Content uploaded by Ruggiero Lovreglio
Author content
All content in this area was uploaded by Ruggiero Lovreglio on May 23, 2024
Content may be subject to copyright.
1 / 40
ComparingtheeffectivenessofARtrainingandslide‐basedtraining:
thecasestudyofmetroconstructionsafety
PeizhenGong1,YingLu1*,RuggieroLovreglio2;XingguangYan g 3;YunxuanDeng2
1SchoolofCivilEngineering,SoutheastUniversity,China,*email:luying_happy@126.com
2SchoolofBuiltEnvironment,MasseyUniversity,Auckland,NewZealand
3ShanghaiJiankeEngineeringConsultingCo.,Ltd.XuhuiDistrict,Shanghai,200032,China
Abstract:Metroconstructionposessignificantrisksandispronetoseveresafetyaccidents.
Safetytrainingiscriticaltomitigaterisksandreducethenumberofaccidents,asitallows
workerstoenhancetheirskillsinriskidentification,riskassessment,andriskresponseand
ultimatelyreduceaccidents.However,traditionaltrainingmethodsusedtoday,suchasslide
presentations,videos,ortext‐basedmaterials,provideapassivelearningenvironment,which
canlimittheireffectivenessinknowledgetransfer.Augmentedreality(AR)hasemergedasa
promisingtoolfortraining,allowingtraineestohaveanactivelearningenvironment.
Nevertheless,researchonassessingtheeffectivenessofARsafetytrainingwhilecomparingit
withtraditionalmethodsisstilllimited.Thisstudyaimstocomparetheshort‐termandlong‐
termeffectsonobjectiveperformancemeasuresandsubjectiveevaluationsofbothtraditional
andARtraining.ThisisdonebydevelopinganewARsafetytrainingformetroconstruction
andcomparingitwithanequivalentslide‐basedsafetytrainingwith72participantsdivided
intotwotraininggroups.Theselectedsafetycasestudycoverstheentiresafetyrisk
managementprocess,includingriskidentification,riskassessment,andriskresponse.Results
indicatethatARtrainingismoreeffectivethantraditionaltrainingintermsofshort‐term
knowledgeacquisitionofriskidentificationandlong‐termknowledgeretentionofrisk
identification,riskassessment,andriskresponse.Thesefindingscanhaveimplicationsforthe
designoffuturesafetytrainingguidelinesandrecommendations.
Keyword:AugmentedReality;SafetyTraining;Construction;Metro.
2 / 40
1.Introduction
Toaddresstheissueofheavysurfacetrafficinmetropolitanareas,thedevelopmentof
metroinfrastructurehasexperiencedrapidgrowthworldwideinrecentyears,particularlyin
developingcountries(Anwaretal.,2023;Fangetal.,2022;Zhangetal.,2023;Zhangetal.,
2022).However,comparedtotraditionalbuildingconstruction,metroconstructionposesa
significantlyhigherriskofserioussafetyaccidentsduetoitsextendedconstructionduration,
variablegeologicalconditions,numeroushazardousfactors,andfrequenthuman‐machine
interactions(Wangetal.,2021).Consequently,ensuringconstructionsafetyinmetroprojects
hasgarneredsignificantattentionfromregulators,industryprofessionals,andscholars.Major
accidentshaveunderscoredtheimportanceofconstructionsafetyinmetroprojects.For
instance,onFebruary18,2003,afireintheDaegumetroresultedinthetragiclossof192lives
and147injuries(Shietal.,2012).OnFebruary8,2018,aconstructionsafetyfailureoccurred
inFoshanmetroLine2betweenHuyongStationandLvdaohuStation,resultingin12fatalities,
8injuries,andanapproximated53.238milliondirecteconomiclosses(Dengetal.,2023).
Theseaccidentshavetriggeredwidespreadpubliccondemnationoftheinadequatesafety
performanceinmetroconstruction.
Pastconstructionexperienceshavedemonstratedthatreachingahighlevelofsafety
requiresthecollectiveeffortsofallstakeholders,includingowners,designers,construction
companies,workers,regulatoryagencies,andeducators(Baoetal.,2022;Jinetal.,2023;Liao
etal.,2015;MitropoulosandMemarian,2012;Wehbeetal.,2016;Xuetal.,2019).However,
thebehaviorofconstructionworkers,includingtheiracceptanceorrejectionofrisks,playsa
crucialroleinensuringtheirownsafety(Gohetal.,2018;Hasanzadehetal.,2018;Menget
al.,2021;Saracinoetal.,2015;Zhangetal.,2016;ZhangRitaetal.,2020).Infact,evidence
showsthatevenwhenconstructioncompaniestakeallnecessaryprecautionsconcerningsite
preparation,complianceenforcement,provisionoftraining,andpersonalprotective
equipment,workersmaystillengageinactivitiesthatputthematrisk(Choietal.,2017a,b;Li
etal.,2020;Liangetal.,2021;MengandChan,2020;Sacksetal.,2013;Shinetal.,2015).Their
abilitiestoidentify,assess,andrespondtoriskssignificantlyinfluencetheirbehaviorandsafety.
Numerousstudieshaveemphasizedthatworkerscangaintheseabilitiesusingeffectivesafety
training(Arifetal.,2021;ChoiandLee,2018;Kimetal.,2017;Panditetal.,2019;Rokooeiet
al.,2023;Wuetal.,2022;Yangetal.,2017;ZhouandGuo,2020).
Traditionaltrainingmethodsrelyonslidepresentations,videos,ortext‐basedmaterials
intheconstructionindustry(Burkeetal.,2006;Dalingetal.,2023;Perlmanetal.,2014;Sacks
etal.,2013;Wuetal.,2023b).Thesetraditionalapproachesexhibitseveraldisadvantagesthat
hindereffectiveknowledgeacquisitionandretention(ButtussiandChittaro,2021;Scorgieet
al.,2024).Thiscritiquestemsfromtheirtendencytofosterapassivelearningenvironment,
whereintraineesarerequiredtopassivelylisten,read,andmemorizeinstructions(Avveduto
etal.,2017;Perlmanetal.,2014;Sacksetal.,2013).Consequently,thelimitedinteractive
3 / 40
natureofthesemethodsresultsindiminishedlearnerengagementandreducedopportunities
foractiveparticipation(Cherrettetal.,2009;Gwynneetal.,2019).ARisarelativelynew
technologythatoverlaysvirtualobjects(i.e.,holograms)ontorealenvironments(Shinand
Jang,2009),anditholdspromiseforsafetytrainingpurposesindifferentfields,including
construction(Kamaletal.,2022;Lietal.,2018;ZhuandLi,2021).
ARoffersahighlyinteractiveandengaginglearningexperience,enablingtraineesto
undergosafetytrainingwithinavirtualenvironmentthataccuratelysimulateshazardous
situations.Traineescanacquireknowledgeofsafetyproceduresandoperatingtechniques,
andhavetheopportunitytopracticethem,thusenhancingtheirsafetyawarenessand
responsecapabilities(Chalhoubetal.,2021;Delgadoetal.,2020;Houetal.,2015;Lietal.,
2018;Shringietal.,2023).AlthoughARtraininghasdemonstratedpotentialeffectivenessin
theshortterm(HouandWang,2013;Murcia‐LopezandSteed,2018),thereislimitedresearch
availableonitslong‐termtrainingeffects.Therefore,thereisaneedfornewresearchto
evaluatetheshort‐andlong‐termimpactsofARsafetytrainingforconstruction,comparing
thisnewgenerationoftrainingwithtraditionaltrainingmethods.
Thepurposeofthisstudyistocomparetheshort‐termandlong‐termeffectsonobjective
performancemeasuresandsubjectiveevaluationsofbothtraditionalandARtraining.Thisis
achievedbydevelopinganewAR‐basedsafetytrainingprototypeformetroconstructionand
bycomparingitwithanequivalenttraditionaltrainingsolutionbasedonslides.TheAR‐based
prototypeemploys3Dmodelstovisualizehazardousscenariosassociatedwithmetro
construction,allowingforflexibletrainingonvarioustopicsinvolvingdangeroussituations.
Specifically,wefocusonsafetyknowledgeacquisitionandretentionasobjectiveperformance
measures,andontaskloadandsystemavailabilityassubjectiveevaluations.Assuch,this
studyisoneofthefirsttoconductacomprehensivecomparisonoftraditionalandARtraining
methodsinthecontextofsafetyriskmanagement,whichincludesriskidentification,risk
assessment,andriskresponse.Thiscomparativeassessmentcontributestoimprovingthe
understandingofAR‐basedsafetytrainingmethodsinthecontextofmetroconstruction,
providingvaluableinsightsforthedevelopmentofmoreeffectivesystemsandenhancing
workersafetyinthisfield.
Thestudyisstructuredasfollows:Section2providesaliteraturereviewoftheapplication
ofARinconstructionsafetytrainingandevaluationcriteriaforAR‐basedconstructionsafety
training.Section3presentsthematerialandmethods.Section4presentstheexperimental
resultsandincludesexploratoryanalyses.Section5discussestheresultsandoutlinesfuture
research.Finally,Section6concludesthestudy.
2.Literaturereview
2.1.ApplicationofARinconstructionsafetytraining
ARisaportabletechnologythatcreatesimmersiveandinteractivelearningexperiences
4 / 40
bysuperimposingvirtualobjectsontoactualenvironments(Alizadehsalehietal.,2020;
Babalolaetal.,2023;Houetal.,2021;Zhangetal.,2021).ByintegratingARintosafetytraining,
itispossibletoenhancetrainees'situationalawareness,hazardidentification,anddecision‐
makingskills,leadingtoimprovedsafetyperformanceduringtheconstructionphase.There
arethreemaincategoriesofwidelyusedARapplications(Schiavietal.,2022).
ThefirstcategoryisstationaryAR,wherevirtualcontentissuperimposedonsurveillance
videocapturedfromfixedcameras(Jiaoetal.,2013).ThistypeofARiscommonlyusedin
surveillanceprocesses.ThesecondcategoryiswearableAR,whichisachievedbyhead‐
mounteddisplays(HMDs)andARgoggles.Researchershavedevelopedvariouscomplex
functionsinARgogglesandHMDs(Grabowskietal.,2018).Forexample,Chenetal.(2021)
provideduserswithacompleteguidanceandvisualizationexperienceforperforming
proceduraltasksliketyingreinforcementbarsandformwork.Wuetal.(2022)createdanAR‐
basedalertsystemforreal‐timemeasurementsofthedistancebetweenaworkerandahazard,
ensuringworkersafetyduringtheconstructionprocess.WearableARdevicesoffersuperior
interactivefeatures,buttheirhighcosthaslimitedwidespreadadoption.Thethirdcategoryis
handheldAR,whichisalightweightapplicationthatcanbedeployedonmobilephones.
HandheldARiseasytoinstallandcanprovidecontinuoussafetyinformation,makingit
suitableforindividualsinsafetytrainingprograms(Chietal.,2022;Lietal.,2018).Researchers
havedevelopedAR‐basedapplicationstoenhancesafetytraining,demonstratingincreased
activelearningbehaviors,engagement,participantinterest,andimprovedlearningoutcomes
andexperiences(Kamaletal.,2021).Theseapplicationsfocusonspecificsafetyelementsand
investigateusers’intentionstoadoptthetechnologyinconstructionsafetytraining(Placencio‐
Hidalgoetal.,2022).ResearchhasdemonstratedthatARvisualizationtechniquesareeffective
intransferringknowledge,andtheinteractivenatureofARhelpstraineesrememberwhat
theyhavelearned(Shringietal.,2023).
DespiteadvancementsinARtechnologyforconstructionsafetytraining,severalresearch
gapspersist.Firstly,moreresearchisneededtoexaminethecost‐effectivenessofARsafety
trainingsolutions,particularlyconcerningwearableARapplications(Ahmed,2019;Eirisetal.,
2018;ElKassisetal.,2023).Thehighcostoftheseapplicationslimitstheirwidespread
adoption.Additionally,understandingpotentialbarriersandchallengestointegratingARinto
existingsafetytrainingpracticesiscriticalforsuccessfulimplementation.Thisnecessitatesthe
developmentofstandardizedguidelinesandbestpracticesfordesigninganddevelopingAR
applicationstailoredtoconstructionsafetytraining(Houwelingetal.,2024;Lietal.,2018).
BridgingthesegapswillfacilitatethesuccessfulimplementationofARtechnologyin
constructionsafetytrainingandenhancesafetypracticesintheindustry.
2.2.EvaluationcriteriaforAR‐basedconstructionsafetytraining
EvaluatingAR‐basedconstructionsafetytrainingcomprehensivelyrequiresthe
establishmentofappropriateevaluationcriteriathatincludebothobjectiveperformance
5 / 40
measuresandsubjectiveevaluations.Objectiveperformancemeasures,suchasknowledge
acquisition(Perlmanetal.,2014;Sacksetal.,2013),arecommonlyemployedtoassessthe
short‐termimpactoftrainingontraineeperformance.Knowledgeacquisitionreferstothe
trainee’sabilitytoswiftlyacquirenewinformationandskillsrelevanttosafetytraining.Many
scholarsutilizeknowledgeacquisitionasanevaluativeindexforassessingtheeffectivenessof
ARtraining.Numerousstudieshavedemonstratedapositivecorrelationbetweenhigher
immersioninARtrainingandimprovedknowledgeacquisition(Dallasegaetal.,2023;Leetal.,
2015;Veraetal.,2018).Leetal.(2015)proposedasystemforexperientialconstructionsafety
trainingusingAR,whichconsistedofsafetyknowledgedissemination(SKD),safetyknowledge
reflection(SKR),andsafetyknowledgeassessment(SKA).Theyemployedknowledge
acquisitiontoevaluatetheeffectivenessofthesystemandfoundthatAReffectivelyenhances
safety.Alshiaretal.(2019)comparedAR‐basedtrainingwithtraditionalpenandpaper‐based
trainingusingknowledgeacquisitionundersimulatedconstructionscenarios,andtheresults
indicatedapreferenceforAR‐basedassessmentamongparticipantsovertraditional
assessment.However,itisworthnotingthatsomestudieshavepresentedcontradictory
results.Liuetal.(2022)discoveredthatforsimplemaintenancetasks,traditionaltraining
methodsoutperformedARtraining.However,asthecomplexityofmaintenancetasks
increased,ARtrainingexhibitedgreaterefficiency,surpassingtraditionaltrainingbyover10%.
