconformtothearchitecturalanddesignphilosophyrepresentedinFig.
2:themissionwasrepresentedasasetofviableprescriptionsoptions;themis-sion,humanoperator,human-deviceinterface,anddeviceautomationweremodeledindependentlyofeachother;andthebehavioroftheautomatedsystemandHDImodelswasdesignedtoaccuratelyreectthebehaviorofthesesystemsasdescribedintheuser'smanual[2]andobservedthroughdirectinteractionwiththedevice.
Anunconstrainedhumanoperatorwasconstructedthatcouldissueanyvalidhumanactiontothehuman-deviceinterfacemodelatanygiventime.
BecausethePCApumpgenerallyoperatesinacon-trolledenvironment,awayfromtemperatureandhumidityconditionsthatmightaffecttheperformanceofthepump'sautomation,noenvironmentalmodelwasincluded.
Finally,becausedocumentationrelatedtotheinternalworkingsofthepumpwaslimited,thesystemautomationmodelwasrestrictedtothatassociatedwiththepumpprogrammingpro-cedure:behaviorthatcouldbegleanedfromtheoperator'smanual[2],correspondenceswithhospitalstaff,anddirectinteractionwiththepump.
3.
2ModelcoordinationModelinfrastructurewasrequiredtoensurethathumanoper-atoractionswereproperlyrecognizedbytheHDImodel.
Inanidealmodelingenvironment,humanactionbehaviororig-inatingfromthehumanoperatormodelcouldhavebothanasynchronousandsynchronousrelationshipwiththeHDImodel.
SynchronousbehaviorwouldallowtheHDImodeltoreacttouseractionsinthesametransitioninwhichtheywereissued/performedbythehumanoperatormodel.
How-ever,boththehumanoperatorandHDImodelsoperateinde-pendentlyofeachother,andmayhavestatetransitionsthataredependentoninternalorexternalconditionsthatarenotdirectlyrelatedtothestateoftheothermodel.
Thissuggestsanasynchronousrelationship.
SALonlyallowsmodelstobecomposedwitheachotherasynchronouslyorsynchronously(butnotboth).
Thus,itwasnecessarytoadaptthemodelstosupportfeaturesassociatedwiththeunusedcomposition.
AsynchronouscompositionwasusedtocomposethehumanoperatorandHDImodels.
ThisnecessitatedsomeadditionalinfrastructuretopreventthehumanoperatormodelfromissuinguserinputsbeforetheHDImodelwasreadytointerpretthemandtopreventthehumanoperatormodelfromterminatingagiveninputbeforetheinterfacecouldrespondtoit.
Thiswasaccomplishedthroughtheaddi-tionoftwoBooleanvariables:oneindicatingthatinputhadbeensubmitted(henceforthcalledSubmitted)andavariableindicatingtheinterfacewasreadytoreceiveactions(hence-forthcalledReady).
Thiscoordinationoccurredasfollows:–IfReadyistrueandSubmittedisfalse,thehumanoperatormodulesetsoneormoreofthehumanactionvariablestoanewinputvalueandsetsSubmittedtotrue.
–IfReadyandSubmittedaretrue,thehuman-deviceinter-facemodulerespondstothevaluesofthehumanactionvariablesandsetsReadytofalse.
–IfReadyisnottrueandSubmittedistrue,thehumanoper-atormodulesetsSubmittedtofalse.
–IfReadyandSubmittedarebothfalseandtheautomatedsystemisreadyforadditionalhumanoperatorinput,thehuman-deviceinterfacemodulesetsReadytotrue.
3.
3VericationresultsAttemptstoverifythismodelusingthespecicationsinEqs.
1and2resultedintwoproblemsrelatedtothefeasibilityandusefulnessofthevericationprocedure.
First,theSAL–SMCprocedurefortranslatingtheSALcodeintoabinarydecisiondiagram(BDD)tookexcessivelylong(morethan24h),atimeframeimpracticalformodeldebugging.
Second,thevericationprocesswhichfollowedtheconstructionoftheBDDeventuallyranoutofmemory,thusnotreturningavericationresult.
4Phase1b:areducedBaxterIpumpmodelAsaresultofthefailedvericationofthemodelproducedinPhase1a,significantrevisionswererequiredtomakethemodelmoretractable.
Thesearediscussedbelow.
4.
1RepresentationofnumericalvaluesToreducethetimeneededtoconverttheSAL-codemodeltoaBDD,anumberofmodicationsweremadetothemodelfromPhase1abyrepresentingmodelconstructsinwaysmorereadilyprocessedbythemodelchecker.
Assuch,themod-icationsdiscussedheredidnotultimatelymaketheBDDrepresentationofthemodelsmaller,butmerelyexpediteditsconstruction.
4.
1.
1RedundantrepresentationofvaluesTwodifferentrepresentationsofthevaluesprogrammedintothepumpbytheoperatorwereusedintheHDIanddeviceautomationmodels.
BecausetheHDIrequiredthehumanoperatortoentervaluesbyscrollingthroughtheavailablevaluesforindividualdigits,anarrayofintegerdigitswasappropriatefortheHDImodel.
However,becausethesystem123224M.
L.
Bolton,E.
J.
Bassautomationwasconcernedwithdynamicallycheckinglimitsandusingenteredvaluestocomputeothervalues,anumer-icalrepresentationoftheactualvaluewasmoreconvenientfortheautomatedsystemmodel.
ThisredundancyburdenedtheBDDtranslator.
Thiswasremediedbyeliminatingthedigitarrayrepresentationsandusingfunctionstoenableactionsfromthehumantaskmodeltoincrementallychangeindividualdigitswithinavalue.
4.
1.
2RealnumbersandintegersInthemodelproducedinPhase1a,allnumericalvalueswererepresentedasrealvalueswithrestrictedranges.
Thiswasdonebecausemostuserspeciedvalueswereeitherinte-gersoroatingpointnumbers(precisetoasingledecimalpoint).
Nodataabstractionswereinitiallyconsideredbecausethenatureofthehumantask(modeledinPhase2)requiredmanipulationofvalues'individualdigits.
However,repre-sentingvaluesthiswayprovedespeciallychallengingfortheBDDtranslator.
Thus,allvaluesweremodiedsothattheycouldberepresentedasrestrictedrangeintegers.
Forintegervariablesrepresentingoatingpointnumbers,thismeantthatthemodelvaluewastentimesthevalueitrepresented.
Thisrepresentationallowedthevaluestostillbemanipulatedattheindividualdigitlevel,whilemakingthemmorereadilyinterpretablebytheBDDtranslator.
4.
1.
3VariablerangesInthePhase1amodel,theupperboundontherangeofallvalue-basedvariableswassettothetheoreticalmaximumofanyvaluethatcouldbeprogrammedintothepump:99,999.
3However,toreducetheamountofworkrequiredfortheBDDconversion,therangeofeachnumericallyvaluedvariablewasgivenaspecicupperboundthatcorrespondedtothemaximumvalueitcouldactuallyassumeinthedevice.
4.
2ModelreductionToreducethesizeofthemodel,avarietyofelementswereremoved.
InallcasesthesereductionsweremeanttoreducethenumberofstatevariablesintheHDIordeviceautoma-tionmodels(slicing),orreducetherangeofvaluesavariablecouldassume(dataabstraction).
Unfortunately,eachofthesereductionsalsoaffectedwhathumantaskscouldultimatelybemodeledandthusveriedinsubsequentmodeliterations.
Allofthefollowingreductionswereundertaken:–InthePhase1amodel,themissionmodelcouldgener-ateaprescriptionfromtheentireavailablerangeofvalidprescriptions.
Thiswaschangedsothatfewerprescription3Alllowerboundsweresetto0.
optionsweregeneratedinPhase1b'smissionmodel:thatofprogrammingaprescriptionwithacontinuousdosagewithtwooptionsforbolusdelivery(0.
0and1.
0ml)andtwocontinuousowrateoptions(1.
0and9.
0ml/h).
Whilethissignificantlyreducedthenumberofmodelstates,italsoreducedthenumberofprescriptionsthatcouldbeusedinvericationprocedures.
–InthePhase1amodel,theHDImodelwouldallowtheoperatortoselectwhatunitstousewhenenteringprescrip-tions(ml,mg,org).
OnlythemlunitoptionwasincludedinthePhase1bmodel.
Thisreducedthenumberofinter-facemessagesinthemodel,allowedfortheremovalofseveralvariables(thoserelatedtotheunitoptionselection,andsolutionconcentrationspecication),andreducedtherangesrequiredforseveralnumericalvaluesrelatedtotheprescription.
Thiseliminatedtheoptionofincludingunitselectionandconcentrationspecicationtaskbehaviorsinthemodel.
–InthePhase1amodel,boththeHDIanddeviceautomationmodelsencompassedbehaviorrelatedtothedeliveryofmedicationsolutionduringtheprimingandbolusadmin-istrationprocedures.
Duringpriming,theHDIallowstheoperatortorepeatedlyinstructthepumptoprimeuntilallairhasbeenpushedoutoftheconnectedtubing.
Duringbolusadministration,theHDIallowstheoperatortotermi-natebolusinfusionbypressingthestopbuttontwice.
ThisfunctionalitywasremovedfromthePhase1bmodels,thuseliminatinginterfacemessagestatesandnumericalvaluesindicatinghowmuchuidhadbeendeliveredinbothpro-cedures.
Thiseliminatedthepossibilityofincorporatingtaskbehaviorrelatedtopumpprimingandbolusadminis-trationinthemodel.
–ThePhase1amodelmimickedthesecurityfeaturesfoundintheoriginaldevicewhichrequiredthehumanoperatortounlockandlockthedeviceonstartupandenterasecu-ritycode.
ThisfunctionalitywasremovedfromthePhase1bmodelwhichreducedthenumberofinterfacemessagesinthemodelandremovedthenumericalvariable(witha0–999range)associatedwithenteringthesecuritycode.
Thiseliminatedthepossibilityofmodelinghumantaskbehaviorrelatedtounlockingandlockingthepumpaswellasenteringthesecuritycodeinthemodel.
–InthePhase1amodel,theinterfacemessagecouldauto-maticallytransitiontobeingblank:mimickingtheactualpump'sabilitytoblankitsscreenafterthreesecondsofoperatorinactivity.
Becausefurtheroperatorinactionwouldresultintheoriginaldeviceissuinga"leftinpro-grammingmode"alert,ablankinterfacemessagecouldautomaticallytransitiontoanalertissuance.
Thisfunction-alitywasremovedfromthePhase1bmodel,eliminatingseveralinterfacemessagesaswellasvariablesthatkepttrackofthepreviousinterfacemessage.
Thus,theoption123Formallyverifyinghuman–automationinteraction225ofmodelingoperatortaskresponsetoscreenblankingandalertswasremovedfromthemodel.
WhilethesereductionsresultedinthePhase1bmodelbeingmuchsmallerandmoremanageablethantheoriginal,theabilitytomodelsomeofthetaskbehaviorsoriginallyasso-ciatedwiththedevicehadtobesacriced.
4.
3ResultsThePhase1bmodelwasabletocompletethevericationprocedurewithEq.
(1)andproduceacounterexamplewithasearchdepthof54inapproximately5.
9h,withthemajorityofthattime(5.
4h)usedforcreatingtheBDDrepresentation[3].
4Notsurprisingly,themodelcheckerranoutofmemorywhenattemptingtoverifyEq.
(2).
5Phase1c:asimplerPCApumpmodelWhilethemodeldevelopedinPhase1bdidproduceusableresultsandhassubsequentlybeenusedinthevericationofadditionalproperties(see[5]),thispowercameattheexpenseofareductioninthescopeofthemissionmodel.
Sincethemissiondirectlyinuenceswhathumanbehaviorwillexe-cute,thislimitedthehumantaskbehaviorthatcouldulti-matelybeveriedaspartofthesystemmodel.
Further,thefactthatthePhase1bmodelwastoocomplexforEq.
(2)tobeveriedpotentiallylimitedanyfuturemodeldevelopmentthatmightaddcomplexity.
Toremedytheseshortcoming,themodelproducedinPhase1bwasfurtherreducedtoonethatencompassedtheprogrammingofthemostbasicPCApumpfunctionalitywhiletherangesofpossiblevaluesfortheremainingmissionmodelvariableswereexpandedtobemorerealistic.
5.
1ModelreductionToobtainasmallerPCAmodel,allofthefollowingwereremoved:theselectionofmodeandtheabilitytospecifyabasalrate,continuousrate,bolusdosage,anduidvolume.
Asaresult,associatedinterfacemessagesandvariableswereremovedalongwiththeabilitytomodeltheirprogrammingaspartofthehumantaskbehaviormodel.
