Stp1555-Eb 28948

Copyright by ASTM Int'l (all rights reserved); Wed Aug 16 21:57:48 EDT 2017
Downloaded/printed by
Universidad Nacional De Colombia (Universidad Nacional De Colombia) pursuant to License Agreement. No further reproductions authorized.
Selected Technical Papers STP1555
Pavement Performance: Current Trends,
Advances, and Challenges
Editor:
Bouzid Choubane
ASTM International
100 Barr Harbor Drive
PO Box C700
West Conshohocken, PA 19438-2959
Printed in the U.S.A.
ASTM Stock #: STP1555
Library of Congress Cataloging-in-Publication Data
ISBN: 978-0-8031-7541-9
This publication has been registered with the Library of Congress.
Library of Congress Control Number: 2012042604
Copyright 2012 ASTM INTERNATIONAL, West Conshohocken, PA. All rights reserved.
This material may not be reproduced or copied, in whole or in part, in any printed, mechanical,
electronic, lm, or other distribution and storage media, without the written consent of the
publisher.
Photocopy Rights
Authorization to photocopy items for internal, personal, or educational classroom use, or the
internal, personal, or educational classroom use of specic clients, is granted by ASTM
International provided that the appropriate fee is paid to ASTM International, 100 Barr Harbor
Drive, P.O. Box C700, West Conshohocken, PA 19428-2959, Tel: 610-832-9634; online:
http://www.astm.org/copyright.
The Society is not responsible, as a body, for the statements and opinions expressed in this
publication. ASTM International does not endorse any products represented in this publication.
Peer Review Policy
Each paper published in this volume was evaluated by two peer reviewers and at least one editor.
The authors addressed all of the reviewers comments to the satisfaction of both the technical
editor(s) and the ASTM International Committee on Publications.
The quality of the papers in this publication reects not only the obvious efforts of the authors
and the technical editor(s), but also the work of the peer reviewers. In keeping with long-standing
publication practices, ASTM International maintains the anonymity of the peer reviewers. The
ASTM International Committee on Publications acknowledges with appreciation their dedication
and contribution of time and effort on behalf of ASTM International.
Citation of Papers
When citing papers from this publication, the appropriate citation includes the paper authors,
paper title, J. ASTM Intl., volume and number, Paper doi, ASTM International, West
Conshohocken, PA, Paper, year listed in the footnote of the paper. A citation is provided as a
footnote on page one of each paper.
Printed in Bay Shore, NY

November, 2012
Foreword
THIS COMPILATION OF Selected Technical Papers, STP1555, Pavement
Performance: Current Trends, Advances, and Challenges, contains 16 papers
presented at a symposium with the same name held in Tampa, FL, USA,
December 5, 2011. The symposium was sponsored by the ASTM International
Committee E17 on Vehicle- Pavement Systems.
The Symposium Chairman and STP Editor is Bouzid Choubane, Florida
Department of Transportation, Gainesville, FL, USA.
Contents
Overview ............................................................ vii
Evaluating Polish Resistance of Tennessee Bituminous Surface Aggregates

L. K. Crouch and H. J. Brown ........................................... 1
Effect of Water, Speed, and Grade on Continuous Friction Measurement
Equipment (CFME)
E. de Len Izeppi, G. Flintsch, and K. McGhee .............................. 20
Ascertaining Parametric Correlations between Lock-Wheel Friction-Testing
andFixed-Slip Friction-Testing Device on Louisiana Highways
K. Gaspard and Z. Zhang .............................................. 40
Pavement Surface Microtexture: Testing, Characterization, and Frictional Interpretation
S. Li, S. Noureldin, K. Zhu, and Y. Jiang .................................... 59
Measuring Rolling Resistance of Typical Swedish Pavements
U. Sandberg ....................................................... 77
Measurement and Geometric and Functional Characterization of Pavement Texture
R. O. Rasmussen and R. C. Sohaney ..................................... 96
Performance of Exposed Aggregate Surface of Composite Pavements at MnROAD
B. Izevbekhai and A. Akkari ............................................ 110
Analyzing Spectral Repeatability of Different Asphalt Pavements Using the OBSI
Method in New Jersey and Massachusetts
J. Hencken, E. H. Haas, and T. Bennert .................................... 135
Estimating Acoustic Performance Trends of Various Asphalt-Surface Mixtures Based
on Field Measurement
Q. Lu ............................................................ 150
Measuring and Visualizing TirePavement Noise Using On-Board Sound Intensity
R. O. Rasmussen and R. C. Sohaney ..................................... 160
Simplified Deflection-based Probabilistic Technique for Network-level
Pavement Analysis
R. Carvalho, R. Stubstad, L. Wiser, and O. Selezneva .......................... 170
Surface Deflections of Perpetual Pavement Sections
D. S. Gedafa, M. Hossain, R. Miller, and D. A. Steele .......................... 192
Case Studies Using Falling Weight Deflectometer Data with Mechanistic-empirical
Design and Analysis
L. M. Pierce, K. D. Smith, J. E. Bruinsma, and N. Sivaneswaran .................. 205
Practical Implementation of Automated Fault Measurement Based on
Pavement Profiles
G. K. Chang, J. Watkins, and R. Orthmeyer ................................. 219
Proactive Pavement Smoothness from the Base Upward
S. N. Goodman, R. Arsenault, and T. Dziedziejko ............................. 238
Golden-Car Simulation Speed and Its Implications to the Relevance of the IRI
S. M. Karamihas .................................................... 248
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Overview
Pavement condition assessment and surface characteristics measurements are
important tools in evaluating the performance and management of roadway
systems. These tools provide information critical to support informed decision-
making for determining cost-effective rehabilitation and preservation strate-
gies, identify potentially hazardous conditions, and monitor the condition and
surface characteristics of the various in-service pavements. This need for such
tangible and quantifiable information has resulted in significant innovations
and advances in pavement evaluation techniques, equipment, and practices.
However, with the ever evolving technologies and related data interpretation
methodologies, more venues for sharing, documenting, and disseminating infor-
mation are needed.
On December 11, 2011, an ASTM International Symposium, addressing a
broad range of topics related to pavement-vehicle interaction and pavement per-
formance/condition assessment, was held in Tampa, Florida. The presentations
at that symposium represented an international effort in both the practical as
well as the developmental aspects of pavement evaluation procedures and tech-
nologies. The symposium also provided a forum for participants and attendees to
gain insight regarding the trends, advances, and challenges in the areas of pave-
ment management, condition survey, pavement-tire interaction, and structural
adequacy from the analytical and experimental points of view.
This Special Technical Publication (STP) is a compilation of selected papers
presented at this International Symposium. The papers provide additional refer-
ence material for those concerned with pavement performance evaluation and
characterization. They cover topics that will be of interest to practitioners as well
as to researchers.
The editor wishes to acknowledge all those who participated in the Sym-
posium, those who contributed to this Special Technical Publication (STP),
and the many reviewers who provided important feedback to the authors. The
editor also wishes to acknowledge the ASTM International Committee E17 on
Vehicle-Pavement Systems for sponsoring the symposium and the ASTM Inter-
national staff for their assistance with the organization of the symposium and
publication of this volume. The editor is grateful for their diligent efforts and
contributing knowledge.
Bouzid Choubane
Florida Department of Transportation
State Materials Office, Gainesville, Florida
Symposium Chair and Editor
vii
Pavement Performance: Current Trends, Advances, and Challenges
STP 1555, 2012
Available online at www.astm.org
DOI:10.1520/STP104347
L. K. Crouch1 and Heather J. Brown2
Evaluating Polish Resistance of Tennessee

