You are on page 1of 12

Calibration of HDM-4 Models for

Estimating the Effect of Pavement


Roughness on Fuel Consumption
for U.S. Conditions
Imen Zaabar and Karim Chatti

Fuel consumption costs are an essential part of life-cycle cost analy- vehicle technology, pavement surface type, pavement condition,
sis. These costs are influenced by vehicle technology, pavement con- roadway geometrics, environment, speed of operation, and other fac-
dition, roadway geometrics, environment, speed, and other factors. tors. Therefore, vehicle operating costs are part of the costs that high-
Many models for the effects of pavement condition on fuel consump- way agencies must consider when evaluating pavement-investment
tion were developed on the basis of data generated years ago in other strategies.
countries for vehicles that vary substantially from those used cur- Reduction in vehicle fuel consumption is one of the main ben-
rently in the United States. Therefore, new information is needed to efits considered in technical and economic evaluations of road
help in refining and developing models that would better apply to U.S. improvements considering its significance. According to the
conditions. The mechanistic model developed as part of the Highway Bureau of Transportation Statistics, the 255 million vehicles in
Development and Management software (HDM-4) is recommended the United States consume approximately 200 billion gal of motor
after calibration for predicting fuel consumption. The results of the fuel annually. With current gas prices, this figure translates to
calibration exercise for U.S. conditions, with field data collected as $400 billion. The fuel consumption of a vehicle is proportional
part of the NCHRP Project 1-45, are presented. The calibrated HDM-4 to the forces acting on the vehicle. These forces are rolling resis-
fuel consumption model was able to predict very adequately the fuel tance, gradient, inertia, curvature, and aerodynamic forces. Pave-
consumption of five different vehicle classes under different operat- ment conditions have been reported to have the greatest significance
ing, weather, and pavement conditions. The better accuracy achieved on rolling resistance forces. A decrease in pavement roughness by
after calibration has improved the prediction of the effect of rough- 3 m/km will result in a 10% decrease in rolling resistance (2). A
ness on fuel consumption. The comparison of sensitivity analyses before 10% reduction in average rolling resistance promises a 1% to 2%
and after calibration has shown that the effect of roughness on fuel decrease in the fuel consumption (3). This decrease would save
consumption increased by 1.75 for the van, 1.70 for the articulated approximately 2 to 4 billion gal of fuel per year of the 200 billion
truck, 1.60 for the medium car, 1.35 for the sport utility vehicle, and gal consumed by the entire vehicle fleet. In this context, a 1%
1.15 for the light truck. to 2% reduction in the fuel consumed would be a meaningful
accomplishment.
A large body of research is available on the effects of pavement
Understanding the costs of highway construction, highway main- condition on vehicle operating costs and on models used to esti-
tenance, and vehicle operation is essential to sound planning and mate these effects. Much of this information and many of the
management of highway investments, especially under increasing models were developed on the basis of data generated some years
infrastructure demands and limited budget resources. Although the ago in other countries for vehicle fleets that vary substantially
infrastructure costs conceived by road agencies are substantial, the from those used currently in the United States and for roadways
cost borne by road users are even greater. In 2009, the American that differ from those built in the United States. However, some
Automobile Association (1) reported an average vehicle operat- relevant information was collected in the United States and
ing cost of 54.9 cents per vehicle mile on the basis of 2008 prices. Canada (4) in recent years that could help in refining these mod-
For conventional vehicles, these costs are related to fuel and oil els or developing models that would better apply to current and
consumption, tire wear, repair and maintenance, and depreciation. future U.S. conditions; this information applies to operating costs
These costs depend on the vehicle class and are influenced by of heavy trucks. Inadequate information is available on the effects
of pavement condition on the operating costs of light vehicles
(automobiles and pickup trucks). Inadequate models for estimat-
Department of Civil and Environmental Engineering, Michigan State University, ing the effects of pavement condition on vehicle operating costs
3546 Engineering Building, East Lansing, MI 48824. Corresponding author: make it difficult to conduct a rational economic analysis. Ade-
K. Chatti, chatti@egr.msu.edu. quate models need to be developed that will provide highway
agencies in the United States with the tools necessary for consid-
Transportation Research Record: Journal of the Transportation Research Board,
No. 2155, Transportation Research Board of the National Academies, Washington,
ering vehicle operating costs in evaluating pavement-investment
D.C., 2010, pp. 105116. strategies and identifying options that yield economic and other
DOI: 10.3141/2155-12 benefits.