ThisresultedinamodelthatonlyencompassedfunctionalityforprogrammingaPCAdose,programmingthedelaybetweenPCAdoses,turningthepumponandoff,andstartingandstoppingtheadministrationoftreatment:functionalitycompatiblewiththemostbasicPCApumpoperations(see[1]).
4Completedmodels,SALoutputs,andcounterexamplescanbefoundathttp://cog.
sys.
virginia.
edu/ISSE2010/.
Valuerangeswerefurtherrestrictedtoreducethesizeofthemodel.
Specifically,theupperboundontheacceptabledelaybetweenPCAdosageswaschangedfrom240to60minutes.
This,coupledwiththeotherreductions,hadtheaddedbenetofallowingthenumberofdigitsrequiredfortheprogrammingofpumpvaluestobereducedto2ratherthantheoriginal4.
ThereductionsinotherareasallowedthescopeofthedelaysandPCAdosagesgeneratedbythemissionmodeltobeexpandedtoamorerepresentativeset.
ForPCAdosages,thefullrangeofvaluesfrom0.
1to9.
9in0.
1mlincrementsweresupported.
Fordelaybetweendosages,veoptionswereavailable:delaysof10,15,30,45,and60min.
Allpumpinterfacefunctionalitywasretainedfromthepreviousmodels.
Thus,theunconstrainedhumantaskmodelwasunchangedaswasthehumantaskandHDImodels'communicationprotocol.
5.
2ResultsThePhase1cmodelranthevericationprocedureforEq.
(1)(withtheeliminatedvariablesremoved)in6swithasearchdepthof22,muchfasterthanthemodelfromPhase1b.
ThevericationofthespecicationinEq.
(2)veriedtotruein129swithasearchdepthof259and78,768,682,750visitedstates.
6Phase2:incorporatingmodelsofhumanbehaviorInthesecondphaseofmodeling,weexpandedourinstantia-tionoftheframeworkbyincorporatingarealistichumantaskbehaviormodel.
WethereforereplacedtheunconstrainedhumanoperatorinthePhase1cmodelwithahumantaskbehaviormodelderivedfrompumpdocumentation[2]andtrainingmaterials.
ThismodelutilizedtheEOFMconceptsandthusrequiredsomeadditionalinfrastructureinordertoincorporateitintotheformalsystemmodel.
Wedescribethebehaviorsthatweremodeled,howtheseweretranslatedintotheformalmodel,andreportvericationresultsforthepro-ducedsystemmodel.
6.
1HumantaskbehaviormodelingandtranslationThepump'smaterialscontainedsixhigh-levelgoaldirectedbehaviorsforperformingavarietyofpumpactivitiesrelevanttothePhase1cmodelasfollows:–Turningonthepump.
–Stoppingtheinfusionofmedication.
–Turningoffthepump.
–EnteringaprescribedvalueforPCAdosagevolumes(inmilliliter).
123226M.
L.
Bolton,E.
J.
Bass–EnteringaprescribedvalueforthedelaybetweenPCAdoses(inminutes),and–Selectingwhethertostartorreviewanenteredprescrip-tion.
TheEOFMmodelsdescribingeachofthesebehaviorsarediscussedbelow.
6.
1.
1TurningOnthepumpThemodelforturningonthepumpisshowninFig.
3.
Here,theEOFMcanexecuteiftheinterfacemessageindi-catesthatthesystemisoff(iInterfaceMessage=SystemOff;aprecondition).
Thishigh-levelactivity(aTurnOnPump)iscompletedbyperformingtheactionofpressingtheon/offbutton(hPressOnOff).
Theorddecompositionoperatorindi-catesthatallofthedecomposedactivitiesoractionsmustbecompletedinsequentialorder.
TheEOFMhasaccomplisheditsgoal(acompletioncondition)whentheinterfacemes-sageindicatesthatthepumpisnolongeroff(iInterfaceMes-sage/=SystemOff).
6.
1.
2StoppinginfusionInfusionofmedicationcanbestopped(Fig.
4)iftheinter-faceindicatesthattreatmentisadministering(iInterfaceMes-sage=TreatmentAdministering).
Thisisaccomplishedbypressingthestopbutton(hPressStop)twiceinquicksuc-cessionwithnootherhumaninputsoccurringinbetween.
Theprocesshascompletedwhentheinterfaceindicatesthattreatmentisnotadministering(iInterfaceMessage/=Treat-mentAdministering).
aTurnOnPumpiInterfaceMessage=SystemOffiInterfaceMessage/=SystemOffordhPressOnOffFig.
3TheEOFMgraphicalrepresentationforturningonthepumpaStopInfusingiInterfaceMessage=TreatmentAdministeringiInterfaceMessage/=TreatmentAdministeringordhPressStophPressStopFig.
4TheEOFMgraphicalrepresentationforstoppinginfusion6.
1.
3TurningOffthepumpThemodelforturningoffthepump(Fig.
5)isrelevantiftheinterfacemessageindicatesthatthesystemisnotoff(iIn-terfaceMessage/=SystemOff).
Thepumpisturnedoffbyperformingtwolowerlevelactivitiesinsequentialorder:stoppinginfusion(aStopInfusion;explainedabove)andpressingthekeysnecessarytoturnoffthepump(aPress-KeysToTurnOffPump).
Thislatteractivityiscompletedbypressingtheon/offbutton(hPressOnOff)twiceinsequence.
Theentireprocessofturningoffthepumpcompleteswhentheinterfaceindicatesthatthepumpisoff(iInterfaceMes-sage=SystemOff).
6.
1.
4ProgrammingavalueintothepumpThevaluesforPCAdosagevolumeanddelaybetweendosagescanbeprogrammedintothepumpusinganEOFMpatternedafterFig.
6.
Thus,foragivenvalueX,thecorre-spondingEOFMbecomesrelevantwhentheinterfaceforsettingthatvalueisdisplayed(iInterfaceMessage=SetX).
Thisisachievedbysequentiallyexecutingtwosub-activities:changingthedisplayedvalue(aChangeXValue)andaccept-ingthedisplayedvalue(aAccept).
Theactivityforchangingthedisplayedvaluecanexecute,andwillrepeatedlyexecute,ifthedisplayedvalueisnotequaltotheprescribedvalue(iCurrentValue/=iPrescribedX).
Thevalueischangedbyexecutingoneormore(denotedbytheor_seqdecomposi-tionoperator)ofthefollowingsub-activities:changingthedigitcurrentlypointedtobythecursor(aChangeDigit:com-pletedbypressingtheupkey(hPressUp)),movingthecursortoadifferentdigit(aNextDigit:completedbypressingonlyoneof(thexordecompositionoperator)theleft(hPress-Left)orright(hPressRight)buttons),orsettingthedisplayedvaluetozero(aClearValue:completedbypressingtheclearbutton(hPressClear)).
Theprocessofchangingthedisplayed123Formallyverifyinghuman–automationinteraction227aTurnOffPumpiInterfaceMessage/=SystemOffiInterfaceMessage=SystemOffordaStopInfusingiInterfaceMessage=TreatmentAdministeringiInterfaceMessage/=TreatmentAdministeringordhPressStophPressStopaPressKeysToTurnOffPumpordhPressOnOffhPressOnOffFig.
5TheEOFMgraphicalrepresentationforturningoffthepumpvaluecompleteswhenthedisplayedvaluematchesthepre-scribedvalue(iCurrentValue=iPrescribedX).
Thedisplayedvalueisacceptedbypressingtheenterkey.
Theentirepro-cessendswhentheinterfaceisnolongerinthestateforacceptingX.
6.
1.
5StartingorreviewingaprescriptionAfteraprescriptionhasbeenprogrammedthehumanopera-torisgiventheoptiontostarttheadministrationofthatpre-scriptionortoreviewit(wheretheoperatorworksthroughtheprogrammingprocedureasecondtimewiththepreviouslyprogrammedoptionsdisplayedateachstep).
TheEOFMforperformingthis(Fig.
7)becomesrelevantatthispoint(iInter-faceMessage=StartBeginsRx).
Itiscompletedbyperform-ingonlyoneoftwoactivities:selectingtheoptiontostarttreatment(aStartRx—performedbypressingthestartbutton(hPressStart))orselectingthereviewoption(aReviewRx—performedbypressingtheenterbutton(hPressEnter)).
6.
2EOFMtranslationTheEOFMsrepresentingthehumantaskmodelweretranslatedintoaSALcodemodule.
ThistranslationwasaccomplishedbycreatingavariableforeachactivityoractionnodeineachEOFM,eachofwhichcouldassumeoneofthreeenumeratedvaluesdescribingitsexecutionstate:ready,exe-cuting,ordone.
Thus,inadditiontohandlingthetransitionallogicforthecoordinationprotocol,thismodulehandledthetransitionlogicforallowingthevariablesrepresentingactivityandactionnodestotransitionbetweenthesethreevalues.
Allactivityandactionvariablesstartinthereadystate.
Theycantransitionbetweenexecutionstatesbasedontheexecutionstateoftheirchildren,parent,andsiblingsintheEOFMstructure;theevaluationoftheirconditions;andtheirpositionwithintheEOFMhierarchy.
Whiletheresultinghumanoperatormoduleanditsassoci-atedunconstrainedoperatormodelbothhadthesameinputsandoutputs,thelogicassociatedwithtraversalofthehumantaskstructuresrequired48additionalvariablesinthehumantaskbehaviormodel.
6.
3ResultsThespecicationinEq.
(1)veried(producedtheexpectedcounterexample)in57swithasearchdepthof42.
Thespec-icationinEq.
(2)veriedtotruein10.
6hwithasearchdepthof437and1,534,862,538visitedstates.
7DiscussionThisworkhasshownthatitispossibleforhuman–automationinteractiontobeevaluatedusingthearchitectureinFig.
1.
However,thiscameasaresultoftradeoffsbetweenthegoalsthearchitectureisdesignedtosupport:123228M.
L.
Bolton,E.
J.
BassaSetXiInterfaceMessage=SetXiInterfaceMessage/=SetXordaChangeXValueiCurrentValue/=iPrescribedXiCurrentValue/=iPrescribedXiCurrentValue=iPrescribedXor_seqaChangeDigitordhPressUpaSelectNextDigitxorhPressLefthPressRightaClearValueordhPressClearaAcceptordhPressEnterFig.
6TheEOFMgraphicalrepresentationofthepatternforprogrammingavalueXintothepump1.
Modelconstructsneedtobeintuitivetohumanfactorsengineerswhowillbebuildingandevaluatingmanyofthemodels;2.
thesub-modelsshouldbedecoupledandmodular(asinFig.
1)inordertoallowforinterchangeabilityofalter-nativesub-models;and3.
theconstructedmodelsneedtobecapableofbeingver-iedinareasonableamountoftime.
Wediscusshoweachofthesegoalswasimpactedandhowrelatedissuesmightbeaddressed.
7.
1Goals1:modelintuitivenessManyofthemodelrevisionswereassociatedwithrepresent-ingmodelconstructsinwaysthatweremorereadilyinter-pretablebythemodelcheckerratherthanthehumanfactorsengineer.
Theseprimarilytooktheformofconvertingoat-ingpointanddigitarrayvaluesintointegersinPhase1b.
Further,theextensivemodelreductionsthatwereundertakeninPhase1cwouldbeverycumbersomeforahumanfactorsengineer.
Therearetwopotentialwaystoaddressthisissue.
Onesolutionwouldbetoimprovethemodelcheckersthemselves.
Giventhatthemodicationswouldnotactuallychangethenumberofreachablestatesinthesystem,thiswouldsuggestthatthemodelcheckerneedonlyoptimizetheBDDconver-sionalgorithms.
Alternatively,additionalmodelingtoolscouldbeusedtohelpmitigatethesituation.
SuchtoolscouldallowhumanfactorsengineerstoconstructorimportHDIprototypes,andtranslatethemintomodelcheckercode.
Thiswouldallowtheunintuitiverepresentationsnecessaryforensuringamodel'sefcientprocessingbythemodelcheckertoberemovedfromthemodeler'sview.
123Formallyverifyinghuman–automationinteraction229aStartOrReviewiInterfaceMessage=StartBeginsRxiInterfaceMessage/=StartBeginsRxxoraStartRxordhPressStartaReviewRxordhPressEnterFig.
7TheEOFMforchoosingtostartorreviewaprescription7.
2Goal2:decouplingofarchitecturesub-modelsBecausetheprotocolusedtocoordinatehumanactionsbetweentheHDIandhumantaskmodels(discussedforPhase1aandusedinallmodelsproducedinallsubsequentphases)assumesaparticularrelationshipbetweenvariablessharedbythesemodels,theyaretightlycoupled.