Moreover,researchonthelong‐termeffectsofARtrainingremainslimited,specificallyin
termsofknowledgeretention,whichreferstotheabilitytorecallandapplykeysafetytraining
pointsoverdaysorweeksinadynamicconstructionenvironment(Doolanietal.,2020).In
assessingtheeffectivenessofsafetytrainingmethods,investigatingthelong‐termeffectsof
safetytrainingprovesvaluableinidentifyingthedurabilityofknowledgeretentionovertime,
enablingamorenuancedexaminationofknowledgeretentionacrossextendedperiods(Huet
al.,2023;Karakostaetal.,2023;Shringietal.,2023;Somerkoskietal.,2022).Forinstance,
Dalingetal.(2023)foundnosignificantdifferencesintheshort‐andlong‐termeffectsofAR
trainingandtraditionaltraining;however,allparticipantsexperiencedasubstantialyet
comparabledeclineinperformance.Paesetal.(2024)comparedanAR‐basedsafetytraining
methodwithavideo‐basedtrainingmethod.TheresultsdemonstratethattheARsystemis
equallyeffectiveastraditionalmethodsintermsofknowledgeacquisitionandretention.
Subjectiveevaluationsoftheuserexperiencearevitalindeterminingtheactual
utilizationandacceptanceofatrainingmethod,inadditiontoobjectiveperformance
measures(Jangetal.,2021;Liuetal.,2023;Xiangetal.,2023).However,currentARtraining
researchhasgivenrelativelylessattentiontosubjectiveevaluations(Hoedtetal.,2017).Tas k
loadandsystemusabilityarecommonlyusedsubjectiveevaluationindexes(Al‐Ahmarietal.,
2018;Houetal.,2015;Koumaditisetal.,2019).Tas k loadreferstothecognitiveandphysical
demandsplacedontraineesduringAR‐basedsafetytraining,encompassingfactorssuchas
mentalload,attentiondemands,andtaskandscenariocomplexity(Leetal.,2015;Liuetal.,
2022;Moesletal.,2023).Assessingtaskloadcanassistinoptimizingtrainingcontentand
6 / 40
deliverytoensurethattraineeseffectivelyprocessandretaininformation.Systemusability
relatestotheaccessibility,reliability,andfunctionalityoftheARtrainingsystemduring
training(Dyrdaetal.,2023;Kamaletal.,2021;Surietal.,2023).Reliableandusablesystems
ensurethattraineescanfullycomprehendtrainingmaterialswhileminimizingpotential
distractionsthatcouldhinderknowledgeacquisition.Severalstudieshavedemonstratedthat
theimplementationofARtrainingnotonlyenhancesobjectiveperformancebutalsohasa
positiveimpactonsubjectiveevaluations,indicatingimproveduserexperienceand
acceptanceofthetrainingmethod(Houetal.,2015;Koumaditisetal.,2019).
2.3.Summary
ThisreviewshowsthatthereisaneedfornewcomprehensiveassessmentsonhowAR
cansupporteffectivesafetytraining.Thereisalackofinvestigationon(a)theeffectivenessof
ARsafetytrainingcombiningobjectiveperformancemeasuresandsubjectiveevaluations;(b)
thecontributionofARtoenhancesafetyriskmanagement,includingriskidentification,risk
assessment,andriskresponseinconstruction.
Thisstudyaimstoaddressthesetworesearchgapsbyexaminingthepotentialbenefits
andeffectivenessofARinsafetytrainingformetroconstruction,aswellasitsroleinthe
broadersafetyriskmanagementprocess.To accomplishtheseobjectives,thestudyposestwo
researchquestions:
QuestionOne(Q1):AretheredifferencesbetweentraditionalandARtrainingintermsof
short‐termandlong‐termobjectiveperformancemeasures,specificallysafetyknowledge
acquisitionandretention?Additionally,doestraditionaltrainingdifferfromARtrainingin
enhancingskillsrelatedtoriskidentification,riskassessment,andriskresponse?
QuestionTwo(Q2):AretheredifferencesbetweentraditionalandARtrainingintermsof
subjectiveevaluations,specificallytaskloadandsystemusability?
3.MaterialandMethods
ThisworkaimsatprototypinganewARsafetytrainingsolutionandcomparingitwith
traditionaltrainingsolutions.Todeveloptheprototype,weidentifyhigh‐riskhazardous
scenariosthroughaccidentreportsandon‐sitesurveys,asreportedinSection3.1.
Subsequently,aprototypeofanAR‐basedmetroconstructionsafetytrainingisdeveloped
andtested.Theobjectiveofthistrainingistoprovideguidancetoconstructionworkers
regardingsafeoperationsduringmetroconstruction,encompassinggeneralsitepractices,
hazard‐specificsafetypractices,andconstructionsafetyprecautions.
Followingthedevelopmentphase,testingisconductedthroughacontrolledbetween‐
subjectsexperiment.The72participantsaredividedintotwogroups,eachassignedtoa
differenttrainingcondition:slide‐basedorAR‐based.Thepurposeofthisexperimentisto
comparetheimpactofthetraditionalslide‐basedtrainingmethodwiththeAR‐basedtraining
7 / 40
methodonparticipantknowledge,taskload,andsystemusability.
3.1.Selectionofhazardscenarios
Theprocessofselectinghazardscenariosinvolvestwoparts:analyzingstatisticalaccident
reportstoidentifyhigh‐riskaccidenttypes,andconductingon‐siteresearchtodetermine
specifichazardscenariosassociatedwiththesehigh‐riskaccidenttypes.Sincethisstudyis
carriedoutinChina,wefocusontheanalysisofaccidentreportsinthiscountryonlyfocusing
onmetroconstructionsites,whichareamongthemostdangerousintheconstructionsector.
Thecollectionofmetroconstructionaccidentreportsforthisstudyencompassestwosources.
Thefirstsourceconsistsofaccidentreportsissuedbynationalorlocalgovernmentsandnon‐
governmentalorganizations(Shaoetal.,2019;Zhouetal.,2021).Thesecondsourceisthe
Internet(Rissolaetal.,2022),wheremetroconstructionaccidentreportscanbeaccessed
fromavarietyofwebsites,includingmediawebsitessuchasTence n t andXinhuaNet,aswell
asinformationwebsiteslikeWikipediaandBaiduEncyclopedia.Atotalof207accidentreports
relatedtometroconstructioninChinabetween2016and2023havebeencollected.Ta b l e 1
presentsasummaryofthenumberofdeathsandnon‐fatalinjuriesforeachaccidenttype.
Statisticalanalysesindicatethatthefollowingsixaccidenttypesposearelativelyhighrisk:
collapse,fall‐from‐height,vehicleinjury,objectstrike,crane‐relatedaccident,andmechanical
injury.ThesefindingsareconsistentwiththeassessmentconductedbyQietal.(2023).
Onthisbasis,specifichazardscenarioscorrespondingtotheaforementionedhigh‐risk
accidenttypesareidentifiedthroughon‐siteresearchconductedatrepresentativemetro
constructionprojects.Thesehazardscenarios,detailedinTable2,willbeutilizedforboth
traditionaltrainingandARtraininginthecontextofmetroconstruction.
Onthisbasis,specifichazardscenarioscorrespondingtotheaforementionedhigh‐risk
accidenttypesareidentifiedthroughon‐siteresearchconductedatrepresentativemetro
constructionprojects.Theon‐siteresearchencompassedathoroughinvestigationofametro
constructionprojectlocatedinSuzhou,China,acityrenownedforitsrobusteconomyand
rankingamongthetop10.Thisparticularmetroprojectspansover40kilometers,comprising
atotalof28stations,andentailsaninvestmentexceeding20billionyuan.Itconstitutesa
significantendeavorintherealmofregionalinfrastructureconstruction.Ourinteractionswith
keystakeholdersattheconstructionsite,includingprojectmanagers,technicalleaders,and
on‐siteworkers,aimedtoextractcrucialinformationandinsightsfromtheirfirsthand
experiences.Theseinvaluablediscussionsfacilitatedtheidentificationanddocumentationof
precisedisasterscenariosassociatedwithhigh‐riskaccidenttypes,subsequentlysummarized
inTab l e 2.ThistablewillbeutilizedforbothtraditionaltrainingandARtraininginthecontext
ofmetroconstruction.Furthermore,wecapturedphotographsofcertainhazardscenarioson‐
site.Itisimportanttonotethatincaseswherespecificscenarioswereinaccessibleon‐site,
alternativemeasureswereemployed,suchasobtainingrelevantimagesfromreputableonline
sources.Thisapproachensuredtheinclusionofacomprehensivearrayofhazardscenariosin
8 / 40
ouranalysis,therebyupholdingthevalidityandintegrityofourstudy.
Tab l e 1‐Statisticsoninjuriesinmetroconstructionaccidentsbytype.
AccidenttypeNumberofdeathsNumberofinjuriesTot a l
Vehicleinjury151429
Electricshock718
Fall‐from‐height35540
Fire617
Mechanicalinjury17118
Crane‐relatedaccident15823
Collapse603191
Objectstrike26329
Poisoning437
Undergroundwaterdamage000
Cablebreakage000
Pipelinerupture000
Otherdamage11011
Otheraccidents404
Asphyxiation101
Struck‐by101
Tab l e 2‐High‐riskaccidenttypesandhazardscenariosidentifiedinarepresentativemetro
project.
No.High‐riskaccident
types
Hazardscenarios
1Vehicle injuryRolloverwhiletransportingheavyloads.
2Constructionworkersindriver’sblindspot.
3Fall‐from‐heightHoleswithoutwarningsignsorprotection.
4Unprotectedouteredgesofupperfloors.
5Workingatheightwithoutproperuseofpersonalsafety
equipment.
6Vent uring intodangerousplacesbyclimbingoversafetyguardrails.
7Fixedscaffoldingwithoutadequatefallprotection.
8ObjectstrikeStrikingagainstfixedorstationaryobjects.
9Improperplacementofmaterialsandequipmentonsiteor
configurationofoperatinglines.
10Crane‐related
accident
Movingcraneswithloadswhereworkersarepresent.
9 / 40
11CollapseImproperlysupportedwallformwork.
12Failuretousepersonalsafetyequipmentproperly.
13Failuretofollowprescribedstepsinconstructionwork.
14MechanicalinjuryFailuretoobservetheconstructionsiteandsurroundingscarefully.
15Failuretousepersonalsafetyequipmentproperly.
3.2.Trainingdesign
BothtraditionalandARtrainingincludethreemainparts,eachwithahalf‐hourtraining
session:
(1)Generalsitespecifications:Thissectioncoverssafetyinstructionsforworkersupon
enteringtheconstructionsite,constructiondresscodes,constructionelectricitycodes,
commonvehicleandmachineryoperationcodes,constructionmaterialstackingcodes,and
work‐at‐heightprotectioncodes.
(2)Hazard‐specificsafetypractices:Thissectionfocusesonspecificsafetypractices
relatedtohazardssuchasvehicleinjury,fall‐from‐height,objectstrike,crane‐relatedaccident,
collapse,andmechanicalinjury.Safetypracticesandproceduresspecifictoeachhazardare
explainedindetail.
(3)Constructionsafetyprotection:Thissectionemphasizestheimportanceofusing
safetyequipment,includingsafetybelts,safetyhelmets,andsafetyshoes.
Itshouldbenotedthatinthesectionsongeneralsitespecificationsandconstruction
safetyprotection,thesametextcontentandpresentationslidesareusedforbothtraditional
andARtraining.Theseslidescontaininformationcompiledfromsafetycodes.However,inthe
hazard‐specificsafetypracticessection,therearedifferencesbetweenthetwotrainings.The
objectiveofthetrainingistoenhancethetrainees’abilitiesinriskidentification,risk
assessment,andriskresponse.Inthetraditionaltraining,thepresentationslidescreatedfor
thehazard‐specificsafetypracticessectionareenrichedwith15setsofpicturesandvideos
fromconstructionsitesandtheInternet,correspondingtoeachofthe15hazardscenariosin
Table2.Thesevisualmaterialswererequiredtobehighdefinitionandfeatureeye‐catching
hazardousbehaviors.Thevisualmaterialsaimtoenhancetheunderstandingandengagement
oftrainees.Traineesreceivepassivelearningthroughtrainerpresentations,involvingactivities
suchaslistening,reading,andmemorizingsafetypracticepoints.
InARtraining,thetextcontentandpresentationslidesremainthesameasintraditional
training.However,thevisualmaterialsarereplacedby15ARhazardscenarios,aligningwith
Table2.ThisstudyadoptstheAR‐QRcodemethodforpresentinghazardscenarios.TheAR‐
QRcodemethodcombinesARandQuickResponse(QR)codestoeffectivelydeliverhazard
scenarioinformation.ThisapproachwasinitiallyproposedbyForoughiSabzevaretal.(2023)
withtheobjectiveofenhancingtheacquisitionofinformationrelatedtobuildingdesignand
constructionprocesses.ThisstudyadaptstheAR‐QRcodemethodforsafetytraininginmetro
10 / 40
construction.
Topresentthehazardousscenariosduringthesafetytraining,thesescenariosareinitially
3DmodeledusingSketchup.Subsequently,thescenesareexportedto.daeformatand
integratedintotheselectedARplatform.ThechosenARplatformisAUGMENT,aversatile
platformthatsupportsvariousapplications,includinge‐commerce,marketing,andfieldsales
(AUGMENT,2023).Itshowsgreatpotentialasatooltosupportsafetytrainingcontent.The
platformautomaticallygeneratesQRcodesthattraineescanscanusingtheirmobilephones
ortabletstoaccessstoredARhazardscenariossuperimposedontherealenvironment.It
shouldbenotedthattheseARhazardscenariosonlyillustratevariouspotentiallyunsafe
situationsanddonotdepicttheconsequencesofaccidents.ARhazardscenesarecreatedfor
eachofthehazardscenariosidentifiedinSection3.1.Fig.1depictstheprototypeofthe
developedARsystem.Fig.2showsexamplesoftraineesreceivingtraditionalandARtraining.