105
106 Transportation Research Record 2155

MODELS FOR ESTIMATING FUEL CONSUMPTION Ptr = power required to overcome traction forces (kW),
Paccs = power required for engine accessories (e.g., fan belt,
A number of major models that have been developed in various alternator; kW),
countries were identified. The most relevant models include the Peng = power required to overcome internal engine friction (kW),
following: = fuel consumption at idling (mL/s),
= engine efficiency (mL/kW/s), and
World Banks HDM-3 and 4 Vehicle Operating Cost (VOC) dFuel = excess fuel conception caused by congestion.
model,
Texas Research and Development Foundation (TRDF) VOC Tables 1 and 2 summarize the equations and submodels of the
model, HDM-4 fuel consumption model. The total power required is divided
Saskatchewan VOC models, into traction power, engine drag, and vehicle accessories. The total
Swedish mechanistic model for simulations on road traffic requirement can be calculated by two alternative methods, depending
(VETO), on whether the traction power is positive or negative. The traction
ARFCOM: Australian Road Fuel Consumption model, power is a function of rolling resistance, gradient, inertial, curvature,
New Zealand VOC model, and and aerodynamic forces.
South African VOC models. The rolling resistance forces are functions of pavement conditions,
tire parameters, and vehicle characteristics. The rolling resistance has
Most of the current VOC models have beneted from the World great signicance on fuel consumption. The presence of water or
Banks HDM research to some extent (5). The basis of HDM research snow on the road increases rolling resistance and consequently fuel
dates back to the seminal study by Weille (6) for the World Bank, consumption.
which led to the development of the Highway Cost Model and sub-
sequently to the most recent HDM IV module. The VOC research
conducted in the United States was primarily initiated by Winfrey (7) FIELD TRIALS
followed by Claffey (8). These initial efforts laid the foundation for
an assembly of VOC data and estimation models in the AASHTO Red To calibrate and validate the HDM-4 fuel consumption model, eld
Book by 1978. In 1982, new VOC models were developed by TRDF tests were conducted to measure the fuel consumption. Five different
(9). The TRDF model was representative of current vehicle tech- locations were selected on the basis of the variability level of their
nology at that time. Zaniewski et al. (9) reported that roughness has no pavement conditions (i.e., roughness, gradient, texture, pavement
effect on fuel consumption. The TRDF models were also incorporated type). All these locations were in Michigan. Table 3 shows the eld
into the MicroBENCOST model, which was intended to be a modern test matrix. The tests were conducted during wet (winter) and dry
replacement for the AASHTO Red Book. International roughness (summer) conditions. The weather condition (temperature and wind
index (IRI) was not an accepted roughness index at that time. The speed) were recorded using a portable weather station. Any change of
TRDF model is based on the empirical approach, which makes the more than 5F in ambient temperature would introduce an error, and
applicability of these models questionable, especially in light of the test would have to be repeated. Table 4 summarizes both winter
emerging technologies in the vehicle industry. Conversely, various and summer conditions.
VOC models are available from extensive research conducted in other The test equipment used during the eld test is an instrument that
parts of the world, particularly Europe and Australia. Most of the cur- could access and log data instantaneously from the vehicles engine
rent work in the eld of VOC models is based on the development of control unit via on board diagnostic connector (Figure 1).
mechanistic-based models; this type of model is more general and is The pavement condition data (raw prole and texture depth) dur-
capable of predicting the outcome for a wide variety of scenarios. ing the test were collected by Michigan Department of Transportation
According to the state of the art, it is clear that the HDM-4 VOC (MDOT) using a rapid travel prolometer and a pavement friction
model has benefited from previous studies and models in differ- tester equipped with a laser-based measurement system to measure
ent countries (e.g., Australia, South Africa). Also, both develop- texture depth.
ing and developed countries adopted HDM-4, and they are trying The slope data surveys were collected by a third party using a high
to calibrate it to their corresponding conditions. Thus, appropriate precision global positioning system. The sampling rate is every 1 s at
existing mechanistic-based VOC models can be calibrated for U.S. highway speed (every 100 ft). The average error is approximately
conditions. twice the error of the total station.
Six different vehicles were used:

