Fueling Cycle Models

WestVirginiaUniversity.
CenterforAlternativeFuels,Engines
andEmissions.
ModelingHeavyduty
VehicleFuel
EconomyBasedon
CycleProperties
NigelClark,Ph.D.;GregoryThompsonPh.D.;Oscar
Delgado,M.Sc.
9/24/2009
ExecutiveSummary
Amethodologyforpredictingheavydutyvehiclefueleconomyduringoperationoverunseenactivity
wasdevelopedbasedonfueleconomydatagatheredfromoperationmeasuredfromvehiclesexercised
overchassisdynamometercyclesandpropertiesofthosecycles.WestVirginiaUniversity(WVU)Center
for Alternative Fuels, Engines, and Emissions (CAFEE) heavyduty chassis dynamometer data for over
theroadtrucksand40foottransitbusesweregatheredfromtheCRCE55/59programandtheWMATA
emission testing program, respectively. A linear model, a black box neural network model, and a
commercialsoftwaremodel(PSAT)wereusedtopredicteitherfueleconomyinadistancetraveledper
volumeoffuelconsumedbasis(milespergallon)orfuelconsumptioninferredfromCO2emissionsmass
rate(gramspersecond)basis.Mostoftheresourcesofthisprojectwerededicatedtothelinearmodel.
Themethodologyallowedforthepredictionoffuel economyfromvehiclesoperatingon anumberof
differentchassisdynamometercyclesbasedonrelativelyfewexperimentalmeasurements.Theresults
oftheapplicationofthelinearmodeltoasetof56heavyheavydutytrucksoperatingoverfivedifferent
cycles showed that the use of average velocity and average positive acceleration as metrics produced
thelowestaveragepercentageerror(lessthan5%).Theresultsoftheapplicationofthelinearmodelto
a set of five buses operating over 16 or 17 different cycles showed again that average velocity and
average positive acceleration were suitable metrics to predict fuel economy with reasonable accuracy
(lessthan10%averagepercentageerror).ItwasalsofoundthatbaselinecyclesmustincludeIdlecycle,
alongwitharelativelyslowtransientcycleandarelativelyhighspeedcycle,preferablywithanaverage
velocityatorabovetheaveragevelocityoftheunseencycle.Basedontheresultsobtainedwithboth
datasets,itwasrecommendedthatthepredictionbemadeintermsofCO2massrate(g/s)andthen
convert to fuel economy (mpg). The results of the application of the black box neural network model
andthecommercialsoftwaremodelproducedaveragepercentageerrorsoftheorderof10%and4%,
respectively.Themaindisadvantagesofthesealternativeapproacheswithrespecttothelinearmodel
were their inherent complexity (application difficulty) and the need to use continuous (secondby
second)data.
ii
TableofContents
ExecutiveSummary.......................................................................................................................................ii
TableofContents.........................................................................................................................................iii
ListofFigures................................................................................................................................................v
ListofTables...............................................................................................................................................vii
1.Objective...................................................................................................................................................1
2.Introduction..............................................................................................................................................1
3.Background...............................................................................................................................................1
3.1RoadLoadEquation............................................................................................................................1
3.2CycleProperties(Metrics)...................................................................................................................2
3.3IntensiveandExtensiveProperties.....................................................................................................2
4.Procedure..................................................................................................................................................3
4.1LinearModel.......................................................................................................................................3
4.2CommercialSoftwareModel..............................................................................................................3
4.3NeuralNetworkBlackBoxModel.....................................................................................................4
5.LinearModelApproach.............................................................................................................................4
5.1Example...............................................................................................................................................4
5.2GeometricExplanation.......................................................................................................................6
5.3PredictedProperties...........................................................................................................................7
6.LinearModelApplicationTruckData.....................................................................................................7
6.1CyclesUsed.........................................................................................................................................8
6.2MetricsUsed.......................................................................................................................................8
6.3CasesUsed........................................................................................................................................10
6.4GoodnessofFitCriteria....................................................................................................................11
6.5TruckDataResultsandAnalysis........................................................................................................11
6.6Extrapolation.....................................................................................................................................20
6.7UDDSTransientInterchange............................................................................................................21
6.8Recommendations(TruckData).......................................................................................................21
7.BusData..................................................................................................................................................22
7.1CyclesUsed.......................................................................................................................................22
7.2MetricsandCasesUsed....................................................................................................................23
iii
7.3BusDataResultsandAnalysis...........................................................................................................23
7.4Extrapolation.....................................................................................................................................24
7.5ComparisonbetweenCO2andFuelEconomyPredictions...............................................................31
7.6BestBusBaselineCycles...................................................................................................................31
7.7BestBusBaselineCombinations.......................................................................................................32
7.8OtherPossibleBusBaselineCycleCombinations.............................................................................32
7.9NonCompatibleBusCycleCombinations.........................................................................................33
7.10BestBusDataCombination:Idle,OCTA,andKCM.........................................................................34
7.11AlternativeMetrics.........................................................................................................................35
7.12UsingStopsperMileasaMetric....................................................................................................36
7.13BestCombination:Idle,NYBus,andKCM.......................................................................................37
7.14BusDataRecommendations...........................................................................................................39
8.BlackBoxModel(NeuralNetworks).......................................................................................................39
8.1NeuralNetworkResults....................................................................................................................40
9.PSATCommercialSoftwareModel.........................................................................................................41
9.1PSATResults......................................................................................................................................43
10.Conclusion.............................................................................................................................................45
11.References............................................................................................................................................46
AppendixAWeightCorrection.................................................................................................................48
AppendixBMethodofuse.......................................................................................................................51
AppendixCBuscycles..............................................................................................................................52
iv
ListofFigures
Figure1.Geometricinterpretationofthelinearmodelmethod.................................................................7
Figure2.Baselinecyclesusedwithtruckdataset........................................................................................8
Figure3.Validationcycleusedwithtruckdataset......................................................................................8
Figure4.ScatterplotofaveragepercenterrorforCO2..............................................................................16
Figure5.Scatterplotofaveragepercenterrorforfueleconomy(mpg)...................................................17
Figure6.ResultsforCO2predictionfor56trucksusingrecommendedbaselinecyclesandmetrics........18
Figure7.Resultsformpgpredictionfor56trucksusingrecommendedbaselinecyclesandmetrics......19
Figure8.Inferredfueleconomybasedonfuelconsumptionprediction.PredictedCO2massratewas
convertedtofueleconomyinmpgusingequations15and16(SeePredictedpropertiessection)..........20
Figure9.Histogramofaverage%Erroramong13or14predictions.........................................................24
Figure10.BestCO2massrateprediction,Bus#32.....................................................................................26
Figure11.Bestfueleconomyprediction,Bus#32.....................................................................................26
Figure12.BestCO2massrateprediction,Bus#35....................................................................................27
Figure18.BestCO2massrateprediction,Bus#41.....................................................................................30
Figure20.CO2massrateandfueleconomyasfunctionsofaveragespeed..............................................31
Figure21.Frequencyofabaselinecycleinpredictionwithaverageerrorbelow10%.............................32
Figure22.Predictionerrorsusingidle,OCTAandKCMasbaselinecyclesandaveragevelocityand
averageaccelerationasmetrics.................................................................................................................34
Figure23.ParityplotforpredictionusingIdle,OCTA,andKCMcyclesandvelocityandacceleration
metrics........................................................................................................................................................35
Figure24.Percentageerrorratioswhenusingothermetricscombinations.............................................36
Figure25.PredictionerrorsusingIdle,NYBus,andKCMasbaselinecyclesandaveragevelocityandstops
permileasmetrics......................................................................................................................................38
Figure26.ParityplotforpredictionusingIdle,NYBus,andKCMusingvelocityandstopspermilemetrics.
....................................................................................................................................................................38
Figure27.Schematicofneuralnetworkarchitecture................................................................................40
Figure28.NeuralnetworkresultsinthepredictionofUDDScycleCO2massrateemissions...................41
Figure29.SummaryofneuralnetworkpredictionofCO2massrate(g/s).Predictionerror10.24%........41
Figure30.ThePSATgraphicaluserinterface.............................................................................................42
Figure31.CO2massratepredictionresultsforTransientcycle.Predictedmassrate:9.30g/s;Measured
massrate:9.76g/s;Predictionerror4.68%.Coefficientofrollingresistance0.0136...............................43
Figure32.CO2massratepredictionresultsforCruisecycle.Predictedmassrate:19.00g/s;Measured
massrate:18.14g/s;Predictionerror4.69%.Aerodynamicdragcoefficient0.37....................................44
Figure33.CO2massratepredictionresultsforUDDScycle.Predictedmassrate:11.68g/s;Measured
massrate:12.13g/s;Predictionerror3.69%.............................................................................................44
Figure33.percentchangeinCO2emissionsperpercentchangeintestweight.......................................49
Figure34.VariationofCO2emissionswithaxlepower(testweight)fortheCreep,Transient,andCruise
cycles...........................................................................................................................................................50
Figure35.Idlecycle.....................................................................................................................................52
Figure36.NewYorkBuscycle....................................................................................................................52
Figure37.Pariscycle...................................................................................................................................53
Figure38.Manhattancycle........................................................................................................................53
Figure39.WashingtonMetroTransitAuthoritycycle................................................................................54
Figure40.NewYorkCompositecycle.........................................................................................................54
Figure41.OrangeCountyTransitAuthoritycycle......................................................................................55
Figure42.CentralBusinessDistrictcycle...................................................................................................55
Figure43.Braunschweigcycle....................................................................................................................56
Figure44.CitySuburbanHeavyVehiclecycle............................................................................................56
Figure45.Beelinecycle..............................................................................................................................57
Figure46.EuropeanTestCycleUrbancycle...............................................................................................57
Figure47.CARBTransientcycle..................................................................................................................58
Figure48.UrbanDynamometerDrivingSchedule(Test_Dcycle)..............................................................58
Figure49.KingCountyMetrocycle............................................................................................................59
Figure50.Arterialcycle..............................................................................................................................59
Figure51.Commutercycle.........................................................................................................................60
vi
ListofTables
Table1.Metricsandmeasuredfueleconomyforthreebaselinecyclesandoneunseencycle..............5
Table2.MetricsvaluesforTruckdata..........................................................................................................9
Table3.Casesusedtruckdata....................................................................................................................11
Table4.CO2massratepredictionresults...................................................................................................12
Table5.Fueleconomy(mpg)predictionresults........................................................................................14
Table6.BestpredictionsforCO2massrate...............................................................................................16
Table7.Bestpredictionforfueleconomy(mpg).......................................................................................16
Table8.Weightingfactorssensitivitytochangesinbaselinecyclesandmetricsused.............................17
Table9.Transitbusesanalyzedinthisresearch.........................................................................................22
Table10.Averagemeasuredpropertiesover5busesfor17cyclesused.................................................23
Table11.BestCO2massrateresultsforeachbus......................................................................................25
Table12.Bestfueleconomy(mpg)resultsforeachbus............................................................................25
Table13.BestbusbaselinecombinationsresultsforCO2(g/s)prediction...............................................32
Table14.Combinationsofbaselinecycleswithaveragepercentageerrorbelow10%............................33
Table15.Resultsforrecommendedcombinationofbaselinecyclesforthebusdata..............................34
Table16.Bestresultswithvelocityandstops/mile...................................................................................36
Table17.Busresultswithlowesterrorusingvelocity,acceleration,andstopspermilemetrics.............37
Table18.BestresultsusingIdleNYBUsandKCM.Averagevelocityandstops/mile.................................37
Table19.SubsetofCRCtruckdataforweightcorrectionanalysis............................................................48
vii
ModelingHeavydutyVehicleFuel
EconomyBasedonCycleProperties
1.Objective
The objective of this project was to develop a suitable methodology for predicting heavy duty vehicle
fuel economy over an unseen speedtime cycle or during unseen onroad activity, based on fuel
economydatafrommeasuredchassisdynamometertestcyclesandpropertiesofthosecycles.
2.Introduction
This work was directed towards developing a methodology for inferring heavyduty vehicle fuel
economy during operation over unseen chassis dynamometer driving cycles based on fuel economy
datawhichhadbeengatheredfromoperationonknownchassisdynamometercyclesandpropertiesof
thosecycles.Themethodologyallowedforthepredictionoffueleconomyfromvehiclesoperatingona
numberofdifferentchassisdynamometercyclesbasedonrelativelyfewexperimentalmeasurements.
Throughthecourseofthiswork,threedifferentapproachesweretakentodefinethebestmethodology
to determine fuel economy for a vehicle exercised over a defined cycle. These approaches included a
mathematicalempiricalbasedlinearmodel,aneuralnetworkbasedmodel,andawholevehiclesystem
simulationmodelwhichincorporatedfueleconomyprediction.
In the modeling efforts presented herein, the West Virginia University (WVU) Center for Alternative
Fuels,Engines,andEmissions(CAFEE)heavydutychassisdynamometerdatawereminedtoidentifytest
campaignsthatcouldprovidesufficientdatatoevaluatethethreedifferentmodelingapproaches.Two
differenttestcampaignswereidentifiedandthesecampaignswerefromtheCRCE55/59programand
theWMATAemissiontestingprogram[1,2].TheCRCE55/59programdataincluded75mediumand
heavyduty dieselfueledovertheroad trucksandtractorsexercisedoversevenchassisdynamometer
test cycles while the WMATA program included twelve transit buses exercised over as many as
seventeenchassisdynamometertestcycles.
3.Background
3.1RoadLoadEquation
Vehicle (or engine) fuel consumption depends on vehicle power demand as the vehicle is driven. The
roadloadequationcanbeusedtocalculatetheinstantaneouspowerrequiredtopropelavehicle.The
power requirement for zero grade includes the rolling resistance which arises from the friction of the
tires, the aerodynamic drag of the vehicle, and the inertial power required to accelerate the vehicle.
Equation1showstheroadloadequationwherePrepresentsthepropulsionpowerdemandedbythe
vehicle at the drive wheels: this is vehicle power, not engine power; engine power would be greater
1
thanvehiclepowerbecausedrivetrainefficienciesarelessthan100%.InEquation1misthemassof
thevehicle,representsthecoefficientofrollingresistance,gistheaccelerationduetogravity,Visthe
instantaneousvelocity,istheambientairdensity,Aisthefrontalcrosssectionalareaofvehicle,CDis
thewinddragcoefficient,andtistime.
(1)
Notethatroadgradewillalsohaveaneffectonvehiclepowerdemand.However,chassisdynamometer
testingisexecutedassuminglevelgradeandroadgradeisexcluded,withrareexception.
3.2CycleProperties(Metrics)
A chassis dynamometer test cycle is defined customarily as a speed versus time array, assuming level
road.Acyclehasadefinedtestdurationandatargetdistance.Thereareadditionalmeanstodescribea
cycle using properties, or metrics, such as average velocity, standard deviation of velocity, average
acceleration,andstopsperunitdistance.Thesemetricsprovidesomeinformationthatthespeedtime
trace cannot give by itself. The most important metric to analyze fuel consumption is believed to be
average velocity, because in Equation 1 velocity appears in each term. Average velocity is a robust
indicatorofthetypeofactivityexhibitedduringagivencycle.A lowaverage velocity canrepresenta
verytransientcyclesimilartowhatisexpectedincitytrafficwhileahighaveragevelocitycanrepresent
amoresteadybehaviorsimilartowhatisexpectedinhighwaydriving.Fewvehiclestravelatasteady,
low speed, unless they are engaged in unusual vocational activity. The road load equation contains
some cycle properties such as velocity and acceleration. However, the road load equation does not
include other important properties such as stops per unit distance and percentage of time idling that
canbetakenintoaccountinordertoanalyzefuelconsumption.
Based on the rolling resistance and wind drag road load equation terms, it can be argued that fuel
consumptionwillbehigherifthevehicleisoperatedthroughahigheraveragespeedcycle.However,a
lowaveragespeedcanrepresentaverytransientcycle.Anothermetricshouldbeintroducedinorderto
account for transient behavior. Standard deviation of speed, average acceleration, and stops per unit
distancearesomeoftheexamplesofmetricsthatcanaccountfortransientbehavior.
Themainhypothesisofthisresearchwasthatcyclemetricsmightbeusedtopredict(withacceptable
accuracy)thefueleconomyofavehicleexercisedthroughanunseenspeedtimetrace.Theroadload
equationsuggeststhataveragevelocityshouldbeoneofthemetricstobeused.Additionalmetricsmay
beselectedtorefinethemodelfurthertoprovideforhigherfidelityintheresultswhileminimizingthe
requiredamountoftestdata,orthenumberofchassisdynamometercyclesneededtobeacquired.
3.3IntensiveandExtensiveProperties
Thedatausedinthesemodelscouldberesolvedbyeitherusingintensiveorextensivecycleproperties.
Extensivepropertiesdependonthesizeofthesystem;inthiscasefactorssuchascycletimelengthor
distance traveled or integrated values of V2 over the duration of the cycle. If a test cycle is run twice
backtoback,andtreatedasonecycle,thevaluesofitsextensivepropertieswouldbedoubled.On
theotherhand,intensivepropertiesdonotdependonthesize(orlength)ofthesystem,andtheyare
2
exemplified by properties such as average speed. The objective of this work was to predict fuel
consumptiononamassrate(grams/second)ortopredictfueleconomyonadistanceperunitvolumeof
fuel consumed (miles/gallon). Both of these sets of units are intensive properties and hence, only
intensive properties were used for the prediction. An equivalent approach would be to use extensive
cycle properties to predict fuel consumption in mass (an extensive property), rather than mass rate.
However,thedesiredunitsforthisworkareintensiveandhenceanintensivepropertysetwasselected.
LowaveragespeedchassisdynamometertestcyclessuchastheCreepcycleorNewYorkBuscycleare
relativelylowdistancecycleswithrelativelyhighamountsofidletime.Thesecyclestranslateintohigh
fuel consumption values in volume per unit distance (gallons/mile) units. Idle cycles are more
problematiconavolumeperunitdistancebecause,bydefinition,thiscyclewouldhaveaninfinitevalue
offuelconsumptionsincethedistancetravelediszero.Aswillbeshownbelow,itispossibletoconvert
fromonesetoffueleconomy,orconsumption,unitstoanothersetofunitsthroughknowledgeofthe
propertiesofthecycle.Assuch,CO2emissionsmassrate,ingramspersecond,isthedesiredintensive
property selected to measure fuel consumption. If a volumespecific fuel economy, miles per gallon
value would be needed, a conversion factor can be used, provided that the carbon content of the
petroleumfuelisestimatedorknown.
4.Procedure
The main features of the three techniques used to predict fuel economy for heavy duty vehicle are
summarizedbelow.
4.1LinearModel
The Linear Model approach involved identifying the most important intensive metrics of a cycle and
developing a technique which calculates the CO2 mass rate emissions for a new cycle based on CO2
mass rate emissions from actual chassis dynamometer test cycle data using those selected metrics as
weightingfactors.Usingthistechnique,heavydutyvehiclesofachosencategorycanbetestedusinga
limited number of chassis dynamometer test cycles, and the data from those tests may be used to
project emissions from an unseen cycle in a wide envelope, within certain bounds. This prediction
approachavoidstheuseofcontinuous(secondbysecond)dataandthepredictionsaremadeapriori
based on the relative cycle statistics. No regression is required. This simple method does not require
trainingamodelasisneededinneuralnetworkmodelingortheneedfordetailedcomponentmodelsas
isneededinvehiclesystemsimulationmodeling.
4.2CommercialSoftwareModel
ModelssuchasADVISORorPSATmaybeusedtopredictthefueleconomyofavehicle,byassembling
modelsofcomponentsofthevehicle,andemployingestimatesforlosses,efficiencyofcomponents,and
vehicle inertia, under constraint of driver behavior. It is difficult to use a pure modeling approach for
actual or comparative fuel consumption prediction using this approach because a great deal of
informationisrequiredforeachvehicle(suchasdrivetraincomponents),andbecauseitisincreasingly
difficult to verify that modeled control strategy (particularly for hybrid vehicles) reflects the inuse
3
control strategy. However, models of this kind may be used readily to translate performance from a
smallsetofrealworldteststoanunseencycleandaccuracyisexpectedtobegood.
4.3NeuralNetworkBlackBoxModel
Anotherpowerfultoolthatcanbeusedtomodelfueleconomyinvolvestrainingofaneuralnetwork,or
othernonclassicalmodels,usingsecondbysecond data,sothattheneuralnetworkcanthenpredict
secondbysecond performance on unseen cycles. Training the neural network using continuous
emissions data requires some skill because one must account for delay and diffusion of data during
measurement, and one must avoid overtraining by selecting input variables and network architecture
suitably. Training should be done with data containing varied vehicle behavior that encompasses the
rangeofvehicleoperationintheunseencycle.Continuousdataarerequiredforthisapproach.
5.LinearModelApproach
WVUhasidentifiedanddevelopedamethodforpredictingemissionsdatafortransientvehicleactivity
(such as a chassis dynamometer test cycle) based on information from other measured chassis
dynamometer test cycles. The technique was presented in a 2004 paper by Taylor et al. [3]. This
approachinvolvesidentifyingthemostimportantproperties(metrics)ofacycle(suchasaveragespeed,
standarddeviationofspeed,andpercentidle)anddevelopingatechniquewhichproportionsemissions
foranunseencyclebasedonemissionsfromrealworldcycledatausingthosemetricsforweighting.
The main assumption in this modeling approach is that for a given vehicle, fuel consumption over an
unseen cycle will be a linear combination of its fuel consumption over other baseline cycles. Each
baseline cycle will contribute to a percentage of the fuel consumption of the unseen cycle. Fuel
consumption depends on cycle properties so the weighing factors (or fractional contributions) of the
different baseline cycles will be obtained based on the selected cycle properties. A set of linear
equations based on cycle properties is posed in order to determine the weighting factors of each
baseline cycle to then estimate the unseen cycle. In each case, the predicted fuel economy (or CO2
emissions)wouldbeaweightedsummationofthefueleconomy(orCO2emissions)fromthebaseline
cycles,withtheweightingcoefficientsconstrainedtosumtounity.
Thenumber(N)ofbaselinetestcyclesdeterminesthenumberofsimultaneousequationsand,hence,
thenumberofpropertiesthatcanbeused.Oneoftheequationswillalwaysconstrainthesumofthe
coefficientstobeequaltoone,sotherewillbeN1propertiesthatareusedtosolvetheNsimultaneous
equations.Thefollowingsectionexplainsthemethodmoreclearlyusingastepbystepexample.
5.1Example
Assumethatfueleconomymeasurementsforthreedifferenttestcyclesareavailableandonewantsto
estimatefueleconomyforafourth,differentcycle.Thethreemeasuredcyclesaretermedthebaseline
cyclesandthepredictedcycleistermedunseencycle.Thebaselinecyclesformthebasistoestimate
fuel economy for the unseen cycle. Each baseline cycle will have a weighting factor that defines the
relative proportion of that cycle to the unseen cycle in terms of the metrics used. It is expected that
these weighting factors also can be used to then estimate the unseen cycle fuel economy. Two cycle
4
propertiesshouldbeusedbecausethenumberofbaselinecyclesisthree.Table1showstwoproperties
(averagevelocityandaverageacceleration)forthreebaselinecycles(termedIdle,Transient,andCruise)
andoneunseencycle(UDDScycle).Theobjectiveistouseinformationfromthethreebaselinecyclesto
predictfueleconomyfromtheUDDScycle.Notethatmetricsotherthanaveragevelocityandaverage
accelerationcouldhavebeenchosentoperformtheanalysiswhichispresentedbelow.
Table1.Metricsandmeasuredfueleconomyforthreebaselinecyclesandoneunseencycle.
Baseline
cycles
Idle
Transient
Cruise
0.00
14.92
39.87
Average
Acceleration
(mph/s)
0.00
0.29
0.12
unseen
cycle
UDDS
18.83
0.32
AverageVelocity
(mph)
Cycle
Measured
FuelEconomy
(mpg)
0.00
3.85
6.58
?
Thenextstepistoposeasetofthreesimultaneousequations,basedontheselectedcycleproperties,
to calculate the weights of each baseline cycle to the unseen cycle. The three unknowns are the
weightingfactorforeachbaselinecycle.TheequationsetisshowninEquations2,3,and4.Thefirst
twoequationsarelinearcombinationsusingthetwodifferentmetrics.Thefirstequationusesaverage
velocity,thesecondequationusesaverageacceleration,andthethirdequationconstrainstheweights
tosumtoone.
(2)
(3)
(4)
ReplacingnumericalvaluesfromTable1inEquations2,3,and4resultswiththefollowingequationset:
14.92
39.87
0.29
0.12
1
18.83
(5)
0.32
(6)
(7)
The next step is to solve the simultaneous equations to obtain the weighting factors. Note that this
solutionisuniqueandforthisexampleis:
0.1446
(8)
1.0744
(9)
0.0702
(10)
5
Finally,theweightingfactorsareusedtocalculatefueleconomyfortheunseencycle:
0.1446
1.0744
0.1446 0
4.59
(11)
(12)
(13)
(14)
0.0702
1.0744 3.85
0.0702 6.58
ThenegativecoefficientfortheIdlecyclesuggeststhatTransientandCruisecyclesalreadycontainmore
IdlethantheUDDScycle(lessIdleshouldbeconsideredintheUDDSthanintheweightedactivemodes
TransientandCruisecycles).TheTransienttermcoefficientequaltoapproximatelyone(1.0744)suggest
that the UDDS is closest to the Transient mode and the low value for the Cruise coefficient (0.0702)
indicates that a relatively small portion of UDDS is at cruise conditions. In this example, the weight
coefficientforIdle,ineffect,isnotbeingusedduetothezerovalueoffueleconomy(mpg)fortheIdle
cycle.RegardlessofthemassoffuelusedperhourofIdling,theIdlecontributionwillbethesame.Itis
recommended that one would predict fuel consumption using CO2 mass rate (g/s) instead of fuel
economy(inmpg)toavoidnegatingtheIdleinformation.
5.2GeometricExplanation
Thepreviousexamplehasanalternativegeometricexplanation.Solvingasystemofthreeequationsand
threeunknownsasshowninthepreviousexampleisequivalenttofindingtheequationofaplaneina
three dimensional space. Figure 1 shows a geometric representation of the previous example. In this
casethedimensionsofthethreedimensionalspaceareaveragevelocity,averageacceleration,andfuel
economy.Informationforeachbaselinecyclerepresentsapointinthatthreedimensionalspace.Once
the three points have been specified, a plane that crosses the three points (unless the points are
collinear)canbedefined.Thisplaneisuniquebecauseoneandonlyoneplanecancrossthroughthese
threepoints.Theplanecanbeusedtopredictothercyclesfueleconomyapriori,justknowingthese
new cycles properties (average velocity and average acceleration in this example). This modeling
approach simplifies the real world surface (which may be curvilinear) to a plane, using minimum
information.Thepredictionmayuseextrapolation(pointsontheplanebutoutsidethetriangleshown)
butaswithotherlinearmodels,extrapolationshouldbeexercisedwithcaution.

