Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 296

Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

Predicting Field Performance of Skid Resistance of Asphalt Concrete Pavement

Hui Wang1, Robert Y. Liang2, F, ASCE

1
Research Assistant, Department of Civil Engineering, University of Akron, Akron, OH 44325-3905;
Email: hw34@zips.uakron.edu.
2
Distinguished Professor (Corresponding Author), Department of Civil Engineering, University of
Akron, Akron, OH 44325-3905; Email: rliang@uakron.edu.

ABSTRACT: Long term abrasion due to traveling vehicle occurring on the asphalt
concrete pavement can reduce the friction, which in term can increase the risk of
highway accident. This work presents a predictive degradation model of skid
resistance for in-service asphalt pavement surface based on laboratory polishing test
results using the commercial grade asphalt polishing machine, characteristics of
aggregate gradation curve, and field traffic conditions represented by average daily
traffic. A nonlinear convex power function is used to form the prediction formula,
within which, two parameters named “time index” and “scale index” are assumed to
have first order relationship with all predictors. Validation of the developed
prediction models are presented by comparisons between the predicted values and
field measured values for five pavement sections from Ohio highway system. The
prediction results show a good consistency between field observations and model
response values. Using the predictive model and assuming that abrasion and polishing
is the controlling process in pavement functionalities and performance; then life
prediction of an asphalt concrete pavement can be determined.

INTRODUCTION

The frictional properties of pavement surfaces play an important role in highway

safety (Henry, 2000). Maintaining adequate friction (or skid resistance) is critical in
reducing crashes on wet-pavement (Bray, 2003). Four major groups of factors
contribute to development of friction at the tire-pavement interface: pavement surface
characteristics, vehicle operating parameters, tire properties, and environmental
factors (Wu et al., 2012). Since the vehicle operating parameters and environmental
factors contain too much uncertainty, many studies have attempted to establish
relationships between pavement friction and inter-surface characteristics (Hall et al.,
2009).

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 297

In pavement engineering, usually the skid resistance is expressed in terms of skid

number (SN). Measuring and monitoring SN using the Locked Wheel Skid Trailer
(LWST) is a practical way to inspect and estimate the performance of hot mix asphalt
pavement. Colony’s research (Colony, 1984; Colony, 1992) indicated that traffic
conditions, properties of construction material, and geological features were strong
contributors to variations of skid number throughout the state highways .
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

Once the measured SN is below certain threshold, resurfacing will be needed to

maintain adequate skid resistance. However, this approach is somewhat passive
toward the problem. A more rational approach would be to evaluate the mix design
and aggregate source during the material selections stage utilizing an accelerated
polishing machine to ensure that friction properties of the construction material can
be adequate for the intended design life of pavement. This proactive idea of ensuring
high polishing-resistant mix design with adequate friction after extended service life
requires that one uses the accelerated polishing and friction test results to forecast the
long-term performance of the mix design in actual pavement service conditions.
The purpose of this paper is to present a laboratory accelerated polishing device for
characterizing the polishing behavior of HMA prepared with gyratory compaction
method, along with descriptions of test results and the accompanied long-term field
SN measurement data for various mix designs and pavement sections. Finally, we
propose a predictive model for relating the laboratory polishing data to the field skid
number degradation behavior, with consideration of traffic counts and aggregate
gradation curves.

EQUIPMENT AND DATA COLLECTION

Laboratory Accelerated HMA Polishing Machine

Liang (2009) developed a research grade laboratory polishing machine. The

picture and the schematic diagram of the research grade polishing machine is
presented in FiG. 1(a) and 1(b), respectively. The polishing machine includes separate
controls for the vertical force on the specimen, the rotational speed of the rubber pad,
and the rate of water spray on the specimen surface during polishing. The guiding
principle in developing this laboratory-scale accelerated polishing machine is to
create a device where the evolution history of friction loss of the asphalt pavement
surface can be accurately replicated and measured in short test duration in a
laboratory controlled environment. In essence, the abrasive action between the rubber
tire of a vehicle and the asphalt concrete pavement surface should be properly
accomplished in the accelerated polishing machine.