HDM-4 FUEL CONSUMPTION MODEL Medium car,


Sport utility vehicle (SUV),
The model adopted is based on the ARFCOM mechanistic model Van,
(5) with alteration to the prediction of engine speed, accessories Light truck (gasoline),
power, and engine drag. The general form of the model is expressed Light truck (diesel), and
conceptually by Equation 1. Heavy truck (articulated).
IFC = f ( Ptr , Paccs + Peng ) = max ( , Ptot (1 + d Fuel )) (1)
Table 5 summarizes the characteristics of these vehicles.
The vehicles were driven over the preselected sections. The data
where
acquisition system was connected to the vehicle. The vehicles were
IFC = instantaneous fuel consumption (mL/s), driven at different speeds with and without cruise control. Multiple
Ptot = total power, and repeated runs were performed.
Zaabar and Chatti 107

TABLE 1 HDM-4 Fuel Consumption Model

Name Description Unit

Ptr
Total power (Ptot) Ptot = + Paccs + Peng for Ptr 0, uphill/level kW
edt
Ptot = edtPtr + Paccs + Peng for Ptr < 0, downhill
edt Drive-train efficiency factor
Engine and accessories power Pengaccs = KPea Pmax ( Paccs_a1 + ( Paccs_a0 Paccs_a1 )) kW
(Pengaccs = Peng + Paccs)
RPM RPMIdle

RPM100 RPMIdle
KPea Calibration factor kW
Pmax Rated engine power
Paccs_a1 b + b2 4 a c
Paccs_a1 =
2a
100 PctPeng
a = b ehp KPea Pmax
2

b = KPea P 100
b max
c =
b Engine efficiency depends on the technology type (gasoline versus diesel) mL/kW/s
ehp Engine horsepower hp
Fuel consumption at idling mL/s
Paccs_a0 Ratio of engine and accessories drag to rated engine power when traveling at 100 km/h
PctPeng Percentage of the engine and accessories power used by the engine %
Engine speed RPM = a0 + a1 SP + a2 SP2 + a3 SP3 RPM
SP = max(20, v)
Vehicle speed m/s
a0 to a3 Model parameter (Table 5)
RPM100 Engine speed at 100 km/h RPM
RPMIdle Idle engine speed RPM
Traction power (Ptr) ( Fa + Fg + Fc + Fr + Fi ) kW
Ptr =
1, 000
Fa Aerodynamic forces N
Fg Gradient forces N
Fc Curvature forces N
Fr Rolling resistance forces N
Fi Inertial forces N

NOTE: b = engine efficiency, dependent on the technology type (gasoline versus diesel); and SP = maximum between 20 km/h and the vehicle
speed.

Tests for trucks were conducted under two different loading condi- where
tions: loaded (Figure 2) and unloaded. The light truck was loaded with
MAF = mass air ow,
two concrete blocks. The total load was 2.82 metric tons. The concrete
14.7 = air-to-fuel ratio, and
blocks were tightly secured to the trailer. The only possible movement
0.74 = density of gasoline in grams per mL.
is in the vertical direction. The trailer of the heavy truck was loaded
with steel sheets. The total load of the steel was 21.32 metric tons. The HDM-4 model overpredicts the engine speed of the vehicle.
These loading conditions are typical in the United States. Therefore, the engine and accessories power will be overpredicted,
and the fuel consumption will be overestimated. Consequently, when
calibrating the fuel consumption model, the traction power (i.e., the
CALIBRATION effect of pavement conditions) will be underestimated. Therefore,
verication of the HDM-4 engine speed model is recommended.
The instantaneous fuel consumption was calculated using Equation 2.
The predicted and measured consumption data were compared with
each other. Calibration of HDM-4 Engine Speed Model

MAF The HDM-4 engine speed model was calibrated for all vehicle classes
IFC = (2)
14.7 0.74 using the data collected during the field tests. Table 6 classifies
TABLE 2 HDM-4 Traction Forces Model

Name Description Unit

Aerodynamic forces (Fa) Fa = 0.5 CDmult CD AF 2 N


CD Drag coefficient
CDmult CD multiplier
AF Frontal area m2
Mass density of the air kg/m3
Vehicle speed m/s
Gradient forces (Fg) Fg = M GR g N
M Vehicle weight kg
GR Gradient radians
g Gravity m/s2
Curvature forces (Fc) M 2 2
N