InARtraining,traineescanplacetheARhazardscenariosonanyflatsurface,suchasthe
groundordesktop.Whilethetrainerusesslidestoteach,theycanfreelyinteractwiththeAR
hazardscenariosbyrotating,zoominginorout,exploringfromdifferentperspectivesand
learningactively.Thisisdifferentfromtraditionaltrainingwhereonlypicturesandvideoscan
beviewed.
Fig.1.PrototypeofthedevelopedARsystem.
11 / 40
Fig.2.Examplesoftraineesreceivingtraditional(left)andARtraining(right):a)
construction
workersindriver’sblindspot;b)workingatheightwithoutproperuseofpersonalsafety
equipment;c)movingcraneswithloadswhereworkersarepresent;d)failuretouse
personalsafetyequipmentproperly.
3.3.Experimentaldesign
Thestudyemploysabetween‐subjectsexperimentaldesigntocomparetheeffectiveness
oftwotrainingmethods:traditionaltrainingandARtraining.Thiswasdonebyrandomly
assigningparticipantsinourstudytotwogroups:TraditionalGroupandARGroup.This
methodisalignedwithmanypreviousstudiesfocusingoncomparingtheeffectivenessof
differentsafetytrainingsolutions(Scorgieetal.,2024).
Theindependentvariableinthestudyisthemodeoftraining,resultingintwo
experimentalconditions.Inthetraditionaltraining,participantsreceivedinstructionfrom
trainersandviewedslidesonaconventionaldisplay.Theseslidesaresupplementedwith
photosandvideosfrommetroconstructionsites,followingamethodcommonlyusedinprior
trainingstudies
(Dalingetal.,2023;Sacksetal.,2013;Wuetal.,2023a).Incontrast,intheAR
training,participantsalsoreceivedguidancefromatrainerandviewedslidesona
conventionalmonitor.However,insteadofviewingphotosandvideos,theyutilizeARto
visualizehazardousscenarios.Bothtrainingsareconductedinagroupsetting.
Thedependentvariableinthisstudyistheparticipants’learningperformance,
specificallyknowledgeacquisitionandretentioninthreeareas:riskidentification,risk
assessment,andriskresponse.TheknowledgedataofthetraditionalgroupandARgroupis
collectedatthreedifferentstages:pre‐training,immediatelyaftertraining,andfourweeks
12 / 40
aftertraining.
Toassessknowledgeacquisition,wecomparetheparticipants’pre‐trainingand
immediatepost‐trainingknowledgelevels.Knowledgeretentionisevaluatedbycomparing
knowledgelevelsimmediatelyaftertrainingwiththosemeasuredfourweekslater.Thechoice
ofafour‐weekintervalalignswithindustrypracticesforassessinglong‐termretentionof
training.Similartimeframeshavebeenusedinpreviousstudiesontrainingretention
(Lovreglioetal.,2021;Sacksetal.,2013).
Fig.3illustratestheexperimentprocess.
Theexperimentconsistsoffoursteps.InStep1,demographicdataandpriorexperience
arecollectedfromparticipants,aswellasobjectiveperformanceassessments(risk
identification,riskassessment,andriskresponseknowledge).Step2involvedproviding
traditionalandARtrainingtobothgroups.Immediatelyaftertraining,inStep3,bothgroups
undergosubjectiveassessmentsregardingtaskloadandsystemusability,aswellasobjective
performanceassessments.Finally,inStep4,objectiveperformanceassessmentsare
administeredtobothgroupsofparticipantsfourweeksafterthetraining.Thedecisionto
measuretaskloadandsystemavailabilityonlyinStep3isbasedonthefactthat,immediately
afterthetraining,participantshaveaclearperceptionofthesubjectiveassessmentsofthe
trainingmethod.Ifthesemeasurementswererepeatedatthefour‐weekmark,participants’
perceptionofthetrainingmethodmayhavediminishedorevenbeenforgotten,whichwould
hinderamoredefinitiveassessmentofthetaskloadandsystemusabilityassociatedwiththe
twotrainingmethods.
Fig.3.Experimentprocess.
3.4.DataCollectionInstruments
Thedatacollectioninstrumentcomprisedthreequestionnairesadministeredatdifferent
periodsoftheexperiment:pre‐training,immediatelyaftertraining,andfourweeksafter
training.Eachquestionnaireisdesignedtogatherspecificinformationandresponsesfromthe
participants.Thecategoriescoveredinthiseffortincludethefollowing:1)demographics,2)
13 / 40
priorexperience,3)knowledge,4)taskload,and5)systemusability.Thepre‐training
questionnaireincludescategories1and2.Thequestionnaireimmediatelyaftertraining
encompassescategories3to5.Thequestionnairefourweeksaftertrainingfocuseson
category3.Thedetailsofeachcategoryaredescribedbelow:
1)Thedemographicssectionconsistsoftwoitemstocollectparticipants’characteristics,
namelygenderandeducationlevel.
2)Thepreviousexperiencesectionincludestwoitemstogatherinformationabout
participants’workexperienceinmetroconstructionandtheirexperiencereceivingsafety
training.Thisisdonetoensurethatbothgroupshavesimilarpriorexperiencesandtoavoid
anybiasesrelatedtoexperience.
3)Theknowledgesectionencompassesthreeitems,eachcontaining10questionsto
assessparticipants’abilitiesinriskidentification,riskassessment,andriskresponse,
respectively.Specifically:
RiskIdentification:Participantsareprovidedwith10riskscenepicturesandasked
toidentifyvariouspotentialhazardsbymarkingthecorrespondingnumbersinTab l e
2.Eachcorrectidentificationearnsonepoint,andthetotalpossiblescoreforthis
taskis10points.
RiskAssessment:Participantsarepresentedwiththesame10riskscenariopictures
andrequiredtoevaluatetherisklevelofeachscenario,categorizingitasverylow,
low,medium,high,orveryhigh.Toassessthecorrectrisklevelsforthesescenarios,
agroupofmetroconstructionsafetyexpertsisaskedtoestablishnormativevalues,
followingthecommonlyusedmethod(Sacksetal.,2013).Subsequently,safety
managersarealsoinvitedtoevaluatetheestablishednormativevaluesbasedon
actualconsequences,andtheoutcomesalignedconsistentlywithreal‐worldrisk
scenarios.Participantsreceiveonepointforeachcorrectassessment,resultingina
totalscoreof10pointsforthistask.
RiskResponse:Participantsaregiventhesame10riskscenariopicturesandasked
tomakebehavioraljudgmentsbyselectingthemostappropriateresponseoption
fromtheprovidedchoices.Theresponseoptionsinclude:(a)continueworkingas
normal;(b)notifythesafetysupervisorandcontinueworkingasnormal;(c)handle
thehazardoussituationindependently;or(d)refusetoworkinthespecificsituation.
Eachcorrectresponseearnsonepoint,andthetotalpossiblescoreforthistaskis
10points.
4)ThetaskloadsectionincludessixitemsderivedfromtheNASATa s k LoadIndex(TLX),
aself‐reporttoolthatmeasuresparticipants’perceivedtaskloadacrosssixaspects:mental
workload,physicalworkload,temporalworkload,trainingeffort,trainingdifficulty,and
stressed/annoyed(Hart,2006;HartandStaveland,1988).Eachdimensionisrepresentedbya
horizontallinewithendpointslabeled“low”(=1)and“high”(=21).Participantsindicatetheir
perceivedtaskloadoneachdimensionbymarkingapoint.
14 / 40
5)Thesystemusabilitysectioncomprisestenitemstoassessparticipants’perceived
abilitytoreceivetrainingandtheirevaluationoftheusabilityofthetrainingmethod.This
sectionemploysawidelyadoptedandvalidatedtoolforassessingtheusabilityofvarious
systems,includingtrainingprograms(Gaoetal.,2020;JohnandBeaconsfield,1996).Afive‐
pointLikertscaleisemployed,with“stronglydisagree”(=1)and“stronglyagree”(=5)asthe
options.Participantsratestatementsrelatedtosystemusability,indicatingwhetherornot
theyagreewiththem.
3.5.ParticipantsandExperimentSession
Thestudycomprisedasampleof72participantswhometthequalificationcriteriaof
beingatleast18yearsoldandpossessingatleastahighschooleducation.Todeterminethe
requiredsamplesize,theSuccess‐Runtheorem(Durivage,2016;Wuetal.,2023a)wasutilized:
𝑛𝑙𝑛 1𝐶
𝑙𝑛 𝑅 (1)
where𝑛 denotesthesamplesize,𝐶 denotestheconfidencelevel,and𝑅 denotesthe
reliability.Thisstudyemployscommonlyusedstatisticalthresholds,namely,95%confidence
and90%reliability(Wuetal.,2023a).Theminimumrequiredsamplesizewascalculatedtobe
29participants,whichalignswiththeseminalworkofCohen(1992).Inthisstudy,thesample
sizewastwicetherequiredamount,consistentwithsimilarstudiesinthefieldthatinvolved
twoexperimentalconditions.Forinstance,Wuetal.(2023a)conductedtworoundsof
experiments,eachwithasamplesizeof16participants.Sacksetal.(2013)conductedtwo
experiments,onewith20participantsandtheotherwith25participants.Seeligeretal.(2023)
recruited35participants.Yipetal.(2019)recruited46participants.Jeelanietal.(2020)
recruited53participants.Abbasetal.(2023)andPerlmanetal.(2014)hadlargersamplesizes
of60and62participants,respectively.
Participantconsentandethicalapprovalwereobtainedpriortothetestpreparation
phase.Participantswereexplicitlyinstructedtorefrainfromconsumingalcoholontheday
precedingtheexperiment.Atotalof72participantswererandomlyassignedtooneoftwo
experimentalconditions,with36participantsineachgroup.
Duringtheformaltestphase,bothgroupsreceivedtheirtraininginasoundproofed
conferenceroom,wherethebrightnesswasadjustedto300to500luxandthetemperature
wasadjustedto20to22°Cinordertoprovideanequallysuitableenvironmentfortraining.
Overall,theexperimentalconditionswereidenticalandconducivetotrainingforbothgroups.
Participantswereaskedtorefrainfromcommunicatingwitheachotherthroughoutthe
training,andgrouptrainingdidnotaffecttheresults.
Inthedatacollectionandprocessingstage,thetestswereadministeredintheformof
questionnaires,asdescribedinSection3.4.
15 / 40
4.Results
Thedataanalysisaimstoexaminewhethertherearesignificantdifferencesbetweenthe
twogroupsofexperimentsinthreeindicators:a)knowledgeinpre‐training,immediatelyafter
training,andfourweeksaftertraining(Section4.2);b)taskload(Section4.3);andc)system
usability(Section4.4).Throughoutthestudy,thenumberofparticipantsremainedconstant
duringallthreedatacollectionphases,includingpre‐training,immediatelyaftertraining,and
fourweeksaftertraining.Therewerenodropouts,andbothgroupsconsistedof36
participantsacrossthethreedatacollectionphases.Itisworthnotingthatnoparticipants
reportedexperiencingmotionsicknessoranydiscomfortwhileusingtheARsystemprototype.
Furthermore,participantswereinstructednottoengageinanyotherrelatedsafetytraining
withinfourweeksaftertraining.
4.1ParticipantDemographics
Thestudyrecruitedatotalof72participants,consistingof21metroconstructionworkers,
19metroconstructionsafetymanagers,and32civilengineeringstudents.Toassessthe
significanceofthedifferencebetweentheARgroupandtheslidegroup,thechi‐squaretest
wasconducted.Theparticipantcharacteristicsandtheresultsofthesignificancetestare
summarizedinTable3.Thechi‐squaretestconfirmedthattherewasnosignificantdifference
(p>0.05)foundbetweenthetwogroupswithrespecttogender,educationlevel,previous
workexperienceinmetroconstruction,andpreviousmetroconstructionsafetytraining
experience.Thisindicatesthatthedistributionoftheseparameterswasrelativelybalanced
acrossthetwogroups.
Tab l e 3‐ParticipantDemographics.
Parameter
Slidetraininggroup(n=36)ARtraininggroup(n=36)
Significance
test
Construction
workers
(n=11)
Safety
directors
(n=9)
Students
(n=16)
Construction
workers
(n=10)
Safety
directors
(n=10)
Students
(n=16)
Gender
𝜒
0,
p=1.000
Woman336435
Man86106711
Educationlevel
𝜒
0.237,
p=0.971
Highschooldegree900800
Undergraduatedegree2692710
Masterdegree035034
Doctoraldegree002002
Previousworkexperienceinmetroconstruction
𝜒
0.090,
16 / 40
Meannumberofyearsatwork51105.210.90p=0.956
Numberofparticipants
injuredatwork
4(36.4%)1(11.1%)0(0%)3(30%)1(10%)0(0%)
Meannumberofwork‐related
accidentswitnessed
11.900.71.80
Previousmetroconstructionsafetytrainingexperience
𝜒
0.001,
p=0.972
Numberofparticipantswho
receivedformalsafetytraining
8(72.7%)9(100%)0(0%)7(70%)10(100%)0(0%)
Meannumberofprevious
safetytraininghours
67006.2690
Theresultsrevealthatconstructionworkersmostlyhavehighschooldegrees,while
safetydirectorsandcivilengineeringstudentspossessundergraduatedegreesorhigher.None
ofthestudentshavepriorexperienceinmetroconstructionorsafetytraining.Safetydirectors
exhibitahighermeannumberofyearsatworkcomparedtoconstructionworkers.Specifically,
theentirepopulationofsafetydirectorshasanaverageof10.9yearsatwork,whereas
constructionworkershaveanaverageof5.1years.Allsafetydirectorshaveundergoneformal
safetytraining,whileonly71.4%ofconstructionworkershavereceivedthesametraining.
Safetydirectorsreceivesafetytrainingforanaverageof11.4timeslongerthanconstruction
workers,with69.5hoursand6.1hours,respectively.Moreover,twosafetydirectorsandseven
constructionworkersexperiencedinjuriesinpreviousconstructionoperations.
4.2AnalysisofKnowledge
Inthissection,wepresenttheknowledgescoresforriskidentificationobtainedby
participantsintheslidegroupandARgroupatdifferenttimepoints:pre‐training,immediately
aftertraining,andfourweeksaftertraining(Section4.2.1).Thescoresforriskassessmentare
reportedinSection4.2.2.Finally,scoresregardingriskresponsesarepresentedinSection4.2.3.