Figure1.Geometricinterpretationofthelinearmodelmethod.
5.3PredictedProperties
As discussed above, CO2 mass rate emissions (g/s) or fuel economy (mpg) could be estimated in the
modelandthentheotherpropertycalculated.Conversionbetweenmassrate(g/s)andfueleconomy
(mpg) is based on average speed. Equation 15 shows the U.S. Environmental Protection Agency
recommended practice to perform the calculation from CO2 mass rate to fuel consumption [10]. This
equationassumesdieselcarboncontentpergallonof2,778gramsandanoxidationfactorof0.99(99
percentofthecarboninthefueliseventuallyoxidized,while1percentremainsunoxidized).Equation
16 shows the calculation from fuel consumption volumetric rate (gal/s) to fuel economy (mpg). This
calculationiscycledependentbecausecycleaveragespeedisusedintheconversion.Notealsothatfor
theIdlecycle,fueleconomy(mpg)iszero.ItisemphasizedagainthatCO2massrate(g/s)shouldbeused
insteadoffueleconomy(mpg)toincludeidleinformationinthemodel.
(15)
(16)
6.LinearModelApplicationTruckData
The linear model was applied to two different heavy duty vehicle types. These types included heavy
duty trucks and transit buses. This section summarizes the linear model application to the truck data.
Chassis dynamometer data for 56 heavy heavyduty trucks operating at a nominal 56,000lbs inertial
masswereused.ThedatausedweregatheredaspartoftheCoordinatingResearchCouncilE55/E59
program, which was created to characterize heavyduty trucks emissions in California. It is noted that
theother19vehiclesintheE55/59programweremediumduty(andnottestedatthe56,000lbsmass)
andorgasolinefueledvehiclesandwereexcludedinthisanalysis.Thatis,onlytheclass8heavyduty
trucksincorporatingdieselenginesweremodeledinthiswork.
6.1CyclesUsed
California Air Resources Board created a fourmode speed versus time heavyheavy duty diesel truck
vehiclechassistestschedule(HHDDT)basedondatagatheredfrompriortruckactivitystudies[12,13].
Idle, Creep, Transient, and Cruise of the HHDDT schedule were used as baseline cycles and the UDDS
(Urban Dynamometer Cycle Schedule) was used as the unseen cycle. The UDDS cycle includes
behaviorthatrepresentsbothfreewayandnonfreewayoperationandislocatedintheCodeofFederal
Regulations [4]. Since actual UDDS data were available, the UDDS was considered a validation cycle.
PredictedandmeasuredUDDSdatacouldbecompared.Figure2showsthetestcyclesusedasbaseline
cyclesandFigure3showsthetestcycleusedforvalidation.
Idle cycle
speed (mph)
speed (mph)
0.8
0.6
0.4
100
200
300
400
500
600
700
800
50
100
150
200
250
300
time (s)
Cruise cycle
60
50
speed (mph)
50
speed (mph)
0
0
900
time (s)
Transient cycle
60
40
30
20
10
0
0
0.2
0
0
Creep cycle
10
40
30
20
10
100
200
300
400
500
600
0
0
700
500
1000
time (s)
1500
time (s)
2000
2500
Figure2.Baselinecyclesusedwithtruckdataset.
UDDS cycle
60
speed (mph)
50
40
30
20
10
0
0
200
400
600
time (s)
800
1000
1200
Figure3.Validationcycleusedwithtruckdataset.
6.2MetricsUsed
Ninedifferentmetricswereevaluatedwiththisdataset.Table2showsmetricsvaluesforthefivecycles
used.Equations17to25showtheformulationoftheproblemusedtocalculatethemetrics.Itisnoted
8
that additional metrics could be defined but from the authors experience these metrics presented in
Table2bestrepresentedinusevehicleactivity.
Table2.MetricsvaluesforTruckdata.
Metric
AverageVelocity(mph)
StandardDeviationofVelocity(mph)
AverageofSquaredVelocity(mph2)
AverageofCubedVelocity(mph3)
AverageAcceleration(mph/s)
InertialPower(mph2/s)
AverageofSquaredAcceleration(mph2/s2)
Stopspermile(stops/mile)
PercentageIdle(%)
BaselineCycles
Creep Transient
1.64
14.92
2.02
13.44
6.76
403
34.78
13044
0.07
0.29
0.23
5.06
0.80
0.86
24.20
1.80
42.30
16.30
Idle
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
100.00
Cruise
39.87
22.01
2074
111410
0.12
3.86
0.26
0.30
8.00
ValidationC.
UDDS
18.83
19.82
747
33992
0.32
6.41
1.80
2.50
33.40
AverageVelocity
(17)
(18)
(19)
(20)
(21)
StandardDeviationofVelocity
AverageofSquaredVelocity
AverageofCubicVelocity
AverageAcceleration
when
A SavitskyGolay filtering method of 2nd degree over 21 data points (2.1 seconds) was applied to the
speedtimetracebefore calculatingacceleration. Thismethodcomputesalocalpolynomialregression
ontheinputdataand ispreferredoverothertechniquessuchasmovingaveragesbecauseittendsto
preserve features of the distribution such as relative maxima, minima and width. Acceleration was
calculatedwithacentraldifferencesscheme.Actualspeedtimetraceswereused.Onlypositivevalues
were taken into account since it was assumed that the engine does not consume fuel when
9
decelerating. Note that the denominator in the formula is the total number of data points, not the
numberofdatapointswithpositiveacceleration.
AverageInertialPower
when
(22)
(23)
(24)
AverageofSquaredAcceleration
when
StopsperMile
A velocity value below 0.5 miles per hour was counted as a stop to account for the resolution in the
chassisdynamometerdataacquisitionsystem.Stopdurationwasnottakenintoaccount,forexample,if
the vehicle remained below 0.5 mph during a long period it was counted as only one stop. The stop
analysiswasdonewithoutfilteringofthespeedtimetrace.
PercentageIdle
Anydatapointwithvelocitybelow0.5mphwasconsideredtobeanidleevent.Theidleanalysiswas
donewithoutfilteringofthespeedtimetrace.
(25)
6.3CasesUsed
Sixtyonecaseswereevaluated,includingfivedifferentcombinationsofbaselinecyclesandanumberof
metrics.Table3showthecasesused.Cycleaveragedvelocitywasusedinallcasesasametric.Cases1
to 4 have eight possible combinations (velocity and another metric) and case 5 has 28 possible
combinations(velocityandtwoothermetrics).
10
Table3.Casesusedtruckdata
CASEID
Baselinecyclesused
0
1
2
3
4
TransientandCruise
Idle,TransientandCruise
Creep,TransientandCruise
Idle,CreepandCruise
Idle,CreepandTransient
Idle,Creep,Transientand
Cruise
#ofmetrics
used
1
2
2
2
2
3
Metricsused
Velocityonly
Velocity+1metric
(allcombinations)
Velocity+2metrics
(allcombinations)
6.4GoodnessofFitCriteria
FourcriteriawereselectedtoevaluatethegoodnessoffitbetweenthemeasuredandpredictedUDDS
data: average percentage error (Equation 26 where
xp are predicted values and xe are experimental
values), maximum absolute error, average absolute error (Equation 27 where x p are predicted values
and
xe areexperimentalvalues),andR2correlationcoefficientsbetweenthemeasuredandpredicted
values.NotethatR2couldbeamisleadingmeasurewhendataareclustered,asmaybethecasewitha
fueleconomymeasure.Therecommendedgoodnessoffitcriterioninthisworkwasaveragepercentage
errororaverageabsoluteerrorbuttheothercriteriaarepresentedbelowaswell.
n
x xe i
x100 (26)
Average%error = 1 p
i =1
( xe )i
n
Averageabsoluteerror = 1 x x
p
e
i
n i =1
(27)
6.5TruckDataResultsandAnalysis
Tables 4 and 5 show the summary of prediction results for CO2 (g/s) and fuel economy (mpg),
respectively. The data were organized based on average percentage error values over 56 predictions
(one prediction per truck). Note that not all of the cases worked well. Some metrics yielded more
suitablethanotherstotranslatefueleconomyorfuelconsumptionamongcycles.Also,somebaseline
cycle combinations were better than others when trying to predict the validation cycle. Note that 27
cases were below 10% error for CO2 mass rate (g/s) and 16 cases were below 10% error for fuel
economy (mpg). This could be because of the loss of information when using the Idle cycle fuel
consumption(mpg)datadiscussedabove.Tables6and7showmoredetailaboutthefourlowesterror
predictionsforCO2(g/s)andfuelconsumption(mpg),respectively.Figures4and5displayscatterplots
oftheaveragepredictionerrorshowingthecaseswheretheaveragepercentageerrorwaslowerthan
12%.Thebestcombinationofaccuracyandeconomy(intermsofnumberofbaselinecyclesused)was
obtained using Idle, Transient, and Cruise as baseline cycles with average velocity and average
accelerationasmetrics.Toincorporatefourbaselinecyclesintotheanalysis,thebestmetrictoaddto
themodelwouldbethenumberofstopsperunitdistance(stops/mile).Itisworthmentioningthatthe
11
useofaveragevelocityastheonlymetricwiththeTransientandCruisebaselinecyclesproducedgood
results for CO2 (g/s) and was even better in terms of economy (number of metrics used) than the
recommendedcombinationofaveragevelocityandacceleration.However,theuseofthismetricalone
shouldbeavoidedbecausetwocycleswithsimilaraveragevelocitycanrepresentverydifferenttypeof
activity and very different fuel consumption patterns. Use of an additional metric is recommended to
accountforthedegreeoftransientbehaviorofthecycle.
Anotherimportantissueisthattheresultswereverysimilarforfueleconomy(mpg)andforCO2(g/s)
evenwiththefueleconomypredictionignoringtheweightingcoefficientfortheIdlecycle.Thiscouldbe
due to the high similarity between the Transient cycle and UDDS cycle, and the fact that only a small
fraction of the fuel was consumed during idling portions of each cycle. Table 8 shows the weighting
factorssensitivitytotheaddition/subtractionofcyclesandmetrics.Notethattheaveragepercentage
error does not show a significant change for fuel economy but the error goes to more than 8% when
usingvelocityaloneasametricwiththeCreepandTransientcycle.Furtherinsightinthistopicwillbe
gainedinthenextsectionofthisreport.
Figures6and7showtheresultsusingtherecommendedbaselinecyclesandmetricsforCO2massrate
and fuel economy, respectively. Idle, Transient, and Cruise cycles with average velocity and average
acceleration as metrics were used to predict the UDDS cycle for 56 trucks. A parity plot between
measuredandpredictedvaluesisshown,aswellastheresultingequationforprediction.Notethatthe
weighting factors are the same for CO2 and fuel economy prediction (using the same metrics and the
samebaselinecycles),buttheIdlecycleinformationislostinthefueleconomypredictionbecauseof
thezeroeconomyvaluefortheIdlecycle.Figure8showstheCO2massrate(g/s)predictionsshownin
Figure 6, converted to fuel economy (mpg) using equations 15 and 16. It is recommended that the
predictionbemadeintermsofCO2massrate(g/s)ortheequivalentunitsofg/soffuelmassflow.One
maythencomputefueleconomy(mpg)usingaveragecyclespeed.
As in the previous example, the Transient cycle weighting factor was nearly one, and the Idle cycle
weightingfactorwasnegative.ThesecoefficientssuggestthattheUDDSisclosesttotheTransientcycle
andthatlessidleshouldbeconsideredintheUDDSthanintheweightedactivemodes(Transientand
Cruisecycles).
Table4.CO2massratepredictionresults.
Case
Used
Average
Percent
Error(%)
Maximum
Error(CO2)
Average
Error(CO2)
R2
Velocity,Acceleration,Stops/mile
4.29
2.00
0.52
0.82
Velocity,Acceleration
4.36
2.02
0.52
0.82
MetricsUsed
Velocity
4.36
2.43
0.54
0.83
4.89
2.16
0.58
0.81
5.49
2.69
0.68
0.83
Velocity,Acceleration,Acceleration
6.68
3.89
0.82
0.65
Velocity,Acceleration,%Idle
7.13
2.61
0.84
0.78
Velocity,Stops/mile
2
12
Velocity,Velocity3,Velocity2
7.31
3.13
0.90
0.80
7.65
3.09
0.90
0.73
Velocity,Acceleration,InertialPower
7.96
2.75
0.94
0.76
Velocity,InertialPower,%Idle
8.01
2.81
0.95
0.77
Velocity,Velocity ,%Idle
8.15
3.20
1.00
0.82
Velocity,InertialPower
8.21
3.02
0.96
0.80
Velocity,Stops/mile
8.26
3.11
0.97
0.80
8.38
3.11
0.98
0.80
Velocity,Velocity ,InertialPower
8.42
5.84
1.04
0.42
8.44
3.14
0.99
0.80
Velocity,InertialPower,Stops/mile
8.45
3.15
0.99
0.80
8.68
3.25
1.06
0.82
8.71
3.29
1.07
0.82
Velocity,Velocity
3
Velocity,Velocity ,Stops/mile
8.82
3.26
1.08
0.82
Velocity,Velocity3
9.06
3.28
1.11
0.82
9.43
3.36
1.15
0.82
Velocity,InertialPower,Acceleration
9.47
3.56
1.11
0.74
Velocity,%Idle
Velocity,Velocity2
2
9.47
3.13
1.11
0.77
9.49
7.19
1.19
0.30
9.65
3.39
1.18
0.82
Velocity,Velocity
10.12
3.44
1.23
0.82
10.36
3.63
1.22
0.71
Velocity,Stops/mile
11.69
3.82
1.37
0.78
Velocity,Std.Dev.Velocity
11.80
7.50
1.47
0.23
13.33
5.40
1.59
0.51
Velocity,Velocity ,Acceleration
13.49
5.43
1.64
0.54
Velocity,%Idle
13.59
4.01
1.65
0.80
15.57
4.31
1.89
0.78
17.33
6.21
2.10
0.46
19.42
13.57
2.40
0.07
Velocity,Velocity2
19.47
11.69
2.35
0.24
19.51
8.86
2.37
0.20
20.30
9.30
2.46
0.17
20.42
9.20
2.48
0.18
20.96
5.72
2.48
0.73
21.65
8.19
2.61
0.25
22.96
5.99
2.72
0.74
2
Velocity,Std.Dev.Velocity,InertialPower
3
Velocity,Stops/mile,%Idle
2
Velocity,Velocity
2
Velocity,Std.Dev.Velocity,Acceleration
2
Velocity,Velocity
2
22.99
7.37
2.77
0.35
Velocity,Std.Dev.Velocity,Stops/mile
23.08
6.01
2.73
0.74
Velocity,%Idle
25.71
5.81
3.10
0.56
26.07
6.41
3.09
0.72
13
Velocity,Stops/mile
26.24
5.96
3.16
0.55
Velocity,Std.Dev.Velocity,%Idle
29.41
6.86
3.49
0.67
31.98
21.60
3.89
0.09
33.52
8.02
3.98
0.70
35.12
8.32
4.17
0.69
Velocity,Velocity
2
2
Velocity,Acceleration ,Stops/mile
Velocity,%Idle
37.64
8.12
4.52
0.19
Velocity,Velocity2,Acceleration
43.46
17.72
5.25
0.01
Velocity,Velocity3,Std.Dev.Velocity
71.85
38.05
8.73
0.00
88.64
33.27
10.62
0.17
106.01
21.31
12.64
0.54
160.07
111.49
19.97
0.01
808.14
162.68
96.50
0.38
1402.10
633.08
169.29
0.03
Velocity,Acceleration ,%Idle
2
Velocity,Std.Dev.Velocity,Velocity
Table5.Fueleconomy(mpg)predictionresults.
MetricsUsed
Case
Used
AveragePercent MaximumError AverageError