M g e
R
Fc = max 0, 10 3
Nw Cs

R Curvature radius m
Superelevation (e) e = max(0, 0.45 0.68 Ln(R)) m/m
Nw Number of wheels
Tire stiffness (Cs) M
2
M
Cs = KCS a 0 + a1 + a2
Nw Nw
KCS Calibration factor
a0 to a2 Model parameter (Table 5)
Rolling resistance (Fr) ( (
Fr = CR 2 FCLIM b11 Nw + CR1 b12 M + b13 2 )) N
CR1 Rolling resistance tire factor
Rolling resistance parameters b11 = 37 Dw
(b11, b12, b13)
{0.067 Dw
b12 = 0.064 Dw

old tires
latest tires
b12 = 0.012 Nw Dw
2

Rolling resistance surface factor (cr2) = Kcr 2 [ a 0 + a1 Tdsp + a 2 IRI + a3 DEF ]


Kcr2 Calibration factor
a0 to a3 Model coefficient (Table 6)
Tdsp Texture depth using sand patch method mm
IRI International roughness index m/km
DEF Benkelman beam rebound deection mm
Climatic factor (FCLIM) FCLIM = 1 + 0.003 PCTDS + 0.002 PCTDW
Inertial forces (Fi) a2
Fi = M a 0 + a1 arctan 3 a

a0 to a2 Model parameter (Table 7)

NOTE: Dw = diameter of wheels (m), PCTDS = percent driving in snow conditions, and PCTDW = percent driving in wet conditions.

TABLE 3 Field Test Matrix

Pavement Type
Section IRI Range Length Speed Limit Test Speed
ID AC PCC (m/km) (km) (km/h) (km/h) Replicates

AB X 1.38.5 1.44 72 56 72 2
BC X 1.77 1.6 72 56 72 2
DE X 3.56 0.48 72 56 72 2
EF X 3.36 0.64 72 56 72 2
GH X 1.12.5 4.8 112 88 104 2
JI X 1.52.6 6.4 80 56 72 2
IJ1 X 1.52.6 0.64 80 72 88 2
IJ2 X 0.84.6 1.6 80 56 72 2
IJ3 X 0.48 80 56 72 2
IJ4 X 1.28 72 56 72 2
Zaabar and Chatti 109

TABLE 4 Weather Conditions

Winter Summer

Day 1 Day 2 Day 1 Day 2 Day 3

Ambient 02 13 28.929.2 27.228.3 22.525.2


temperature
(C)
Wind speed 1.72.4 0.41 2.12.9 1.42.4 1.72.4
(m/s)

vehicle classes into categories. Figures 3 and 4 show the results of the
calibration. Table 7 summarizes the new coefficients for the engine
speed model.

FIGURE 1 Fuel consumption data acquisition system.


Calibration of HDM-4 Fuel
Consumption Model
The data collected during eld tests were used to calibrate the
The HDM-4 fuel consumption model provides two calibration fac-
HDM-4 fuel consumption model. With cruise control, low consump-
tors as described in Bennett and Greenwood (10). The two factors
are as follows: tion was underestimated, but high consumption was overestimated
(Figure 5). During the test, when the vehicle is driven over a steep pos-
Kcr2: a calibration factor for the rolling resistance and itive slope, the cruise control will be invalidated and the vehicle speed
Kpea: a calibration factor for the accessories and engine power. will decrease resulting in a decrease in fuel consumption. However,
when the vehicle is driven over a steep negative slope, the HDM-4
We calculated the least sum of square differences between the model yields negative traction power, which is equivalent to the vehi-
observed and estimated values using HDM-4 model without calibra- cle mobilizing by itself without any need for traction force provided by
tion. Then the values that minimize least sum of square were obtained. the engine. Thus, the predicted amount of fuel consumed will decrease.
The methodology that was used is summarized as follows: This phenomena never occurred during the tests. Instead, we noted that
the speed increased because the cruise control never sets a negative
1. A random value was assigned to Kcr2, and then the value of force to the engine. Figure 6 supports these observations.
Kpea yielding the lower least square value is determined. However, for some vehicles, it is difficult to maintain constant speed
2. This process is repeated until the lowest least square is obtained. without cruise control, especially when the roads are very rough. This

