Professional Documents
Culture Documents
Aleadelat2016, Development of Serviceability Prediction Model
Aleadelat2016, Development of Serviceability Prediction Model
To cite this article: Waleed Aleadelat, Promothes Saha & Khaled Ksaibati (2016): Development
of serviceability prediction model for county paved roads, International Journal of Pavement
Engineering, DOI: 10.1080/10298436.2016.1176167
Article views: 24
Download by: [University of Tennessee, Knoxville] Date: 09 June 2016, At: 02:12
International Journal of Pavement Engineering, 2016
http://dx.doi.org/10.1080/10298436.2016.1176167
roughness index, pavement condition index (PCI) and rut depth for flexible pavements only. Ten panellists prediction model; local
from Wyoming rated 30 pavement sections that were randomly selected at different distresses’ levels; agencies
using two vehicles (SUV and Sedan). The statistical analysis indicated that the seating position, age and
gender are not significant to the rating process. However, the vehicle’s type found to be significant. The
newly developed model from this study explains 80% of the variations in the PSI values of county roads
(adjusted R2 = 0.80). In addition, the new model seems to provide more realistic representation of the
conditions of county roads than the statewide model used on the state’s highway system.
420
17%
1166 Miles
46%
1658 Miles 760
68% 187 30%
8%
County Secondary
48 4 29 2 31 2
288 0% 0%
2% 1% 0% 2%
10% 246
Downloaded by [University of Tennessee, Knoxville] at 02:12 09 June 2016
13%
837
30% 1543 Miles
1631 Miles 85%
57%
Interstate
Primary
PSI model. It is important when presenting the condition of the Irick 1960, Nakamura 1962, Nair et al. 1985). For example, in
county roads to legislators to compare models that are similar to 1962 a comprehensive study was performed by Velma Nakamura
the state’s system. This way the condition of the roadway network in the State of Louisiana (Nakamura 1962). The scope of this
can be reasonably deduced when compared to the state’s system. study included 60 pavement sections that varied between rigid
The newly developed model is geared more to the decision-mak- and flexible pavements. The road roughness was measured using
ers of the local county roads so that they can allocate funds to Indiana State Highway Roughometer. Thirty raters with diverse
the roads in the worst shape. backgrounds were requested to drive vehicles similar to their
own vehicles on the selected sections and to provide their rating
regarding the ride quality only. Different models with high cor-
Previous research
relation were developed using the analysis of variance (ANOVA)
The first major attempt to find a relation between pavement’s pro- and linear regression that relate the PSR to the measured rough-
file and different roadway user’s perceptions was conducted by ness. It was concluded that the rating panel method of evaluating
the American Association of State Highway Officials (AASHO) pavement serviceability is practical and applicable for rigid, over-
in 1960 (Carey and Irick 1960). A panel of 100 individuals rated lay and flexible pavements. Also, variations of knowledge and
different sections on a scale from 0 to 5. These ratings were used experience in the highway engineering field are not important
to develop two models (one for asphaltic concrete and the other when selecting members for rating panels.
for Portland cement). The developed models correlate the differ- Nair et al. (1985) conducted a ride quality experiment to
ent physical measurements with the panel ratings. incorporate the changing trends in passenger vehicles, and public
The AASHO research effort led to the concept of pavement opinions of the quality of the ride of Texas highways. Twenty
serviceability; and considered the travelling public as the high- panel’s members were selected to rate 171 sections (flexible and
way customers. The actual mean panel ratings were considered rigid pavements) using 5 vehicles (2 subcompact cars and 3
as the present serviceability rating (PSR) and the predicted values mid-size cars). The road roughness was measured using three
of PSR through statistical models were considered as the PSI different roughometers (Mays Ride Meter, SIO meter and 690D
(Carey and Irick 1960). Surface Dynamic Profiler). It was found that surface roughness,
After the AASHO Road Test, several studies were performed vehicle size, vehicle type, vehicle wheelbase length, rater fatigue,
using different devices to measure road roughness (Carey and pavement type and maintenance were significant to the rating
International Journal of Pavement Engineering 3
process. It was contradictory to the expectation that the vehicle Table 1. A sample comprehensive roadway condition database.