Error(%)
(mpg)
(mpg)
R2
4.59
0.67
0.20
0.82
4.83
0.57
0.21
0.82
4.94
0.60
0.22
0.79
Velocity,Acceleration,Stops/mile
4.98
0.70
0.21
0.82
Velocity
5.02
0.55
0.22
0.84
5.13
0.56
0.23
0.84
Velocity,Velocity
5.40
0.58
0.24
0.83
5.48
0.59
0.24
0.83
Velocity,Stops/mile
3
Velocity,Velocity
6.65
0.93
0.29
0.65
6.90
0.80
0.30
0.80
Velocity,Velocity3,Stops/mile
7.94
0.78
0.36
0.83
Velocity,%Idle
8.43
1.02
0.38
0.51
8.70
0.83
0.39
0.82
Velocity,Velocity
9.81
0.90
0.44
0.82
9.82
0.91
0.43
0.80
Velocity,Stops/mile
10.00
0.92
0.44
0.80
Velocity,InertialPower,Stops/mile
10.20
0.93
0.44
0.80
Velocity,Velocity2
10.57
0.95
0.48
0.82
Velocity,%Idle
10.75
1.02
0.49
0.66
Velocity,Stops/mile
13.61
1.11
0.60
0.78
Velocity,Stops/mile,%Idle
14.21
1.19
0.64
0.77
Velocity,%Idle
14.92
1.23
0.68
0.78
14
Velocity,Velocity2,%Idle
16.77
1.33
0.76
0.79
16.99
1.34
0.77
0.79
18.07
2.45
0.80
0.29
Velocity,Acceleration,%Idle
22.69
1.64
1.02
0.69
Velocity,InertialPower,%Idle
23.11
1.66
1.04
0.68
Velocity,%Idle
23.80
1.70
1.07
0.65
Velocity,Stops/mile
24.83
1.85
1.12
0.52
Velocity,Std.Dev.Velocity,Stops/mile
24.94
1.89
1.10
0.73
26.05
1.95
1.15
0.73
28.75
1.99
1.29
0.61
Velocity,Velocity ,Velocity
30.34
2.12
1.36
0.64
Velocity,Std.Dev.Velocity,%Idle
31.80
2.30
1.42
0.39
37.08
2.71
1.65
0.67
2
Velocity,Acceleration,InertialPower
3
Velocity,Acceleration ,Stops/mile
2
37.86
2.75
1.68
0.68
Velocity,Std.Dev.Velocity,Acceleration2
43.97
3.05
1.95
0.73
45.71
2.97
2.04
0.30
49.72
3.33
2.20
0.74
50.97
3.40
2.26
0.74
Velocity,InertialPower,Acceleration
2
2
Velocity,Acceleration,Acceleration
57.07
3.70
2.52
0.72
60.73
3.88
2.69
0.70
62.59
3.97
2.77
0.68
2
65.35
4.10
2.89
0.66
Velocity,Velocity2
102.41
6.40
4.54
0.62
Velocity,Std.Dev.Velocity,InertialPower
131.90
8.43
5.85
0.64
139.03
8.41
6.19
0.15
142.47
8.89
6.32
0.60
148.27
9.25
6.57
0.59
152.99
9.54
6.78
0.59
185.66
11.25
8.26
0.24
Velocity,Velocity
225.89
14.17
10.03
0.58
Velocity,Velocity2
252.09
15.27
11.21
0.29
294.59
18.15
13.07
0.52
Velocity,Acceleration ,%Idle
341.75
31.76
15.02
0.00
346.92
21.57
15.41
0.54
Velocity,Velocity
506.02
30.91
22.50
0.38
544.22
32.90
24.18
0.34
Velocity,Velocity ,Std.Dev.Velocity
820.60
50.52
36.45
0.48
2835.30
175.90
126.03
0.48
Velocity,Std.Dev.Velocity,Velocity2
12207.33
746.53
542.24
0.43
Velocity,Velocity2,Acceleration
2
15
Table6.BestpredictionsforCO2massrate.
BaselineCyclesUsed
Metric(s)Used
Cruise
TransientandCruise
Velocity,Accelerationand
Stops/mile
VelocityandAcceleration
Velocity
Average
%Error
Average
Error
(g/s)
Max.
Error
(g/s)
R2
4.29
0.52
2.00
0.82
4.36
4.36
4.89
0.52
0.54
0.58
2.02
2.43
2.16
0.82
0.83
0.81
4.59
4.83
4.94
Average
Error
(mpg)
0.20
0.21
0.22
Max.
Error
(mpg)
0.67
0.57
0.60
0.82
0.82
0.79
4.98
0.21
0.70
0.82
Table7.Bestpredictionforfueleconomy(mpg).
BaselineCyclesUsed
Average
%Error
MetricsUsed
Cruise
VelocityandInertialPower
Velocity,Accelerationand
Stops/mile
R2
Scatterplot of Average % Error vs CASE USED