TABLE 5 Characteristics of Vehicles Used in Field Trials

Vehicle Class

Medium Car, SUV, Van, Light Truck, Heavy Truck,


Mitsubishi, Nissan, Ford, GMC, International,
Galant, Pathnder, E350, W4500, 9200 6 4,
Characteristic 2008 2009 2008 2006 2005

Drag coefficient 0.4 0.5 0.5 0.6 0.8


Frontal area (m2) 1.9 2.9 2.9 4.2 9
Tare weight (t) 1.46 2.5 2.9 3.7 13.6
Maximum allowable load (t) 2.9 22.7
GVW (t) 6.6 36.3
Weight of the load (t) 2.8 21.3
Gas type Gas Gas Gas Diesel Diesel
Tire diameter (m) 0.38 0.4 0.4 0.4 0.57
Tire pressure (psi) 35 39 43 75 110
Tire type Radial Radial Radial Radial Bias
Cargo length (m) 4.88 15.85
Other 4WD 15 seats Flat bed

NOTE: GVW = gross vehicle weight, 4WD = four-wheel drive, and SUV = sport utility vehicle.
110 Transportation Research Record 2155

(a) (b)

(c) (d)

FIGURE 2 Different loading conditions: (a) loading of light truck, (b) loaded light truck, (c) loading of heavy truck, and (d ) loaded heavy truck.

TABLE 6 Vehicle Classification Used in Engine Speed Model observation is important, especially when looking at the effect of pave-
ment conditions on fuel consumption. The observations regarding the
Categories Vehicle Classes Vehicle Used effect of cruise control were not applicable to the medium car, van, and
SUV, because they are light vehicles. Therefore, for calibration pur-
Passenger car Small car Medium car
poses, data collected during tests with cruise control were used for
Medium car
Large car these vehicles. The data collected during tests without cruise control
Minibus were used for the light and heavy truck. Figure 7 shows the results after
Light commercial vehicle Light delivery vehicle Van calibration of the HDM-4 fuel consumption model for all vehicle
Light goods vehicle classes. Table 8 summarizes the calibration coefficients. Statistical
Four-wheel drive (FWD) FWD SUV analysis showed that there is no difference between the observed and
Light truck Light truck Light truck the estimated fuel consumption at 95% condence level.
Light bus
Heavy truck Medium truck Articulated
Heavy truck truck Discussion of Results
Articulated truck
Medium bus From previous studies and our sensitivity analysis of the HDM-4
Heavy bus
Coach model, the effect of roughness on fuel consumption was estimated at
approximately 5%. If the error of the estimate exceeds 5%, the model
Zaabar and Chatti 111

y = -0.0007x3 + 0.2006x2 + 2000


3000 0.868x + 720.05

Predicted Engine Speed (rpm)


Engine Speed (rpm)
2500 R2 = 0.99
2000 1500
1500
1000 1000
500
0
500
0 20 40 60 80 100 120
Speed (Km/h)

measured engine speed-wet condition 0


engine speed model (HDM 4) 0 500 1000 1500 2000
measured engine speed-dry condition Measured Engine Speed (rpm)
calibrated model
(b)
(a)

y = 0.0062x3 - 0.3018x2+ 6.7795x


2500 2000
+ 671.98
Engine Speed (rpm)

Predicted Engine Speed (rpm)


2
2000 R = 0.96
1500
1500

1000
1000
500

0
500
0 20 40 60
Speed (Km/h)

measured engine speed-wet condition 0


engine speed model (HDM 4) 0 500 1000 1500 2000
measured engine speed-dry condition Measured Engine Speed (rpm)
calibrated model
(d)
(c)

FIGURE 3 Calibration of HDM-4 engine speed model for van and small car: (a) calibration
procedure for small car, (b) measured versus predicted for small car, (c) calibration procedure
for van, and (d) measured versus predicted for van.

y = 0.0019x3 - 0.1331x2 +
3000 3.6701x + 982.37 2000
Engine Speed (rpm)

Predicted Engine Speed (rpm)

2500 R2 = 0.99
2000 1500
1500
1000
1000
500
0
0 20 40 60 80 100 120 500
Speed (Km/h)

measured engine speed-wet condition 0


engine speed model (HDM 4) 0 500 1000 1500 2000
measured engine speed-dry condition Measured Engine Speed (rpm)
calibrated model
(b)
(a)