speed has no effect on the rating process. County Route BegMP* EndMP* Rut(in) IRI(in/mile) PCI
Regardless of the different roughness measurement devices Laramie ML1102B 0.620 2.110 0.14 128 73
that were used in previous studies; none of these devices were Laramie ML1102B 2.110 2.610 0.16 108 71
able to give a measurement that could be standardised to an Laramie ML1108B 1.354 1.499 0.15 60 100
Laramie ML1108B 1.499 1.616 0.09 115 100
acceptable common scale (Hudson et al. 1985). In order to estab-
lish a standard roughness measurement, the World Bank in 1982 *BegMp: Beginning Mile Post (Miles).
EndMp: Ending Mile Post (Miles).
sponsored research experiment. This experiment resulted in the
development of the IRI. The IRI is determined by measuring the
actual road profile, and then processing it through a mathemati- roads. The Wyoming T2/LTAP Center analysed the collected
cal algorithm. This algorithm, known as the quarter-car simula- distress data for each roadway segment. The analysed distresses
tion (Gillespie 1985), simulates the response of a reference vehicle data-sets were merged into a single comprehensive database rep-
to road roughness. The accumulated suspension deflections of resenting the average pavement condition by segment. A sample
the reference vehicle can be divided on the travelling distance comprehensive database is presented in Table 1. The IRI was
to provide an index in the units of slope (Shafizadeh et al. 2002). measured using the South Dakota profiler. This device is a laser-
Accordingly, IRI is considered a geometric property of the road. type profiler manufactured according to ASTM E950 specifica-
Hence, it is a time stable index which generates the same values tions and meeting class 1 requirements (Inertial Road Profiler
Downloaded by [University of Tennessee, Knoxville] at 02:12 09 June 2016
when applied to the same road (Sayers and Karamihas 1998). 2016). The longitudinal pavement profile for both wheel paths
Later on, there were significant research attempts to develop was measured and analysed using the quarter-car simulation to
relationships between IRI and PSR (Al Omari and Darter 1994, generate the actual IRI value. The average IRI of the right and
Gulen et al. 1994, Pologruto 1999, Shafizadeh et al. 2002, Hernán the left wheel paths was considered the final IRI. The PCI was
de Solminihac et al. 2003). Al Omari and Darter (1994) per- calculated based on visual inspection and observation of the road
formed a study on 378 sections of different types from 6 different surface; through a video log that represents the real road surface.
states: Indiana, Louisiana, Michigan, New Mexico, New Jersey For every mile of roadway section, random 1000-ft. sub-seg-
and Ohio. They developed non-linear relationships with reason- ments were surveyed. After running the video logs through the
able R2 values. However, the developed models are biased and Pathways Services software, all the distresses’ features and their
invalid statistically. Because they were forced to pass through severities were defined visually and input into the software.
PSI = 5 when IRI is zero (Gulen et al. 1994). In the same year, However, the cracking was only considered in calculating the
Gulen et al. (1994) conducted a study in the State of Indiana. PCI of county roads. The software itself was used to define the
Twenty sections (9 bituminous and 11 concrete) were rated by 10 deduct values; and perform the required calculations to deter-
randomly chosen raters. Twenty-one different prediction models mine the PCI value for each sub-segment. The average PCI of all
were developed for bituminous, concrete or both pavement types. sub-segments was considered the final PCI for the entire roadway
Consequently, previous studies developed models without section (Shahin 2005, Huntington et al. 2013).
considering the particularity of the local roads. Moreover, there
are major differences between the state and local roadway sys- Riding quality survey
tems when it comes to the expectation of serviceability. These
A riding quality survey experiment was designed taking into
differences are crucial to develop a specific PMS for managing
account the particularity of the local county roads. Road sections
county paved roads.
with various roughness levels were included in the experiment.