Property Predicted = CO2
12
11
Average % Error
10
10
9
8
7
6
5
4
0
2
3
CASE USED
METRIC USED
Velocity
Velocity , %Idle
Velocity , Acceleration
Velocity , Acceleration^2
Velocity , Inertial Power
Velocity , Std.Dev.Velocity
Velocity , Stops/mile
Velocity , Velocity^2
Velocity ,Velocity^3
Velocity, Acceleration, %Idle
Velocity, Acceleration, Acceleration^2
Velocity, Acceleration, Inertial Power
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Acceleration, Stops/mile
Inertial Power, %Idle
Inertial Power, Acceleration^2
Inertial Power, Stops/mile
Velocity^2, %Idle
Velocity^2, Inertial Power
Velocity^2, Stops/mile
Velocity^3, %Idle
Velocity^3, Inertial Power
Velocity, Velocity^3, Stops/mile

Velocity, Velocity^3, Velocity^2
Figure4.ScatterplotofaveragepercenterrorforCO2.
16
Scatterplot of Average % Error vs CASE USED

Property Predicted = mpg
12
METRIC USED
Velocity
Velocity , %Idle
11
Velocity , Acceleration
Average % Error
10
10
Velocity , Acceleration^2
Velocity , Inertial Power
Velocity , Std.Dev.Velocity
Velocity , Stops/mile
Velocity , Velocity^2
Velocity ,Velocity^3
Velocity, Acceleration, %Idle
Velocity, Acceleration, Acceleration^2
7
6
5
4
0
2
3
CASE USED
Velocity,
Velocity,
Velocity,
Velocity,
Acceleration, Inertial Power

Acceleration, Stops/mile
Inertial Power, %Idle
Inertial Power, Acceleration^2
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity,
Velocity^2,
Velocity^2,
Velocity^2,
Velocity^3,
Velocity^3,
Velocity^3,
Velocity^3,
Inertial Power, Stops/mile

%Idle
Inertial Power
Stops/mile
%Idle
Inertial Power
Stops/mile
Velocity^2
Figure5.Scatterplotofaveragepercenterrorforfueleconomy(mpg).
Table8.Weightingfactorssensitivitytochangesinbaselinecyclesandmetricsused.
Weights
Idle
Creep
Transient
Cruise
0.1519
0.0243
1.0488
0.0788
0.1326
0.1672
0.2944
1.0552
1.0995
0.8433
1.2944
MetricsUsed
Velocity,Acceleration,
Stops/mile
0.0774 VelocityandAcceleration
0.0677 VelocityandAcceleration
0.1567
Velocity
Velocity
Fuel
CO2
Economy
Average
Average
Error(%)
Error(%)
4.98
4.29
4.59
4.83
5.02
4.97
4.36
4.89
4.36
8.67
17
Fue
el Consu
umption Predicttion
16
data
parity Line
e
Predicted CO2 (g/s)
14
12
10
10
12
14
16
Measu
ured CO2
2 (g/s)
BasselineCyclesUsed
MetricsUssed
Average
A
%
%Error
Avverage
Errror(g/s)
Max.
M
E
Error
(
(g/s)
R2
Id
dle,Transientand
Cruise
Velocityand
Acceleration
4.36
0.52
2.02
0.82
2
Figgure6.ResultssforCO2predicctionfor56tru
ucksusingreco
ommendedbaaselinecyclesaandmetrics.
18
Fu
uel Econ
nomy Predictio
P
on
6
data
parity line
Predicted mpg
Mea
asured mpg
m
elineCyclesU
Used
Base
MetricsUssed
Average
A
%
%Error
Avverage
E
Error
(m
mpg)
Max.
M
E
Error
(m
mpg)
R2
Idle,TTransientandCruise
Velocityand
Acceleration
4.59
0
0.20
0
0.67
0.82
Figgure7.Resultsformpgprediictionfor56trrucksusingreccommendedbaaselinecyclesandmetrics.
19
Inferred Fuel Economy from Fuel Consumption Prediction
Inferred fuel economy (mpg)
6
5
data
parity line
4
3
2
1
0
0
Measured fuel economy (mpg)
BaselineCyclesUsed
MetricsUsed
Average
%Error
Average
Error
(mpg)
Max.
Error
(mpg)
R2
Idle,Transientand
Cruise
Velocityand
Acceleration
4.40
0.19
0.68
0.76
Figure8.Inferredfueleconomybasedonfuelconsumptionprediction.PredictedCO2massratewasconverted
tofueleconomyinmpgusingequations15and16(SeePredictedpropertiessection).
6.6Extrapolation
Other predictions performed using the recommended combination of Idle, Transient and Cruise as
baseline cycles and average velocity and average acceleration as metrics were tested with the High
SpeedCruisecycleoftheHDDTscheduleastheunseencycle.Thispredictionshowedthatthemethod
did not extrapolate well (above 30% error), tending to overpredict fuel economy. Care must be taken
when predicting cycles with average speeds higher than the maximum average speed of the baseline
cycles.
PredictionoftheCruisecycleusingIdle,Creep,andUDDScycles,alongwithvelocityandaccelerationas
metrics,resultedinhigherrors(38%error).Oncemore,themethoddidnotextrapolatewell.
The inaccuracy in the prediction using extrapolation could only be circumvented when the functional
form assumed by the method (in this case a plane) accurately represented the nature of the function
beingextrapolated.Thenatureoftheaerodynamicdragtermintheroadloadequation(cubicvelocity
dependence) could affect the results when using baseline cycles without significant aerodynamic drag
contribution (perhaps at speeds below 40 mph or so). Also, the High Speed Cruise cycle involved
steadierengineoperationthantheTransientorCruisecycles,afactdiscussedindetailinrecentpapers
[15,16,17]. A recommendation is to always include a relatively high speed cycle in the baseline cycles
20
withthecaveatthathighspeedisarelativetermandengineeringjudgmentplaysanimportantrolein
thelinearmodeldevelopment.
6.7UDDSTransientInterchange
ThesuggestedsimilaritybetweentheUDDScycleandTransientcyclewasinvestigatedbyinterchanging
thesecyclesintheanalysis(usingtheUDDSasabaselinecyclealongwiththeIdleandCruisecycleswith
velocityandaccelerationmetrics).Thepredictionworkedwellwitha4.8%averageerror.Thefollowing
equationsshowthevaluesofthebaselinecyclesusedinbothcases.
PredictingUDDScyclewithIdle,Transient,andCruiseasbaselinecycles:
0.1326
1.0552
0.0774
PredictingtransientcyclewithIdle,UDDS,andCruisecycles
0.1257
0.9476
0.0733
As expected, the weight coefficient for UDDS was also nearly one. Note also that the sign of idle and
cruisewereinterchangedmeaningthatUDDShaslessIdlecontributionthantheTransientcycleandthe
TransientcyclehaslessCruisecyclecontributionthantheUDDScycle.
6.8Recommendations(TruckData)
Based on the results, the recommendation is to use average velocity and average acceleration as
metrics.Ifanothermetric(andbaselinecycle)isgoingtobeaddedtothemodel,itisrecommendedto
usestopsperunitdistanceastheadditionalmetric.
Itisnecessarytohavebaselinecycleswhicharesufficientlydissimilarsothattheyprovideawidebasis
for establishing the metricdependent behavior. The best combination in terms of accuracy and
economyistheuseofIdle,Transient,andCruisecycleinthiswork.Basedonactualtruckoperationas
found in the field, it is believed that these three cycles, or cycles with similar characteristics, would
providesufficientdatatopredictothercycles(oractualvehicleactivity).However,itisrecognizedthat
there are always exceptions and engineering judgment will need to be used to determine the most
appropriate baseline cycles that should be used. This combination of cycles has a good diversity of
characteristicsintermsofthemetrics used,ithasazerovelocity,zeroaccelerationcycle(Idle),alow
velocity, high acceleration cycle (Transient), and a high velocity low acceleration cycle (Cruise). Cycles
shouldcoverabroadrangeofoperationinordertobesuitableasbaselinecycles.
21
DuetothefactthattheIdlecyclefueleconomy(mpg)waszeroandsomeinformationwaslostduring
themodeling,itisrecommendedthatthepredictionbemadeintermsofCO2massrate(g/s)andthen
converttofueleconomy(mpg).Extrapolationshouldbeusedwithcare.
7.BusData
The second vehicle type chosen was 40 foot transit buses. Chassis dynamometer data from two
conventionaldieselbuses,twocompressednaturalgasbuses,andonehybridbuswereusedinthispart
oftheresearch.TheU.S.DepartmentofEnergy(DOE)andtheU.S.DepartmentofTransportation(DOT)
sponsoredtheCenterforAlternativeFuels,Engines,andEmissions(CAFEE)ofWestVirginiaUniversity
(WVU) to conduct the program in cooperation with Washington Metropolitan Area Transit Authority
(WMATA)[14].Table9showsrelativeinformationofthebuses.
Table9.Transitbusesanalyzedinthisresearch.
Bus
WMATA
Technology
Number BusNo.
Manufacturer
BusType
&Model
Year
EngineType&
ModelYear
RG6081
280hp/206kW
2005
CumminsCG
280hp/206kW
2005
GVW
(kg)
Available
Cycles
19,334
16
19,334
16
32
2640
CNG
JohnDeere
Orion
2005
35
2503
CNG
Cummins
Orion
2005
Cummins
ISL2802005
18,416
16
37
6003
Hybrid
Allison
New
Flyer
2005
39
9654
Diesel
DDC
Orion
1992
DDCS50
275hp/202kW
2003
17,896
17
41
6150
Diesel
Cummins
New
Flyer
2006
Cummins
ISM2802006
18,416
17
7.1CyclesUsed
Seventeendifferentcycleswereavailableforthedieselbusesandsixteencycleswereavailableforthe
CNGandhybridbuses.Table10showstheavailablecyclesandaveragemeasuredpropertiesofvelocity,
acceleration and stops/mile. The measured Idle cycle average velocity was not equal to zero because
sometestspresentedwheelspeednoiseduringmeasurementasdescribedaboveforthemotivationto
defineidleandstopatvelocitiesunder0.5mph.Targetspeedtimetracesforthesecyclesareshownin
AppendixC.
22
Table10.Averagemeasuredpropertiesover5busesfor17cyclesused.
#
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
CycleName
CARBIdleCycle
NewYorkBusCycle
ParisCycle
ManhattanCycle
WashingtonMetroTransitAuthority
Cycle
NewYorkCompositeCycle
OrangeCountyTransitAuthorityCycle
CentralBusinessDistrictCycle
BraunschweigCycle
CitySuburbanHeavyVehicleCycle
BeelineCycle
EuropeanTestCycleUrban
CARBTransientCycle
UrbanDynamometerDrivingSchedule
KingCountyMetroCycle
ArterialCycle
CommuterCycle
CycleID
Idle
NYBus
PARIS
Manhattan
AverageMeasuredProperties
(over5buses)
Velocity
Acceleration
Stops/mile
(mph)
(mph/s)
0.06
0.00
0.00
3.54
0.25
18.71
6.68
0.32
12.40
6.86
0.37
9.76
WMATA
8.43
0.30
6.26
NY-Comp
OCTA
CBD
BRAUN
CSHVC
Beeline
ETCURBAN
TRANS
Test_D
KCM
ART
COMM
8.75
12.22
13.08
13.88
14.02
14.03
14.11
15.33
18.71
23.35
25.58
44.37
0.15
0.41
0.52
0.45
0.27
0.43
0.28
0.28
0.20
0.41
0.55
0.18
7.21
4.87
6.80
4.19
2.29
3.73
1.92
1.69
2.04
1.85
1.47
0.17
7.2MetricsandCasesUsed
Basedontheresultsfromthetruckstudy,averagevelocityandaverageaccelerationwereselectedto
perform the analysis. For each bus, all possible combinations of three baseline cycles were used to
predict CO2 mass rate (g/s) over the remaining thirteen or fourteen cycles. That is a total of 3040
predictions were made including 560 combinations among 16 cycles for three buses and 680
combinationsamong17cyclesfortwobuses.
7.3BusDataResultsandAnalysis
As anticipated, not all combinations of baseline cycles were suitable to perform a good prediction.
Figure 9 shows a histogram of average percent error lower than 200%. Approximately 44% of the
combinationsproducederrorslessthan20%.Notethatsomecombinationsofcyclesproducedveryhigh
predictionerrors.Therewereerrorsabovethe200%upperscalevalueshowninthefigurebutwerenot
plotted.Themethodisnotsuitableifthecyclesareillchosen.Forexample,usingtwoverylowspeed
cycles plus idle to predict high speed behavior is fraught with difficulty, and is exacerbated if a poor
combination of metrics is chosen. Note that average percentage error in this case is among 13 or 14
predicted cycles for the same vehicle while in the previous data set (truck data) the average percent
errorwasamong56vehiclesoverthesamecycle(UDDS).Oneshouldnotattempttocomparetheerrors
fromthetwodatasets;theseerrorsarecomparablewithinthesamedatasetonly.
23
Histogram of % Error for CO2 (g/s) prediction.