Measurement of Friction Values

The friction values (i.e., British Pendulum Number (BPN)) of laboratory polished
HMA specimens were measured using the British pendulum tester (BPT). Typically,
after each one hour of polishing action, the specimen was removed from the polishing
machine to measure the friction value.

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 298

3 4.25

A DJU ST H EIG H T OF
RU BBER PA DS 20.00
ELECTRIC BO X
W ITH CRAN K

14.25
1/2 N PT N IPPLE
FOR FLUID

H M A SAM PLE

42.625
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

SH RO UD
B1 B1
B B 26.375

BU ILT FO RM

9.437

DO LLY
28.00

(a) Overall view (b) machine details

FIG. 1. Accelerated polishing machine.

Table 1. Information of hot mix asphalt material properties

Optimum Normal
Air
Polish Asphalt Asphalt Maximum
Gmb Gmn Void Designation
susceptibility Binder Content Aggregate
(%)
(%) Size (in)
Possible low
PG 64-
polish 6.3 2.324 2.390 3.5 0.375 L1
22
(Gravel)
Possible low
PG 64-
polish (Trap 5.6 2.118 2.618 4.0 0.5 L2
22
Rock)
Possible low
PG 64-
polish 6.3 2.302 2.386 3.5 0.375 L3
22
(Gravel)
Possible
medium PG 70-
5.9 2.352 2.435 3.5 0.5 M1
polish 22
(Limestone)
Possible
medium PG 64-
6.1 2.352 2.452 4.0 0.375 M2
polish 22
(Limestone)
Possible
medium PG 70-
5.6 2.451 2.549 4.0 0.375 M3
polish 22
(Dolomite)
Possible
medium PG 70-
5.9 2.417 2.521 4.0 0.375 M4
polish 22M
(Dolomite)
Possible high
PG 64-
polish 5.6 2.361 2.450 3.5 0.5 H1
22
(Limestone)

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 299

80
L1 Decrease 20%
L2 Decrease 23%
75 L3 Decrease 18%
M1 Decrease 27%
M2 Decrease 29%
70 M3 Decrease 28%
M4 Decrease 31%
H1 Decrease 35%
65

BPN
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

60

55

50

45
0 1 2 3 4 5 6 7 8
Time (hour)

FIG. 2. Experimental results of polishing test.

Based on polish susceptibility in lab condition using the research grade polishing
machine, we classified these aggregate sources roughly into three categories: low (L),
medium (M), and high (H). Table 1 provides information of jot mix material
properties.
FiG. 2 presents the decreasing trend of BPN along with 8 hours polishing process
and the percentage decrease in BPN between initial and final (After 8 hours) values
for different HMA mixes. Based on the testing results, it is concluded that there is a
nonlinear relationship between polishing time and BPN value. Materials with similar
polish susceptibility form a cluster. After 5 to 6 hours, all the materials can be easily
separated into 3 clusters (L, M, H). In this test program, the difference between H1
and others is so significant that it can be quickly identified. The materials designated
as L and M, however, take several hours of polishing to be able to discern their
polishing behavior differences.

Description of Macrotexture Using Aggregate Gradation Curve

Pavement surface texture is affected by the aggregate size and size distribution as
well as by the aggregate shape. The macrotexture is determined by the overall
properties of the pavement surface and the microtexture is mainly a function of the
surface texture of the aggregate particles. Pavement macrotexture can be modified by
changing the size and size distribution of the aggregates, while microtexture can only
be modified by changing the types of aggregates used in the mixture (Kowalski et al.,
2010). Based on these previous understanding, the particle size analysis is a proper
way to describe the pavement texture.

Long-Term Field Friction Data Collection

The field measurement program was designed to measure friction and texture of
pavement surfaces at the selected pavement locations that used the same mix
materials as tested in the laboratory test program. Friction and texture of all pavement
sections were measured annually over a three-year period from 2007 to 2010.