A representative panel from the locals of Wyoming rated the
Methodology selected sections using two different vehicles on a scale from 0 to
5. The mean panel rating (MPR) for each segment was considered
This section presents the formation of the PSI model used in this
as the PSR for that segment. General driving styles and transport
research. The developed model relates the present serviceability
vehicle selection were very important in representing the normal
ratings (PSR) of county paved roads to the different roadway
driving scenario on Wyoming county roads. The riding quality
condition indices (IRI, rut depth and PCI). This process can be
survey should be conducted within the same time frame as the
divided into three major steps: (1) building a comprehensive
pavement condition data collection (ASTM (E1927-98)).
roadway distress database, (2) conducting a riding quality survey
and (3) regression analysis. Each step is described briefly in the
following subsections. Regression analysis
This research, like many previous studies, supported the idea Different diagnostic analysis techniques were performed to check
that the roads serviceability identification cannot be deduced the normality of ratings, outliers and homogeneity of variances
only from the physical measurements. The assessment of the real among the raters.
roadway customers (driving public) is vital to inferring the actual To test the normality of samples, the Shapiro–Wilk test was
serviceability conditions (Shafizadeh et al. 2002). used. This test calculates the W-statistics (Equation (2)) to specify
if a sample comes from a normal distribution or not. (Anderson-
Building a comprehensive distresses database Darling and Shapiro-Wilk tests 2015)
∑n
( i=1 ai x(i) )2
In 2014, a comprehensive data collection effort was conducted W = ∑n
by Pathway services Inc. on all 2444 miles of Wyoming county ( i=1 xi − x)̄ 2 (2)
4 W. A. EADELAT ET AL.
where the x(i) are the ordered sample values (in ascending order) Table 2. ASTM approximation of MPR from IRI measurements.
and the ai are constants generated from the means, variances and IRI measurement (inch/mile) Approximate MPR
covariances of the order statistics of a sample of size n. 25 4.5
Shapiro–Wilk tables were used to define the P-value related to 50 4.0
the calculated W-statistic. The defined P-Value should be greater 75 3.5
125 3.0
than the critical P-value (i.e. 0.05) to consider the normality of 200 2.5
the tested samples. 300 2.0
The Bartlett’s test was used to test the homogeneity of the 500 1.5
800 1.0
variances between the different raters. The analysis tests the fol-
lowing hypothesis:
Ho: σ21 = σ22 = ……… σ29; where σ2n is the rater’s variance.
The test calculates the T-Statistics using the following equation: Table 3. Roadway condition indices used in sections of test selection.
where Si2 is the variance of the ith group, N is the sample size, k is Error (MPR units) Non-normal distribution Normal distribution
the number of groups and Sp2 is the pooled variance. The hypoth- 0.1 319 138
0.2 80 35
esis can be rejected and the variances will not be considered 0.3 36 15
homogeneous, if the value of T > X1−𝛼,k−1
2
, where X1−𝛼,k−1
2
is the 0.4 20 9
critical value of the chi-square distribution with k − 1 degrees 0.5 13 6
0.6 9 4
of freedom and significance level of α. 0.7 7 3
After conducting the diagnostic analysis, linear regression 0.8 5 –
analysis with multiple predictor transformed variables was 0.9 4 –
1.0 3 –
used in the development of the PSI Model. The general multiple
regression model, that was used in this study, is shown in the
following equation:
should be similar to the posted speed limit. According to the
Yi = 𝛽0 + 𝛽1 Xi1 + 𝛽2 Xi2 + 𝛽3 Xi3 + 𝜀i (5) ASTM ( ASTM E1927-98 2012 ) procedure, a minimum number
where: of 20 sections should be selected for each pavement type. In this
Yi is the response variable; Xi1, Xi2, Xi3, are the predictor var- study, a total of 30 flexible pavement sections were randomly
iables; β1, β2, β3, are the regression coefficients; εi is the random selected, 3 at each pavement condition index level; to achieve a
error term. representative sample. Table 3 shows the different roadway con-
In order to investigate the performance of the developed dition indices that were considered in the study.