Unlessamodelcheckercanbemadetosupportbothasynchronousandsynchronousrelationshipsbetweenmodelsmoreelegantly,thiscoordinationinfrastructurecannotbeeliminated.
However,asolutionmaybefoundinanadditionallevelofabstraction.
AtoolsetfortranslatingaHDIprototypeintomodelcheckingcode,couldhandletheconstructionofthecoordinationprotocol,makingthisprocesseffectivelyinvis-ibletothemodeler.
SuchaprocesscouldalsoallowformoreefcientmeansofcoordinatingtheHDIandhumantaskmod-els:onethatmightnotrequiretheuseofseparatemodelsintheactualmodelcheckercode.
WhiletheextensivemodelreductionsfromPhase1greatlydiminishedthedelitywithwhichthemodelrepresentedtheactualPCApump,thisprovidessomeadvantages.
SincethemodelfromPhase2doesnotsufferfromthememoryusageproblemsencounteredinPhase1,thisopensthedoortotheadditionofothermodelconstructstobeaddedallowingforamorecompletesystemanalysis.
FutureworkcanexpandthemodeldevelopedinPhase2withenvironmentalanddeviceautomationmodelsthatarecompatiblewiththeformalPCApumpreferencemodeldescribedin[1].
7.
3Goal3:modelveriabilityWearepredominantlyconcernedwithexploringhowformalmethodscanbeusedtoprovideinsightsintohumanfactorsandsystemsengineeringconcerns.
Ifourgoalwastofor-mallyverifypropertiesoftheBaxterIpump,themodelingcompromiseswemadeinordertoobtainaveriablemodelmightnecessitateachangeinmodelingphilosophyorveri-cationapproach.
Therearemanybarrierstotheveriabilityofmodelsofrealisticsystems.
Theseincludelargenumbersofparallelprocesses,largerangesofdiscretevaluedvariables,andnon-discretelyvaluedvariables.
Themodelingeffortsdescribedhereweresochallengingbecausethetargetsystemwasdependentonalargenumberofuserspeciednumericalvalues,allofwhichhadverylargeacceptableranges.
Thisresultedinthescopeofthemodelbeingreducedtothepointwhereitcouldnolongerbeusedforverifyingalloftheorigi-nalhumanoperatortaskbehaviors:withthemodelproducedinPhase1bmakingminorcompromisesandthemodelpro-ducedinPhase1conlyallowingforbehaviorsassociatedwithbasicPCApumpfunctionality.
AswasdemonstratedinPhase2,theveriabilityofthemodelactuallyincreasedwiththeinclusionofthehumantaskbehaviorasindicatedbythe98%reductioninthereportedstatespacefromthePhase1ctothePhase2model.
However,thiscameattheexpenseoftheverica-tionprocesstaking284timesaslong.
Thus,inacontextwherevericationtimeislessimportantthanthesizeofthemodel'sstatespace,theinclusionofthehumantaskbehaviormodelmaygenerallyprovetobeadvantageousinthefor-malvericationofsystemsthathaveahuman–automationinteractioncomponent,irrespectiveofwhetherthehumanbehaviorisofconcerninthevericationprocess.
Futureeffortsshouldinvestigatethedifferentfactorsthataffectthistradeoff.
Evenexploitingthisadvantage,therelativesimplicityofthedevicethatwasmodeledinthisworkmakesitclearthattherearemanysystemsthatdependofhuman–auto-mationinteractionthatwouldbeevenmorechallengingtoverify,ifnotimpossible,usingthesetechniques.
Whiletheuseofboundedmodelcheckersmayprovidesomeverica-tioncapabilitiesforcertainsystems,thereislittlethatcanbedonewithouteitherusingadditionalabstractiontechniquesorfutureadvancesinmodelcheckingtechnologyandcom-putationpower.
Itiscommonpracticeintheformalmethodscommunitytousemoreadvancedformsofdataabstractionthanthoseemployedinthisworktomitigatethecomplexityofvari-ableswithlargevalueranges(anoverviewofthesemethodscanbefoundin[20]).
Becausethenatureofthemodeledhumantaskbehaviorinthisworkwasconcernedwiththedigitleveleditingofthedatavalues,suchabstractionswere123230M.
L.
Bolton,E.
J.
Bassnotappropriateforthisparticularendeavor.
Additionally,automaticpredicateabstractiontechniqueslikethoseusedincounterexample-guidedabstractionrenement[6]couldpotentiallyalleviatesomeofthecomplexityproblemsenco-unteredinthisworkwithoutrequiringchangestothemodelsthemselves.
Futureworkshouldinvestigatehowthesedif-ferentabstractiontechniquescouldbeusedwhenmodelingsystemsthatdependonhuman-automationinteractioninwaysthatareintuitivetohumanfactorsengineers.
Itisclearthatthemultiple,large-value-rangedvariableswerethesourceofmostofthemodelcomplexityproblemsinthepumpexample,asshowninthedrasticdecreaseinver-icationperformancetimebetweenthemodelsproducedinPhases1band1c.
Thus,hadthetargetsystembeenmorecon-cernedwithproceduralbehaviorsandlessontheinterrela-tionshipsbetweennumericalvalues,thesystemmodelwouldhavebeenmuchmoretractable.
Futureworkshouldidentifyadditionalpropertiesofsystemsdependentonhuman–auto-mationinteractionthatlendthemselvestobeingmodeledandveriedusingtheframeworkdiscussedhere.
Finally,someoftheperformanceissuesweencounteredcanbeattributedtoouruseofSAL.
Forexample,modelcheckerssuchasSPIN[15]donotperformthelengthypro-cessofconstructingtheBDDrepresentationbeforestartingthecheckingprocess.
Futureworkshouldinvestigatewhichmodelcheckerisbestsuitedforevaluatinghuman–automa-tioninteraction.
8ConclusionTheworkpresentedherehasshownthatitispossibletocon-structmodelsofhuman–automationinteractionaspartofalargersystemforuseinformalvericationprocesseswhileadheringtosomeofthearchitecturalgoalsinFig.
1.
Ithasalsoshownthattheincorporationofhumantaskbehaviormodelsintosystemmodelsmayhelpalleviatethestateexplosionprobleminsomesystemsthatdependonhuman–automationinteraction.
However,thissuccesswastheresultofanumberofcompromisesthatproducedamodelthatwasnotasrep-resentative,understandable,ormodularasdesired.
Thus,inorderforformalmethodstobecomemoreusefulfortheHFEcommunity,thevericationtechnologywillneedtobeabletosupportamorediversesetofsystems.
Further,newmodelingtoolsmayberequiredtosupportrepresentationsthathumanfactorsengineersuse.
Theseadvanceswillultimatelyallowformalmethodstobecomeamoreusefultoolforhumanfactorsengineersworkingwithsafetycriticalsystems.
AcknowledgmentsTheresearchdescribedwassupportedinpartbyGrantNumberT15LM009462fromtheNationalLibraryofMedicineandResearchGrantAgreementUVA-03-01,sub-award2623-VAfromtheNationalInstituteofAerospace(NIA).
Thecontentissolelytheresponsibilityoftheauthorsanddoesnotnecessarilyrepresenttheof-cialviewsoftheNIA,NASA,theNationalLibraryofMedicine,ortheNationalInstitutesofHealth.
TheauthorswouldliketothankRaduI.
SiminiceanuoftheNIAandBenDiVitooftheNASALangleyResearchCenterfortheirtechnicalhelp.
TheywouldliketothankDianeHaddon,JohnKnapp,PaulMerrel,KathrynMcGough,andSherryWoodoftheUniversityofVirginiaHealthSystemfordescribingthefunctionalityoftheBaxterIpumpandforprovidingdocumentation,trainingmaterials,anddeviceaccess.
OpenAccessThisarticleisdistributedunderthetermsoftheCreativeCommonsAttributionNoncommercialLicensewhichpermitsanynoncommercialuse,distribution,andreproductioninanymedium,providedtheoriginalauthor(s)andsourcearecredited.
References1.
ArneyD,JetleyR,JonesP,LeeI,SokolskyO(2007)Formalmeth-odsbaseddevelopmentofaPCAinfusionpumpreferencemodel:genericinfusionpump(GIP)project.
In:Proceedingsofthe2007jointworkshoponhighcondencemedicaldevices,software,andsystemsandmedicaldeviceplug-and-playinteroperability.
IEEEComputerSociety,Washington,DC,pp23–332.
BaxterHealthCareCorporation(1995)Ipumppainmanage-mentsystemoperator'smanual.
BaxterHeathCareCorporation,McGawPark3.
BoltonML,BassEJ(2009)Buildingaformalmodelofahuman-interactivesystem:insightsintotheintegrationofformalmethodsandhumanfactorsengineering.
In:ProceedingsoftherstNASAformalmethodssymposium.
NASAAmesResearchCenter,Moff-ettField,pp6–154.
BoltonML,BassEJ(2009)Enhancedoperatorfunctionmodel:agenerichumantaskbehaviormodelinglanguage.
In:Proceed-ingsoftheIEEEinternationalconferenceonsystems,man,andcybernetics.
IEEE,Piscataway,pp2983–29905.
BoltonML,BassEJ(2009)Amethodfortheformalvericationofhuman-interactivesystems.
In:Proceedingsofthe53rdannualmeetingofthehumanfactorsandergonomicssociety.
HumanFac-torsandErgonomicsSociety,SantaMonica,pp764–7686.
ClarkeE,GrumbergO,JhaS,LuY,VeithH(2003)Counterexam-ple-guidedabstractionrenementforsymbolicmodelchecking.
JACM50(5):752–7947.
CrowJ,JavauxD,RushbyJ(2000)Modelsandmechanizedmeth-odsthatintegratehumanfactorsintoautomationdesign.
In:Pro-ceedingsofthe2000internationalconferenceonhuman-computerinteractioninaeronautics.
AssociationfortheAdvancementofArticialIntelligence,MenloPark,pp163–1688.
CurzonP,RuksenasR,BlandfordA(2007)Anapproachtoformalvericationofhuman–computerinteraction.
FormalAspComput19(4):513–5509.
DeMouraL,OwreS,ShankarN(2003)TheSALlanguagemanual.
Technicalreport,ComputerScienceLaboratory,SRIInternational,MenloPark10.
DeganiA(1996)Modelinghuman–machinesystems:onmodes,error,andpatternsofinteraction.
PhDthesis,GeorgiaInstituteofTechnology,Atlanta11.
DeganiA,KirlikA(1995)Modesinhuman–automationinterac-tion:Initialobservationsaboutamodelingapproach.
In:Proceed-ingsoftheIEEEinternationalconferenceonsystems,manandcybernetics.
IEEE,Piscataway,pp3443–345012.
FieldsRE(2001)Analysisoferroneousactionsinthedesignofcriticalsystems.
PhDthesis,UniversityofYork,York13.
HeymannM,DeganiA(2007)Formalanalysisandautomaticgen-erationofuserinterfaces:approach,methodology,andanalgo-rithm.
HumFactors49(2):311–330123Formallyverifyinghuman–automationinteraction23114.
HeymannM,DeganiA,BarshiI(2007)Generatingproceduresandrecoverysequences:aformalapproach.
In:Proceedingsofthe14thinternationalsymposiumonaviationpsychology.
AssociationforAviationPsychology,Dayton,pp252–25715.
HolzmannGJ(2003)Thespinmodelchecker,primerandreferencemanual.
Addison-Wesley,Reading16.
JavauxD(2002)Amethodforpredictingerrorswheninteractingwithnitestatesystems.
Howimplicitlearningshapestheuser'sknowledgeofasystem.
ReliabEngSystSaf75(2):147–16517.
KirwanB,AinsworthLK(1992)Aguidetotaskanalysis.
TaylorandFrancis,Philidelphia18.
KohnLT,CorriganJ,DonaldsonMS(2000)Toerrishuman:build-ingasaferhealthsystem.
NationalAcademyPress,Washington19.
KreyN(2007)2007Nallreport:accidenttrendsandfactorsfor2006.
Technicalreport.
http://download.
aopa.
org/epilot/2007/07nall.
pdf20.
Mansouri-SamaniM,PasareanuCS,PenixJJ,MehlitzPC,O'MalleyO,VisserWC,BratGP,MarkosianLZ,PressburgerTT(2007)Programmodelchecking:apractitioner'sguide.
Techni-calreport,IntelligentSystemsDivision,NASAAmesResearchCenter,MoffettField21.