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 300

At each measurement point, measurements include: the skid number, which was
measured using LWST; the friction number was measured using DFT; and the mean
profile depth, was measured using CTM. In addition, BPN was also measured at the
approximately same location over the three years of field work. In general, all field
measurements were taken on the left wheel path. The DFT and CTM measurements
are the average of two runs on the left wheel path.
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

In this work, the SN(64)R is selected as the indicator of friction in the development
of predictive model.

MODEL DEVELOPMENT

Model framework and Parameter Analysis

The first step in developing the degradation model for an in-service asphalt
pavement surface is to examine the trend of friction degradation curves obtained from
lab test results using an accelerated polishing machine. Based on the trend of friction
degradation curves observed from lab test data, the convex power function was
considered as the suitable mathematical form for curve fitting of the decaying trend.
Thus, the time-dependent friction degradation curve can be expressed as Eq. 1.
t
BPN  BPN 0 (1  ) m (1)
t0
where BPN0 is the initial number of BPN, t0 and m are two parameters to be referred
to as “time index” and “scale index,” respectively.
Assuming that this mathematical equation can be extended to describe the
degradation of SN(64)R for an in-service asphalt pavement surface, then it is
necessary to determine how the time index and scale index would affect the
relationship between friction and time.
The next step in model development is to consider the influencing factors of the in-
service asphalt pavement surface that control the two scale parameters, namely time
index and scale index. Based on a literature review of relevant research work and
judging the available field data in this research, five predictors and two responses are
selected, as given in Eq. 2 and Eq. 3.
Time index: t0  1 ADT   2 PV   3tstable   4   5 (2)
Scale index: m  1 ADT   2 PV  3tstable   4  5 (3)
where ADT is Average daily traffic; PV is Polishing value; tstable is the time lasted
until the BPN is stable during the lab test; κ is scale parameter of Aggregate
Gradation Curve; λ is shape parameter of Aggregate Gradation Curve.
BPN 0  BPN8
PV 
BPN 0
BPN0: British pendulum friction number before polishing
BPN8: British pendulum friction number after 8 hours of polishing
grain size 
Gradation Curve Fitting Equation: percentage passing = 1  exp(( ) )


Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 301

Determining the Model Coefficients

The model coefficients are determined using field measurement values of SN(64)R
at different service years for the five pavement materials. The field data for
determining model coefficients are taken from the materials labeled as M1 to M4 and
L3, and the actual field traffic count. From nonlinear regression algorithm, the two
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

responses tˆ0 and m̂ can be determined using any optimization technique such as
Gauss–Newton algorithm, Newton–Raphson method, and Levenberg–Marquardt
algorithm.(Fletcher, 2013)
Set P as the factor Matrix, which contains the information of highway traffic data
(ADT) and the pavement material features, the structure of the factor matrix is as
follows:
ADT PV t stable κ λ
 b11 b12 b13 b14 b15  Material 1
P  b21 b22 b23 b24 b25  Material 2 (4)
 ...  ...
 
bn1 bn 2 bn 3 bn 4 bn 5  Material n
Once we have the factor matrix and the nonlinear regression responses: tˆ0 and m̂ ,
a multiple regression analysis can be performed to determine the coefficient vectors
of t0 and m.
  P 1 *tˆ0 (5)
  P 1 * mˆ (6)
-1
P is the general inverse of factor matrix P and vectors α and β are model
coefficients.
The friction degradation model for SN(64)R for the in-service asphalt pavement
surface is as follows:
t
SN  SN 0 (1  ) m (7)
t0
where SN 0 is the skid number at the time when the asphalt concrete pavement was
newly constructed. Based on the collected field data of this study, the multiple
regression result is given below.
t 0  6.0110 4 ADT  0.638PV  3.247t stable  61.123  387.51
m  8.87 10 5 ADT  0.006 PV  0.164t stable  2.683  12.288
It is noted that in case there is no such record of SN 0 , The relationship between
SN and BPN proposed by Kissoff and given in Eq. 8 can be employed to convert the
BPN0 of any new pavement specimen to SN0 (Kissoff, 1988).
SN  0.862 ( BPN )  9.690 (8)

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 302

MODEL APPLICATION

To validate the proposed degradation model, it is used to predict SN(64)R for three
types of pavement sections of Ohio state routes by comparisons with the measured
data collected from field work. The measurement processes and techniques are
thoroughly described in previous section and the measurements of some predictors
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

and the estimation of aggregate gradation parameters are shown in Table 2.