700
600
Frequency
500
400
300
200
100
0
25
50
75
100
% Error
125
150
175
200
Figure9.Histogramofaverage%Erroramong13or14predictions.
Tables 11 and 12 show the best predictions for each bus for CO2 mass rate and fuel economy,
respectively.NotethattheIdlecycleispresentinthebestpredictionsforallvehicles.Figures10to19
showparityplotsofmeasuredversuspredictedvaluesofCO2massrateandfueleconomyforthebest
predictionsforbuses32,35,37,39,and41.Byvisualinspectionoftheslopeofthelinearregressionline
and R2 correlation coefficients values it can be seen that CO2 predictions were better than the fuel
economyprediction.Moreover,CO2predictionerrorswerebelow8.88%whilefueleconomyprediction
errors were 15.89% or worst. The best baseline cycle combination for CO2 was usually not the best
baselinecyclecombinationforfueleconomy,buttheIdlecyclewaspresentinboththebestCO2mass
rate and best fuel economy models. Most of the combinations included a cycle with relatively high
averagevelocitysuchasCOMM,ART,KCM,orTest_Dandatransientcyclewithrelativelylowaverage
velocity such as WMATA, Paris, or OCTA. The KCM cycle appears to be valuable because it worked
relatively well both as a low speed cycle or as a high speed cycle depending on the baseline cycle
combination.TheKCMcycleaveragevelocitywasapproximately50%ofthehighestaveragevelocityof
the cycles used (COMM cycle has an average velocity of 44.37mph) and the KCM cycle average
acceleration was approximately 75% of the highest average acceleration (ART cycle with 0.55mph/s).
Overall, values of average velocity above 18.7 mph produced average percentage errors below 10%
when usedashighspeed,lowaccelerationbaselinecycleandvaluesofaccelerationabove0.3 mph/s
produced average percentage errors below 10% when used as low speed, high acceleration baseline
cycles.BasedontheresultsshowninFigure14,hybridvehiclesappeartobeabletobemodeledusing
the linear modeling methodology presented here. However, additional hybrid vehicles and hybrid
vehiclearchitecturewillneedtobeinvestigatedtosubstantiatethisclaim.
7.4Extrapolation
Fordieseltransitbuses(Figure16andFigure18)fuelconsumptionwasunderpredictedfortheCOMM
cycleduetoextrapolation(predictingacyclewithhigheraveragevelocitythanthebaselinecyclewith
highest average velocity) and this brought the linear regression line down for the CO2 mass rate plot
24
causingtheslopeintheregressionlinetobelessthanone.However,itappearsthataslongasahigh
speed cycle was used as a baseline cycle, the extrapolation would not produce significant deviations
between the measured and estimated values. CNG bus 35 (Figure 12) illustrated an exception to this
observation. The best results for bus 35 were obtained when using baseline cycles idle, Paris and
WMATA.Thehighestaveragevelocitybaselinecycleinthiscasewas8.43mphfortheWMATAcycleand
wasabletopredicttheCOMMcyclewithanaveragevelocityof44.37mphwithin1.12g/s.Inthiscase
higheraveragespeedextrapolationispossible.
Table11.BestCO2massrateresultsforeachbus.
CycleName
BusID
32
35
37
39
41
CycleA
CycleB
CycleC
%Error
Idle
Idle
Idle
Idle
Idle
Beeline
PARIS
OCTA
KCM
PARIS
KCM
WMATA
KCM
ART
KCM
8.48
7.53
8.17
8.70
8.88
Average
Error(g/s)
0.54
0.53
0.51
0.83
0.76
CO2(g/s)
Max.Error Std.Dev.
(g/s)
Error(g/s)
1.12
0.37
1.20
0.37
1.29
0.40
2.45
0.68
2.33
0.65
R2
0.92
0.95
0.98
0.93
0.93
Table12.Bestfueleconomy(mpg)resultsforeachbus.
CycleName
BusID
32
35
37
39
41
CycleA
Idle
Idle
Idle
Idle
Idle
CycleB
CycleC
%Error
Manhattan
BRAUN
OCTA
Manhattan
Manhattan
OCTA
KCM
COMM
KCM
KCM
15.89
19.73
22.26
20.34
22.17
FuelEconomy(mpg)
Average
Max.Error Std.Dev.
Error(mpg)
(mpg)
Error(mpg)
0.67
2.61
0.67
0.70
1.57
0.58
1.19
2.43
0.93
0.85
2.12
0.76
1.01
2.19
0.77
R2
0.77
0.60
0.18
0.38
0.39
25
Best Prediction CO2(g/s) CNG Bus #32

16
Data
Parity line
Linear fit
14
12
Predicted
10
10
12
14
16
Measured
CycleName
BusID
32
CycleA
CycleB
CycleC
%Error
Idle
Beeline
KCM
8.48
Average
Error(g/s)
0.54
CO2 (g/s)
Max.Error
(g/s)
1.12
Std.Dev.
Error(g/s)
0.37
R2
0.92
Figure10.BestCO2massrateprediction,Bus#32.
Best Prediction MPG(miles/gallon) CNG Bus #32
10
Data
Parity line
Linear fit
Predicted
10
Measured
CycleName
BusID
CycleA
32 Idle
CycleB
CycleC
Manhattan OCTA
%Error
15.89
FuelEconomy(mpg)
Average
Max.Error
Std.Dev.
Error(mpg)
(mpg)
Error(mpg)
0.67
2.61
0.67
R2
0.77
Figure11.Bestfueleconomyprediction,Bus#32
26
Best Prediction CO2(g/s) CNG Bus #35

18
Data
Parity line
Linear fit
16
14
Predicted
12
10
10
12
14
16
18
Measured
CycleName
BusID
35
CycleA
CycleB
CycleC
%Error
Idle
PARIS
WMATA
7.53
Average
Error(g/s)
0.53
CO2 (g/s)
Max.Error
(g/s)
1.20
Std.Dev.
Error(g/s)
0.37
R2
0.95
Best Prediction MPG(miles/gallon) CNG Bus #35
7
Data
Parity line
Linear fit
Predicted
Measured
CycleName
BusID
CycleA
35 Idle
CycleB
BRAUN
CycleC
KCM
%Error
19.73
FuelEconomy(mpg)
Average
Max.Error
Std.Dev.
Error(mpg)
(mpg)
Error(mpg)
0.70
1.57
0.58
R2
0.60
Figure13.Bestfueleconomyprediction,Bus#35.
27
Best Prediction CO2(g/s) Hybrid Bus #37

20
Data
Parity line
Linear fit
18
16
Predicted
14
12
10
10
12
14
16
18
20
Measured
Bus
ID
37
CycleName
CycleA
Idle
CO2(g/s)
CycleB CycleC
OCTA
%Error
KCM
8.17
AverageError
(g/s)
0.51
Max.Error
(g/s)
1.29
Std.Dev.Error
(g/s)
0.40
R2
0.98
Best Prediction MPG(miles/gallon) Hybrid Bus #37
9
Data
Parity line
Linear fit
Predicted
Measured
CycleName
BusID
CycleA
37 Idle
CycleB
OCTA
CycleC
COMM
%Error
22.26
FuelEconomy(mpg)
Average
Max.Error
Std.Dev.
Error(mpg)
(mpg)
Error(mpg)
1.19
2.43
0.93
R2
0.18
28
Best Prediction CO2(g/s) Diesel Bus #39

Data
Parity line
Linear fit
20
Predicted
15
10
10
15
20
Measured
CycleName
BusID
39
CycleA
CycleB
CycleC
%Error
Idle
KCM
ART
8.70
Average
Error(g/s)
0.83
CO2 (g/s)
Max.Error
(g/s)
2.45
Std.Dev.
Error(g/s)
0.68
R2
0.93
Best Prediction MPG(miles/gallon) Diesel Bus #39
8
Data
Parity line
Linear fit
Predicted
Measured
CycleName
BusID
CycleA
39 Idle
CycleB
CycleC
Manhattan KCM
%Error
20.34
FuelEconomy(mpg)
Average
Max.Error
Std.Dev.
Error(mpg)
(mpg)
Error(mpg)
0.85
2.12
0.76
R2
0.38
29
Best Prediction CO2(g/s) Diesel Bus #41