MitchellCM,MillerRA(1986)Adiscretecontrolmodelofoper-atorfunction:amethodologyforinformationdislaydesign.
IEEETransSystManCybernASystHum16(3):343–35722.
PerrowC(1984)Normalaccidents.
BasicBooks,NewYork23.
RushbyJ(2002)Usingmodelcheckingtohelpdiscovermodeconfusionsandotherautomationsurprises.
ReliabEngSystSaf75(2):167–17724.
SchraagenJM,ChipmanSF,ShalinVL(2000)Cognitivetaskanalysis.
LawrenceErlbaumAssociates,Mahwah25.
StantonN(2005)Humanfactorsmethods:apracticalguideforengineeringanddesign.
AshgatePublishing,Brookeld26.
ThurmanDA,ChappellAR,MitchellCM(1998)AnenhancedarchitectureforOFMspert:adomain-independentsystemforintentinferencing.
In:ProceedingsoftheIEEEinternationalconferenceonsystems,man,andcybernetics.
IEEE,Piscataway,pp3443–345027.
VicenteKJ(1999)Cognitiveworkanalysis:towardsafe,produc-tive,andhealthycomputer-basedwork.
LawrenceErlbaumAsso-ciates,Mahwah28.
WellsAT,RodriguesCC(2004)Commercialaviationsafety,4thedn.
McGraw-Hill,NewYork29.
WickensCD,LeeJ,LiuYD,Gordon-BeckerS(2003)Introductiontohumanfactorsengineering.
Prentice-Hall,UpperSaddleRiver123
1007/s11334-010-0129-9ORIGINALPAPERFormallyverifyinghuman–automationinteractionaspartofasystemmodel:limitationsandtradeoffsMatthewL.
Bolton·EllenJ.
BassReceived:8March2010/Accepted:25March2010/Publishedonline:9April2010TheAuthor(s)2010.
ThisarticleispublishedwithopenaccessatSpringerlink.
comAbstractBoththehumanfactorsengineering(HFE)andformalmethodscommunitiesareconcernedwithimprovingthedesignofsafety-criticalsystems.
Thisworkdiscussesamodelingeffortthatleveragedmethodsfrombotheldstoperformformalvericationofhuman–automationinter-actionwithaprogrammabledevice.
Thiseffortutilizesasystemarchitecturecomposedofindependentmodelsofthehumanmission,humantaskbehavior,human-deviceinter-face,deviceautomation,andoperationalenvironment.
ThegoalsofthisarchitectureweretoallowHFEpractitionerstoperformformalvericationsofrealisticsystemsthatdependonhuman–automationinteractioninareasonableamountoftimeusingrepresentativemodels,intuitivemodelingcon-structs,anddecoupledmodelsofsystemcomponentsthatcouldbeeasilychangedtosupportmultipleanalyses.
Thisframeworkwasinstantiatedusingapatientcontrolledanal-gesiapumpinatwophasedprocesswheremodelsineachphasewereveriedusingacommonsetofspecications.
Therstphasefocusedonthemission,human-deviceinterface,anddeviceautomation;andincludedasimple,unconstrainedhumantaskbehaviormodel.
Thesecondphasereplacedtheunconstrainedtaskmodelwithonerepresentingnormativepumpprogrammingbehavior.
Becausemodelsproducedintherstphaseweretoolargeforthemodelcheckertoverify,anumberofmodelrevisionswereundertakenthataffectedthegoalsoftheeffort.
Whiletheuseofhumantaskbehaviormodelsinthesecondphasehelpedmitigatemodelcomplex-ity,vericationtimeincreased.
AdditionalmodelingtoolsM.
L.
Bolton(B)·E.
J.
BassDepartmentofSystemsandInformationEngineering,UniversityofVirginia,151Engineer'sWay,Charlottesville,VA,USAe-mail:mlb4b@virginia.
eduE.
J.
Basse-mail:ejb4n@virginia.
eduandtechnologicaldevelopmentsarenecessaryformodelcheckingtobecomeamoreusabletechniqueforHFE.
KeywordsHuman–automationinteraction·Taskanalysis·Formalmethods·Modelchecking·Safetycriticalsystems·PCApump1IntroductionBothhumanfactorsengineering(HFE)andformalmethodsareconcernedwiththeengineeringofrobustsystemsthatwillnotfailunderrealisticoperatingconditions.
Thetraditionaluseofformalmethodshasbeentoevaluateasystem'sauto-mationunderdifferentoperatingand/orenvironmentalcon-ditions.
However,humanoperatorscontrolanumberofsafetycriticalsystemsandcontributetounforeseenproblems.
Forexample,humanbehaviorhascontributedtobetween44,000and98,000deathsnationwideeveryyearinmedicalpractice[18],74%ofallgeneralaviationaccidents[19],atleasttwo-thirdsofcommercialaviationaccidents[28],andanumberofhighproledisasterssuchastheincidentsatThreeMileIslandandChernobyl[22].
HFEfocusesonunderstandinghumanbehaviorandapplyingthisknowledgetothedesignofhuman–automationinteraction:makingsystemseasiertousewhilereducingerrorsand/orallowingrecoveryfromthem[25,29].
ByleveragingtheknowledgeofbothHFEandformalmethods,researchershaveidentiedthecognitiveprecon-ditionsformodeconfusionandautomationsurprise[7,10,16,23];automaticallygenerateduserinterfacespecications,emergencyprocedures,andrecoverysequences[13,14];andidentiedhumanbehaviorsequences(normativeorerrone-ous)thatcontributetosystemfailures[8,12].
123220M.
L.
Bolton,E.
J.
BassWhileallofthisworkhasproducedusefulresults,themodelshavenotincludedallofthecomponentsnecessarytoanalyzehuman–automationinteraction.
ForHFEanalysesofhuman–automationinteraction,theminimalsetofcom-ponentsarethegoalsandproceduresofthehumanopera-tor;theautomatedsystemanditshumaninterface;andtheconstraintsimposedbytheoperationalenvironment.
Cogni-tiveworkanalysisisconcernedwithidentifyingconstraintsintheoperationalenvironmentthatshapethemissiongoalsofthehumanoperator[27];cognitivetaskanalysisiscon-cernedwithdescribinghowhumanoperatorsnormativelyanddescriptivelyperformgoalorientedtaskswheninteract-ingwithanautomatedsystem[17,24];andmodelingframe-workssuchas[11]seektonddiscrepanciesbetweenhumanmentalmodels,human-deviceinterfaces(HDIs),anddeviceautomation.
Inthiscontext,problemsrelatedtohuman–automationinteractionmaybeinuencedbythehumanoperator'smission,thehumanoperator'staskbehavior,theoperationalenvironment,theHDI,thedevice'sautomation,andtheirinterrelationships.
Wearedevelopingmethodsandtoolstoallowhumanfactorsengineerstoexploittheirexistinghumantaskmodelingconstructswiththepowerfulvericationcapabil-itiesofformalmethodsinordertoidentifypotentialprob-lemswithhuman–automationinteractioninsafetycriticalsystemsthatmayberelatedtohumantaskbehavior,theautomateddevice,theoperationalenvironment,ortheirinter-action.
Tothisend,wearedevelopingacomputationalframe-work(Fig.
1)fortheformalmodelingofhuman-automationinteraction.
Thisframeworkutilizesconcurrentmodelsofhumanoperatortaskbehavior,humanmission(thegoalstheoperatorwishestoachieveusingthesystem),deviceauto-mation,andtheoperationalenvironmentwhicharecom-posedtogethertoformalargersystemmodel.
Inter-modelinteractionisrepresentedbyvariablessharedbetweenmod-els.
Environmentvariablescommunicateinformationaboutthestateoftheenvironmenttothedeviceautomation,mis-sion,andhumantaskmodels.
Missionvariablescommuni-catethemissiongoalstothehumantaskmodel.
InterfacevariablesconveyinformationaboutthestateoftheHDI(dis-playedinformation,thestateofinputwidgets,etc.
)tothehumantaskmodel.
ThehumantaskmodelindicateswhenandwhatactionsahumanoperatorwouldperformontheHDI.
TheHDIcommunicatesitscurrentstatetothedeviceauto-mationviatheinterfacevariables.
TheHDIreceivesinfor-mationaboutthestateofthedeviceautomationmodelviatheautomationstatevariables.
Forbroaderapplicability,theanalysisframeworkmustsupportmodelingconstructsintuitivetothehumanfac-torsengineerinordertoallowhimtoeffectivelymodelhumanmissions,humantasks,andHDIs.
Becauseanengi-neermaywishtorerunvericationsusingdifferentmissions,taskmodels,HDIs,environments,orautomationbehaviors,thesecomponentsshouldremaindecoupled(asisthecaseinFig.
1).
Finally,themodelingtechniquemustbecapableofrepresentingthetargetsystemswithenoughdelitytoallowtheengineertoperformthedesiredverication,anddosoinareasonableamountoftime(thiscouldmeanseveralhoursforasmallproject,orseveraldaysforamorecomplicatedone).
ThispaperdescribesaninstantiationofthisframeworkusingamodelofaBaxterIpumpPainManagementSystem[2],apatientcontrolledanalgesia(PCA)pumpthatadminis-terspainmedicationinaccordancewithconstraintsdenedbyahealthcaretechnician(describedinSect.
2.
1).
Modelsweredevelopedintwophases.
TherstphaseinvolvedtheconstructionanddebuggingoftheHDI,deviceautomation,andhumanmissionmodels(anenvironmentalmodelwasnotSystemModelHumanTaskModelDeviceAutomationModelHuman-DeviceInterfaceModelMissionModelEnvironmentModelMissionVariablesHumanActionVariablesInterfaceVariablesInterfaceVariablesAutomationStateVariablesEnvironmentVariablesFig.
1Frameworkfortheformalmodelingofhuman–automationinteraction.
Arrowsbetweenmodelsrepresentvariablesthataresharedbetweenmodels.
Thedirectionofthearrowindicateswhethertherep-resentedvariablesaretreatedasinputsoroutput.
Ifthearrowissourcedfromamodel,therepresentedvariablesareoutputsofthatmodel.
Ifthearrowterminatesatamodel,therepresentedvariablesareinputstothatmodel123Formallyverifyinghuman–automationinteraction221includedbecauseofthegeneralstabilityoftheenvironmentinwhichanactualpumpoperates)withanunconstrainedhumantaskmodelservingasaplaceholderforamorereal-istichumantaskmodel.
Thesecondextendedthemodelpro-ducedinPhase1witharealisticmodelofthehumantask,completingtheframework.
Eventhoughthetargetdeviceinthismodelingeffortwasseeminglysimple,thesystemmodelthatwasinitiallydevel-opedinPhase1(Phase1a)wastoodifcultforthemodelcheckertoprocessquicklyandtoocomplexforittoverify.
Thusanumberofrevisionswereundertaken[3].
InPhase1bareducedandabstractedmodeloftheBaxterIpumpwasproducedwhich,whilecapableofbeingusedinsomever-ications,didsoattheexpenseoflimitingthenumberofgoalsrepresentedinthehumanmissionmodel.
ThisPhase1bmodellimitedtheusefulnessofincorporatinghumantaskbehaviorinPhase2.
Thus,inPhase1c,thesystemmodelwasreducedtoencompasstheprogrammingprocedureforamuchsimplerPCApump.
InPhase2,theincorporationofthemorerealistichumantaskbehavioractuallyresultedinareductionofthetotalsystemmodel'scomplexity,butdidsoattheexpenseofanincreaseinvericationtime.
Thispaperdiscussesthesemodelingphases,thevericationresultspro-ducedinthem,andtheirassociatedcompromisesinrelationtothegoalsofthemodelingarchitecture.
Theseareusedtodrawconclusionsaboutthefeasibilityofusingformalmeth-odstoinformhuman–automationinteraction.
2Methods2.
1ThetargetsystemTheBaxterIpumpisanautomatedmachinethatcontrolsdeliveryofsedative,analgesic,andanestheticmedicationsolutions[2].
Solutiondeliveryviaintravenous,subcutane-ous,andepiduralroutesissupported.
Medicationsolutionsaretypicallystoredinbagslockedinacompartmentonthebackofthepump.
Pumpbehaviorisdictatedbyinternalautomation,whichcandependonhowthepumpisprogrammedbyahumanoperator.
PumpprogrammingisaccomplishedviaitsHDI(Fig.
2)whichcontainsadynamicLCDdisplay,asecu-ritykeylock,andeightbuttons.
Whenprogrammingthepump,theoperatorisabletospecifyallofthefollowing:whethertouseperiodicorcontinuousdosesofmedications(i.
e.