Substituting the predictors with the corresponding numbers in Table 2, we can
calculate the time index and scale index of each highway section. Then, by plugging
the indexes into Eq. 7, we can predict degradation of SN(64)R with in-service years.

Table 2. The predictor and responses value of five highways

PV
Highway Material ADT tstable κ λ t0 m
(%)
Wood 22 M4 11000 30.9 5 0.990772 0.164573 13.31674 -0.60463
Huron 162 M1 6000 27.2 7 1.150266 0.18598 0.002258 -0.02157
Huron 250 M2 9290 28.6 7 1.150266 0.18598 2.872943 -0.30473
Lucas 64 M3 4390 28.4 6 1.092368 0.169079 0.03801 -0.08742
Harrison
L3 1430 17.6 4 1.086712 0.17634 1.01777 -0.11387
250

The comparison of the predicted degradation curve with measured data is shown in
FiG. 3. The quality of prediction is considered acceptable considering R-squared
value and the scattering of measured data. It is noticed that the prediction curve
always goes through the initial point, because the initial SN is a constraint to the
model function. Therefore, a relatively accurate estimation of initial SN is important
for searching right trend of the empirical curve. Due to some uncertainty effects, such
as weather condition, variance of traffic volumes as well as the inevitable random
error when operating the measurement devices, the field measurements fall on both
sides of the prediction curve. Of course, using test information collected from the
laboratory test results for predicting field pavement performance may also contribute
to errors.
70 72

2 70
65
R =0.876 R2=0.885
68

60 66

64
55
62
SN

SN

60
50
58

45 56

54
40
52

35 50
0 5 10 15 20 25 0 5 10 15 20 25
t (year) t (year)

(a)Wood 22 M4 (b)Huron 162 M1

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 303

60 70

2
55
R =0.805
65
R2=0.992

50 60

45 55

SN
SN

40 50
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

35 45

30 40

25 35
0 5 10 15 20 25 0 5 10 15 20 25
t (year) t (year)

(c) Huron 250 M2 (d)Lucas 64 M3

65

R2=0.979
60

55
SN

50

45

40
0 5 10 15 20 25
t (year)

(e)Harrison 250 L3

FIG. 3. SN prediction curve of different pavement sections.

DISCUSSIONS

The Influence of Average Daily Traffic Volume

Let's focus on the prediction results of Huron 162 (M1) and Huron 250 (M2), here
both of the pavement materials contain medium polishing aggregate. The laboratory
polishing test result shows the degradation of BPN has similarity and the gradation
parameters are exactly the same. The only difference is the average daily traffic
volume. Huron 250 has a relatively higher traffic volume; therefore, it takes a long
time to reach the stable plateau and the rate of decreasing is still significant after 25
years. Huron 162, however, reaches the stable status quickly after around 10 years,
and the skid number is still about 60. Hence, the average daily traffic volume exerts
effect on the shape of degradation curve.

Interaction effect of predictors

The predictors are considered as independent with each other. This assumption is
employed to simplify the model framework. However, some interactions between
model parameters are known to exist (Table 3). Although these interactions between
predictors are likely to lower the model reliability, they were not considered in model
development.