PSI model, a comparison analysis was then conducted between
the county and the WYDOT model. In addition, the developed Transport vehicles
county model was compared to a basic model that was developed
according to the ASTM approximation of MPR values (Table 2), In this study, two vehicles were selected: a sedan vehicle (2011
to confirm its validity. Ford Fusion) and SUV vehicle (2014 Ford Explorer). The four-
wheel drive (4WD) SUV vehicle was selected since the majority
of Wyoming’s locals tend to drive this type of vehicle due to the
Experiment design severe snowy weather conditions in the State of Wyoming.
The ASTM (E1927-98) standard guide was adapted in defining
the elements of this experiment. The experiment design was Panel selection
divided into three sections: (1) pavement test sections’ selection,
(2) transport vehicles and (3) panel selection. The panel size was selected based on the maximum allowed error
criteria defined by ASTM as shown in Table 4. The panel size
was defined to be 10 based on 0.4 MPR maximum error with
Pavement test sections selection normal distribution for the SUV vehicle and 6 based on 0.5 MPR
The pavement sections were selected based on homogeneous for the sedan vehicle. The panellists were chosen from different
physical properties that cover a wide range of roughness. The backgrounds as shown below:
sections were straight and long enough to maintain a fixed speed • 4 panellists: 1 woman and 3 men were selected from the
with a minimum of 25 s exposure time. Usually, the driving speed Wyoming T²/LTAP Center with an average age of 44 years.
International Journal of Pavement Engineering 5
• 3 panellists: 1 woman and 2 men were selected from the Table 6. Bartlett’ test results.
Civil Engineering Graduate Department of the University Parameters SUV vehicle Sedan vehicle
of Wyoming with an average age of 24 years. T-Statistics 13.797 9.215
• 3 panellists: 3 men were selected from the WYDOT X21−α,k−1 20.09 13.3
Design Squad with an average age of 22 years. α 0.01 0.01
roughness. For the rating process, 10 raters rated the 30 sections Mean 3.348 2.716
in 3 trips; using the SUV vehicle and 6 raters, out of the previous Variance 0.403 0.323
group, rated the same sections using the sedan vehicle in 2 trips. Observation 26 26
Pooled variances 0.363
Table 5 shows a summary of Wyoming county roads’ ride Hypothesised mean difference 0
quality study. df 50
t stat 3.779
P(T ≤ t) one-tail 0.000209
Data analysis t critical one-tail 2.403
P(T ≤ t) two-tail 0.0004
The obtained ratings for the 30 sections were entered into a t critical two-tail 2.678
spreadsheet along with the different distresses for every pave-
ment section. One section was excluded from the analysis; as it
was under reconstruction during the rating process. The individ- The Bartlett’s Test was used to test the homogeneity of vari-
ual rater’s performance was examined by plotting the individual ances between the different raters. Table 6 shows the Bartlett’s test
panel ratings vs the mean panel rating. analysis results for both vehicles. Since the values of T-Statistic
After examination of each rater’s performance when com- were less than the values of X1−𝛼,k−1
2
, the hypothesis was accept-
pared to the group, it was found that rater No. R8 rated the able and the variances were homogeneous at significance level
sections higher than others. Therefore, he was considered as an of α = 0.01 for both vehicles.
outlier and was excluded from the data analysis (Figure 2). For The variance analysis results using F-Test showed that raters’
other raters, there was no major difference or discrepancy since age, gender and different seating positions are not significant to
all the points were almost lying on the equality line. the rating process (F < F critical) at significance level of α = 0.01.
The Shapiro–Wilk’s Test was used to test the normality of However, the T-test analysis showed that there is significant dif-
the MPR. The obtained P-Values from the test were 0.0688 and ference between the MPR’s using different vehicles (t > t critical)
0.0673 for the SUV and sedan vehicles, respectively. Since these as shown in Table 7.