,themodewhichcanbePCA,Basal+PCA,orContinu-ous),whethertouseprescriptioninformationpreviouslypro-grammedintothepump,theuidvolumecontainedinthemedicationbag,theunitsofmeasureusedfordosage(ml,mg,org),whetherornottoadministerabolus(aninitialdoseofmedication),dosageamounts,dosageowrates(foreitherbasalorcontinuousratesasdeterminedbythemode),StartStopEnterOn/OffClearCLCDDisplayInterfaceMessageDisplayedValueCursorSecurityKeyLeftButtonUp(Scroll)ButtonRightButtonClearButtonStartButtonStopButtonEnterButtonOn/OffButtonFig.
2AsimpliedrepresentationoftheBaxterIpump'shuman-deviceinterface.
Notethattheactualpumpcontainsadditionalcontrolsandinformationconveyancesthedelaytimebetweendosages,and1hlimitsontheamountofdeliveredmedication.
Duringprogramming,thesecuritykeyisusedtolockandunlockthecompartmentcontainingthemedicationsolution.
Theunlockingandlockingprocessisalsousedasasecuritymeasuretoensurethatanauthorizedpersonisprogrammingthepump.
Thestartandstopbuttonsareusedtostartandstopthedeliveryofmedicationatspecictimesduringpro-gramming.
Theon–offbuttonisusedtoturnthedeviceonandoff.
TheLCDdisplaysupportspumpoperationoptions.
Whentheoperatorchoosesbetweentwoormoreoptions,theinter-facemessageindicateswhatisbeingchosen,andtheinitialordefaultoptionisdisplayed.
Pressingtheupbuttonallowstheprogrammertoscrollthroughtheavailableoptions.
Whenanumericalvalueisrequired,theinterfacemessageconveysitsnameandthedisplayedvalueispresentedwiththecursorunderoneofthevalue'sdigits.
Theprogrammercanmovethepositionofthecursorbypressingtheleftandrightbuttons.
Heorshecanpresstheupbuttontoscrollthroughthedifferentdigitvaluesavailableatthatcursorposition.
Theclearbuttonsetsthedisplayedvaluetozero.
Theenterbuttonisusedtoconrmvaluesandtreatmentoptions.
Asidefromtheadministrationoftreatment,thepump'sautomationsupportsdynamiccheckingandrestrictionofoperatorenteredvalues.
Thus,inadditiontohavinghardlim-itsonvalueranges,theextremacanchangedynamicallyinresponsetootheruserspeciedvalues.
2.
2ApparatusAllformalmodelswereconstructedusingtheSymbolicAnalysisLaboratory(SAL)language[9]becauseofitsasso-ciatedanalysisanddebuggingtools,anditssupportforboththeasynchronousandsynchronouscompositionofdifferent123222M.
L.
Bolton,E.
J.
Bassmodels(modulesusingSAL'sinternalsemantics).
ThetaskmodelrepresentationsdescribednextweretranslatedintotheSALlanguageasasinglemoduleusingacustom-builtjavaprogram[5].
AllvericationsweredoneusingSAL-SMC3.
0,theSALsymbolicmodelchecker.
1Vericationswereconductedona3.
0GHzdual-coreIntelXeonprocessorwith16GBofRAMrunningtheUbuntu9.
04desktop.
Humantaskmodelswerecreatedusinganintermediarylanguagecalledenhancedoperatorfunctionmodel(EOFM)[4],anXML-based,generichumantaskmodelinglan-guagebasedontheoperatorfunctionmodel(OFM)[21,26].
EOFMsarehierarchicalandheterarchicalrepresentationsofgoalormissiondrivenactivitiesthatdecomposeintolowerlevelactivities,andnally,atomicactions—whereactionscanrepresentanyobservable,cognitive,orperceptualhumanbehavior.
EOFMsexpresstaskknowledgebyexplicitlyspec-ifyingtheconditionsunderwhichhumanoperatoractivi-tiescanbeundertaken:whatmustbetruebeforetheycanexecute(preconditions),whentheycanrepeat(repeatcon-ditions),andwhentheyhavecompleted(completioncondi-tions).
Anyactivitycandecomposeintooneormoreotheractivitiesoroneormoreactions.
Adecompositionoperatorspeciesthetemporalrelationshipsbetweenandthecardi-nalityofthedecomposedactivitiesoractions(whentheycanexecuterelativetoeachotherandhowmanycanexecute).
EOFMscanberepresentedvisuallyasatree-likegraphstructure(examplescanbeseeninFigs.
3,4,5,6,7).
Intheserepresentations,actionsarerepresentedasrect-anglesandactivitiesarerepresentedasroundedrectan-gles.
Anactivity'sdecompositionispresentedasanarrow,labeledwiththedecompositionoperator,extendingbelowitthatpointstoalargeroundedrectanglecontainingthedecomposedactivitiesoractions.
Inthiswork,threedecompositionoperatorsareused:(1)ord(allactivitiesoractionsinthedecompositionmustexecuteintheordertheyappear);(2)or_seq(oneormoreoftheactivitiesoractionsinthedecompositionmustexecute);and(3)xor(exactlyoneactivityoractioninthedecompositionmustexecute).
Conditionsonactivitiesarerepresentedasshapesorarrows(annotatedwiththeconditionlogic)con-nectedtotheactivitythattheyconstrain.
Theform,posi-tion,andcoloroftheshapearedeterminedbythetypeofcondition.
Apreconditionispresentedasayellow,downward-pointingtriangleconnectedtotheleftsideoftheactivity.
Acompletionconditionisrepresentedbyamagenta,upward-pointingtriangleconnectedtotherightsideoftheactivity.
Arepeatconditionisdepictedasanarrowrecur-sivelypointingtothetopoftheactivity.
Moredetailscanbefoundin[4].
1SomemodeldebuggingwasalsoconductedusingSAL'sboundedmodelchecker.
2.
3VericationspecicationTwospecicationswereemployedineachofthemodelingphases:bothwerewritteninlineartemporallogicandeval-uatedusingSAL–SMC.
Therst(Eq.
1),usedformodeldebugging,veriesthatavalidprescriptioncouldbepro-grammedintothepump:GiInterfacemessage=TreatmentAdministering∧iMode=iPrescribedMode∧lFluidVolume=iPrescribedFluidVolume∧lPCADose=iPrescribedPCADose∧lDelay=iPrescribedDelay∧lBasalRate=iPrescribedBasalRate∧l1HourLimit=iPrescribed1HourLimit∧lBolus=iPrescribedBolus∧lContinuousRate=iPrescribedContinuousRate(1)Here,ifthemodelisabletoenterastateindicatingthattreatmentisadministering(iInterfaceMessage=Treat-mentAdministering)withtheentered(orprogrammed)prescriptionvalues(iMode,lFluidVolume,…,lContinuous-Rate)matchingtheprescriptionvaluesgeneratedbythemissionmodel(variableswiththeiPrescribedprex),acounterexamplewouldbeproducedillustratinghowthatpre-scriptionwasprogrammed.
Variableswithaniprexindicatethatthevariableisaninputtothehumantaskmodel.
Vari-ableswithanlprexindicatethatthevariableislocaltoagivenmodel.
Thesecondspecication(Eq.
2)representedasafetyprop-ertythatwasexpectedtoverifytotrue,thusallowingthemodelcheckertotraversetheentirestatespaceofeachphase'smodel.
BecausesuchavericationallowsSALtoreportthesizeofamodel'sstatespace,vericationsusingthisspecicationwouldprovidesomemeansofcomparingthecomplexityofthemodelsproducedineachphase.
GiInterfaceMessage=TreatmentAdministering∧iMode=Continuous∧lDelay=0(2)Here,thespecicationisassertingthatthemodelshouldneverenterastatewheretreatmentisadministeringinthePCAorBasal+PCAmodes(iMode=Continuous)whenthereisnodelaybetweendoses.
2Thus,ifEq.
(2)veriestotrue,thepumpwillneverallowaprogrammertoenterprescriptionsthatwouldallowpatientstocontinuouslyadministerPCAdosestothemselves[2].
2AdelaycanonlybeensetwhenthePCAorBasal+PCAmodeshavebeenselectedbythehumanoperator.
TherearenodelaysbetweendoseswhenthepumpisintheContinuousmode.
123Formallyverifyinghuman–automationinteraction2233Phase1a:arepresentativemodeloftheIpump3.
1ModeldescriptionAninitialmodelwascreatedtoconformtothearchitecturalanddesignphilosophyrepresentedinFig.
2:themissionwasrepresentedasasetofviableprescriptionsoptions;themis-sion,humanoperator,human-deviceinterface,anddeviceautomationweremodeledindependentlyofeachother;andthebehavioroftheautomatedsystemandHDImodelswasdesignedtoaccuratelyreectthebehaviorofthesesystemsasdescribedintheuser'smanual[2]andobservedthroughdirectinteractionwiththedevice.
Anunconstrainedhumanoperatorwasconstructedthatcouldissueanyvalidhumanactiontothehuman-deviceinterfacemodelatanygiventime.
BecausethePCApumpgenerallyoperatesinacon-trolledenvironment,awayfromtemperatureandhumidityconditionsthatmightaffecttheperformanceofthepump'sautomation,noenvironmentalmodelwasincluded.
Finally,becausedocumentationrelatedtotheinternalworkingsofthepumpwaslimited,thesystemautomationmodelwasrestrictedtothatassociatedwiththepumpprogrammingpro-cedure:behaviorthatcouldbegleanedfromtheoperator'smanual[2],correspondenceswithhospitalstaff,anddirectinteractionwiththepump.
3.
2ModelcoordinationModelinfrastructurewasrequiredtoensurethathumanoper-atoractionswereproperlyrecognizedbytheHDImodel.
Inanidealmodelingenvironment,humanactionbehaviororig-inatingfromthehumanoperatormodelcouldhavebothanasynchronousandsynchronousrelationshipwiththeHDImodel.
SynchronousbehaviorwouldallowtheHDImodeltoreacttouseractionsinthesametransitioninwhichtheywereissued/performedbythehumanoperatormodel.
How-ever,boththehumanoperatorandHDImodelsoperateinde-pendentlyofeachother,andmayhavestatetransitionsthataredependentoninternalorexternalconditionsthatarenotdirectlyrelatedtothestateoftheothermodel.
Thissuggestsanasynchronousrelationship.
SALonlyallowsmodelstobecomposedwitheachotherasynchronouslyorsynchronously(butnotboth).
Thus,itwasnecessarytoadaptthemodelstosupportfeaturesassociatedwiththeunusedcomposition.
AsynchronouscompositionwasusedtocomposethehumanoperatorandHDImodels.
ThisnecessitatedsomeadditionalinfrastructuretopreventthehumanoperatormodelfromissuinguserinputsbeforetheHDImodelwasreadytointerpretthemandtopreventthehumanoperatormodelfromterminatingagiveninputbeforetheinterfacecouldrespondtoit.
Thiswasaccomplishedthroughtheaddi-tionoftwoBooleanvariables:oneindicatingthatinputhadbeensubmitted(henceforthcalledSubmitted)andavariableindicatingtheinterfacewasreadytoreceiveactions(hence-forthcalledReady).
Thiscoordinationoccurredasfollows:–IfReadyistrueandSubmittedisfalse,thehumanoperatormodulesetsoneormoreofthehumanactionvariablestoanewinputvalueandsetsSubmittedtotrue.
–IfReadyandSubmittedaretrue,thehuman-deviceinter-facemodulerespondstothevaluesofthehumanactionvariablesandsetsReadytofalse.
–IfReadyisnottrueandSubmittedistrue,thehumanoper-atormodulesetsSubmittedtofalse.
–IfReadyandSubmittedarebothfalseandtheautomatedsystemisreadyforadditionalhumanoperatorinput,thehuman-deviceinterfacemodulesetsReadytotrue.
3.
3VericationresultsAttemptstoverifythismodelusingthespecicationsinEqs.
1and2resultedintwoproblemsrelatedtothefeasibilityandusefulnessofthevericationprocedure.
First,theSAL–SMCprocedurefortranslatingtheSALcodeintoabinarydecisiondiagram(BDD)tookexcessivelylong(morethan24h),atimeframeimpracticalformodeldebugging.
Second,thevericationprocesswhichfollowedtheconstructionoftheBDDeventuallyranoutofmemory,thusnotreturningavericationresult.
4Phase1b:areducedBaxterIpumpmodelAsaresultofthefailedvericationofthemodelproducedinPhase1a,significantrevisionswererequiredtomakethemodelmoretractable.
Thesearediscussedbelow.
4.