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 304

Table 3. Correlation matrix of influencing factors

ADT PV (%) t_stable κ λ
ADT 1 0.84011 0.393985 -0.2959 -0.10469
PV (%) 1 0.601857 -0.17387 -0.19206
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

t_stable 1 0.676838 0.614159

κ Sym 1 0.911784
λ 1

Impact of correlation between factors

Further study is performed on the model. FiG. 4 shows the main effects of
different predictors. It demonstrates that the main effect of ADT and PV have the
same pattern. Similarly, the same observation can be made for parameters κ and λ.
This scenario indicates that ADT-PV and κ-λ are aliases. Therefore, it is hard to
distinguish the effects contributed to the model from ADT and PV, as well as, κ and
λ.

(b) Predictors
Predictors
(a) ADT PV (%) t_stable
ADT PV (%) t_stable 0.00
15
-0.15
10 -0.30
-0.45
5 -0.60
3 3 3 E3 E4 .6 .2 .4 .6 .9 4 5 6 7
4E 9E 6E
m

0 1 . 4.3 29 1. 1 17 27 28 28 30
9.
t_0

1430 4390 6000 9290 11000 17.6 27.2 28.4 28.6 30.9 4 5 6 7

κ λ κ λ
15 0.00
-0.15
10 -0.30
-0.45
5
-0.60
0 7 2 0 5 9 6 6
0 99 08 09 15 16 16 17 18
0.99 1.086 1.092 1.150 0.165 0.169 0.176 0.186 0. 1. 1. 1. 0. 0. 0. 0.

FIG. 4. Main effect plots for (a) t0 and (b) m.

CONCLUSIONS

The conclusions from this study can be summarized as follows.

• The friction and texture data measured in the field for the HMA pavement
sections shows significant scattering, even though efforts were made to conduct
measurements at the same mile markers and same physical spots on the wheel path.
The scattering can be attributed to high spatial variability as well as to uncertainties of
environmental effects.
• The functionalities of the new commercial grade accelerated polishing machine
for gyratory compacted 152.4 mm diameter samples have been validated. The use of
research grade polishing machine or other equivalent machines that satisfy the design
requirements can be accepted in the lab for routine testing purposes.

Pavement Materials, Structures, and Performance

 Pavement Materials, Structures, and Performance GSP 239 © ASCE 2014 305

• The prediction models for degradation of friction have been developed and
compared with field measured data. It appears that the predictive equations could be
used to forecast in service performance of asphalt pavement surface friction.

REFERENCES
Downloaded from ascelibrary.org by California Department of Transportation on 04/25/19. Copyright ASCE. For personal use only; all rights reserved.

Bray, J. (2003). "Skid Accident Reduction Program (Skarp): Targeted Crash

Reductions." Institute of Transportation Engineers (ITE) 2003 Technical
Conference and Exhibit.
Colony, D. C. (1984). "Skid Resistance of Bituminous Surfaces in Ohio." NASA
STI/Recon Technical Report N, 85, 13086.
Colony, D. C. (1992), "Influence of Traffic, Surface Age and Environment on Skid
Number"
Fletcher, R. (2013), Practical Methods of Optimization, John Wiley & Sons.
Hall, J., Smith, K., Titus-Glover, L., Wambold, J., Yager, T. & Rado, Z. (2009),
"Nchrp Web-Only Document 108: Guide for Pavement Friction." Transportation
Research Board of the National Academies, Washington, DC.
Henry, J. J. (2000), "Evaluation of Pavement Friction Characteristics."
Transportation Research Board.
Kissoff, N. V. (1988), "Investigation of Regional Differences in Ohio Pavement Skid
Resistance through Simulation Modelling." University of Toledo.
Kowalski, K. J., McDaniel, R. S. & Olek, J. (2010), "Identification of Laboratory
Technique to Optimize Superpave Hma Surface Friction Characteristics."
Liang, R. Y. (2009), "Continuing Investigation of Polishing and Friction
Characteristics of Limestone Aggregate in Ohio."
Wu, Z., King, B., Abadie, C. & Zhang, Z. (2012), "Development of Design Procedure
to Predict Asphalt Pavement Skid Resistance." Transportation Research Record:
Journal of the Transportation Research Board, 2306(1), 161-170.

Pavement Materials, Structures, and Performance