FIGURE 4 Calibration of HDM-4 engine speed model for sport utility vehicle (SUV), light
and heavy trucks: (a) calibration procedure for SUV, (b) measured versus predicted for SUV.
(continued on next page)
112 Transportation Research Record 2155

y = -0.0018x3 + 0.3798x2 -

Engine Speed (rpm)

Predicted Engine Speed (rpm)


3000 3.0722x + 550.08
2500
R2 = 0.99
2000 2000

1000 1500

0 1000
0 20 40 60 80 100 120
Speed (Km/h) 500

measured engine speed-wet condition 0


0 500 1000 1500 2000 2500
engine speed model (HDM 4)
measured engine speed-dry condition Measured Engine Speed (rpm)
calibrated model (d)
(c)

y = 6E-05x3 + 0.2077x2 -
Engine Speed (rpm)

Predicted Engine Speed


5.3791x + 799.6 2000
3000
2
R = 0.99 1500
2000

(rpm)
1000 1000
0
0 50 100 500
Speed (Km/h) 0
0 1000 2000
measured engine speed-dry condition
engine speed model (HDM 4) Measured Engine Speed (rpm)
calibrated model (f)
(e)

FIGURE 4 (continued) Calibration of HDM-4 engine speed model for SUVs and light and
heavy trucks: (c) calibration procedure for light truck, (d) measured versus predicted
for light truck, (e) calibration procedure for articulated truck, and (f ) measured versus
predicted for articulated truck.

will not be able to estimate the effect of roughness correctly. In our


case, the error ranges from 2.5% for articulated trucks to 8% for
TABLE 7 New Coefficients for Engine Speed Model medium cars. Therefore, more detailed analysis was conducted to
by Vehicle Class
look for the sources of error.
Engine Speed Coefficients and Statistics

a0 a1 a2 a3 EFFECT OF ROUGHNESS
ON FUEL CONSUMPTION
Small car 720.05 0.868 0.2006 0.0007
Medium car 720.05 0.868 0.2006 0.0007 To estimate the effect of roughness on fuel consumption, a more
Large car 720.05 0.868 0.2006 0.0007 detailed analysis was required. The analysis consists of the following
Light delivery car 589.6 0.5145 0.0168 0.0019 operations:
Light goods vehicle 589.6 0.5145 0.0168 0.0019
Four-wheel drive 943.51 0.0861 0.0069 0.0007
1. Range discretization. The grade data were divided into equal
ranges. The width of the discretization interval was selected to be equal
Light truck 550.08 3.0722 0.3798 0.0018
to 0.1% (on the basis of the sensitivity of fuel consumption to grade).
Minibus 720.05 0.868 0.2006 0.0007
2. Analysis of covariance. The grade was treated as a xed fac-
Light bus 550.08 3.0722 0.3798 0.0018 tor, IRI as a covariate variable (IRI_SI) and the fuel consumption
Medium truck 833.7 17.717 0.9671 0.0055 as the dependent variable (FC_mLKm). The groups that have at
Heavy truck 833.7 17.717 0.9671 0.0055 least three points were selected to be used in this analysis.
Articulated truck 833.7 17.717 0.9671 0.0055 3. Linear regression analysis. A linear function was tted to the
Medium bus 833.7 17.717 0.9671 0.0055 data within each group of grade.
Heavy bus 833.7 17.717 0.9671 0.0055
Coach 833.7 17.717 0.9671 0.0055
The results of the regression and the lack of t analysis are pre-
sented in the table that follows and Table 9. The lack of t test
Zaabar and Chatti 113

2.50

Fuel Consumption (mL/Km)


2.00

1.50

1.00
`

0.50

0.00
0 2 4 6 8 10
Distance (miles)

Predicted FC (mL/Km) Measured FC (mL/Km)

(a)

350
Fuel Consumption (mL/Km)

300

250

200

150
`
100

50

0
0 2 4 6 8 10
Distance (Km)

Predicted Fuel Consumption Measured Fuel Consumption

(b)

FIGURE 5 Predicted and measured fuel consumption versus distance:


(a) van and (b) light diesel truck (FC fuel consumption).