Data
Parity line
Linear fit
20
18
16
Predicted
14
12
10
10
12
14
16
18
20
Measured
CycleName
BusID
41
CO2(g/s)
CycleA
CycleB
CycleC
%Error
Idle
PARIS
KCM
8.88
AverageError Max.Error
(g/s)
(g/s)
0.76
2.33
Std.Dev.
Error(g/s)
0.65
R2
0.93
Best Prediction MPG(miles/gallon) Diesel Bus #41
8
Data
Parity line
Linear fit
Predicted
Measured
CycleName
BusID
CycleA
41 Idle
CycleB
CycleC
Manhattan KCM
%Error
22.17
FuelEconomy(mpg)
Average
Max.Error
Std.Dev.
Error(mpg)
(mpg)
Error(mpg)
1.01
2.19
0.77
R2
0.39
30
7.5ComparisonbetweenCO2andFuelEconomyPredictions
ThepredictionofCO2massratetendstobeabettermethodofpredictionthanadirectfueleconomy
prediction. This stems from the fuel economy at idle being equal for all vehicles (a value of zero by
definition),whereasCO2massratefortheIdlecyclewasdifferent(andnonzero)foreachvehicle.Using
CO2massrateprovidesadditionalinformationinthemodel.Anotherpossibleexplanationisthemore
lineardependencebetweenCO2andspeedthanfueleconomyandspeedasisshownbycomparisonof
R2correlationcoefficientsoflinearfitsinFigure20.Thenonlinearityoffuelconsumptionorofdistance
specificemissionslevelswithrespecttoaveragespeediswelldocumentedinspeedcorrectionfactors
whichareoftenusedinemissionsinventorymodels,suchasEMFAC.
CO2(g/s)ormpg
20
18
y=0.357x+2.888
R=0.954
16
14
12
10
8
6
4
y=0.112x+2.255
R=0.611
2
0
10
20
30
40
50
AverageSpeed(mph)
mpg
CO2(g/s)
Linear(mpg)
Linear(CO2(g/s))
Figure20.CO2massrateandfueleconomyasfunctionsofaveragespeed.
7.6BestBusBaselineCycles
The key to obtaining a prediction with low error is the selection of suitable combinations of baseline
cycles.Agivenbaselinecycleisnotgoodorbadbyitself.Rather,itisthecombinationofindividual
baselinecyclesthatmakesthepredictiveapproachsuitableornot.Also,abaselinecyclecanhavelow
errorswithsomemetricsbuthavehigherrorswithothermetrics.Further,thesuccessoftheapproach
mayvaryaccordingtotheapplicationbecauseinsomecasesthebestaveragepredictionforafleetof
vehiclesmaybetheobjective,whereasinotherapplicationstheremaybeaneedtoconstraintheworst
individual error in the prediction. To this end, an analysis was performed to identify which cycles are
moreassociatedwithpredictionswithlowerrorsusingaveragevelocityandaverageacceleration.Figure
21showsahistogramofthefrequencyofagivencycleinthecombinationswithpercentageerrorless
than10%.TheanalysiswasdoneusingtheCO2massratedata.TheplotsshowthattheIdlecycleshould
be used in every prediction. Note also that cycles like ETCURBAN or NYBus are not present in any
combination with error less than 10%. These two cycles are transient, low speed but their values of
accelerationarebelow0.3mph/s.
31

Histogram of Baseline Cycles in cases with %Error less than 10%
160
140
Frequency
120
100
80
60
40
20
0
p
A IS
D
A
S
N
C
N
D M
T ne
M an us
le
A R eli RA U C B OM S HV RBA Id KC att YB om C T AR est_ RAN AT
O
C
M
e
P
N
h
C
B
T W
T
C CU
B
an
NY
M
ET
Baseline Cycles
Figure21.Frequencyofabaselinecycleinpredictionwithaverageerrorbelow10%.
7.7BestBusBaselineCombinations
Table13showsthethreecombinationsofbaselinecycleswiththelowesterrorforthepredictionofCO2
massrateforthefivebusesconsidered.ThecombinationofIdlecycle,arelativelylowaveragespeed
with high average acceleration cycle, and a relatively high speed cycle was present in the three
combinations. Note that these combinations produced prediction results with less than 10% average
errorforallbusesoveraverywiderangeofcycles.
Table13.BestbusbaselinecombinationsresultsforCO2(g/s)prediction.
AveragePercentageError(%)
BaselineCycles
Used
IdleOCTAKCM
IdlePARISKCM
IdleOCTACOMM
CNG#32 CNG#35 Hybrid#37 Diesel#39 Diesel#41

8.61
9.11
8.52
7.78
7.55
7.80
8.17
8.89
8.38
8.75
8.72
9.21
9.26
8.88
9.80
Average
Max.
Percent.
Absolute
Error(%)
Error(g/s)
(5buses)
8.51
3.95
8.63
3.18
8.74
2.20
7.8OtherPossibleBusBaselineCycleCombinations
Otherbaselinecombinationsidentifiedinthebusdatawithaveragepercentageerrorbelow10%among
fivebusesaresummarizedinTable14.AllcombinationsincludeIdlecycle,andmostofthemincluded
one intermediate transient cycle and one relatively high speed cycle. Baseline cycles should cover a
broadrangeofoperationencompassingtheunseencycle.
32
Table14.Combinationsofbaselinecycleswithaveragepercentageerrorbelow10%.
AveragePercentageError(overpredictedcycles)
Bus32 Bus35 Bus37
Bus39
Bus41
IdleKCMOCTA
8.61
7.78
8.17
8.75
9.26
IdleKCMPARIS
9.11
7.55
8.89
8.72
8.88
IdleKCMManhattan
8.51
8.08
NA
8.84
9.14
IdleManhattanART
NA
8.10
NA
8.86
9.21
IdleOCTACOMM
8.52
7.80
8.38
9.21
9.80
IdlePARISART
NA
7.54
9.95
8.76
9.05
IdleWMATAKCM
9.00
7.56
8.20
9.73
10.42
IdleOCTATest_D
10.16
8.41
8.24
9.56
9.54
IdleARTCOMM
NA
7.79
9.89
9.61
9.80
IdleWMATACOMM
9.02
7.74
8.67
10.68
12.06
IdleWMATATest_D
10.18
8.13
8.38
10.35
11.13
IdleWMATAART
NA
7.53
10.45
10.17
10.45
IdleKCMCOMM
8.48
8.57
8.43
11.14
11.67
IdleBRAUNCOMM
9.41
8.74
11.17
9.34
9.75
IdlePARISTest_D
10.94
8.15
9.55
10.07
9.84
IdleOCTAART
NA
7.80
11.52
8.76
10.76
IdleBeelineKCM
8.48
9.15
8.38
9.66
13.38
IdleTest_DART
NA
9.03
10.28
10.59
9.45
IdlePARISCOMM
9.75
7.85
10.47
10.44
10.93
IdleBeelineCOMM
8.58
8.52
8.77
9.97
13.71
IdleOCTATRANS
8.94
10.55
8.72
11.28
10.37
IdleManhattanCSHVC 9.66
9.53
NA
9.74
11.00
IdleBeelineTest_D
9.95
8.56
8.52
9.81
13.13
NANotAvailable.
Cycles
AveragePercentageError
8.51
8.63
8.64
8.72
8.74
8.82
8.98
9.18
9.27
9.63
9.63
9.65
9.66
9.68
9.71
9.71
9.81
9.84
9.89
9.91
9.97
9.98
9.99
7.9NonCompatibleBusCycleCombinations
Anoncompatiblepairisdefinedasapairofcyclesthatproducepredictionaveragepercentageerrors
above50%whenusedasbaselinecycles.Somenoncompatiblecyclepairswereidentified.Beelinecycle
wasnotcompatiblewith Braun,CBD,ETCURBANorOCTAcycles.ETCURBANwasnotcompatiblewith
CSHVC, COMM, KCM or ART cycles. It is clear that it is necessary to have cycles which are sufficiently
dissimilar so that they provide an accurate basis for establishing the metricdependent behavior. For
exampleCSHVCandETCURBANresultedinveryhighaveragepredictionerrorsofabove2000%.Thisis
believed to be due to the similarity between their metrics (average velocities of 14.02mph and
14.11mph and average accelerations of 0.27mph/s and 0.28mph/s) so that the ability to extrapolate
usingthedataisseverelycurtailed.Definingaplanebasedonthreepointswhentwopointsarein(or
very near) the same location, results in an illdefined surface. Although matrix inversion (solving of
simultaneousequations)ispossibleinthatscenario,thesolutionbecomessensitivetochangesinone
particular metric. It is speculated that experimental error then becomes a first order effect in the
analysis.
33
7.10BestBusDataCombination:Idle,OCTA,andKCM
Table 15 shows the results for the combination of baseline cycles with the lowest error. Figure 22
furtherillustrateshowthiscombinationpredictedeachcycletoahighfidelityandcouldbeusefultosee
whenfuelconsumptionforonecycleisconsistentlyoverpredictedorunderpredicted.Forexampleitcan
beseenthatCBD,CSHVC,ETCURBAN,andTransientcyclesfuelconsumptionwereoverpredictedforall
ofthebuses.TheCOMMcyclefuelconsumptionwasmostlyunderpredicted(particularlyforthediesel
buses)becauseextrapolationappearedtobeunabletoaccountfortheaerodynamicdragportionofthe
road load equation. Figure 23 shows a parity plot of measured versus predicted values for all the
predictions using the recommended combination. Note that extrapolation caused subtle
underpredictions.
Table15.Resultsforrecommendedcombinationofbaselinecyclesforthebusdata.
BusID
35
37
32
39
41
CycleName
CO2(g/s)
Cycle1 Cycle2 Cycle3 %Error AverageError Max.Error Std.Dev.Error

Idle
Idle
Idle
Idle
Idle
OCTA
OCTA
OCTA
OCTA
OCTA
KCM
KCM
KCM
KCM
KCM
7.78
8.17
8.61
8.75
9.26
0.53
0.51
0.50
0.81
0.72
1.14
1.29
1.39
2.11
2.09
0.35
0.40
0.35
0.61
0.57
R2
0.96
0.98
0.95
0.95
0.94
2.5
PredictionErrorCO2(g/s)
2
1.5
1
0.5
0
0.5
1
1.5
2
2.5
CNG#32
CNG#35
Hybrid#37
Diesel#39
Diesel#41
Figure22.Predictionerrorsusingidle,OCTAandKCMasbaselinecyclesandaveragevelocityandaverage
accelerationasmetrics.
34
PredictedCO2(g/s)
25
CNG#32
CNG#35
20
Hybrid#37
15
Diesel#39
Diesel#41
10
ParityLine
5
0
0
10
15
20
25
MeasuredCO2(g/s)
Figure23.ParityplotforpredictionusingIdle,OCTA,andKCMcyclesandvelocityandaccelerationmetrics.
7.11AlternativeMetrics
An exploratory analysis was completed to discern if other metrics had the potential to replace
acceleration or to be used in conjunction with velocity and acceleration to obtain improved results.
Figure24showstheratioofaveragepercentageerrorforalternativemetricsoveraveragepercentage
errorusingvelocityandacceleration.Avaluebelowoneforthisratiomeantthatthealternativemetrics
combination had the potential to have lower errors than the velocity and acceleration combination
alone.Itcanbeseenthatstopspermileandaverageof(velocitysquared)aremetricsthatcouldreduce
theerrorwhencomparedtousingjustthevelocityandaccelerationmetrics.Theplotalsoshowsthat
theaccuracyofpredictioncouldbeimprovedbyaddingonemoremetrictotheanalysisbutatthecost
ofaddingonemorebaselinecycle.Onlythebestpossiblecase(thecasewithminimumaveragepercent
error) was used for this analysis and further research should be done in order to validate these
conclusions. The next section shows a more detailed analysis using stops per mile as an alternative
metric.
35
%Error/%ErrorusingVandA
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
VAIP
VAV^2
Bus39
VAStd.V
VA%Idle
Bus37
VAStops
Bus35
V%Idle
VStops
VIP
VV^2
VStd.V
VA
Bus32
Bus41
Figure24.Percentageerrorratioswhenusingothermetricscombinations.
7.12UsingStopsperMileasaMetric
BasedontheresultsshowninFigure24,thebusdataweremodeledusingthevelocityandstopsper
milemetricswithdifferentcombinationsofthreebaselinecycles.Theresultswiththelowesterrorsare
shown in Table 16. The data was evaluated using velocity, acceleration, and stops per mile using
differentcombinationsoffourbaselinecycles.Table17showstheresultswiththelowesterrorsforthis
case.Itisworthmentioningthattheidlecyclewasstillpresentinallofthecases.
Table16.Bestresultswithvelocityandstops/mile.
BusID
CycleA
CycleB
CycleC
%Error
Average
Error(g/s)
32
35
37
39
41
Idle
Idle
Idle
Idle
Idle
NYBus
NYBus
NYBus
PARIS
NYBus
CBD
KCM
WMATA
Test_D
COMM
6.08
8.99
6.56
9.55
7.82
0.47
0.77
0.46
0.94
0.59
Max.Abs.
Error(g/s)
1.89
2.55
1.23
1.85
1.42
Std.
Dev.
Error
(g/s)
0.49
0.63
0.41
0.45
0.51
R2
0.97
0.94
0.98
0.95
0.89
36
Table17.Busresultswithlowesterrorusingvelocity,acceleration,andstopspermilemetrics.
Bus
ID
CycleA
32
35
37
39
41
Idle
Idle
Idle
Idle
Idle
CycleB
CycleC
CycleD
NYBus
BRAUN
NYBus
WMATA
NYBus
PARIS
NYBus
BRAUN
Manhattan OCTA
COMM
COMM
Test_D
KCM
Test_D
Average
%Error
Error
(g/s)
4.92
5.65
6.97
6.82
8.15
0.32
0.40
0.48
0.63
0.63
Max.
Abs.
Error
(g/s)
0.55
1.13
1.23
1.63
1.33
Std.
Dev.
Error
(g/s)
0.19
0.33
0.42
0.42
0.45
R2
0.91
0.94
0.98
0.98
0.97
7.13BestCombination:Idle,NYBus,andKCM
TheIdle,NYBus,andKCMcombinationofbaselinecyclesresultedinthelowesterrorforthebusdata
whenthestopspermilemetricwasused.Notethatusingvelocityandaccelerationasmetrics,thebest
combinationofbaselinecycleswasIdle,OCTAandKCMandwhenusingvelocityandstopspermileas
metrics,thebestcombinationofbaselinecycleswasIdle,NYBus,andKCM.TheNYBuscyclewasnota
goodbaselinecyclewhenusingaverageaccelerationbutwasoneofthebestcycleswhenusingstops
permile.Table18,Figure25,andFigure26showtheseresultinmoredetail.
Table18.BestresultsusingIdleNYBUsandKCM.Averagevelocityandstops/mile.
CycleName
CO2massrate(g/s)
Bus
ID
CycleA
CycleB
CycleC
%Error
32
37
41
35
39
Idle
Idle
Idle
Idle
Idle
NYBus
NYBus
NYBus
NYBus
NYBus
KCM
KCM
KCM
KCM
KCM
6.32
6.78
7.92
8.99
9.63
Average Max.Abs. Std.Dev.