1RepresentationofnumericalvaluesToreducethetimeneededtoconverttheSAL-codemodeltoaBDD,anumberofmodicationsweremadetothemodelfromPhase1abyrepresentingmodelconstructsinwaysmorereadilyprocessedbythemodelchecker.
Assuch,themod-icationsdiscussedheredidnotultimatelymaketheBDDrepresentationofthemodelsmaller,butmerelyexpediteditsconstruction.
4.
1.
1RedundantrepresentationofvaluesTwodifferentrepresentationsofthevaluesprogrammedintothepumpbytheoperatorwereusedintheHDIanddeviceautomationmodels.
BecausetheHDIrequiredthehumanoperatortoentervaluesbyscrollingthroughtheavailablevaluesforindividualdigits,anarrayofintegerdigitswasappropriatefortheHDImodel.
However,becausethesystem123224M.
L.
Bolton,E.
J.
Bassautomationwasconcernedwithdynamicallycheckinglimitsandusingenteredvaluestocomputeothervalues,anumer-icalrepresentationoftheactualvaluewasmoreconvenientfortheautomatedsystemmodel.
ThisredundancyburdenedtheBDDtranslator.
Thiswasremediedbyeliminatingthedigitarrayrepresentationsandusingfunctionstoenableactionsfromthehumantaskmodeltoincrementallychangeindividualdigitswithinavalue.
4.
1.
2RealnumbersandintegersInthemodelproducedinPhase1a,allnumericalvalueswererepresentedasrealvalueswithrestrictedranges.
Thiswasdonebecausemostuserspeciedvalueswereeitherinte-gersoroatingpointnumbers(precisetoasingledecimalpoint).
Nodataabstractionswereinitiallyconsideredbecausethenatureofthehumantask(modeledinPhase2)requiredmanipulationofvalues'individualdigits.
However,repre-sentingvaluesthiswayprovedespeciallychallengingfortheBDDtranslator.
Thus,allvaluesweremodiedsothattheycouldberepresentedasrestrictedrangeintegers.
Forintegervariablesrepresentingoatingpointnumbers,thismeantthatthemodelvaluewastentimesthevalueitrepresented.
Thisrepresentationallowedthevaluestostillbemanipulatedattheindividualdigitlevel,whilemakingthemmorereadilyinterpretablebytheBDDtranslator.
4.
1.
3VariablerangesInthePhase1amodel,theupperboundontherangeofallvalue-basedvariableswassettothetheoreticalmaximumofanyvaluethatcouldbeprogrammedintothepump:99,999.
3However,toreducetheamountofworkrequiredfortheBDDconversion,therangeofeachnumericallyvaluedvariablewasgivenaspecicupperboundthatcorrespondedtothemaximumvalueitcouldactuallyassumeinthedevice.
4.
2ModelreductionToreducethesizeofthemodel,avarietyofelementswereremoved.
InallcasesthesereductionsweremeanttoreducethenumberofstatevariablesintheHDIordeviceautoma-tionmodels(slicing),orreducetherangeofvaluesavariablecouldassume(dataabstraction).
Unfortunately,eachofthesereductionsalsoaffectedwhathumantaskscouldultimatelybemodeledandthusveriedinsubsequentmodeliterations.
Allofthefollowingreductionswereundertaken:–InthePhase1amodel,themissionmodelcouldgener-ateaprescriptionfromtheentireavailablerangeofvalidprescriptions.
Thiswaschangedsothatfewerprescription3Alllowerboundsweresetto0.
optionsweregeneratedinPhase1b'smissionmodel:thatofprogrammingaprescriptionwithacontinuousdosagewithtwooptionsforbolusdelivery(0.
0and1.
0ml)andtwocontinuousowrateoptions(1.
0and9.
0ml/h).
Whilethissignificantlyreducedthenumberofmodelstates,italsoreducedthenumberofprescriptionsthatcouldbeusedinvericationprocedures.
–InthePhase1amodel,theHDImodelwouldallowtheoperatortoselectwhatunitstousewhenenteringprescrip-tions(ml,mg,org).
OnlythemlunitoptionwasincludedinthePhase1bmodel.
Thisreducedthenumberofinter-facemessagesinthemodel,allowedfortheremovalofseveralvariables(thoserelatedtotheunitoptionselection,andsolutionconcentrationspecication),andreducedtherangesrequiredforseveralnumericalvaluesrelatedtotheprescription.
Thiseliminatedtheoptionofincludingunitselectionandconcentrationspecicationtaskbehaviorsinthemodel.
–InthePhase1amodel,boththeHDIanddeviceautomationmodelsencompassedbehaviorrelatedtothedeliveryofmedicationsolutionduringtheprimingandbolusadmin-istrationprocedures.
Duringpriming,theHDIallowstheoperatortorepeatedlyinstructthepumptoprimeuntilallairhasbeenpushedoutoftheconnectedtubing.
Duringbolusadministration,theHDIallowstheoperatortotermi-natebolusinfusionbypressingthestopbuttontwice.
ThisfunctionalitywasremovedfromthePhase1bmodels,thuseliminatinginterfacemessagestatesandnumericalvaluesindicatinghowmuchuidhadbeendeliveredinbothpro-cedures.
Thiseliminatedthepossibilityofincorporatingtaskbehaviorrelatedtopumpprimingandbolusadminis-trationinthemodel.
–ThePhase1amodelmimickedthesecurityfeaturesfoundintheoriginaldevicewhichrequiredthehumanoperatortounlockandlockthedeviceonstartupandenterasecu-ritycode.
ThisfunctionalitywasremovedfromthePhase1bmodelwhichreducedthenumberofinterfacemessagesinthemodelandremovedthenumericalvariable(witha0–999range)associatedwithenteringthesecuritycode.
Thiseliminatedthepossibilityofmodelinghumantaskbehaviorrelatedtounlockingandlockingthepumpaswellasenteringthesecuritycodeinthemodel.
–InthePhase1amodel,theinterfacemessagecouldauto-maticallytransitiontobeingblank:mimickingtheactualpump'sabilitytoblankitsscreenafterthreesecondsofoperatorinactivity.
Becausefurtheroperatorinactionwouldresultintheoriginaldeviceissuinga"leftinpro-grammingmode"alert,ablankinterfacemessagecouldautomaticallytransitiontoanalertissuance.
Thisfunction-alitywasremovedfromthePhase1bmodel,eliminatingseveralinterfacemessagesaswellasvariablesthatkepttrackofthepreviousinterfacemessage.
Thus,theoption123Formallyverifyinghuman–automationinteraction225ofmodelingoperatortaskresponsetoscreenblankingandalertswasremovedfromthemodel.
WhilethesereductionsresultedinthePhase1bmodelbeingmuchsmallerandmoremanageablethantheoriginal,theabilitytomodelsomeofthetaskbehaviorsoriginallyasso-ciatedwiththedevicehadtobesacriced.
4.
3ResultsThePhase1bmodelwasabletocompletethevericationprocedurewithEq.
(1)andproduceacounterexamplewithasearchdepthof54inapproximately5.
9h,withthemajorityofthattime(5.
4h)usedforcreatingtheBDDrepresentation[3].
4Notsurprisingly,themodelcheckerranoutofmemorywhenattemptingtoverifyEq.
(2).
5Phase1c:asimplerPCApumpmodelWhilethemodeldevelopedinPhase1bdidproduceusableresultsandhassubsequentlybeenusedinthevericationofadditionalproperties(see[5]),thispowercameattheexpenseofareductioninthescopeofthemissionmodel.
Sincethemissiondirectlyinuenceswhathumanbehaviorwillexe-cute,thislimitedthehumantaskbehaviorthatcouldulti-matelybeveriedaspartofthesystemmodel.
Further,thefactthatthePhase1bmodelwastoocomplexforEq.
(2)tobeveriedpotentiallylimitedanyfuturemodeldevelopmentthatmightaddcomplexity.
Toremedytheseshortcoming,themodelproducedinPhase1bwasfurtherreducedtoonethatencompassedtheprogrammingofthemostbasicPCApumpfunctionalitywhiletherangesofpossiblevaluesfortheremainingmissionmodelvariableswereexpandedtobemorerealistic.
5.
1ModelreductionToobtainasmallerPCAmodel,allofthefollowingwereremoved:theselectionofmodeandtheabilitytospecifyabasalrate,continuousrate,bolusdosage,anduidvolume.
Asaresult,associatedinterfacemessagesandvariableswereremovedalongwiththeabilitytomodeltheirprogrammingaspartofthehumantaskbehaviormodel.
ThisresultedinamodelthatonlyencompassedfunctionalityforprogrammingaPCAdose,programmingthedelaybetweenPCAdoses,turningthepumponandoff,andstartingandstoppingtheadministrationoftreatment:functionalitycompatiblewiththemostbasicPCApumpoperations(see[1]).
4Completedmodels,SALoutputs,andcounterexamplescanbefoundathttp://cog.
sys.
virginia.
edu/ISSE2010/.
Valuerangeswerefurtherrestrictedtoreducethesizeofthemodel.
Specifically,theupperboundontheacceptabledelaybetweenPCAdosageswaschangedfrom240to60minutes.
This,coupledwiththeotherreductions,hadtheaddedbenetofallowingthenumberofdigitsrequiredfortheprogrammingofpumpvaluestobereducedto2ratherthantheoriginal4.
ThereductionsinotherareasallowedthescopeofthedelaysandPCAdosagesgeneratedbythemissionmodeltobeexpandedtoamorerepresentativeset.
ForPCAdosages,thefullrangeofvaluesfrom0.
1to9.
9in0.
1mlincrementsweresupported.
Fordelaybetweendosages,veoptionswereavailable:delaysof10,15,30,45,and60min.
Allpumpinterfacefunctionalitywasretainedfromthepreviousmodels.
Thus,theunconstrainedhumantaskmodelwasunchangedaswasthehumantaskandHDImodels'communicationprotocol.
5.
2ResultsThePhase1cmodelranthevericationprocedureforEq.
(1)(withtheeliminatedvariablesremoved)in6swithasearchdepthof22,muchfasterthanthemodelfromPhase1b.
ThevericationofthespecicationinEq.
(2)veriedtotruein129swithasearchdepthof259and78,768,682,750visitedstates.
6Phase2:incorporatingmodelsofhumanbehaviorInthesecondphaseofmodeling,weexpandedourinstantia-tionoftheframeworkbyincorporatingarealistichumantaskbehaviormodel.
WethereforereplacedtheunconstrainedhumanoperatorinthePhase1cmodelwithahumantaskbehaviormodelderivedfrompumpdocumentation[2]andtrainingmaterials.
ThismodelutilizedtheEOFMconceptsandthusrequiredsomeadditionalinfrastructureinordertoincorporateitintotheformalsystemmodel.
Wedescribethebehaviorsthatweremodeled,howtheseweretranslatedintotheformalmodel,andreportvericationresultsforthepro-ducedsystemmodel.
6.
1HumantaskbehaviormodelingandtranslationThepump'smaterialscontainedsixhigh-levelgoaldirectedbehaviorsforperformingavarietyofpumpactivitiesrelevanttothePhase1cmodelasfollows:–Turningonthepump.
–Stoppingtheinfusionofmedication.
–Turningoffthepump.
–EnteringaprescribedvalueforPCAdosagevolumes(inmilliliter).
123226M.
L.
Bolton,E.
J.
Bass–EnteringaprescribedvalueforthedelaybetweenPCAdoses(inminutes),and–Selectingwhethertostartorreviewanenteredprescrip-tion.
TheEOFMmodelsdescribingeachofthesebehaviorsarediscussedbelow.
6.
1.
1TurningOnthepumpThemodelforturningonthepumpisshowninFig.
3.
Here,theEOFMcanexecuteiftheinterfacemessageindi-catesthatthesystemisoff(iInterfaceMessage=SystemOff;aprecondition).
Thishigh-levelactivity(aTurnOnPump)iscompletedbyperformingtheactionofpressingtheon/offbutton(hPressOnOff).
Theorddecompositionoperatorindi-catesthatallofthedecomposedactivitiesoractionsmustbecompletedinsequentialorder.
TheEOFMhasaccomplisheditsgoal(acompletioncondition)whentheinterfacemes-sageindicatesthatthepumpisnolongeroff(iInterfaceMes-sage/=SystemOff).
6.
1.
2StoppinginfusionInfusionofmedicationcanbestopped(Fig.
4)iftheinter-faceindicatesthattreatmentisadministering(iInterfaceMes-sage=TreatmentAdministering).
Thisisaccomplishedbypressingthestopbutton(hPressStop)twiceinquicksuc-cessionwithnootherhumaninputsoccurringinbetween.