Thefiguresrepresentingtheknowledgescoresforeachsectionuserectanglestoindicatethe
meanscore,withdotsrepresentingindividualdatapoints.Incaseswheredifferent
participantsreceivedthesamescore,thedotsoverlap.
Toanalyzethesignificantstatisticaldifferencesbetweenthetwoexperimentalgroups,
wefirstconductedanormalitytestandahomogeneityofvariancetest.Ifbothtestsyielded
p‐valuesgreaterthan0.05,indicatingthatthenormalityandhomogeneityassumptionswere
met,weusedtheindependentsamplet‐testforsignificanceanalysis.Otherwise,weemployed
theMann‐WhitneyUtest.Itisimportanttonotethatinboththeindependentsamplet‐test
andtheMann‐WhitneyUtest,ap‐valuelessthan0.05indicatesasignificantdifference,while
ap‐valuegreaterthanorequalto0.05suggestsnosignificantdifference.
4.2.1RiskIdentification
17 / 40
Fig.4andTable4presenttheresultsoftheriskidentificationknowledgescores.The
scoresdidnotpassthenormalitytest(p<0.05),necessitatingtheuseoftheMann‐Whitney
Utesttodeterminesignificantdifferencesbetweenthetwoexperimentalgroups.Comparing
theriskidentificationknowledgescorespre‐trainingandimmediatelyaftertraining,both
groupsdemonstratedasignificantincrease.Thissuggeststhatbothtrainingmethods
effectivelyenhancedparticipants’riskidentificationknowledge.Whencomparingscores
immediatelyaftertrainingandfourweeksaftertraining,onlytheslidegroupexhibiteda
significantdecrease,whiletherewasnoevidenceofasignificantdeclineintheARgroup.In
fact,whiletheslidegrouphaddecreasedvaluesof0.78,theARgrouphadalowerdecrease
valueof0.67.ThedatapresentedinTable4indicatethatparticipantsinthetwogroupsdid
notsignificantlydifferintheirriskidentificationknowledge.Thissuggeststhatbothgroups
hadthesamelevelofriskidentificationknowledgeatthestartoftheexperiment.Additionally,
thescoresoftheARgroupimmediatelyaftertrainingandfourweeksaftertrainingwere
significantlyhigherthanthoseoftheslidegroup.
Fig.4.RiskidentificationknowledgescoreforslidetrainingandARtraining.
Tab l e 4‐Comparisonofriskidentificationknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginARandslidetrainingconditions.
TimepointsParameterARSlideARvsSlide
Pre‐training
N3636Mann‐WhitneyU=615.500
Z=‐0.372
p=0.710
M4.7204.830
SD1.8301.682
18 / 40
Immediatelyaftertraining
N3636Mann‐WhitneyU=456.000
Z=‐2.231
p=0.026*
M7.3606.810
SD1.0991.238
Fourweeksaftertraining
N3636Mann‐WhitneyU=415.500
Z=‐2.727
p=0.006**
M6.6906.030
SD1.6001.134
Pre‐trainingvs
Immediatelyaftertraining
Mann‐WhitneyU147.500256.500/
Z‐5.703‐4.491/
p<0.001***<0.001***/
Immediatelyaftertraining
vsFourweeksafter
training
Mann‐WhitneyU504.000466.000/
Z‐1.690‐2.130/
p0.0910.033*/
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.Twoasterisks(**)indicatestatistically
significantresultswithp<0.01.Threeasterisks(***)indicatestatisticallysignificantresultswithp<0.001.
Table5displaystheresultsofriskidentificationknowledgescoreswithintheARgroup.
Comparingthepre‐trainingandimmediatelyaftertrainingscores,allthreecategoriesof
participants(constructionworkers,safetydirectorsandstudents)showedsignificant
improvements.Whencomparingscoresimmediatelyaftertrainingandfourweeksafter
training,noneofthethreecategoriesofparticipantsshowedsignificantdecreases.Moreover,
thedatarevealedsignificantdifferencesinthescoresofsafetydirectorsandstudentsboth
pre‐trainingandimmediatelyaftertraining.Table6presentstheresultsofriskidentification
knowledgescoreswithintheslidegroup.Comparingthepre‐trainingandimmediatelyafter
trainingscores,exceptforsafetydirectors,allothercategoriesofparticipants(construction
workersandstudents)exhibitedsignificantimprovements.Whencomparingscores
immediatelyaftertrainingandfourweeksaftertraining,noneofthethreecategoriesof
participantsshowedsignificantdecreases.Furthermore,thedataalsoindicatedsignificant
differencesinthepre‐trainingscoresbetweenconstructionworkersandsafetydirectors,as
wellasbetweensafetydirectorsandstudents.
Tab l e 5‐Comparisonofriskidentificationknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginARtrainingconditions.
TimepointsParameter
Construction
workers
Safety
directors
Students
Construction
workersvs
Construction
workersvs
Safetydirectors
vsStudents
19 / 40
SafetydirectorsStudents
Pre‐training
N101016Mann‐Whitney
U=25.000
Z=‐1.929
p=0.054
Mann‐Whitney
U=77.000
Z=‐0.161
p=0.872
Mann‐Whitney
U=36.000
Z=‐2.368
p=0.018*
M4.2006.1004.188
SD1.6871.9691.424
Immediately
aftertraining
N101016Mann‐Whitney
U=26.000
Z=‐1.874
p=0.061
Mann‐Whitney
U=72.500
Z=‐0.418
p=0.676
Mann‐Whitney
U=38.500
Z=‐2.314
p=0.021*
M7.1008.0007.125
SD1.3700.9430.885
Fourweeks
aftertraining
N101016Mann‐Whitney
U=36.000
Z=‐1.119
p=0.263
Mann‐Whitney
U=71.000
Z=‐0.490
p=0.624
Mann‐Whitney
U=73.000
Z=‐0.384
p=0.701
M6.6007.2006.438
SD1.0750.9192.128
Pre‐trainingvs
Immediately
aftertraining
Mann‐
WhitneyU
7.00023.00012.000///
Z‐3.299‐2.096‐4.439///
p<0.001***0.036*<0.001***///
Immediately
aftertraining
vsFourweeks
aftertraining
Mann‐
WhitneyU
41.00026.000116.000///
Z‐0.710‐1.890‐0.475///
p0.4780.0590.635///
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.Threeasterisks(***)indicate
statisticallysignificantresultswithp<0.001.
Tab l e 6‐Comparisonofriskidentificationknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginslidetrainingconditions.
TimepointsParameter
Construction
workers
Safety
directors
Students
Construction
workersvs
Safetydirectors
Construction
workersvs
Students
Safetydirectors
vsStudents
Pre‐training
N11916Mann‐Whitney
U=21.000
Z=‐2.222
p=0.026*
Mann‐Whitney
U=85.000
Z=‐0.152
p=0.879
Mann‐Whitney
U=12.000
Z=‐3.457
p<0.001***
M4.3646.4444.250
SD2.0141.0141.125
Immediately
aftertraining
N11916Mann‐Whitney
U=34.000
Mann‐Whitney
U=87.000
Mann‐Whitney
U=53.500
M6.6367.2226.688
20 / 40
SD1.1201.2021.352
Z=‐1.245
p=0.213
Z=‐0.051
p=0.959
Z=‐1.080
p=0.280
Fourweeks
aftertraining
N11916Mann‐Whitney
U=40.500
Z=‐0.720
p=0.472
Mann‐Whitney
U=81.000
Z=‐0.363
p=0.717
Mann‐Whitney
U=56.000
Z=‐0.956
p=0.339
M6.0916.3335.813
SD0.8311.2251.276
Pre‐trainingvs
Immediately
aftertraining
Mann‐
WhitneyU
26.00027.50022.500///
Z‐2.316‐1.235‐4.052///
p0.021*0.217<0.001***///
Immediately
aftertraining
vsFourweeks
aftertraining
Mann‐
WhitneyU
45.50027.50091.000///
Z‐1.041‐1.234‐1.431///
p0.2980.2170.152///
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.Threeasterisks(***)indicate
statisticallysignificantresultswithp<0.001.
4.2.2RiskAssessment
TheresultsofriskassessmentknowledgescoresaredepictedinFig.5andTable7.Since
thenormalitytestyieldedap‐valuelessthan0.05,indicatingthatthescoresdidnotpassthe
normalitytest,weemployedtheMann‐WhitneyUtesttodeterminesignificantdifferences
betweentheexperimentalgroups.Bothgroupsshowedasignificantincreaseinrisk
assessmentknowledgescoresaftertraining,indicatingtheeffectivenessofbothtraining
methodsinimprovingparticipants’riskassessmentknowledge.Whenresultswerecompared
immediatelyaftertrainingandfourweeksaftertraining,onlytheslidegroupshoweda
significantdrop;theARgroupdidnotexhibitanysignificantdeclineinscores.Accordingto
Tab l e 7,therewerenosignificantdifferencesinriskassessmentknowledgebetweenthetwo
groups,indicatingthatbothgroups’knowledgebaseswerecomparable.Furthermore,there
werenosignificantdifferencesinscoresbetweenthetwogroupsimmediatelyaftertraining
andfourweeksaftertraining.
21 / 40
Fig.5.RiskassessmentknowledgescoreforslidetrainingandARtraining.
Tab l e 7‐Comparisonofriskassessmentknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginARandslidetrainingconditions.
TimepointsParameterARSlideARvsSlide
Pre‐training
N3636Mann‐WhitneyU
=624.500
Z=‐0.276
p=0.783
M4.314.25
SD1.2831.204
Immediatelyafter
training
N3636Mann‐WhitneyU
=620.000
Z=‐0.322
p=0.747
M6.866.81
SD1.3971.411
Fourweeksafter
training
N3636Mann‐WhitneyU
=567.500
Z=‐0.933
p=0.351
M6.335.97
SD1.6211.158
Pre‐trainingvs
Immediatelyafter
training
Mann‐WhitneyU131.500108.000/
Z‐5.922‐6.212/
p<0.001***<0.001***/
Immediatelyafter
trainingvsFourweeks
Mann‐WhitneyU524.500443.500/
Z‐1.416‐2.362/
22 / 40
aftertrainingp0.1570.018*/
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.Threeasterisks(***)indicate
statisticallysignificantresultswithp<0.001.
WithintheARgroup,Table8presentstheresultsofriskassessmentknowledgescores.
Participantsinallthreecategoriesshowedsignificantimprovementswhencomparingpre‐
trainingandimmediatelyaftertrainingscores.Whencomparingscoresimmediatelyafter
trainingandfourweeksaftertraining,noneofthethreecategoriesshowedsignificant
decreases.Moreover,thedatashowedsignificantdifferencesinthepre‐trainingscores
betweenconstructionworkersandsafetydirectors,aswellasbetweensafetydirectorsand
students.Table9displaystheresultsofriskassessmentknowledgescoreswithintheslide
group.Whencomparingpre‐trainingandimmediatelypost‐trainingresults,participantsinall
threecategoriesdemonstratedstatisticallysignificantgains.Therewerenosignificant
decreasesinanyofthethreecategorieswhencomparingscoresimmediatelyaftertraining
andfourweeksaftertraining.Additionally,theresultsshowedstatisticallysignificant
differencesinthepre‐trainingandfourweeksaftertrainingscoresofconstructionworkers
andsafetydirectors,aswellassafetydirectorsandstudents.
23 / 40
Tab l e 8‐Comparisonofriskassessmentknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginARtrainingconditions.
TimepointsParameter
Construction
workers
Safety
directors
Students
Construction
workersvs
Safetydirectors
Construction
workersvs
Students
Safetydirectors
vsStudents
Pre‐training
N101016Mann‐Whitney
U=12.000
Z=‐3.089
p=0.002**
Mann‐Whitney
U=78.000
Z=‐0.109
p=0.913
Mann‐Whitney
U=28.000
Z=‐2.859
p=0.004**
M3.9005.4003.875
SD1.1970.5161.310
Immediately
aftertraining
N101016Mann‐Whitney
U=29.000
Z=‐1.628
p=0.103
Mann‐Whitney
U=78.500
Z=‐0.082
p=0.935
Mann‐Whitney
U=46.500
Z=‐1.810
p=0.070
M6.6007.6006.563
SD1.1741.5061.365
Fourweeks
aftertraining
N101016Mann‐Whitney
U=32.000
Z=‐1.388
p=0.165
Mann‐Whitney
U=78.500
Z=‐0.083
p=0.934
Mann‐Whitney
U=53.000
Z=‐1.447
p=0.148
M6.1007.0006.063
SD1.4491.6331.692
Pre‐trainingvs
Immediately
aftertraining
Mann‐
WhitneyU
6.00011.00023.000///
Z‐3.434‐3.068‐4.019///
p<0.001***0.002**<0.001***///
Immediately
aftertraining
vsFourweeks
aftertraining
Mann‐
WhitneyU
37.00039.000105.000///
Z‐1.007‐0.851‐0.888///
p0.3140.3950.375///
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Twoasterisks(**)indicatestatisticallysignificantresultswithp<0.01.Threeasterisks(***)indicate
statisticallysignificantresultswithp<0.001.
Tab l e 9‐Comparisonofriskassessmentknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginslidetrainingconditions.
TimepointsParameter
Construction
workers
Safety
directors
Students
Construction
workersvs
Safetydirectors
Construction
workersvs
Students
Safetydirectors
vsStudents
Pre‐training
N11916Mann‐Whitney
U=5.000
Mann‐Whitney
U=79.000
Mann‐Whitney
U=20.000
M3.8185.4443.875
24 / 40
SD0.7510.5271.310
Z=‐3.510
p<0.001***
Z=‐0.463
p=0.643
Z=‐3.190
p=0.001**
Immediately
aftertraining
N11916Mann‐Whitney
U=31.000
Z=‐1.432
p=0.152
Mann‐Whitney
U=81.000
Z=‐0.356
p=0.721
Mann‐Whitney
U=33.000
Z=‐2.258
p=0.024
M6.7277.6676.375
SD1.6181.4141.088
Fourweeks
aftertraining
N11916Mann‐Whitney
U=19.000
Z=‐2.426
p=0.015*
Mann‐Whitney
U=69.500
Z=‐0.967
p=0.334
Mann‐Whitney
U=27.500
Z=‐2.582
p=0.010*
M5.8186.8895.563
SD0.8740.9281.209
Pre‐trainingvs
Immediately
aftertraining
Mann‐
WhitneyU
3.0008.50016.000///
Z‐3.842‐2.922‐4.356///
p<0.001***0.003**<0.001***///
Immediately
aftertraining
vsFourweeks
aftertraining
Mann‐
WhitneyU
41.00023.50078.000///
Z‐1.334‐1.551‐1.953///
p0.1820.1210.051///
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.Twoasterisks(**)indicatestatistically
significantresultswithp<0.01.Threeasterisks(***)indicatestatisticallysignificantresultswithp<0.001.