Error(g/s) Error(g/s) Error(g/s)
0.49
0.45
0.59
0.77
0.96
2.28
1.31
1.37
2.55
1.98
0.57
0.45
0.51
0.63
0.52
R2
0.98
0.98
0.96
0.94
0.94
37
PredictionErrorCO2(g/s)
3
CNG#32
CNG#35
Hybrid#37
Diesel#39
Diesel#41
Figure25.PredictionerrorsusingIdle,NYBus,andKCMasbaselinecyclesandaveragevelocityandstopsper
mileasmetrics.
25
PredictedCO2(g/s)
20
CNG#32
15
CNG#35
Hybrid#37
10
Diesel#39
Diesel#41
ParityLine
0
0
10
15
20
25
MeasuredCO2(g/s)
Figure26.ParityplotforpredictionusingIdle,NYBus,andKCMusingvelocityandstopspermilemetrics.
38
7.14BusDataRecommendations
Average velocity, average acceleration, and stops per mile were the metrics identified to predict CO2
massratefromunseencycles.TheIdlecyclemustbepresentasabaselinecycle,alongwitharelatively
slowtransientcycle(about25%ofthemaximumaveragespeedandaverageaccelerationhigherthan
0.3mph/s)andarelativelyhighspeedcycle,preferablywithanaveragevelocityatorabovetheaverage
velocityoftheunseencycle.Caremustbeusedwhenextrapolatinghigheraveragespeedcyclesrelative
tothecyclesusedintheprediction.
ItisinterestingtonotethatKCMcyclewasidentifiedasasuitablecycleusingbothaveragevelocityand
averageaccelerationasmetricsandaveragevelocityandstopspermileasmetrics.Itappearsasthough
thiscyclehasadvantageouscharacteristicsthatminimizetheerrorsinthelinearmodel.
8.BlackBoxModel(NeuralNetworks)
The second approach used to predict fuel consumption used the training of a neural network with
continuoussecondbyseconddata(instantaneouspropertiesandinstantaneousfuelconsumption)from
baselinecycles.Themodelwasthenusedtopredictsecondbysecondfuelconsumptionoveranunseen
cycle, and the cycleaveraged CO2 emissions mass rate could be determined. Training the neural
networkrequiressomeskillbecauseonemustselectthetrainingdataappropriately.Neuralnetworks
requirealargediversityoftraininginordertocaptureallofthedetailsofthephysicalsystem.However
onemustavoidovertrainingoftheneuralnetwork.
Chassisdynamometerdatafor56heavyheavydutytrucksoperatingatanominal56,000lbswereused
in this part of the research. This data were the same as used to perform the linear modeling for the
truck data. This data were gathered as part of the Coordinating Research Council E55/E59 program,
whichwascreatedtocharacterizeheavydutytrucksemissionsinCalifornia.
Trainingoftheneuralnetworkwasdonewiththesecondbyseconddatafromallofthetransientcycles
and cruise cycles that were available from a given vehicle. Input data included properties such as
instantaneous velocity, instantaneous acceleration, instantaneous square of velocity, instantaneous
cubeofvelocity,andinstantaneousinertialpower.TheoutputvariablewasCO2massrateemissions,in
g/s. One neural network was created and trained for each of the 56 vehicles considered. The neural
network then was used to predict secondbysecond CO2 mass rate emissions for the UDDS cycle
(validationcycle).
Several network architectures were evaluated using the Matlab Neural Network Toolbox [5]. These
architecturesincludeddifferentnumberofhiddennodes,numberofhiddenlayers,anddifferenttypes
of transfer function of the nodes. Different combinations of metrics were also used. The best results
wereobtainedusingabackpropagationneuralnetworkwithtwohiddenlayers,100neuronsinthefirst
layer with a tansigmoid transfer function and one linear neuron in the output layer. The input layer
usedinstantaneousvaluesofvelocity,velocitycubed,andinertialpowerasmetrics.Figure27showsa
schematic of the neural network architecture. The scope present program did not include the
optimizationoftheneuralnetarchitectureforthespecificapplication.
39
Figure27.Schematicofneuralnetworkarchitecture.
8.1NeuralNetworkResults
Figure28showsarepresentativeexampleoftheresultsobtainedforoneparticularvehicle.Itappears
that the neural net was able to capture the overall trend with an acceptable accuracy. However,
between500secondsand800secondoftheUDDScycle,theneuralnetunderpredictedthemeasured
response.Causesforsecondbyseconddifferencesmightbeadscribedtothebinarynatureofcooling
fanload,theuseofadifferentgearthananticipated,ortheuseofanenginecontrolstrategythatnot
anticipatedbythetrainingdataset.Theoverallresultswerecalculatedbyintegratingtheinstantaneous
resultsandaresummarizedinFigure29whereeachdatapointrepresentsoneUDDScycle(Thereare
morethan56datapointsbecauseofrepeatedtests).Themodelwassystematicallyunderestimatingthe
averagefuelconsumptionoftheUDDScycle.
Neuralnetworkanalysishassomedisadvantageswithrespecttothelinearmodel.Themainoneisthat
secondbyseconddataisrequiredandthattherearethousandsofdatapoints(at1Hz)comparedwith
only a few properties used in the linear model. Another disadvantage is that the prediction is not
unique, because the neural network model depends on the selection of training parameters, the
networkarchitecture,andthealgorithmused.
40
PredictedCO2massrate(g/s)
Figure28.NeuralnetworkresultsinthepredictionofUDDScycleCO2massrateemissions.
18
16
14
12
10
8
6
4
2
0
R=0.413
10
15
20
MeasuredCO2massrate(g/s)
Figure29.SummaryofneuralnetworkpredictionofCO2massrate(g/s).Predictionerror10.24%.
9.PSATCommercialSoftwareModel
PSAT is a forwardlooking model that simulates fuel economy and performance in a realistic manner
taking into account transient behavior and control strategy [6]. PSAT is also called a commandbased
model[7].Thenecessarywheeltorquetoreachthedesiredspeedisestimatedbypassinginformation
fromthedrivermodelto thevehiclecontroller(different components),such asthrottlecommandfor
theengine,gearnumberfortransmission,andmechanicalbrakingforwheels[8].Therearethreemain
components losses and they include the inertia of the vehicle, the aerodynamic drag, and the rolling
resistance.Theselossesareaddedtogethertoproducearoughestimateforthetorquedemandatthe
vehicle'swheels.Thendecisionsabouthowdifferentcomponentsworkaremadebasedonthedriver
41
demand and the latest information from the components sensors. Eventually, the vehicle controller
commands (i.e., engine torque) are transformed and can then be used by the respective component
models (i.e. throttle). Briefly, the forwardlooking method works by modeling the command of the
driver which in turn causes the appropriate components response to meet the desired vehicle speed
[7]. As components react to the commands as if a person were driving the vehicle, the user can
implement advanced component models by taking into account transient effects (such as engine
starting, clutch engagement/disengagement, or shifting) or developing realistic control strategies [7].
Figure30showsthePSATgraphicaluserinterface.
Figure30.ThePSATgraphicaluserinterface.
APSATmodelwasdevelopedfora1998Peterbilttruck,witha410hpC12Caterpillarengine.CO2mass
ratewaspredictedfollowingtheprocedurebelow:
1)APSATmodelwasdeveloped(Figure30)forthetrucklistedabove.
2)ThemodelwasrunoverIdleandTransientcycles,andtheresultswerecomparedtomeasuredvalues
inordertofindtheeffectivecoefficientofrollingresistance.
3)Keepingtherollingresistancefixed,themodelwasevaluatedovertheCruisecycle,andtheresults
werecomparedtomeasuredvaluesinordertofindtheaerodynamicdragcoefficient.
4) Steps 2 and 3 were repeated in order to find the best fit between measured and PSAT simulated
results.
5)Finally,themodelwasevaluatedovertheUDDScycleandCO2massratepredictionswerecompared
tomeasuredvalues.
42
9.1PSATResults
Figures31and32showtheparityplotofmeasuredversuspredictedvaluesofCO2massratein(g/s)for
TransientcycleandCruisecycle,respectively.Thecoefficientofrollingresistanceforthispredictionwas
0.0136,andtheaerodynamicdragcoefficient0.37.Thesearebestfitvalues,butthedragcoefficientis
lowandtherollingresistanceishighincomparisontocommonwisdom.Figure33showstheparityplot
of measured versus predicted values of CO2 mass rate in (g/s) for UDDS cycle. Secondby second
prediction of the UDDS cycle is good with a percentage error below 4%. However, the procedure of
selecting components and assembling the model should be repeated for each different vehicle that
needstobesimulated.
TRANS
PSAT Simulated CO2 mass rate (g/s)
70
60
50
40
y = 1.093x - 1.366
R = 0.940
30
20
10
0
-10
10
20
30
40
Measured CO2 mass rate (g/s)
50
60
70
Figure31.CO2massratepredictionresultsforTransientcycle.Predictedmassrate:9.30g/s;Measuredmass
rate:9.76g/s;Predictionerror4.68%.Coefficientofrollingresistance0.0136.
43
CRUISE
70
60
50
40
y = 1.121x - 1.346
R = 0.951
30
20
10
0
-10
10
20
30
40
50
60
70
Figure32.CO2massratepredictionresultsforCruisecycle.Predictedmassrate:19.00g/s;Measuredmassrate:
18.14g/s;Predictionerror4.69%.Aerodynamicdragcoefficient0.37.
UDDS
70
60
50
40
y = 1.073x - 1.334
R = 0.937
30
20
10
0
-10
10
20
30
40
50
60
70
Figure33.CO2massratepredictionresultsforUDDScycle.Predictedmassrate:11.68g/s;Measuredmassrate:
12.13g/s;Predictionerror3.69%.
44
10.Conclusion
Alinearmodelmethodologyforthepredictionofheavydutyvehiclefueleconomybasedonmeasured
chassis dynamometer test cycles and properties of those cycles was developed and verified. The
methodology allowed for the prediction of fuel economy from vehicles operating on a number of
differentchassisdynamometercyclesbasedonrelativelyfewexperimentalmeasurements.Theresults
of the application of the linear model to a set of fiftysix heavy heavyduty trucks operating over five
different cycles showed that the use of average velocity and average positive acceleration as metrics
produced the best results in terms of average percentage error (less than 5%). The results of the
application of the linear model to a set of five buses operating over up to seventeen different cycles
showedagainthataveragevelocityandaveragepositiveaccelerationweresuitablemetricstopredict
fueleconomywithreasonableaccuracy(lessthan10%averagepercentageerror).Ifanothermetric(and
baselinecycle)isgoingtobeaddedtothemodel,itisrecommendedtousestopsperunitdistanceas
the additional metric. It was also found that baseline cycles must include Idle cycle, along with a
relativelyslowtransientcycleandarelativelyhighspeedcycle,preferablywithanaveragevelocityator
abovetheaveragevelocityoftheunseencycle.Basedontheresultsobtainedwithbothdatasets,itwas
recommended that the prediction be made in terms of CO2 mass rate (g/s) and then convert to fuel
economy(mpg).
TwoalternativeapproachesusingneuralnetworksandthecommercialsimulationsoftwarePSATwere
also developed and verified. The results of the application of these modeling strategies produced
average percentage errors of the order of 10% and 4% respectively. The main disadvantages of these
alternative approaches with respect to the linear model were their inherent complexity (application
difficulty)andtheneedtousecontinuous(secondbysecond)data.
45
11.References
[1] Clark, N.N.; Gautam, M.; Wayne, W.S.; Lyons, D.W.; Thompson, G.J.; Zielinska, B., HeavyDuty
Vehicle Chassis Dynamometer Testing for Emissions Inventory, Air Quality Modeling, Source
ApportionmentandAirToxicsEmissionsInventory,CRCReportNo.E55/59,2007.
[2] Melendez, M.; Taylor, J.; Zuboy, J.; Wayne, W.S.; Smith, D., Emission Testing of Washington
MetropolitanAreaTransitAuthority(WMATA)NaturalGasandDieselTransitBuses,TechnicalReport
NREL/TP54036355,2005.
[3]Taylor,S.T.;Clark,N.N.;Gautam,M.;WayneW.S.,DieselEmissionsPredictionfromDissimilarCycle
Scaling,JournalofAutomobileEngineering,Volume218,pp.341352,2004.
[4] Code of Federal Regulations, Title 40, Part 86, Subpart N and Appendix D. Office of the Federal
Register,NationalArchivesandRecordsAdministration,1991.
[5]Demuth,H.;Beale,M.;Hagan,M.,MATLABNeuralNetworkToolbox6UsersGuide.
[6]Ciccarelli,T.;Toossi,R.,FinalReport:AssessmentofHybridConfigurationandControlStrategiesin
PlanningFutureMetropolitan/UrbanTransitSystem,CaliforniaStateUniversityLongBeach,2002.
[7]Slezak,L.AnnualProgressReport:AdvancedVehicleTechnologyAnalysisandEvaluationActivities,
2004.
[8]Rousseau,A.;Pasquier,M.,ValidationofaHybridModelingSoftware(PSAT)UsingItsExtensionfor
Prototyping(PSATPRO),GlobalPowertrainCongress,2001.
[9] The Mathworks, Argonne National Laboratory Develops Powertrain System Analysis Toolkit with
MathWorksTMTools,91552v00,2008.
[10]EnvironmentalProtectionAgency,AverageCarbonDioxideEmissionsResultingfromGasolineand
DieselFuel.Retrievedfromhttp://www.epa.gov/otaq/climate/420f05001.htmonAugust2009.
[11] Gajendran P.; Clark, N.N., Effect of Truck Operating Weight on HeavyDuty Diesel Emissions.
EnvironmentalScience&Technology,Vol.37,No.18,2003.
[12]Gautam,M.;Clark,N.N.;Riddle,W.;Nine,R.;Wayne,W.S.;Maldonado,H.;Agrawal,A.;Carlock,M.,
Development and Initial Use of a Heavy Duty Diesel Truck Test Schedule for Emissions
Characterization,SAETransactionsJournalofFuels&Lubricants,Vol.111pp.812825,2002.
[13] Clark, N.N.; Gautam, M.; Riddle, W. ; Nine, R.; Wayne W.S., Examination of a HeavyDuty Diesel
Truck Chassis Dynamometer Schedule. SAE Powertrain Conference, Tampa, Florida, SAE Paper 2004
012904,2004.
[14] Wayne, W.S.; Clark, N.N.; Khan, A.S.; Gautam, M.; Thompson, G.; Lyons, D.,Regulated and Non
Regulated Emissions and Fuel Economy from Conventional Diesel, HybridElectric Diesel, and Natural
GasTransitBuses.JournalofTransportationResearchForum,Vol.47,No.3,2008.
46