values are greater than 0.05; the MPR data is considered to be Using Excel regression analysis techniques, and after consid-
normally distributed for the riding quality ratings using both ering several linear and non-linear models with various types
vehicles. of transformation, the following two models were separately
developed with a logarithmic transformation for each vehicle
type:
Sedan vehicle:
INDIVIDUAL PANEL RATINGS
5.00
4.00
( )
PCI
PSI = 1.185 + 2.892e−0.003IRI − 1.469Rut2 − 1.247 1 − 100
3.00
R2 = 0.80, MSE = 0.08 (6)
2.00
SUV vehicle:
1.00
( )
PCI
0.00 PSI = 1.219 + 3.3e−0.002IRI − 4.122Rut2 − 0.475 1 − 100
0.00 1.00 2.00 3.00 4.00 5.00
R2 = 0.76, MSE = 0.09 (7)
MEAN PANEL RATING
R8 Performance Unity Line
where:
Figure 2. Performance of rater R8 compared to the group. IRI = International Roughness Index (in/miles)
6 W. A. EADELAT ET AL.
Figure 3. New models vs WYDOT model. Figure 4. WYDOT model vs ASTM model.
4
To compare between the developed models and the WYDOT
PSR ACCORDING TO
NEW MODELS
Model, both models were applied on the Wyoming county roads’ 3
indicates that model (6) can be used for rehabilitation analysis’ Recommendations
purposes; since this range of PSR values is the prime concern of
The newly developed PSI model (Equation (6)) provides more
the rehabilitation analysis (Gulen et al. 1994).
realistic representation of the current condition of county paved
Figure 6 shows the current condition of the county roads using
roads. In addition, it reflects the general satisfaction of Wyoming
the newly developed model (Equation (6)). It can be noticed that
locals with the majority of their local county roads. Hence, it
it gives better distribution of the county roads’ condition among
is recommended to use this new model in the PMS of county
all the PSI levels. It decreases the total number of sections that
roads. However, when comparing the condition of county roads
require reconstruction (PSI < 2) and increases the number of
to the state’s highway system, the state model should be used for
sections with acceptable conditions (2 < PSI < 3.5). This wide
both county and state roads. The methodology developed and
distribution of PSI values, gives reasonable reflection of the gen-
implemented in this research can be implemented in other states
eral satisfaction of Wyoming locals with the majority of their
with minor changes.
county roads. This has great importance in the implementation
of PMS and the funding allocation process for the county roads’
network. It can be challenging when 68% of the roads are in very Disclosure statement
poor condition to choose projects for rehabilitation, when only No potential conflict of interest was reported by the authors.
15% of the roadway network is in very poor condition according
to the perceptions of the real county roads customers (Wyoming
Downloaded by [University of Tennessee, Knoxville] at 02:12 09 June 2016
locals). Funding
This study was supported by the Wyoming County Commissioner
Association (WCCA), Wyoming Department of Transportation (WYDOT)
Conclusions and the State Transportation Innovation Council (STIC).
The State of Wyoming does not currently have a model to pre-
dict the serviceability status (PSR) of county roads. In this study, References
a PSI model (Equation (6)) was developed for county paved
Al-Omari, B. and Darter, I.M., 1994. Relationships between international
roads; based on the perceptions of Wyoming locals. The newly roughness index and present serviceability rating. In: Transportation
developed model is statistically valid and may provide better research record: journal of the transportation research board, No. 1435.
representation of the county roads’ condition; compared to the Washington, DC: Transportation Research Board, 132–136.
Wyoming statewide model. The main conclusions drawn from Anderson-Darling and Shapiro-Wilk Tests 2015. Information technology
laboratory. Retrieved from National Institute of Standards and
this study are:
Technology. [Online]. Available from: http://www.itl.nist.gov/div898/
(1) The mean panel ratings (MPR) of local county roads handbook/prc/section2/prc213.htm [Accessed 10 September 2015].
ASTM E1927-98, 2012. Standard guide for conducting subjective pavement
can be predicted reasonably from a wide range of dif- ride quality ratings. West Conshohocken, PA: ASTM International.
ferent roadway distresses. Available from: www.astm.org.
(2) The seating position, age and gender of the rater are Carey, J.N.W. and Irick, P.E., 1960. The pavement serviceability-
not significant factors; while vehicle’s type is signifi- performance concept. Highway Research Board Bulletin No. 250, 40–58.
cant in the rating process. Gillespie, T.D. 1985. Measuring Road Roughness and its Effects on User Cost
and Comfort: A Symposium. vol. 884., ASTM International.