4.2.3RiskResponse
TheresultsoftheriskresponseknowledgescoresarepresentedinFig.6andTabl e 10.
Asthenormalitytestyieldedap‐valuelessthan0.05,indicatingafailuretopassthenormality
test,theMann‐WhitneyUtestwasemployedtodeterminestatisticallysignificantdifferences
betweentheexperimentalgroups.Whencomparingpre‐trainingandimmediatelyafter
trainingriskresponseknowledgeratings,bothgroupsshowedstatisticallysignificantgains.
Onlytheslidegrouphadasignificantlossinscoreswhencomparingimmediatelyaftertraining
andfourweeksaftertraining;theARgroupdidnotexhibitanysignificantdecline.According
toTable10,therewerenosignificantdifferencesinthetwogroups’scoresofriskresponse
knowledge,indicatingthattheirstartingpointsfortheexperimentwereaboutthesame.
Additionally,therewerenosignificantdifferencesinthetwogroups’resultsimmediatelyafter
trainingandfourweeksaftertraining.
25 / 40
Fig.6.RiskresponseknowledgescoreforslidetrainingandARtraining.
Tab l e 10‐Comparisonofriskresponseknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginARandslidetrainingconditions.
TimepointsParameterARSlideARvsSlide
Pre‐training
N3636Mann‐WhitneyU
=610.000
Z=‐0.436
p=0.663
M6.366.31
SD2.1931.721
Immediatelyafter
training
N3636Mann‐WhitneyU
=582.500
Z=‐0.751
p=0.452
M7.837.64
SD1.6481.437
Fourweeksafter
training
N3636Mann‐WhitneyU
=532.500
Z=‐1.325
p=0.185
M7.256.72
SD1.5921.892
Pre‐trainingvs
Immediatelyafter
training
Mann‐WhitneyU384.500368.000/
Z‐3.005‐3.230/
p0.003**0.001**/
Immediatelyafter
trainingvsFourweeks
Mann‐WhitneyU503.000447.000/
Z‐1.659‐2.300/
26 / 40
aftertrainingp0.0970.021*/
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.Twoasterisks(**)indicatestatistically
significantresultswithp<0.01.
Table11displaystheriskresponseknowledgescoreswithintheARgroup.Onlythe
studentsshowedsignificantimprovementwhencomparingpre‐trainingandimmediatelyafter
trainingscores.Furthermore,nosignificantdeclineswereobservedinanyofthethree
participantcategorieswhencomparingscoresimmediatelyaftertrainingandfourweeksafter
training.Furthermore,thedataindicatednosignificantdifferencesinthescoresofthethree
typesofparticipantsatthethreetimepointsintheirrespectivepairwisecomparisons.The
findingsoftheslidegroup’sriskresponseknowledgescoresareshowninTable12.Participants
inallthreecategoriesdisplayedsignificantimprovementswhencomparingpre‐trainingand
immediatelyaftertrainingscores.Similarly,whencomparingscoresimmediatelyaftertraining
andfourweeksaftertraining,noneofthethreecategoriesofparticipantsshowedsignificant
decreases.Additionally,thedatarevealednosignificantdifferencesinthescoresofthethree
typesofparticipantsatthethreetimepointsintheirrespectivepairwisecomparisons.
Tab l e 11‐Comparisonofriskresponseknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginARtrainingconditions.
TimepointsParameter
Construction
workers
Safety
directors
Students
Construction
workersvs
Safetydirectors
Construction
workersvs
Students
Safetydirectors
vsStudents
Pre‐training
N101016Mann‐Whitney
U=36.500
Z=‐1.044
p=0.296
Mann‐Whitney
U=71.000
Z=‐0.481
p=0.631
Mann‐Whitney
U=47.000
Z=‐1.757
p=0.079
M6.3007.0006.000
SD2.0032.5822.098
Immediately
aftertraining
N101016Mann‐Whitney
U=39.000
Z=‐0.876
p=0.381
Mann‐Whitney
U=77.500
Z=‐0.134
p=0.893
Mann‐Whitney
U=68.500
Z=‐0.618
p=0.537
M7.7008.2007.688
SD1.4181.3981.957
Fourweeks
aftertraining
N101016Mann‐Whitney
U=33.000
Z=‐1.361
p=0.174
Mann‐Whitney
U=67.500
Z=‐0.683
p=0.494
Mann‐Whitney
U=54.000
Z=‐1.397
p=0.162
M7.1007.8007.000
SD0.9941.2292.033
Pre‐trainingvs
Immediately
aftertraining
Mann‐
WhitneyU
28.50033.50071.000///
Z‐1.656‐1.290‐2.183///
27 / 40
p0.0980.1970.029*///
Immediately
aftertraining
vsFourweeks
aftertraining
Mann‐
WhitneyU
34.00038.000101.500///
Z‐1.249‐0.936‐1.021///
p0.2120.3490.307///
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.
Tab l e 12‐Comparisonofriskresponseknowledgescorespre‐training,immediatelyafter
training,and4weeksaftertraininginslidetrainingconditions.
TimepointsParameter
Construction
workers
Safety
directors
Students
Construction
workersvs
Safetydirectors
Construction
workersvs
Students
Safetydirectors
vsStudents
Pre‐training
N11916Mann‐Whitney
U=39.500
Z=‐0.814
p=0.416
Mann‐Whitney
U=71.000
Z=‐0.874
p=0.382
Mann‐Whitney
U=46.500
Z=‐1.513
p=0.130
M6.3647.0005.875
SD1.4331.1182.094
Immediately
aftertraining
N11916Mann‐Whitney
U=40.500
Z=‐0.713
p=0.476
Mann‐Whitney
U=86.500
Z=‐0.075
p=0.940
Mann‐Whitney
U=57.500
Z=‐0.839
p=0.402
M7.5468.0007.500
SD1.2141.1181.751
Fourweeks
aftertraining
N11916Mann‐Whitney
U=31.500
Z=‐1.402
p=0.161
Mann‐Whitney
U=88.000
Z=0
p=1.000
Mann‐Whitney
U=53.000
Z=‐1.094
p=0.274
M6.5467.3336.500
SD1.5081.2252.394
Pre‐trainingvs
Immediately
aftertraining
Mann‐
WhitneyU
34.50020.00070.500///
Z‐1.753‐1.881‐2.220///
p0.080*0.060*0.026*///
Immediately
aftertraining
vsFourweeks
aftertraining
Mann‐
WhitneyU
34.00028.50095.000///
Z‐1.777‐1.103‐1.262///
p0.0760.2700.207///
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.Anasterisk(*)indicatesstatisticallysignificantresultswithp<0.05.
28 / 40
4.3AnalysisofTaskLoad
Inthissection,thetaskloadexperiencedbytheparticipantswasanalyzed.Fig.7displays
thescoresobtainedforthesixitemsassessedinthetaskload.Thedatadidnotpassthe
normalitytest(p<0.05),andtheMann‐WhitneyUtestwasusedtodeterminestatistically
significantdifferencesbetweenthetwoexperimentalgroups.TheresultspresentedinTable
13indicatethattherewerenosignificantdifferencesinthetaskloadobservedacrossthe
differenttrainingconditions.
Fig.7.Tas kloadforARtrainingandslidetraining.
Tab l e 13‐Ta s kloadscoresimmediatelyaftertraininginARandslidetrainingconditions.
ARvsSlide
Mental
Workload
Physical
Workload
Tempora l
Workload
Training
Effort
Training
Difficulty
Stressed/An
noyed
Mann‐WhitneyU550.000617.000608.000635.500645.000647.000
Z‐1.111‐0.351‐0.453‐0.142‐0.034‐0.011
p0.2670.7250.6510.8870.9730.991
4.4AnalysisofSystemUsability
TheanalysisofsystemusabilityisdepictedinFig.8.Thedatapassedthenormalitytest
(p>0.05)andhomogeneityofvariancetest(p>0.05),andtheindependentsamplet‐testwas
usedtoanalyzesignificantstatisticaldifferencesbetweenthetwoexperimentalgroups.The
analysispresentedinTab l e14demonstratesthattherearenosignificantdifferencesinthe
systemusabilityobservedacrossdifferenttrainingconditions.
29 / 40
Fig.8.SystemusabilityforARtrainingandslidetraining.
Tab l e 14‐SystemusabilityscoresimmediatelyaftertraininginARandslidetrainingconditions.
ARvsSlideNMSD95%‐CItdfp
SystemUsability
363.9000.529
[‐0.263,0.302]0.137700.891
363.8810.666
Note:Nindicatesthesamplesize.Mindicatesthemeanofthescores.SDindicatesthestandarddeviationofthe
scores.CIindicatestheconfidenceinterval.
5.Discussions
Thisstudyaimstocomparetheeffectsofslide‐basedtrainingandAR‐basedtrainingon
objectiveperformancemeasuresandsubjectiveevaluations.Twopost‐trainingtestswere
carriedout,separatedbyfourweeks,toevaluatetheeffectsoftrainingovertimeand
comparetheshort‐andlong‐termtrainingeffects.Theresultsofthisstudyprovideevidence
supportingtheeffectivenessofARindeliveringmetroconstructionsafety‐relatedknowledge.
Thesefindingsareconsistentwithpriorresearch,demonstratingthatARhasamoderately
positiveinfluenceonbothknowledgeacquisitionandretention(GarzónandAcevedo,2019).
Thefindingsofthisstudyshowthat,intermsofshort‐termacquisitionofrisk
identificationknowledge,theARgroupexhibitedasignificantlyhighertrainingeffect
comparedtotheslidegroup.ThissuggeststhatARtechnologyisparticularlybeneficialforrisk
identificationtraining.Previousstudieshaverevealedsimilarfindings,whichhaveshownthat
ARtrainingenhancestrainingefficiency(HouandWang,2013;Koumaditisetal.,2019)and
30 / 40
improvesperformanceinmaintenancetasks,withthepositiveeffectsofARtraining
increasingastaskcomplexityrises(Liuetal.,2022;ZhangandRobb,2021).Onepossible
explanationforthisdifferenceisthatAR,comparedtotraditionalmethods,enhances
informationclarity,facilitatesfasterknowledgeacquisition,providesgreaterstimulation,and
offersadequateguidance(Wolfetal.,2021).However,intermsofshort‐termacquisitionof
riskassessmentandriskresponseknowledge,thisstudyfoundnosignificantdifference
betweentheARgroupandslidegroup.Thissuggeststhatextendedreality(XR)technologyis
notnecessarilysuperiortotraditionalmethodsforshort‐termknowledgeacquisitioninrisk
assessmentandriskresponse(Kaplanetal.,2020;Theesetal.,2020).Thisobservationmay
alsobeattributedtothefactthattheARvisualizationusedinthisstudysolelyconsistedof
staticmodels.Futureresearchcouldinvestigatewhetherdynamicmodelscontributetothe
short‐termacquisitionofriskassessmentandriskresponseknowledge.Nevertheless,itis
importanttonotethatbothgroupsdemonstratedasignificantimprovementinshort‐term
safetyknowledgefollowingthetraining,highlightingtheimportanceofsafetytraining.This
emphasizesthatsafetytrainingpriortoengaginginmetroconstructionoperationscan
effectivelyenhanceriskidentification,riskassessment,andriskresponsecapabilities,thereby
reducingon‐siteaccidents(Sacksetal.,2013;Shringietal.,2023).
Intermsoflong‐termretentionofriskidentification,riskassessment,andriskresponse
knowledge,theslidegroupexhibitedsignificantdecreasesinknowledgefourweeksafter
training.Severalfactorscouldcontributetothisdecline,includingthecomplexityofthe
knowledgebeingretained,naturalforgetfulnessovertime,orthelackofexplicitinformation
providedtoparticipantsregardingthefollow‐uptestafterfourweeks.However,theARgroup
didnotshowasignificantdecrease.Thisshowsthat,incontrasttotraditionalslide‐based
training,individualsmaybelesslikelytoforgetinformationafterreceivingARtraining.AR
trainingmayofferadvantagesintermsofknowledgeretention.Thismaybeduetothefact
thatARtrainingcanprovideimmersiveandinteractivelearningexperiences,enhance
engagementandimproveknowledgeretention,helpingtopromotebettermemory.Dalinget
al.(2023)conductedastudycomparingtheshort‐andlong‐termeffectsofARtrainingand
traditionaltrainingandfoundnosignificantdifferencesbetweenthetwoapproaches.
However,whencontrollingfortheparticipants’age,ARtrainingyieldedrelativelybetter
outcomes.Furthermore,thetimeintervalforassessingthelong‐termeffectivenessofthe
trainingswasonlytwoweeksintheirstudy,whichmayexplainthelackofsignificant
differencesinknowledgeretentionbetweenthetwoapproachesandtheinconsistencywith
theresultspresentedinthispaper.Thefindingsofthispaperareconsistentwithevidencefrom
otherpreviousstudies(Gargrishetal.,2021;Lametal.,2021).Therefore,Q1hasbeen
answered.