Theprocesshascompletedwhentheinterfaceindicatesthattreatmentisnotadministering(iInterfaceMessage/=Treat-mentAdministering).
aTurnOnPumpiInterfaceMessage=SystemOffiInterfaceMessage/=SystemOffordhPressOnOffFig.
3TheEOFMgraphicalrepresentationforturningonthepumpaStopInfusingiInterfaceMessage=TreatmentAdministeringiInterfaceMessage/=TreatmentAdministeringordhPressStophPressStopFig.
4TheEOFMgraphicalrepresentationforstoppinginfusion6.
1.
3TurningOffthepumpThemodelforturningoffthepump(Fig.
5)isrelevantiftheinterfacemessageindicatesthatthesystemisnotoff(iIn-terfaceMessage/=SystemOff).
Thepumpisturnedoffbyperformingtwolowerlevelactivitiesinsequentialorder:stoppinginfusion(aStopInfusion;explainedabove)andpressingthekeysnecessarytoturnoffthepump(aPress-KeysToTurnOffPump).
Thislatteractivityiscompletedbypressingtheon/offbutton(hPressOnOff)twiceinsequence.
Theentireprocessofturningoffthepumpcompleteswhentheinterfaceindicatesthatthepumpisoff(iInterfaceMes-sage=SystemOff).
6.
1.
4ProgrammingavalueintothepumpThevaluesforPCAdosagevolumeanddelaybetweendosagescanbeprogrammedintothepumpusinganEOFMpatternedafterFig.
6.
Thus,foragivenvalueX,thecorre-spondingEOFMbecomesrelevantwhentheinterfaceforsettingthatvalueisdisplayed(iInterfaceMessage=SetX).
Thisisachievedbysequentiallyexecutingtwosub-activities:changingthedisplayedvalue(aChangeXValue)andaccept-ingthedisplayedvalue(aAccept).
Theactivityforchangingthedisplayedvaluecanexecute,andwillrepeatedlyexecute,ifthedisplayedvalueisnotequaltotheprescribedvalue(iCurrentValue/=iPrescribedX).
Thevalueischangedbyexecutingoneormore(denotedbytheor_seqdecomposi-tionoperator)ofthefollowingsub-activities:changingthedigitcurrentlypointedtobythecursor(aChangeDigit:com-pletedbypressingtheupkey(hPressUp)),movingthecursortoadifferentdigit(aNextDigit:completedbypressingonlyoneof(thexordecompositionoperator)theleft(hPress-Left)orright(hPressRight)buttons),orsettingthedisplayedvaluetozero(aClearValue:completedbypressingtheclearbutton(hPressClear)).
Theprocessofchangingthedisplayed123Formallyverifyinghuman–automationinteraction227aTurnOffPumpiInterfaceMessage/=SystemOffiInterfaceMessage=SystemOffordaStopInfusingiInterfaceMessage=TreatmentAdministeringiInterfaceMessage/=TreatmentAdministeringordhPressStophPressStopaPressKeysToTurnOffPumpordhPressOnOffhPressOnOffFig.
5TheEOFMgraphicalrepresentationforturningoffthepumpvaluecompleteswhenthedisplayedvaluematchesthepre-scribedvalue(iCurrentValue=iPrescribedX).
Thedisplayedvalueisacceptedbypressingtheenterkey.
Theentirepro-cessendswhentheinterfaceisnolongerinthestateforacceptingX.
6.
1.
5StartingorreviewingaprescriptionAfteraprescriptionhasbeenprogrammedthehumanopera-torisgiventheoptiontostarttheadministrationofthatpre-scriptionortoreviewit(wheretheoperatorworksthroughtheprogrammingprocedureasecondtimewiththepreviouslyprogrammedoptionsdisplayedateachstep).
TheEOFMforperformingthis(Fig.
7)becomesrelevantatthispoint(iInter-faceMessage=StartBeginsRx).
Itiscompletedbyperform-ingonlyoneoftwoactivities:selectingtheoptiontostarttreatment(aStartRx—performedbypressingthestartbutton(hPressStart))orselectingthereviewoption(aReviewRx—performedbypressingtheenterbutton(hPressEnter)).
6.
2EOFMtranslationTheEOFMsrepresentingthehumantaskmodelweretranslatedintoaSALcodemodule.
ThistranslationwasaccomplishedbycreatingavariableforeachactivityoractionnodeineachEOFM,eachofwhichcouldassumeoneofthreeenumeratedvaluesdescribingitsexecutionstate:ready,exe-cuting,ordone.
Thus,inadditiontohandlingthetransitionallogicforthecoordinationprotocol,thismodulehandledthetransitionlogicforallowingthevariablesrepresentingactivityandactionnodestotransitionbetweenthesethreevalues.
Allactivityandactionvariablesstartinthereadystate.
Theycantransitionbetweenexecutionstatesbasedontheexecutionstateoftheirchildren,parent,andsiblingsintheEOFMstructure;theevaluationoftheirconditions;andtheirpositionwithintheEOFMhierarchy.
Whiletheresultinghumanoperatormoduleanditsassoci-atedunconstrainedoperatormodelbothhadthesameinputsandoutputs,thelogicassociatedwithtraversalofthehumantaskstructuresrequired48additionalvariablesinthehumantaskbehaviormodel.
6.
3ResultsThespecicationinEq.
(1)veried(producedtheexpectedcounterexample)in57swithasearchdepthof42.
Thespec-icationinEq.
(2)veriedtotruein10.
6hwithasearchdepthof437and1,534,862,538visitedstates.
7DiscussionThisworkhasshownthatitispossibleforhuman–automationinteractiontobeevaluatedusingthearchitectureinFig.
1.
However,thiscameasaresultoftradeoffsbetweenthegoalsthearchitectureisdesignedtosupport:123228M.
L.
Bolton,E.
J.
BassaSetXiInterfaceMessage=SetXiInterfaceMessage/=SetXordaChangeXValueiCurrentValue/=iPrescribedXiCurrentValue/=iPrescribedXiCurrentValue=iPrescribedXor_seqaChangeDigitordhPressUpaSelectNextDigitxorhPressLefthPressRightaClearValueordhPressClearaAcceptordhPressEnterFig.
6TheEOFMgraphicalrepresentationofthepatternforprogrammingavalueXintothepump1.
Modelconstructsneedtobeintuitivetohumanfactorsengineerswhowillbebuildingandevaluatingmanyofthemodels;2.
thesub-modelsshouldbedecoupledandmodular(asinFig.
1)inordertoallowforinterchangeabilityofalter-nativesub-models;and3.
theconstructedmodelsneedtobecapableofbeingver-iedinareasonableamountoftime.
Wediscusshoweachofthesegoalswasimpactedandhowrelatedissuesmightbeaddressed.
7.
1Goals1:modelintuitivenessManyofthemodelrevisionswereassociatedwithrepresent-ingmodelconstructsinwaysthatweremorereadilyinter-pretablebythemodelcheckerratherthanthehumanfactorsengineer.
Theseprimarilytooktheformofconvertingoat-ingpointanddigitarrayvaluesintointegersinPhase1b.
Further,theextensivemodelreductionsthatwereundertakeninPhase1cwouldbeverycumbersomeforahumanfactorsengineer.
Therearetwopotentialwaystoaddressthisissue.
Onesolutionwouldbetoimprovethemodelcheckersthemselves.
Giventhatthemodicationswouldnotactuallychangethenumberofreachablestatesinthesystem,thiswouldsuggestthatthemodelcheckerneedonlyoptimizetheBDDconver-sionalgorithms.
Alternatively,additionalmodelingtoolscouldbeusedtohelpmitigatethesituation.
SuchtoolscouldallowhumanfactorsengineerstoconstructorimportHDIprototypes,andtranslatethemintomodelcheckercode.
Thiswouldallowtheunintuitiverepresentationsnecessaryforensuringamodel'sefcientprocessingbythemodelcheckertoberemovedfromthemodeler'sview.
123Formallyverifyinghuman–automationinteraction229aStartOrReviewiInterfaceMessage=StartBeginsRxiInterfaceMessage/=StartBeginsRxxoraStartRxordhPressStartaReviewRxordhPressEnterFig.
7TheEOFMforchoosingtostartorreviewaprescription7.
2Goal2:decouplingofarchitecturesub-modelsBecausetheprotocolusedtocoordinatehumanactionsbetweentheHDIandhumantaskmodels(discussedforPhase1aandusedinallmodelsproducedinallsubsequentphases)assumesaparticularrelationshipbetweenvariablessharedbythesemodels,theyaretightlycoupled.
Unlessamodelcheckercanbemadetosupportbothasynchronousandsynchronousrelationshipsbetweenmodelsmoreelegantly,thiscoordinationinfrastructurecannotbeeliminated.
However,asolutionmaybefoundinanadditionallevelofabstraction.
AtoolsetfortranslatingaHDIprototypeintomodelcheckingcode,couldhandletheconstructionofthecoordinationprotocol,makingthisprocesseffectivelyinvis-ibletothemodeler.
SuchaprocesscouldalsoallowformoreefcientmeansofcoordinatingtheHDIandhumantaskmod-els:onethatmightnotrequiretheuseofseparatemodelsintheactualmodelcheckercode.
WhiletheextensivemodelreductionsfromPhase1greatlydiminishedthedelitywithwhichthemodelrepresentedtheactualPCApump,thisprovidessomeadvantages.
SincethemodelfromPhase2doesnotsufferfromthememoryusageproblemsencounteredinPhase1,thisopensthedoortotheadditionofothermodelconstructstobeaddedallowingforamorecompletesystemanalysis.
FutureworkcanexpandthemodeldevelopedinPhase2withenvironmentalanddeviceautomationmodelsthatarecompatiblewiththeformalPCApumpreferencemodeldescribedin[1].
7.
3Goal3:modelveriabilityWearepredominantlyconcernedwithexploringhowformalmethodscanbeusedtoprovideinsightsintohumanfactorsandsystemsengineeringconcerns.
Ifourgoalwastofor-mallyverifypropertiesoftheBaxterIpump,themodelingcompromiseswemadeinordertoobtainaveriablemodelmightnecessitateachangeinmodelingphilosophyorveri-cationapproach.
Therearemanybarrierstotheveriabilityofmodelsofrealisticsystems.
Theseincludelargenumbersofparallelprocesses,largerangesofdiscretevaluedvariables,andnon-discretelyvaluedvariables.
Themodelingeffortsdescribedhereweresochallengingbecausethetargetsystemwasdependentonalargenumberofuserspeciednumericalvalues,allofwhichhadverylargeacceptableranges.
Thisresultedinthescopeofthemodelbeingreducedtothepointwhereitcouldnolongerbeusedforverifyingalloftheorigi-nalhumanoperatortaskbehaviors:withthemodelproducedinPhase1bmakingminorcompromisesandthemodelpro-ducedinPhase1conlyallowingforbehaviorsassociatedwithbasicPCApumpfunctionality.
AswasdemonstratedinPhase2,theveriabilityofthemodelactuallyincreasedwiththeinclusionofthehumantaskbehaviorasindicatedbythe98%reductioninthereportedstatespacefromthePhase1ctothePhase2model.
However,thiscameattheexpenseoftheverica-tionprocesstaking284timesaslong.
Thus,inacontextwherevericationtimeislessimportantthanthesizeofthemodel'sstatespace,theinclusionofthehumantaskbehaviormodelmaygenerallyprovetobeadvantageousinthefor-malvericationofsystemsthathaveahuman–automationinteractioncomponent,irrespectiveofwhetherthehumanbehaviorisofconcerninthevericationprocess.
Futureeffortsshouldinvestigatethedifferentfactorsthataffectthistradeoff.
Evenexploitingthisadvantage,therelativesimplicityofthedevicethatwasmodeledinthisworkmakesitclearthattherearemanysystemsthatdependofhuman–auto-mationinteractionthatwouldbeevenmorechallengingtoverify,ifnotimpossible,usingthesetechniques.
Whiletheuseofboundedmodelcheckersmayprovidesomeverica-tioncapabilitiesforcertainsystems,thereislittlethatcanbedonewithouteitherusingadditionalabstractiontechniquesorfutureadvancesinmodelcheckingtechnologyandcom-putationpower.
Itiscommonpracticeintheformalmethodscommunitytousemoreadvancedformsofdataabstractionthanthoseemployedinthisworktomitigatethecomplexityofvari-ableswithlargevalueranges(anoverviewofthesemethodscanbefoundin[20]).
Becausethenatureofthemodeledhumantaskbehaviorinthisworkwasconcernedwiththedigitleveleditingofthedatavalues,suchabstractionswere123230M.
L.
Bolton,E.