[15] Clark, N.N.; Bedick, C.R.; Wang, L.; Thompson, G.; McKain, D.L.; Ralston, R., Emissions from a
Legacy Diesel Engine Exercised through the ACES Engine Test Schedule, SAE Powertrain Meeting,
Shanghai,China,SAEPaper2008011751,2008.
[16]Zhen,F.;Clark,N.N.;Bedick,C.;Gautam,M.;Wayne,W.S.;Thompson,G.;Lyons,D.,"Development
ofaHeavyHeavyDutyDieselEngineScheduleforRepresentativeMeasurementofEmissions,"Journal
of the Air & Waste Management Association, Vol. 59, No. 8, pp. 950959, 2009.
[17] Bedick, C.; Clark, N.N.; Zhen, F.; Atkinson, R.; McKain, D., "Testing of a Heavy HeavyDuty Diesel
Engine Schedule for Representative Measurement of Emissions," Journal of the Air & Waste
ManagementAssociation,Vol.59,No.8,pp.960971,2009.
47
AppendixAWeightCorrection
Twotypesofweightcorrectiontechniqueswereappliedtoasubsetofthetruckdata.Table19shows
informationoftheeleventrucksused.Testdataforthreedifferentweights(30,000lbs,56,000lbs,and
66,000lbs)overIdle,Creep,Transient,andCruisecycleswereavailableforthesetrucks.
ThefirstweightcorrectiontechniqueconsistedofcalculatingthepercentchangeinCO2emissionsper
percentchangeintestweight.Figure33showsaplotofCO2massratepercentagechangeasafunction
of percentage change in weight for the Creep, Transient, and Cruise cycles. Each point represents the
averagechangeamongthesubsetoftrucks.Theaccompanyingtableshowstheslopesofthelinearfit
foreachcycle.Inthisway,a10%increaseinweightproduceda6%increaseinCO2massrate(g/s)over
thetransientcycle,1.8%increaseinCO2massrateoverthecreepcycle,anda3.7%increaseinCO2mass
rateoverthecruisecycle.NotethatIdlecycleanalysisisnotrelevantbecauseemissionsshouldbethe
sameregardlessoftheweightifthevehicleisnotmoving.
ItcanbeseenthatforeachcycletherewasanincreaseintheCO2emissionswithincreasingweight.The
CO2 emissions are a measure of the energy expended as they directly correspond to the fuel
consumptionofthevehicle.TherelationshipbetweentestweightandCO2emissionsiscycledependant
due to differences in transient behavior. Transient operation involves extra energy spent during
accelerations, which is lost during decelerations (braking). This means that the Transient cycle will
requirethevehicletoexpendhigherenergywithincreasingweightthantheCruiseorCreepcyclesasit
isseeninthevaluesofslopesinthetableaccompanyingFigure33.
It is believed that the change in slope for the Cruise cycle is due to the fact that wind drag losses for
heavydutytrucksbecomeasubstantialcontributiononlyatsustainedspeedsofover50mph[11].
Table19.SubsetofCRCtruckdataforweightcorrectionanalysis.
E55CRC
(truck)
Engine
Vehicle
Vehicle
model
Manufacturer
year
Manufacture
Engine
Engine
Power(hp)
Model
Engine
Disp.
Odometer
(Liter)
Reading
(mile)
27
2000
Freightliner
Detroit
Series60
500
12.7
420927
28
1999
Freightliner
Detroit
Series60
500
12.7
645034
29
2000
Volvo
Cummins
1SX475ST2
450
14.9
120000
30
1999
Freightliner
Detroit
Series60
500
12.7
138625
31
1998
Kenworth
Cummins
N14460E+
460
14
587389
32
1992
Volvo
Caterpillar
3406B
280
14.6
595242
33
1985
Freightliner
Caterpillar
3406
310
14.6
988726
48
34
2004
Freightliner
Detroit
Series60
500
14
19094
35
2001
Sterling
Detroit
Series60
470
12.7
106377
36
2001
Peterbilt
Caterpillar
C15
475
14.6
284553
38
2003
Volvo
Cummins
ISX
530
14.9
2829
Change in CO2 emissions (%)
-50
15
10
5
0
-40
-30
-20
-10
-5
10
Cycle
Slope
Creep
0.18
20
-10
-15
Transient 0.60
-20
Cruise
0.37
-25
-30
Change in weight (%)
Creep
Transient
Cruise
Figure33.percentchangeinCO2emissionsperpercentchangeintestweight
A more theoretical approach makes use of axle horsepower to correlate it with CO2 emissions. Axle
horsepower(AHP)isgivenbytheroadloadequationforalevelroad,asfollows:
(28)
Wheremisthemassofthevehicle(kg),Visthevehiclevelocity(m/s),Aisthefrontalareaofthevehicle
(m2),gistheaccelerationduetogravity(m/s2),CDistheaerodynamicdragcoefficientofthevehicle,is
thetirerollingresistancecoefficient,andisthedensityofair(kg/m3).
Chassisdynamometertestdata canbeused toobtainreliablecorrelationsbetweenaxle horsepower
andCO2massrateemissionsoftheform:
49
Iftheequationabove(valuesofC1andC2)remainsreliableforawidevarietyoftruckapplications,itis
possible to apply these relationships to the axle power known for a given cycle (integration of
instantaneousAHPoverthetestduration)and,therefore,topredictthetotalCO2massarisingfromthat
cycle.Figure34showsvariationofCO2emissionswithaxlepower(testweight)fortheCreep,Transient,
andCruisecycles.Again,therelationshipbetweentestweightandemissionswillbecycledependent;a
highly transient cycle with high loads is likely to emphasize the effect of weight on CO2 production in
comparison to a steadystate operation with the same average speed. Conversely, if a test schedule
contains long periods of idle, the idle CO2 emissions contribution may become significant and will be
weightinsensitive[11].
Scatterplot of CO2 (g/s) vs Axle Energy (hp-hr)

Cy cle
Creep
Cruise
Idle
Transient
20
CO2 (g/s)
15
10
0
0
10
20
30
40
Axle Energy (hp-hr)
50
60
Figure34.VariationofCO2emissionswithaxlepower(testweight)fortheCreep,Transient,andCruisecycles.
50
AppendixBMethodofuse
1. Identifyfleetofvehiclesubsetfromafleettobeanalyzed.
2. Obtainvehiclesetdrivingcharacteristics.Thiscouldbeeithervehiclespeedtimecyclesoron
roadactivity.Thisiscalledunseencycle.
3. Specify cycle metrics that can be used to translate fuel economy between cycles: Average
velocityandaverageacceleration.
4. Calculatemetricsfortheselectedvehiclesetdrivingcharacteristics.
AverageVelocity:Summationofinstantaneousvelocityovernumberofdatapoints.
Average Acceleration: Summation of instantaneous positive acceleration over number of data

points. Instantaneous positive acceleration is calculated using a central differences scheme.
Filteringofspeedtimetracebeforecalculationofaccelerationisrecommended.
when
5. Selectrepresentativechassisdynamometerbaselinecycles.
a. Idlecycle.
b. Low average speed (50% to 100% of average speed of fleet), relatively high acceleration
(100%to150%ofaverageaccelerationoffleet)cycle.
c. Highaveragespeed(morethan150%ofaveragespeedoffleet),relativelylowacceleration
cycle(50%to100%ofaverageaccelerationoffleet)cycle.
6. Perform chassis dynamometer test for each vehicle/drivetrain/engine configuration. Obtain
integratedCO2massemissions,testduration,anddistancetraveled.
7. Pose a linear set of equations using average velocity and average acceleration of the three
baselinecyclesandtheunseencycle.
"
"

"
"
8. Obtainthethreeweightingfactorsbysolvingthelinearsetofequations.
9. ApplytheweightingfactorstopredictthevehicleunseencyclesCO2emissions.
2
"
"
10. ConverttheCO2massrateemissionstofueleconomy.
51
AppendixCBuscycles
Idle cycle
speed (mph)
0.8
0.6
0.4
0.2
0
0
100
200
300
400
500
600
700
800
time (s)
900
Figure35.Idlecycle.
New York Bus cycle
50
speed (mph)
40
30
20
10
0
0
100
200
300
400
500
600
time (s)
Figure36.NewYorkBuscycle.
52
Paris cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
time (s)
Figure37.Pariscycle.
Manhattan cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
time (s)
1000
1200
1400
Figure38.Manhattancycle.
53
WMATA cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
time (s)
Figure39.WashingtonMetroTransitAuthoritycycle.
New York Composite cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
time (s)
1000
1200
Figure40.NewYorkCompositecycle.
54
OCTA cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
time (s)
Figure41.OrangeCountyTransitAuthoritycycle.
CBD cycle
50
speed (mph)
40
30
20
10
0
0
100
200
300
400
time (s)
500
600
Figure42.CentralBusinessDistrictcycle.
55
Braunschweig cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
1000
1200
1400
1600
1800
time (s)
Figure43.Braunschweigcycle.
CSHVC cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
1000
1200
1400
1600
1800
time (s)
Figure44.CitySuburbanHeavyVehiclecycle.
56
Beeline cycle
50
speed (mph)
40
30
20
10
0
0
200
400
600
800
1000
1200
1400
1600
1800
time (s)
Figure45.Beelinecycle.
ETC Urban cycle
50
speed (mph)
40
30
20
10
0
0
100
200
300
400
time (s)
500
600
Figure46.EuropeanTestCycleUrbancycle.
57
Transient cycle
50
speed(mph)
40
30
20
10
0
0
100
200
300
400
500
600
700
time (s)
Figure47.CARBTransientcycle.
UDDS cycle (Test D)
60
speed (mph)
50
40
30
20
10
0
0
200
400
600
800
1000
time (s)
1200
Figure48.UrbanDynamometerDrivingSchedule(Test_Dcycle).
58
King County Metro cycle
60
speed (mph)
50
40
30
20
10
0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
time (s)
Figure49.KingCountyMetrocycle.
Arterial cycle
50
speed (mph)
40
30
20
10
0
0
50
100
150
time (s)
200
250
300
Figure50.Arterialcycle.
59
Commuter cycle
60
speed (mph)
50
40
30
20
10
0
0
50
100
150
200
time (s)
250
300
350
Figure51.Commutercycle.
60

Fueling Cycle Models

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fueling Cycle Models

Uploaded by

Copyright:

Available Formats

WestVirginiaUniversity.

xp are predicted values and xe are experimental

AveragePercent MaximumError AverageError

Scatterplot of Average % Error vs CASE USED

Velocity, Velocity^3, Stops/mile

Scatterplot of Average % Error vs CASE USED

Velocity, Acceleration, Acceleration^2

Acceleration, Inertial Power

Inertial Power, Stops/mile

Predicted CO2 (g/s)

Inferred Fuel Economy from Fuel Consumption Prediction

Inferred fuel economy (mpg)

Measured fuel economy (mpg)

Histogram of % Error for CO2 (g/s) prediction.

Best Prediction CO2(g/s) CNG Bus #32

Best Prediction CO2(g/s) CNG Bus #35

Best Prediction CO2(g/s) Hybrid Bus #37

Best Prediction CO2(g/s) Diesel Bus #39

Best Prediction CO2(g/s) Diesel Bus #41

CNG#32 CNG#35 Hybrid#37 Diesel#39 Diesel#41

Cycle1 Cycle2 Cycle3 %Error AverageError Max.Error Std.Dev.Error

Average Max.Abs. Std.Dev.

PSAT Simulated CO2 mass rate (g/s)

Measured CO2 mass rate (g/s)

PSAT Simulated CO2 mass rate (g/s)

Measured CO2 mass rate (g/s)

PSAT Simulated CO2 mass rate (g/s)

Measured CO2 mass rate (g/s)

Change in CO2 emissions (%)

Scatterplot of CO2 (g/s) vs Axle Energy (hp-hr)

Average Acceleration: Summation of instantaneous positive acceleration over number of data

New York Bus cycle

New York Composite cycle

ETC Urban cycle

UDDS cycle (Test D)

King County Metro cycle

You might also like