Intermsofsubjectiveevaluations,specificallytaskloadandsystemusability,thisstudy
didnotfindsignificantdifferencesbetweenthetwotrainingconditions.Thisindicatesthat
theARapplicationwaswell‐designed,andparticipantsexpressedquiteconfidenceinbeing
31 / 40
trainedusingthetechnology.Thetaskloadresultsindicatethatneithertrainingmethod
imposedsignificantphysicalormentaldemandsandthatbothwereequallyeffectivein
promotinglearning.Additionally,participantsdidnotexperiencestressordistressduringthe
training,whichmayexplaintheirsimilarperformanceintermsofshort‐termknowledge
acquisition.ThesefindingsdifferfrompreviousresearchthatreportedparticipantsinAR
trainingexperiencedlessfrustrationcomparedtotraditionaltraining,withasignificant
differenceobserved(Wuetal.,2023b).Thesystemusabilityscoreswerehighforbothtraining
groups,indicatingthatparticipantsfoundboththeslidesandARtrainingtobeefficientand
useful.Furthermore,thedatapresentedinthispaperdemonstratethattheARapplication
receivedhighersystemusabilityscoresthantraditionaltraining,whilealsoyieldinglowertask
loadscores.Thisfindingsuggests,tosomeextent,thattheARapplicationoffersimproved
usabilityandreducedtaskload.Therefore,Q2hasbeenanswered.
Thisstudyoffersseveralsignificantcontributions.Firstly,itdevelopedanovelprototype
ofanARsafetytrainingsolutiontailoredspecificallyformetroconstruction.Theeffectiveness
ofthissolutionwasevaluatedthroughauser‐centeredcomparativeassessment,providing
valuableinsightsintoitssuitabilityandefficacy.Importantly,theproposedARsystemutilizes
commonlyavailablemobilephonesandtablets,makingitacost‐effectivetrainingtool.This
allowsforAR‐basedgrouptrainingwithoutthelimitationsimposedbytheavailabilityof
specializedARequipment,suchasHMDs.Thisaccessibilityenhancesthefeasibilityand
scalabilityofimplementingARtraininginthemetroconstructionenvironment.TheARsystem
employs3DmodelscreatedinSketchupandimportedintotheARplatformtovisualize
hazardousscenariosrelatedtovarioustopics,includingmetroconstruction.Thisflexibility
facilitateseffectivetrainingondifferentsubjectsinvolvingdangeroussituations,extending
thepotentialapplicationsofthedevelopedARsystembeyondmetroconstruction.Secondly,
thisstudyevaluatesobjectiveperformancemeasuresandsubjectiveevaluationsatthree
stages:pre‐training,immediatelyaftertraining,andfourweeksaftertraining.Itcoversthe
entiresafetyriskmanagementprocess,includingriskidentification,riskassessment,andrisk
response.Byevaluatingknowledgeacquisition,retention,andsubjectiveperception,this
studyprovidesacomprehensiveunderstandingoftheeffectivenessofAR‐basedmetro
constructionsafetytraining.Incorporatingriskidentification,riskassessment,andrisk
responseintotheevaluationprocessfurtherenhancesthepracticalrelevanceofthe
evaluationresults.Thisstudyexpandstheunderstandingoftheapplicabilityandeffectiveness
ofAR‐basedmetroconstructionsafetytrainingfrombothuserandsafetyriskmanagement
perspectives.Itprovidesvaluableinsightsforthedevelopmentandimplementationofmore
effectivesystemsaimedatenhancingworkersafetyinmetroconstruction,contributingtothe
existingbodyofknowledgeinthisfield.Byincreasingtrainees’awarenessandunderstanding
ofsafetymeasuresandimprovingtheirperformance,theseAR‐basedtrainingsystemshave
thepotentialtoenhanceoverallsafetyandincreasethelikelihoodofsurvivalinemergency
situations.Ultimately,researchonuser‐centeredAR‐basedmetroconstructionsafetytraining
32 / 40
systemshasthepotentialtodrivetheadvancementofincreasinglyeffectivetoolsforspecific
usersandsafetyriskmanagement.Thefindingsandrecommendationspresentedinthisstudy
canguidefutureresearchanddevelopmenteffortstowardcreatingmoreeffectiveand
customizedARtrainingsolutionstailoredtothespecificneedsofusersandsafetyrisk
managementrequirements.
Thisstudyhasseverallimitationsthatcouldbeaddressedinfutureresearch.Firstly,the
limitedsamplesizemayrestricttheabilitytofullyinvestigatethepotentialadvantagesofAR
trainingfordifferenttypesofusers.Futurestudiescouldaimtoincreasethesamplesize,
diversity,andreliabilityofparticipantstoobtainmorerobustfindingsandgeneralizethe
resultstoabroaderpopulation.Secondly,theARtraininginthisstudyonlypresentedstatic
unsafescenariosthroughmobilephonesandtablets,withoutillustratingtheconsequencesof
accidents.FutureresearchcouldincorporatedynamicARscenariosthatdemonstratethe
severeconsequencesofunsafebehavioraswellasuseimmersiveopticalseethrough
solutions(Lovreglioetal.,2024).BycomparingtheeffectivenessofstaticARtraining(without
accidentconsequences)anddynamicARtraining(withaccidentconsequences),researchers
canevaluatetheimpactofincorporatingconsequencesonimprovingriskidentification,risk
assessment,andriskresponsecapabilities.Additionally,investigatingpotentialdifferences
betweenusingARimmersivesolutions(e.g.,opticalseethough)andmobilephones/tablets
wouldbeworthwhile.Furthermore,thisstudyfallsshortofadequatelyaddressingthe
hierarchyofhazardcontrol,includingelimination,isolation,engineeringcontrols,
administrativecontrols,andpersonalprotectiveequipment,withinthecontextofrisk
responseduringtrainingandassessment.Futureresearchcouldincorporatethiscontentinto
trainingandassessment.Afinallimitationofthisstudyisthatitreliessolelyonsubjective
ratingstoassesstaskloadandsystemusabilitymaylimittherobustnessoftheresults.Future
researchcouldincludeincorporatingadditionalphysiologicaldatasources,suchaseye
tracking,electroencephalography(EEG)measurements,andcomputervisiontechniques
(BlaesingandBornewasser,2021;Drouotetal.,2022;Shietal.,2020).
6.Conclusion
Thisstudyaimstocomparetheshort‐termandlong‐termeffectsonobjective
performancemeasuresandsubjectiveevaluationsofbothtraditionalandARtraining.To
achievethis,anovelARsafetytrainingprototypewasdevelopedformetroconstruction,and
itwascomparedtoanequivalentslide‐basedsafetytraining.Atotalof72participantswere
separatedintotwotraininggroupsfortheinvestigation.Thesafetycasestudychosen
encompassedtheentiresafetyriskmanagementprocess,includingriskidentification,risk
assessment,andriskresponse.
TheresultsdemonstratethatARtrainingismoreeffectivethantraditionaltrainingin
regardstoshort‐termknowledgeacquisitionofriskidentification,aswellaslong‐term
33 / 40
knowledgeretentionofriskidentification,riskassessment,andriskresponse.Regarding
acquiringriskassessmentandriskresponseknowledgeintheshort‐term,aswellastheoverall
learningexperience,ARtrainingperformssimilarlytotraditionaltraining.Thesefindings
highlighttheeffectivenessofARtraininginenhancingknowledgeacquisitionandretentionin
safetytraining.Whileitmaynotexhibitsignificantadvantagesovertraditionalmethodsinall
aspectsofassessment,itstilloffersavaluablealternativethatcanimprovetrainees’learning
experienceandoutcomes.Thesefindingshaveimplicationsforthedesignoffuturesafety
trainingguidelinesandrecommendations.
DataAvailabilityStatement
Alldatathatsupportthefindingsofthisstudyareavailablefromthecorresponding
authoruponreasonablerequest.
Acknowledgements
ThisresearchwassupportedbytheMinistryofEducationofHumanitiesandSocial
ScienceprojectofChina(GrantNo.23YJA630069)andtheJiangsuProvinceConstruction
SystemScienceandTech n o log y Project(GrantNo.2023JH04004).
References
Abbas,A.,Seo,J.,Ahn,S.,Luo,Y., WyllieMitchell,J.,Lee,G.,Billinghurst,M.2023."HowImmersive
VirtualRealitySafetyTrain ingSystemFeaturesImpactLearningOutcomes:AnExperimentalStudy
ofForkliftTraining."J.Manage.Eng.,39(1),04022068.https://doi.org/10.1061/(ASCE)ME.1943‐
5479.0001101.
Ahmed,S.2019."AReviewonUsingOpportunitiesofAugmentedRealityandVirtualRealityin
ConstructionProjectManagement."Org.Technol . Manag.Constr.,11(1),1839‐1852.
https://doi.org/10.2478/otmcj‐2018‐0012.
Al‐Ahmari,A.,Ameen,W., Abidi,M.H.,Mian,S.H.2018."Evaluationof3Dprintingapproachformanual
assemblytraining."Int.J.Ind.Ergon.,66,57‐62.https://doi.org/10.1016/j.ergon.2018.02.004.
Alizadehsalehi,S.,Hadavi,A.,Huang,J.C.2020."FromBIMtoextendedrealityinAECindustry."Autom.
Constr.,116.https://doi.org/10.1016/j.autcon.2020.103254.
Alshiar,M.,Holtkamp,B.,Biediger,D.,Wilson,M.,Yun,C.,Kim,K.,Ieee,2019.SMACK:Subjective
MeasureofAppliedContextualKnowledge,2019IEEEGamesEntertain.MediaConf.
Anwar,A.H.M.M.,Toa sinOakil,A.,Muhsen,A.,Arora,A.2023."Whatwouldittakeforthepeopleof
Riyadhcitytoshiftfromtheircarstotheproposedmetro?"Cas.Stud.Transp.Policy,12,101008.
https://doi.org/https://doi.org/10.1016/j.cstp.2023.101008.
Arif,M.,Nasir,A.R.,Thaheem,M.J.,Khan,K.I.A.2021."ConSafe4All:Aframeworkforlanguagefriendly
safetytrainingmodules."Saf.Sci.,141.https://doi.org/10.1016/j.ssci.2021.105329.
AUGMENT2023."TheAugmentedRealityPlatformthatGeneratesBusinessandCommercial
Benefits."https://www.augment.com/,(accessedNovember8,2023).
34 / 40
Avveduto,G.,Tanca, C.,Lorenzini,C.,Tecchia,F., Carrozzino,M.,Bergamasco,M.,2017.SafetyTraini ng
UsingVirtualReality:AComparativeApproach,AugmentedReality,VirtualReality,AndComputer
Graphics,AVR2017,PTI,pp.148‐163.
Babalola,A.,Manu,P. , Cheung,C.,Yunusa‐Kaltungo,A.,Bartolo,P. 2023."Applicationsofimmersive
technologiesforoccupationalsafetyandhealthtrainingandeducation:Asystematicreview."Saf.
Sci.,166,106214.https://doi.org/https://doi.org/10.1016/j.ssci.2023.106214.
Bao,L.,Tran , S.V.T.,Nguyen,T. L . , Pham,H.C.,Lee,D.,Park,C.2022."Cross‐platformvirtualrealityfor
real‐timeconstructionsafetytrainingusingimmersivewebandindustryfoundationclasses."
Autom.Constr.,143.https://doi.org/10.1016/j.autcon.2022.104565.
Blaesing,D.,Bornewasser,M.2021."InfluenceofIncreasingTaskComplexityandUseofInformational
AssistanceSystemsonMentalWorkload."BrainSci.,11(1).
https://doi.org/10.3390/brainsci11010102.
Burke,M.J.,Sarpy,S.A.,Smith‐Crowe,K.,Chan‐Serafin,S.,Salvador,R.O.,Islam,G.2006."Relative
effectivenessofworkersafetyandhealthtrainingmethods."Am.J.PublicHealth,96(2),315‐324.
https://doi.org/10.2105/AJPH.2004.059840.
Buttussi,F.,Chittaro,L.2021."AComparisonofProceduralSafetyTraininginThreeConditions:Virtual
RealityHeadset,Smartphone,andPrintedMaterials."IEEETrans .Learn.Technol.,14(1),C‐15.
https://doi.org/10.1109/TLT.2020.3033766.
Chalhoub,J.,Ayer,S.K.,Ariaratnam,S.T.2021."AugmentedRealityforEnablingUn‐andUnder‐Tra ine d
IndividualstoCompleteSpecialtyConstructionTasks."J.Inf.Te chnol.Construct.,26,128‐143.
https://doi.org/10.36680/j.itcon.2021.008.
Chen,H.,Hou,L.,Zhang,G.,Moon,S.2021."DevelopmentofBIM,IoTandAR/VRtechnologiesforfire
safetyandupskilling."Autom.Constr.,125.https://doi.org/10.1016/j.autcon.2021.103631.
Cherrett,T., Wills,G.,Price,J.,Maynard,S.,Dror,I.E.2009."Makingtrainingmorecognitivelyeffective:
Makingvideosinteractive."Br.J.Educ.Technol.,40(6),1124‐1134.https://doi.org/10.1111/j.1467‐
8535.2009.00985.x.
Chi,H.L.,Kim,M.K.,Liu,K.Z.,Thedja,J.P.P.,Seo,J.,Lee,D.E.2022."Rebarinspectionintegrating
augmentedrealityandlaserscanning."Autom.Constr.,136.
https://doi.org/10.1016/j.autcon.2022.104183.
Choi,B.,Ahn,S.,Lee,S.2017a."ConstructionWorkers'GroupNormsandPersonalStandardsRegarding
SafetyBehavior:SocialIdentityTheoryPerspective."J.Manage.Eng.,33(4).
https://doi.org/10.1061/(ASCE)ME.1943‐5479.0000511.
Choi,B.,Ahn,S.,Lee,S.2017b."RoleofSocialNormsandSocialIdentificationsinSafetyBehaviorof
ConstructionWorkers.I:TheoreticalModelofSafetyBehaviorunderSocialInfluence."J.Constr.
Eng.Manage.,143(5).https://doi.org/10.1061/(ASCE)CO.1943‐7862.0001271.
Choi,B.,Lee,S.2018."AnEmpiricallyBasedAgent‐BasedModeloftheSociocognitiveProcessof
ConstructionWorkers'SafetyBehavior."J.Constr.Eng.Manage.,144(2).
https://doi.org/10.1061/(ASCE)CO.1943‐7862.0001421.
Cohen,J.1992."Apowerprimer."PsychologicalBulletin,v112(n1).
Daling,L.M.,Te n b r o c k, M.,Isenhardt,I.,Schlittmeier,S.J.2023."Assembleitlikethis!–IsAR‐orVR‐
basedtraininganeffectivealternativetovideo‐basedtraininginmanualassembly?"Appl.Ergon.,
110,104021.https://doi.org/https://doi.org/10.1016/j.apergo.2023.104021.