400 400
Predicted Fuel Rate (mL/Km)
Predicted Fuel Rate (mL/Km)

300 300

200 200

100 100

0 0
0 100 200 300 400 0 100 200 300 400
Measured Fuel Rate (mL/Km) Measured Fuel Rate (mL/Km)
(a) (b)

FIGURE 6 Observed versus estimated fuel consumption (a) with cruise control and (b) without cruise control.
114 Transportation Research Record 2155

y=x y=x

Calibrated Fuel Rate (mL/Km)


Calibrated Fuel Rate (mL/Km)
100 200
R2 = 0.90 R2 = 0.83
80 SSE = 4.09 SSE = 9.58
150
60
100
40
50
20

0 0
0 20 40 60 80 100 0 50 100 150 200
Measured Fuel Rate (mL/Km) Measured Fuel Rate (mL/Km)
(a) (b)

y=x y=x
Calibrated Fuel Rate (mL/Km)

Calibrated Fuel Rate (mL/Km)


120 300
R2 = 0.89 R2 = 0.82
100 250
SSE = 4.19 SSE = 10.16
80 200
60 150
40 100

20 50

0 0
0 20 40 60 80 100 120 0 50 100 150 200 250 300
Measured Fuel Rate (mL/Km) Measured Fuel Rate (mL/Km)
(c) (d)

y=x
Calibrated Fuel Rate (mL/Km)

400
R2 = 0.88
SSE = 5.29
300

200

100

0
0 100 200 300 400
Measured Fuel Rate (mL/Km)
(e)

FIGURE 7 Observed fuel consumption versus estimated using HDM-4 model for all vehicles:
(a) medium car, (b) SUV, (c) van, (d) light truck, and (e) articulated truck.

(detailed in the following table) conrms that the selected model ts


the data very well.
TABLE 8 Summary of Model Performance
Sum of Degrees of Mean
Source Squares Freedom Square F Significance
SSE Number of Data
Kcr2 Kpea (mL/km) Considered Lack of Fit 46.867 119 .394 .650 .792
Pure Error 1.818 3 .606
Medium car 0.5 0.25 4.09 456
Results summarized in Table 9 showed the following:
SUV 0.58 0.56 9.58 250
Light truck 0.99 0.61 10.16 356 1. Grade is statistically signicant,
Van 0.67 0.49 4.19 352 2. IRI is statistically signicant, and
Articulated truck 1.1 0.35 5.29 456 3. Texture is statistically not signicant.

NOTE: SSE = least sum of square, Kcr2 = calibration factor for rolling resistance, Because the effect of roughness is statistically signicant, our focus
and Kpea = calibration factor for accessories and engine power. was to check the accuracy of the calibrated model. Figure 8a shows the
Zaabar and Chatti 115

TABLE 9 Tests of Between-Subjects Effects

Type III
Source Sum of Squares df Mean Square F Sig.

Corrected model 4,300.769a 14 307.198 769.817 .000


Intercept 19,697.944 1 19,697.944 49,361.721 .000
IRI 23.557 1 23.557 59.032 .000
Texture .147 1 .147 .368 .545
Grade 3,796.846 12 316.404 792.887 .000
Error 48.684 122 .399
Total 351,401.815 137
Corrected total 4,349.454 136

a
R-squared = .989 (adjusted R-squared = .988).
Change in Fuel Consumption (%)

4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
1 2 3 4 5
IRI (m/km)

Medium car - HDM 4 Medium car - Regression


SUV - HDM 4 SUV - Regression
Van - HDM 4 Van - Regression
Light truck - HDM 4 Light truck - Regression
Articulated truck - HDM 4 Articulated truck - Regression
(a)
Change in Fuel Consumption (%)

4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
1 2 3 4 5
IRI (m/km)

Medium car
SUV
Van
Light truck
Articulated truck
(b)

FIGURE 8 Effect of roughness on fuel consumption: (a) comparison of


HDM-4 with regressed data and (b) HDM-4 without calibration.
116 Transportation Research Record 2155

change in fuel consumption as a function of IRI using the calibrated Hanna. The authors also are grateful for the technical support from
HDM-4 model and the results from the regression analysis described MDOT in conducting pavement condition testing, coordinated by
previously mentioned. As can be seen in the gure, the results match Tom Hynes. Finally the authors acknowledge the help received by
very well. Figure 8b shows the trends predicted by HDM-4 before cal- several MSU graduate students and Anshu Manik in conducting
ibration. The comparison of sensitivity analyses before and after the eld tests.
calibration showed that the effect of roughness on fuel consumption
increased by 1.75 for the van, 1.70 for the articulated truck, 1.60 for
the medium car, 1.35 for the SUV, and 1.15 for the light truck. REFERENCES