J.
Bassnotappropriateforthisparticularendeavor.
Additionally,automaticpredicateabstractiontechniqueslikethoseusedincounterexample-guidedabstractionrenement[6]couldpotentiallyalleviatesomeofthecomplexityproblemsenco-unteredinthisworkwithoutrequiringchangestothemodelsthemselves.
Futureworkshouldinvestigatehowthesedif-ferentabstractiontechniquescouldbeusedwhenmodelingsystemsthatdependonhuman-automationinteractioninwaysthatareintuitivetohumanfactorsengineers.
Itisclearthatthemultiple,large-value-rangedvariableswerethesourceofmostofthemodelcomplexityproblemsinthepumpexample,asshowninthedrasticdecreaseinver-icationperformancetimebetweenthemodelsproducedinPhases1band1c.
Thus,hadthetargetsystembeenmorecon-cernedwithproceduralbehaviorsandlessontheinterrela-tionshipsbetweennumericalvalues,thesystemmodelwouldhavebeenmuchmoretractable.
Futureworkshouldidentifyadditionalpropertiesofsystemsdependentonhuman–auto-mationinteractionthatlendthemselvestobeingmodeledandveriedusingtheframeworkdiscussedhere.
Finally,someoftheperformanceissuesweencounteredcanbeattributedtoouruseofSAL.
Forexample,modelcheckerssuchasSPIN[15]donotperformthelengthypro-cessofconstructingtheBDDrepresentationbeforestartingthecheckingprocess.
Futureworkshouldinvestigatewhichmodelcheckerisbestsuitedforevaluatinghuman–automa-tioninteraction.
8ConclusionTheworkpresentedherehasshownthatitispossibletocon-structmodelsofhuman–automationinteractionaspartofalargersystemforuseinformalvericationprocesseswhileadheringtosomeofthearchitecturalgoalsinFig.
1.
Ithasalsoshownthattheincorporationofhumantaskbehaviormodelsintosystemmodelsmayhelpalleviatethestateexplosionprobleminsomesystemsthatdependonhuman–automationinteraction.
However,thissuccesswastheresultofanumberofcompromisesthatproducedamodelthatwasnotasrep-resentative,understandable,ormodularasdesired.
Thus,inorderforformalmethodstobecomemoreusefulfortheHFEcommunity,thevericationtechnologywillneedtobeabletosupportamorediversesetofsystems.
Further,newmodelingtoolsmayberequiredtosupportrepresentationsthathumanfactorsengineersuse.
Theseadvanceswillultimatelyallowformalmethodstobecomeamoreusefultoolforhumanfactorsengineersworkingwithsafetycriticalsystems.
AcknowledgmentsTheresearchdescribedwassupportedinpartbyGrantNumberT15LM009462fromtheNationalLibraryofMedicineandResearchGrantAgreementUVA-03-01,sub-award2623-VAfromtheNationalInstituteofAerospace(NIA).
Thecontentissolelytheresponsibilityoftheauthorsanddoesnotnecessarilyrepresenttheof-cialviewsoftheNIA,NASA,theNationalLibraryofMedicine,ortheNationalInstitutesofHealth.
TheauthorswouldliketothankRaduI.
SiminiceanuoftheNIAandBenDiVitooftheNASALangleyResearchCenterfortheirtechnicalhelp.
TheywouldliketothankDianeHaddon,JohnKnapp,PaulMerrel,KathrynMcGough,andSherryWoodoftheUniversityofVirginiaHealthSystemfordescribingthefunctionalityoftheBaxterIpumpandforprovidingdocumentation,trainingmaterials,anddeviceaccess.
OpenAccessThisarticleisdistributedunderthetermsoftheCreativeCommonsAttributionNoncommercialLicensewhichpermitsanynoncommercialuse,distribution,andreproductioninanymedium,providedtheoriginalauthor(s)andsourcearecredited.
References1.
ArneyD,JetleyR,JonesP,LeeI,SokolskyO(2007)Formalmeth-odsbaseddevelopmentofaPCAinfusionpumpreferencemodel:genericinfusionpump(GIP)project.
In:Proceedingsofthe2007jointworkshoponhighcondencemedicaldevices,software,andsystemsandmedicaldeviceplug-and-playinteroperability.
IEEEComputerSociety,Washington,DC,pp23–332.
BaxterHealthCareCorporation(1995)Ipumppainmanage-mentsystemoperator'smanual.
BaxterHeathCareCorporation,McGawPark3.
BoltonML,BassEJ(2009)Buildingaformalmodelofahuman-interactivesystem:insightsintotheintegrationofformalmethodsandhumanfactorsengineering.
In:ProceedingsoftherstNASAformalmethodssymposium.
NASAAmesResearchCenter,Moff-ettField,pp6–154.
BoltonML,BassEJ(2009)Enhancedoperatorfunctionmodel:agenerichumantaskbehaviormodelinglanguage.
In:Proceed-ingsoftheIEEEinternationalconferenceonsystems,man,andcybernetics.
IEEE,Piscataway,pp2983–29905.
BoltonML,BassEJ(2009)Amethodfortheformalvericationofhuman-interactivesystems.
In:Proceedingsofthe53rdannualmeetingofthehumanfactorsandergonomicssociety.
HumanFac-torsandErgonomicsSociety,SantaMonica,pp764–7686.
ClarkeE,GrumbergO,JhaS,LuY,VeithH(2003)Counterexam-ple-guidedabstractionrenementforsymbolicmodelchecking.
JACM50(5):752–7947.
CrowJ,JavauxD,RushbyJ(2000)Modelsandmechanizedmeth-odsthatintegratehumanfactorsintoautomationdesign.
In:Pro-ceedingsofthe2000internationalconferenceonhuman-computerinteractioninaeronautics.
AssociationfortheAdvancementofArticialIntelligence,MenloPark,pp163–1688.
CurzonP,RuksenasR,BlandfordA(2007)Anapproachtoformalvericationofhuman–computerinteraction.
FormalAspComput19(4):513–5509.
DeMouraL,OwreS,ShankarN(2003)TheSALlanguagemanual.
Technicalreport,ComputerScienceLaboratory,SRIInternational,MenloPark10.
DeganiA(1996)Modelinghuman–machinesystems:onmodes,error,andpatternsofinteraction.
PhDthesis,GeorgiaInstituteofTechnology,Atlanta11.
DeganiA,KirlikA(1995)Modesinhuman–automationinterac-tion:Initialobservationsaboutamodelingapproach.
In:Proceed-ingsoftheIEEEinternationalconferenceonsystems,manandcybernetics.
IEEE,Piscataway,pp3443–345012.
FieldsRE(2001)Analysisoferroneousactionsinthedesignofcriticalsystems.
PhDthesis,UniversityofYork,York13.
HeymannM,DeganiA(2007)Formalanalysisandautomaticgen-erationofuserinterfaces:approach,methodology,andanalgo-rithm.
HumFactors49(2):311–330123Formallyverifyinghuman–automationinteraction23114.
HeymannM,DeganiA,BarshiI(2007)Generatingproceduresandrecoverysequences:aformalapproach.
In:Proceedingsofthe14thinternationalsymposiumonaviationpsychology.
AssociationforAviationPsychology,Dayton,pp252–25715.
HolzmannGJ(2003)Thespinmodelchecker,primerandreferencemanual.
Addison-Wesley,Reading16.
JavauxD(2002)Amethodforpredictingerrorswheninteractingwithnitestatesystems.
Howimplicitlearningshapestheuser'sknowledgeofasystem.
ReliabEngSystSaf75(2):147–16517.
KirwanB,AinsworthLK(1992)Aguidetotaskanalysis.
TaylorandFrancis,Philidelphia18.
KohnLT,CorriganJ,DonaldsonMS(2000)Toerrishuman:build-ingasaferhealthsystem.
NationalAcademyPress,Washington19.
KreyN(2007)2007Nallreport:accidenttrendsandfactorsfor2006.
Technicalreport.
http://download.
aopa.
org/epilot/2007/07nall.
pdf20.
Mansouri-SamaniM,PasareanuCS,PenixJJ,MehlitzPC,O'MalleyO,VisserWC,BratGP,MarkosianLZ,PressburgerTT(2007)Programmodelchecking:apractitioner'sguide.
Techni-calreport,IntelligentSystemsDivision,NASAAmesResearchCenter,MoffettField21.
MitchellCM,MillerRA(1986)Adiscretecontrolmodelofoper-atorfunction:amethodologyforinformationdislaydesign.
IEEETransSystManCybernASystHum16(3):343–35722.
PerrowC(1984)Normalaccidents.
BasicBooks,NewYork23.
RushbyJ(2002)Usingmodelcheckingtohelpdiscovermodeconfusionsandotherautomationsurprises.
ReliabEngSystSaf75(2):167–17724.
SchraagenJM,ChipmanSF,ShalinVL(2000)Cognitivetaskanalysis.
LawrenceErlbaumAssociates,Mahwah25.
StantonN(2005)Humanfactorsmethods:apracticalguideforengineeringanddesign.
AshgatePublishing,Brookeld26.
ThurmanDA,ChappellAR,MitchellCM(1998)AnenhancedarchitectureforOFMspert:adomain-independentsystemforintentinferencing.
In:ProceedingsoftheIEEEinternationalconferenceonsystems,man,andcybernetics.
IEEE,Piscataway,pp3443–345027.
VicenteKJ(1999)Cognitiveworkanalysis:towardsafe,produc-tive,andhealthycomputer-basedwork.
LawrenceErlbaumAsso-ciates,Mahwah28.
WellsAT,RodriguesCC(2004)Commercialaviationsafety,4thedn.
McGraw-Hill,NewYork29.
WickensCD,LeeJ,LiuYD,Gordon-BeckerS(2003)Introductiontohumanfactorsengineering.
Prentice-Hall,UpperSaddleRiver123
- assoubuntu9.04相关文档
- resolveubuntu9.04
- manageubuntu9.04
- integratingubuntu9.04
- 模式ubuntu9.04
- rulesubuntu9.04
- GAubuntu9.04
RAKsmart便宜美国/日本/中国香港VPS主机 低至月$1.99 可安装Windows
RAKsmart 商家这几年还是在做事情的，虽然他们家顺带做的VPS主机并不是主营业务，毕竟当下的基础云服务器竞争过于激烈，他们家主营业务的独立服务器。包括在去年开始有新增多个数据中心独立服务器，包括有10G带宽的不限流量的独立服务器。当然，如果有需要便宜VPS主机的他们家也是有的，比如有最低月付1.99美元的美国VPS主机，而且可选安装Windows系统。这里商家有提供下面六款六月份的活动便宜V...
速云：广州移动/深圳移动/广东联通/香港HKT等VDS,9折优惠,最低月付9元;深圳独立服务器1050元/首月起
速云怎么样？速云，国人商家，提供广州移动、深圳移动、广州茂名联通、香港hkt等VDS和独立服务器。现在暑期限时特惠，力度大。广州移动/深圳移动/广东联通/香港HKT等9折优惠，最低月付9元；暑期特惠，带宽、流量翻倍，深港mplc免费试用！点击进入：速云官方网站地址速云优惠码：全场9折优惠码：summer速云优惠活动：活动期间,所有地区所有配置可享受9折优惠,深圳/广州地区流量计费VDS可选择流量翻...
提速啦(24元/月）河南BGP云服务器活动 买一年送一年4核 4G 5M
提速啦的来历提速啦是 网站 本着“良心 便宜 稳定”的初衷 为小白用户避免被坑 由赣州王成璟网络科技有限公司旗下赣州提速啦网络科技有限公司运营 投资1000万人民币 在美国Cera 香港CTG 香港Cera 国内 杭州 宿迁 浙江 赣州 南昌 大连 辽宁 扬州 等地区建立数据中心 正规持有IDC ISP CDN 云牌照 公司。公司购买产品支持3天内退款 超过3天步退款政策。提速啦的市场定位提速啦主...
ubuntu9.04为你推荐
-
摩拜超15分钟加钱怎么领取摩拜单车免费卷对对塔为什么不能玩天天擂台？（对对塔）同ip域名同IP网站具体是什么意思，能换独立的吗同一ip网站如何用不同的IP同时登陆一个网站百度指数词百度指数我创建的新词www.ijinshan.com桌面上多了一个IE图标,打开后就链接到009dh.com这个网站,这个图标怎么删掉啊?dadi.tvApple TV是干嘛的？怎么用？多少钱？www.k8k8.com谁能给我几个街污网站我去自己学www.zzzcn.com哪里有免费看书的网站m.yushuwu.com至尊影视网www.xuexiyu.com 怎么只收录首页啊