(3) Using sedan vehicles is more preferable than SUVs in Gulen, et al., 1994. Correlation of present serviceability ratings with
conducting riding quality surveys, especially when it international roughness index. In: Transportation research record:
comes to the rehabilitation analysis purposes. journal of the transportation research board, No.1435. Washington, DC:
(4) Equation (6) can be used to predict with high cer- Transportation Research Board, 27–37.
tainty the PSR values for the county roads in Hernán de Solminihac, T., et al., 2003. Analysis of pavement serviceability
for the ASHTO design method: the Chilean case. Arabian Journal for
Wyoming; in order to identify the maintenance’s and Science and Engineering, 28 (2; Part B), 143–160.
different rehabilitation’s needs. Hudson, W., et al., 1985. Root-mean-square vertical acceleration as a
(5) PSI is a subjective parameter that is affected by dif- summary roughness statistic. In: Measuring Road Roughness and its
ferent experimental factors (i.e. vehicle type, vehicle Effects on User Cost and Comfort. ASTM International.
speed). However, it surpasses the solely objective Huntington, G.H., et al., 2013. Mitigating impacts of oil and gas traffic
on southeastern Wyoming county roads. Cheyenne, Laramie, WY:
parameters assessment (i.e. IRI and PCI) by consid- University of Wyoming.
ering the driving public’s perceptions. This difference Inertial Road Profiler 2016. Pathway Services Inc. Retrieved from Pathway
is crucial when it comes to maintenance’s and rehabil- Services Inc. [Online] Available from: http://www.pathwayservices.
itation’s selection process. com/inertial_road_profiler.shtml [Accessed 3 January 2016].
(6) Wyoming locals have lower expectations when it Janisch, D., 2009. An overview of Mn/DOT’s pavement condition rating
procedures and indices. [Online]. Available from: http://www.dot.
comes to the ride quality of roads at the local system state.mn.us/materials/pvmtmgmtdocs/Rating_Overview_State.pdf
compared to the statewide system. [Accessed 10 September 2015].
(7) In general, WYDOT Model (Equation1) gives Nair, S.K., Hudson, W.R., and Lee, C.E., 1985. Realistic pavement
lower PSR values for the county roads; when serviceability equations using the 690D surface dynamics profilometer.
compared to a higher roadway classification which Research Report 354-1F, Center for Transportation Research, University
of Texas at Austin, Austin, TX.
can be misrepresentative in the PMS process of Nakamura, V.F., 1962. Serviceability ratings of highway pavements. Thesis
county roads. (Masters). Purdue University.
8 W. A. EADELAT ET AL.
Pologruto, M., 1999. Vermont agency of transportation ride quality study. of Pavement Engineering, doi:http://dx.doi.org/10.1080/10298436.2015.
[Online]. Available from: https://www.google.com/url?sa=t&rct=j&q= 1065992.
&esrc=s&source=web&cd=1&cad=rja&uact=8&ved= Sayers, M.W. and Karamihas, S.M., 1998.The little book of profiling, Ann
0ahUKEwjpja7vmJbKAhVPz2MKHc2_CD0QFggcMAA&url=http% Arbor, MI: University of Michigan.
3A%2F%2Fresearch.transportation.org%2F_layouts%2 Shafizadeh, K.R., Mannering, F.L., and Pierce, L.M., 2002. A statistical
FA A SH TOR AC % 2 F F i l e D ow n l o a d. a spx % 3 F Su r ve y F i l e I D % analysis of factors associated with driver-perceived road roughness on
3D242&usg=AFQjCNHu0z501xWKkO52Byo7-Olxhyv5gA&bvm= urban highways. Research Report WA-RD538.1. Washington State
bv.110151844,d.cGc [Accessed 6 January 2016]. Transportation Center, University of Washington, Seattle, WA.
Saha, P. and Ksaibati, K., 2015. A risk-based optimisation methodology for Shahin, M.Y., 2005. Pavement management for airports, roads, and parking
pavement management system of county roads. International Journal lots. vol. 501. New York, NY: Springer.
Downloaded by [University of Tennessee, Knoxville] at 02:12 09 June 2016