1. American Automobile Association. Your Driving Costs. AAA, Heathrow,


CONCLUSION Fla., 2009. http://www.aaaexchange.com/Assets/Files/200948913570.
DrivingCosts2009.pdf. Accessed April 21, 2010.
In this paper, we showed that the calibrated HDM-4 fuel consump- 2. Lu, X. P. Effects of Road Roughness on Vehicular Rolling Resistance.
Measuring Road Roughness and Its Effects on User Cost and Comfort
tion model was able to adequately predict the fuel consumption of
(T. D. Gillespie and M. Sayers, eds.), ASTM Special Technical Publica-
ve different vehicle classes under different operating, weather, and tion 884. American Society for Testing and Materials, Philadelphia, Pa.,
pavement conditions. The good quality of the data obtained in this 1985, pp. 143161.
study allowed the calibration and validation of the HDM-4 fuel con- 3. Special Report 286: Tires and Passenger Vehicle Fuel Economy: Inform-
sumption and engine speed models. Also, we were able to better ing Consumers, Improving Performance. Transportation Research Board
estimate the effect of roughness on fuel consumption. of the National Academies, Washington, D.C., 2006.
The better accuracy achieved after calibration of the HDM-4 fuel 4. de Solminihac, H., R. Harrison, J. P. Covarrubias, and A. Altamira. Cal-
ibration of Fuel Consumption Model in HDM4 Model: Application to
consumption model to U.S. conditions has improved the prediction Observed Consumption in Canada and Chile. Presented at 83rd Annual
of the effect of roughness on fuel consumption. The comparison of Meeting of the Transportation Research Board, Washington, D.C.,
sensitivity analyses before and after calibration showed that the 2004.
effect of roughness on fuel consumption increased by 1.75 for the 5. Bennett, C. R., and I. D. Greenwood. Volume 7: Modeling Road User and
van, 1.70 for the articulated truck, 1.60 for the medium car, 1.35 for Environmental Effects in HDM-4. Version 3.0. International Study of
the SUV, and 1.15 for the light truck. Highway Development and Management Tools (ISOHDM), World Road
Association (PIARC), Cedex, France, 2003.
Also, the key characteristics of representative vehicles used in the 6. Weille, D. Quantication of Road User Savings. World Bank Staff Occa-
HDM-4 model vary substantially from those of U.S. vehicles. There- sional Paper No. 2. World Bank, Washington, D.C., 1966.
fore, the predicted fuel consumption using the model without cal- 7. Winfrey, R. Economic Analysis for Highways. International Textbook
ibration was lower than the actual consumption. For example, the Company, Scranton, Pa., 1969.
amount of fuel consumed by a light truck in the United States is 8. Claffey, P. J. NCHRP Report 111: Running Cost of Motor Vehicles as
equivalent to the predicted consumption by a medium truck using the Affected by Road Design and Traffic. HRB, National Research Council,
Washington, D.C., 1971.
HDM-4 default characteristics of representative vehicles. Therefore, 9. Zaniewski, J. P., B. C. Butler, G. Cunningham, G. E. Elkins, M. S.
we recommend the use of the HDM-4 fuel consumption model after Paggi, and R. Machemehl. Vehicle Operating Costs, Fuel Consump-
calibration and adjustment to U.S. conditions. tion, and Pavement Type and Condition Factors. FHWA-PL-82-001.
Texas Research and Development Foundation, Austin, 1982.
10. Bennett, C. R., and I. D. Greenwood. Volume 5: HDM-4 Calibration Ref-
ACKNOWLEDGMENTS erence Manual. International Study of Highway Development and Man-
agement Tools (ISOHDM), World Road Association (PIARC), Cedex,
France, 2003.
This research was conducted as part of NCHRP Project 1-45. The
authors acknowledge the nancial support of NCHRP and the input
of the technical panel and the TRB senior program officer, Amir The Surface PropertiesVehicle Interaction Committee peer-reviewed this paper.

You might also like