Dallasega,P. , Schulze,F., Revolti,A.2023."AugmentedRealitytoovercomeVisualManagement
implementationbarriersinconstruction:aMEPcasestudy."Construct.Manage.Econ.,41(3),232‐
35 / 40
255.https://doi.org/10.1080/01446193.2022.2135748.
Delgado,J.M.D.,Oyedele,L.,Demian,P., Beach,T.2020."Aresearchagendaforaugmentedandvirtual
realityinarchitecture,engineeringandconstruction."Adv.Eng.Inf.,45.
https://doi.org/10.1016/j.aei.2020.101122.
Deng,Y. , Liu,Z.,Song,L.,Ni,G.,Xu,N.2023."Exploringthemetroconstructionaccidentsandcausations
forimprovingsafetymanagementbasedondataminingandnetworktheory."Eng.Constr.Archit.
Manage.https://doi.org/10.1108/ECAM‐06‐2022‐0603.
Doolani,S.,Owens,L.,Wessels,C.,Makedon,F.2020."vIS:AnImmersiveVirtualStorytellingSystemfor
VocationalTraining."Appl.Sci.‐Basel,10(22).https://doi.org/10.3390/app10228143.
Drouot,M.,LeBigot,N.,Bricard,E.,deBougrenet,J.‐L.,Nourrit,V. 2022."Augmentedrealityon
industrialassemblyline:Impactoneffectivenessandmentalworkload."Appl.Ergon.,103.
https://doi.org/10.1016/j.apergo.2022.103793.
Durivage,M.2016."HowToEstablishSampleSizesForProcessValidationUsingTheSuccess‐Run
Theorem."https://www.bioprocessonline.com/doc/how‐to‐establish‐sample‐sizes‐for‐process‐
validation‐using‐the‐success‐run‐theorem‐0001,(accessedNovember8,2023).
Dyrda,D.,Klusmann,J.,Rudolph,L.,Stieglbauer,F. ,Amougou,M.,Pantfoerder,D.,Vogel‐Heuser,B.,
Klinker,G.,2023.SpecifyingVolumesofInterestforIndustrialUseCases,2023IEEEInt.Symp.
MixedAugment.Reality,ISMAR,pp.771‐779.
Eiris,R.,Gheisari,M.,Esmaeili,B.2018."PARS:UsingAugmented360‐DegreePanoramasofRealityfor
ConstructionSafetyTraini ng." Int.J.Environ.Res.PublicHealth,15(11).
https://doi.org/10.3390/ijerph15112452.
ElKassis,R.,Ayer,S.K.,ElAsmar,M.2023."AugmentedRealityApplicationsforSynchronized
CommunicationinConstruction:AReviewofChallengesandOpportunities."Appl.Sci.‐Basel,
13(13).https://doi.org/10.3390/app13137614.
Fang,M.,Zhang,Y. , Zhu,M.,Chen,S.2022."CauseMechanismofMetroCollapseAccidentBasedon
RiskCoupling."Int.J.Environ.Res.PublicHealth,19(4).https://doi.org/10.3390/ijerph19042102.
ForoughiSabzevar,M.,Gheisari,M.,Lo,L.J.2023."AR‐QRcodeforimprovingcrewaccesstodesignand
constructioninformation."Autom.Constr.,154,105017.
https://doi.org/https://doi.org/10.1016/j.autcon.2023.105017.
Gao,M.,Kortum,P. , Oswald,F. L . 2020."Multi‐LanguageToolkitfortheSystemUsabilityScale."Int.J.
Hum.Comput.Interact.,36(20),1883‐1901.https://doi.org/10.1080/10447318.2020.1801173.
Gargrish,S.,Kaur,D.P.,Mantri,A.,Singh,G.,Sharma,B.2021."Measuringeffectivenessofaugmented
reality‐basedgeometrylearningassistantonmemoryretentionabilitiesofthestudentsin3D
geometry."Comput.Appl.Eng.Educ.,29(6),1811‐1824.https://doi.org/10.1002/cae.22424.
Garzón,J.,Acevedo,J.2019."Meta‐analysisoftheimpactofAugmentedRealityonstudents’learning
gains."Educ.Res.Rev.,27,244‐260.
Goh,Y. M . , Ubeynarayana,C.U.,Wong ,K.L.X.,Guo,B.H.W.2018."Factorsinfluencingunsafebehaviors:
Asupervisedlearningapproach."Accid.Anal.Prev.,118,77‐85.
https://doi.org/https://doi.org/10.1016/j.aap.2018.06.002.
Grabowski,M.,Rowen,A.,Rancy,J.‐P. 2018."Evaluationofwearableimmersiveaugmentedreality
technologyinsafety‐criticalsystems."Saf.Sci.,103,23‐32.
https://doi.org/https://doi.org/10.1016/j.ssci.2017.11.013.
Gwynne,S.M.V.,Kuligowski,E.D.,Boyce,K.E.,Nilsson,D.,Robbins,A.P.,Lovreglio,R.,Thomas,J.R.,Roy‐
Poirier,A.2019."Enhancingegressdrills:Preparationandassessmentofevacueeperformance."
36 / 40
FireMater.,43(6),613‐631.https://doi.org/10.1002/fam.2448.
Hart,S.G.2006."Nasa‐Ta s k LoadIndex(NASA‐TLX);20Years Later."Proc.Hum.FactorsErgon.Soc.Annu.
Meet.,50(9),904‐908.https://doi.org/10.1177/154193120605000909.
Hart,S.G.,Staveland,L.E.,1988.DevelopmentofNASA‐TLX(TaskLoadIndex):ResultsofEmpiricaland
TheoreticalResearch,In:Hancock,P.A . , Meshkati,N.(Eds.),AdvancesinPsychology.North‐Holland,
pp.139‐183.https://doi.org/https://doi.org/10.1016/S0166‐4115(08)62386‐9.
Hasanzadeh,S.,Esmaeili,B.,Dodd,M.D.2018."ExaminingtheRelationshipbetweenConstruction
Workers'VisualAttentionandSituationAwarenessunderFallandTrippingHazardConditions:
UsingMobileEyeTra c kin g ."J.Constr.Eng.Manage.,144(7).
https://doi.org/10.1061/(ASCE)CO.1943‐7862.0001516.
Hoedt,S.,Claeys,A.,VanLandeghem,H.,Cottyn,J.2017."Theevaluationofanelementaryvirtual
trainingsystemformanualassembly."Int.J.Prod.Res.,55(24),7496‐7508.
https://doi.org/10.1080/00207543.2017.1374572.
Hou,L.,Wang,X.2013."Astudyonthebenefitsofaugmentedrealityinretainingworkingmemoryin
assemblytasks:Afocusondifferencesingender."Autom.Constr.,32,38‐45.
https://doi.org/10.1016/j.autcon.2012.12.007.
Hou,L.,Wang ,X.,Tr uije ns,M.2015."UsingAugmentedRealitytoFacilitatePipingAssembly:An
Experiment‐BasedEvaluation."J.Comput.Civ.Eng.,29(1).https://doi.org/10.1061/(ASCE)CP.1943‐
5487.0000344.
Hou,L.,Wu,S.,Zhang,G.,Ta n , Y. , Wang,X.2021."LiteratureReviewofDigitalTwins Applicationsin
ConstructionWorkforceSafety."Appl.Sci.‐Basel,11(1).https://doi.org/10.3390/app11010339.
Houweling,K.P.,Mallam,S.C.,vandeMerwe,K.,Nordby,K.2024."TheeffectsofAugmentedRealityon
operatorSituationAwarenessandHead‐DownTime."Appl.Ergon.,116.
https://doi.org/10.1016/j.apergo.2023.104213.
Hu,Z.,Chan,W.T. ,Hu,H.,Xu,F. 2023."CognitiveFactorsUnderlyingUnsafeBehaviorsofConstruction
WorkersasaToolinSafetyManagement:AReview."J.Constr.Eng.Manage.,149(3).
https://doi.org/10.1061/JCEMD4.COENG‐11820.
Jang,J.,Ko,Y. , Shin,W. S . , Han,I.2021."AugmentedRealityandVirtualRealityforLearning:An
ExaminationUsinganExtendedTechnologyAcceptanceModel."IEEEAccess,9,6798‐6809.
https://doi.org/10.1109/ACCESS.2020.3048708.
Jeelani,I.,Han,K.,Albert,A.2020."Developmentofvirtualrealityandstereo‐panoramicenvironments
forconstructionsafetytraining."Eng.Constr.Archit.Manag.,27(8),1853‐1876.
https://doi.org/10.1108/ECAM‐07‐2019‐0391.
Jiao,Y. , Zhang,S.,Li,Y. , Wang,Y. , Ya n g ,B.2013."TowardscloudAugmentedRealityforconstruction
applicationbyBIMandSNSintegration."Autom.Constr.,33,37‐47.
https://doi.org/10.1016/j.autcon.2012.09.018.
Jin,Z.,Kang,S.,Lee,Y. , Jung,Y. 2023."Standardtermsasanalyticalvariablesforcollectivedatasharing
inconstructionmanagement."Autom.Constr.,148,104752.
https://doi.org/https://doi.org/10.1016/j.autcon.2023.104752.
John,B.,Beaconsfield,W.1996."SUS:AQuickandDirtyUsabilityScale."
http://hell.meiert.org/core/pdf/sus.pdf.
Kamal,A.A.,Junaini,S.N.,Hashim,A.H.2022."EvaluatingtheEffectivenessandUsabilityofAR‐based
OSHApplication:HazHunt."Int.J.Adv.Comput.Sci.Appl.,13(5),99‐106.
Kamal,A.A.,Junaini,S.N.,Hashim,A.H.,Sukor,F.S ., Said,M.F.2021."TheEnhancementofOSHTrai nin g
37 / 40
withanAugmentedReality‐BasedApp."Int.J.OnlineBiomed.Eng.,17(13),120‐134.
https://doi.org/10.3991/ijoe.v17i13.24517.
Kaplan,A.D.,Cruit,J.,Endsley,M.,Beers,S.M.,Sawyer,B.D.,Hancock,P. A . 2020."TheEffectsofVirtual
Reality,AugmentedReality,andMixedRealityasTra ini n gEnhancementMethods:AMeta‐
Analysis."Hum.Factors,63(4),706‐726.https://doi.org/10.1177/0018720820904229.
Karakosta,A.,Ve lentza, A.‐M.,Pasalidou,C.,Fachantidis, N.,Ieee,2023.SociallyAssistiveRobotics
optimizingAugmentedRealityEducationalApplicationforTe a c h i n g Tr aff icSafetyinKindergarten,
202332ndIEEEInt.Conf.RobotHumanInteract.Commun.RO‐MAN,pp.1210‐1215.
Kim,H.,Ahn,C.R.,Yang ,K.2017."IdentifyingSafetyHazardsUsingCollectiveBodilyResponsesof
Workers."J.Constr.Eng.Manage.,143(2).https://doi.org/10.1061/(ASCE)CO.1943‐7862.0001220.
Koumaditis,K.,Venck ute,S.,Jensen,F. S . , Chinello,F. , Ieee,2019.ImmersiveTra i nin g:Outcomesfrom
SmallScaleAR/VRPilot‐Studies,201926thIEEEConferenceonVirtualRealityand3DUser
Interfaces(VR),pp.1894‐1898.
Lam,M.C.,Sadik,M.J.,Elias,N.F.2021."Theeffectofpaper‐basedmanualandstereoscopic‐based
mobileaugmentedrealitysystemsonknowledgeretention."VR,25(1),217‐232.
https://doi.org/10.1007/s10055‐020‐00451‐9.
Le,Q.T.,Pedro,A.,Lim,C.R.,Park,H.T.,Park,C.S.,Kim,H.K.2015."AFrameworkforUsingMobileBased
VirtualRealityandAugmentedRealityforExperientialConstructionSafetyEducation."Int.J.Eng.
Educ.,31(3),713‐725.
Li,S.,Wu,X.,Wang,X.,Hu,S.2020."RelationshipbetweenSocialCapital,SafetyCompetency,andSafety
BehaviorsofConstructionWorkers."J.Constr.Eng.Manage.,146(6).
https://doi.org/10.1061/(ASCE)CO.1943‐7862.0001838.
Li,X.,Yi,W. , Chi,H.‐L.,Wang,X.,Chan,A.P.C.2018."Acriticalreviewofvirtualandaugmentedreality
(VR/AR)applicationsinconstructionsafety."Autom.Constr.,86,150‐162.
https://doi.org/10.1016/j.autcon.2017.11.003.
Liang,Q.,Leung,M.‐y.,Ahmed,K.2021."Howadoptionofcopingbehaviorsdeterminesconstruction
workers'safety:Aquantitativeandqualitativeinvestigation."Saf.Sci.,133.
https://doi.org/10.1016/j.ssci.2020.105035.
Liao,P.‐C.,Lei,G.,Xue,J.,Fang,D.2015."InfluenceofPerson‐OrganizationalFitonConstructionSafety
Climate."J.Manage.Eng.,31(4),04014049.https://doi.org/10.1061/(ASCE)ME.1943‐
5479.0000257.
Liu,X.‐W., Li,C.‐Y., Dang,S.,Wang,W. , Qu,J.,Chen,T., Wang,Q.‐L.2022."ResearchonTrain i ng
EffectivenessofProfessionalMaintenancePersonnelBasedonVirtualRealityandAugmented
RealityTechnology."Sustain.,14(21).https://doi.org/10.3390/su142114351.
Liu,Y. , Ye, G.,Xiang,Q.,Yan g , J.,Goh,Y. M . , Gan,L.2023."Antecedentsofconstructionworkers'safety
cognition:Asystematicreview."Saf.Sci.,157.https://doi.org/10.1016/j.ssci.2022.105923.
Lovreglio,R.,Duan,X.,Rahouti,A.,Phipps,R.,Nilsson,D.2021."Comparingtheeffectivenessoffire
extinguishervirtualrealityandvideotraining."VR,25(1),133‐145.
https://doi.org/10.1007/s10055‐020‐00447‐5.
Lovreglio,R.,Paes,D.,Feng,Z.,Zhao,X.,2024.DigitalTechnolog i e s forFireEvacuations,In:Huang,X.,
Tam , W. C . (Eds.),IntelligentBuildingFireSafetyandSmartFirefighting.SpringerNatureSwitzerland,
Cham,pp.439‐454.https://doi.org/10.1007/978‐3‐031‐48161‐1_18.
Meng,X.,Chan,A.H.S.2020."Demographicinfluencesonsafetyconsciousnessandsafetycitizenship
behaviorofconstructionworkers."Saf.Sci.,129,104835.
38 / 40
https://doi.org/https://doi.org/10.1016/j.ssci.2020.104835.
Meng,X.,Chan,A.H.S.,Lui,L.K.H.,Fang,Y. 2021."Effectsofindividualandorganizationalfactorson
safetyconsciousnessandsafetycitizenshipbehaviorofconstructionworkers:Acomparativestudy
betweenHongKongandMainlandChina."Saf.Sci.,135,105116.
https://doi.org/https://doi.org/10.1016/j.ssci.2020.105116.
Mitropoulos,P. , Memarian,B.2012."TeamProcessesandSafetyofWorkers:Cognitive,Affective,and
BehavioralProcessesofConstructionCrews."J.Constr.Eng.Manage.,138(10),1181‐1191.
https://doi.org/10.1061/(ASCE)CO.1943‐7862.0000527.
Moesl,B.,Schaffernak,H.,Vor rabe r,W. , Braunstingl,R.,Koglbauer,I.V.2023."MultimodalAugmented
RealityApplicationsforTrainingofTraffic ProceduresinAviation."MultimodalTechn o l . Interact.,
7(1).https://doi.org/10.3390/mti7010003.
Murcia‐Lopez,M.,Steed,A.2018."AComparisonofVirtualandPhysicalTrainingTra n sfer ofBimanual
AssemblyTasks."IEEETrans.VisualComput.Graphics,24(4),1574‐1583.
https://doi.org/10.1109/TVCG.2018.2793638.
Paes,D.,Feng,Z.,King,M.,KhorramiShad,H.,Sasikumar,P. , Pujoni,D.,Lovreglio,R.2024."Opticalsee‐
throughaugmentedrealityfiresafetytrainingforbuildingoccupants."Autom.Constr.,162,105371.
https://doi.org/https://doi.org/10.1016/j.autcon.2024.105371.
Pandit,B.,Albert,A.,Patil,Y. , Al‐Bayati,A.J.2019."Impactofsafetyclimateonhazardrecognitionand
safetyriskperception."Saf.Sci.,113,44‐53.https://doi.org/10.1016/j.ssci.2018.11.020.
Perlman,A.,Sacks,R.,Barak,R.2014."Hazardrecognitionandriskperceptioninconstruction."Saf.Sci.,
64,22‐31.https://doi.org/https://doi.org/10.1016/j.ssci.2013.11.019.
Placencio‐Hidalgo,D.,Alvarez‐Marin,A.,Castillo‐Vergara,M.,Sukno,R.2022."Augmentedrealityfor
virtualtrainingintheconstructionindustry."Wo rk,71(1),165‐175.https://doi.org/10.3233/WOR‐
205049.
Qi,H.,Zhou,Z.,Yua n,J.,Li,N.,Zhou,J.2023."Accidentpatternrecognitioninsubwayconstructionfor
theprovisionofcustomizedsafetymeasures."TunnellingUndergroundSpaceTe c h n ol . ,137,
105157.https://doi.org/https://doi.org/10.1016/j.tust.2023.105157.
Rissola,E.A.,Aliannejadi,M.,Crestani,F.2022."Mentaldisordersononlinesocialmediathroughthe
lensoflanguageandbehaviour:Analysisandvisualisation."Inf.Process.Manag.,59(3).
https://doi.org/10.1016/j.ipm.2022.102890.
Rokooei,S.,Shojaei,A.,Alvanchi,A.,Azad,R.,Didehvar,N.2023."Virtualrealityapplicationfor
constructionsafetytraining."Saf.Sci.,157.https://doi.org/10.1016/j.ssci.2022.105925.
Sacks,R.,Perlman,A.,Barak,R.,2013.Constructionsafetytrainingusingimmersivevirtualreality,9ed.
Tay l o r &Francis,GreatBritain,pp.1005‐1017.
Saracino,A.,Curcuruto,M.,Antonioni,G.,Mariani,M.G.,Guglielmi,D.,Spadoni,G.2015."Proactivity‐
and‐consequence‐basedsafetyincentive(PCBSI)developedwithafuzzyapproachtoreduce
occupationalaccidents."Saf.Sci.,79,175‐183.https://doi.org/10.1016/j.ssci.2015.06.011.
Schiavi,B.,Havard,V. , Beddiar,K.,Baudry,D.2022."BIMdataflowarchitecturewithAR/VRtechnologies:
Usecasesinarchitecture,engineeringandconstruction."Autom.Constr.,134.
https://doi.org/10.1016/j.autcon.2021.104054.
Scorgie,D.,Feng,Z.,Paes,D.,Parisi,F. , Yiu,T.W. ,Lovreglio,R.2024."Virtualrealityforsafetytraining:A
systematicliteraturereviewandmeta‐analysis."Saf.Sci.,171,106372.
https://doi.org/https://doi.org/10.1016/j.ssci.2023.106372.
Seeliger,A.,Cheng,L.,Netland,T.2023."Augmentedrealityforindustrialqualityinspection:An
39 / 40
experimentassessingtaskperformanceandhumanfactors."Comput.Ind.,151,103985.
https://doi.org/https://doi.org/10.1016/j.compind.2023.103985.
Shao,B.,Hu,Z.,Liu,Q.,Chen,S.,He,W.2019."Fatalaccidentpatternsofbuildingconstructionactivities
inChina."Saf.Sci.,111,253‐263.https://doi.org/10.1016/j.ssci.2018.07.019.
Shi,C.,Zhong,M.,Nong,X.,He,L.,Shi,J.,Feng,G.2012."Modelingandsafetystrategyofpassenger
evacuationinametrostationinChina."Saf.Sci.,50(5),1319‐1332.
https://doi.org/10.1016/j.ssci.2010.07.017.
Shi,Y. , Du,J.,Zhu,Q.2020."Theimpactofengineeringinformationformatontaskperformance:Gaze
scanningpatternanalysis."Adv.Eng.Inf.,46.https://doi.org/10.1016/j.aei.2020.101167.
Shin,D.‐P., Gwak,H.‐S.,Lee,D.‐E.2015."Modelingthepredictorsofsafetybehaviorinconstruction
workers."Int.J.Occup.Saf.Ergon.,21(3),298‐311.
https://doi.org/10.1080/10803548.2015.1085164.
Shin,D.H.,Jang,W. ‐S.2009."UtilizationofubiquitouscomputingforconstructionARtechnology."
Autom.Constr.,18(8),1063‐1069.https://doi.org/https://doi.org/10.1016/j.autcon.2009.06.001.
Shringi,A.,Arashpour,M.,Golafshani,E.M.,Dwyer,T. , Kalutara,P. 2023."EnhancingSafetyTrai n ing
PerformanceUsingExtendedReality:AHybridDelphi‐AHPMulti‐AttributeAnalysisinaType‐2
FuzzyEnvironment."Buildings,13(3).https://doi.org/10.3390/buildings13030625.
Somerkoski,B.,Tarkkanen,K.,Oliva,D.,Lehto,A.,Luimula,M.,2022.Pedagogicsolutionsandresultsin
designingamobilegameforfiresafetyteaching,CEURWo rkshop Proceedings,pp.44‐53.
Suri,P. A . , Fajar,M.,Nurrahman,I.,Surjadi,A.A.K.,Arifa,Z.S.,2023.Tow a r d s AugmentedReality(AR)
ApplicationsonDifferentGenerations:ACaseStudyonTheShipFamiliarization,2023International
ConferenceonInformatics,Multimedia,CyberandInformationSystems,ICIMCIS2023,pp.665‐
669.
Thees,M.,Kapp,S.,Strzys,M.P.,Beil,F. , Lukowicz,P. , Kuhn,J.2020."Effectsofaugmentedrealityon
learningandcognitiveloadinuniversityphysicslaboratorycourses."Comput.Hum.Behav.,108,
106316.https://doi.org/https://doi.org/10.1016/j.chb.2020.106316.
Vera,L.,Gimeno,J.,Casas,S.,Garcia‐Pereira,I.,Portales,C.,2018.AHybridVirtual‐AugmentedSerious
GametoImproveDrivingSafetyAwareness,Adv.Comput.Entertain.Techn o l . ACE,pp.293‐310.
Wang,Q.,Kang,X.,Zhu,K.2021."CouplingEvaluationMethodoftheConstructionRiskforSubway
DeepFoundationPit."JournalofNortheasternUniversity.NaturalScience,42(8),1152‐1158.
Wehbe,F., Hattab,M.A.,Hamzeh,F.2016."Exploringassociationsbetweenresilienceandconstruction
safetyperformanceinsafetynetworks."Saf.Sci.,82,338‐351.
https://doi.org/https://doi.org/10.1016/j.ssci.2015.10.006.
Wolf,J.,Wolfer,V. , Halbe,M.,Maisano,F.,Lohmeyer,Q.,Meboldt,M.2021."Comparingthe
effectivenessofaugmentedreality‐basedandconventionalinstructionsduringsingleECMO
cannulationtraining."Int.J.Comput.Assist.Radiol.Surg.,16(7),1171‐1180.
https://doi.org/10.1007/s11548‐021‐02408‐y.
Wu,S.,Hou,L.,Chen,H.,Zhang,G.,Zou,Y. , Tusha r,Q.2023a."Cognitiveergonomics‐basedAugmented
Realityapplicationforconstructionperformance."Autom.Constr.,149.
https://doi.org/10.1016/j.autcon.2023.104802.
Wu,S.,Hou,L.,Chen,H.,Zhang,G.,Zou,Y., Tush a r, Q.2023b."Cognitiveergonomics‐basedAugmented
Realityapplicationforconstructionperformance."Autom.Constr.,149,104802.
https://doi.org/https://doi.org/10.1016/j.autcon.2023.104802.
Wu,S.,Hou,L.,Zhang,G.,Chen,H.2022."Real‐timemixedreality‐basedvisualwarningforconstruction
40 / 40
workforcesafety."Autom.Constr.,139.https://doi.org/10.1016/j.autcon.2022.104252.
Xiang,Q.,Ye,G.,Liu,Y., Goh,Y. M . , Wang,D.,He,T.2023."Cognitivemechanismofconstructionworkers'
unsafebehavior:Asystematicreview."Saf.Sci.,159.https://doi.org/10.1016/j.ssci.2022.106037.
Xu,Q.,Chong,H.‐Y., Liao,P. ‐C.2019."Collaborativeinformationintegrationforconstructionsafety
monitoring."Autom.Constr.,102,120‐134.
https://doi.org/https://doi.org/10.1016/j.autcon.2019.02.004.
Yan g , K.,Ahn,C.R.,Vuran,M.C.,Kim,H.2017."Collectivesensingofworkers'gaitpatternstoidentify
fallhazardsinconstruction."Autom.Constr.,82,166‐178.
https://doi.org/10.1016/j.autcon.2017.04.010.
Yip,J.,Won g,S.‐H.,Yick,K.‐L.,Chan,K.,Wong, K.‐H.2019."Improvingqualityofteachingandlearning
inclassesbyusingaugmentedrealityvideo."Comput.Educ.,128,88‐101.
https://doi.org/10.1016/j.compedu.2018.09.014.
Zhang,B.,Robb,N.2021."AComparisonoftheEffectsofAugmentedRealityN‐BackTr ain i ngand
Tra diti o n alTwo‐DimensionalN‐BackTrainingonWorkingMemory."SAGEOpen,11(2),
21582440211014507.https://doi.org/10.1177/21582440211014507.
Zhang,J.,Xia,X.,Liu,R.,Li,N.2021."Enhancinghumanindoorcognitivemapdevelopmentand
wayfindingperformancewithimmersiveaugmentedreality‐basednavigationsystems."Adv.Eng.
Inf.,50,101432.https://doi.org /https://doi.org/10.1016/j.aei.2021.101432.
Zhang,L.,Liu,Q.,Wu,X.,SkibniewskiMiroslaw,J.2016."PerceivingInteractionsonConstructionSafety
Behaviors:Workers’Perspective."J.Manage.Eng.,32(5),04016012.
https://doi.org/10.1061/(ASCE)ME.1943‐5479.0000454.
Zhang,P. , Yan g,X.,Wu,J.,Sun,H.,Wei,Y. , Gao,Z.2023."Couplinganalysisofpassengerandtrainflows
foralarge‐scaleurbanrailtransitsystem."Front.Eng.Manag.,10(2),250‐261.
https://doi.org/10.1007/s42524‐021‐0180‐2.
ZhangRita,P. , Lingard,H.,Oswald,D.2020."ImpactofSupervisorySafetyCommunicationonSafety
ClimateandBehaviorinConstructionWorkgroups."J.Constr.Eng.Manage.,146(8),04020089.
https://doi.org/10.1061/(ASCE)CO.1943‐7862.0001881.
Zhang,S.,Loosemore,M.,SunindijoRiza,Y. , Galvin,S.,Wu,J.,Zhang,S.2022."AssessingSafetyRisk
ManagementPerformanceinChineseSubwayConstructionProjects:AMultistakeholder
Perspective."J.Manage.Eng.,38(4),05022009.https://doi.org/10.1061/(ASCE)ME.1943‐
5479.0001062.
Zhou,Z.,Guo,W.2020."Applicationsofitemresponsetheorytomeasuringthesafetyresponse
competencyofworkersinsubwayconstructionprojects."Saf.Sci.,127.
https://doi.org/10.1016/j.ssci.2020.104704.
Zhou,Z.,Irizarry,J.,Zhou,J.2021."Developmentofadatabaseexclusivelyforsubwayconstruction
accidentsandcorrespondinganalyses."TunnellingUndergroundSpaceTechnol.,111.
https://doi.org/10.1016/j.tust.2021.103852.
Zhu,Y., Li,N.2021."Virtualandaugmentedrealitytechnologiesforemergencymanagementinthebuilt
environments:Astate‐of‐the‐artreview."J.Saf.Sci.Resilience,2(1),1‐10.
https://doi.org/https://doi.org/10.1016/j.jnlssr.2020.11.004.