Bituminous Surface Aggregates
REFERENCE: Crouch, L. K. and Brown, Heather J., Evaluating Polish Re-

sistance of Tennessee Bituminous Surface Aggregates, Pavement Perform-
ance: Current Trends, Advances, and Challenges on December 5, 2011 in
Tampa, FL; STP 1555, B. Choubane, Editor, pp. 119, doi:10.1520/
STP104347, ASTM International, West Conshohocken, PA 2012.
ABSTRACT: The Tennessee Dept. of Transportation (TDOT), as part of an
overall plan to achieve minimum HMA resurfacing cycle goals of 8 yr on inter-
states and 12 yr on state routes, initiated a project to pair aggregate polish
resistance with pavement functional needs. A new pre-evaluation procedure
for aggregate polish resistance was developed and named the Tennessee
Terminal Textural Condition Method (T3CM). Thirty-two bituminous surface
projects were selected for the T3CM evaluation. The projects selected
included both test strips and in-service bituminous surfaces. Further, several
projects contained coarse aggregates from the same source. When possible,
the coarse aggregates used in the projects were sampled at the HMA cold-
feed stockpile rather than the aggregate source to ensure that the exact
aggregates placed in the bituminous surface course were tested. The aggre-
gates were subjected to two physical tests: T3CM and British polishing
wheel/British pendulum, as well as three chemical/mineralogical tests: Per-
cent silica ASTM C25, percent silica by x-ray diffraction, and loss-on-ignition
(LOI). In addition, the bituminous surface courses containing these aggre-
gates were monitored for skid resistance (i.e., polish resistance) by TDOT
personnel using the locked-wheel trailer. A plot of project skid data versus
time was considered to have a terminal skid number if the later skid results
were stable or increasing. No project was considered to have had a terminal
skid number unless more than three skid data points were available. Correla-
tion coefficients between laboratory test methods and terminal skid numbers
of bituminous surface aggregates were 0.705, 0.065, 0.076, 0.366, and
Manuscript received September 16, 2011; accepted for publication March 26, 2012; published
online October 2012.
1
P.E., Ph.D., Dept. of Civil and Environmental Engineering, Tennessee Technological Univ.,
Cookeville, TN 38505, e-mail: lcrouch@tntech.edu
2
Ph.D., Concrete Industry Management Program, Middle Tennessee State Univ., Murfreesboro,
TN 37132, e-mail: hjbrown@mtsu.edu
Copyright V
C 2012 by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA
19428-2959.
1
2 STP 1555 ON PAVEMENT PERFORMANCE
0.116 for the T3CM, the British Pendulum and British Wheel Method, ASTM
C25 percent silica, LOI, and x-ray diffraction percent silica, respectively. Infor-
mation is also presented on aggregate variability characterized by the T3CM,
the British Pendulum, and British Wheel Method, and ASTM C25 percent
silica.
KEYWORDS: aggregate, loss-on-ignition, bituminous, silica, polish resistance
Introduction
Long-lasting asphalt pavement surfaces require coarse aggregates that resist pol-
ishing sufficiently to provide adequate skid resistance throughout the life of the
surface course. If unsatisfactory materials can be eliminated, savings in accident
costs, maintenance, and reconstruction will result. However, there is ambiguity
in the term unsatisfactory. An aggregate, which is satisfactory for most aver-
age daily traffic (ADT) levels, may be unsatisfactory for high ADT interstate
applications. In eliminating unsatisfactory materials, caution is often consid-
ered prudent and, subsequently, some materials, which could provide adequate
performance in many ADT applications, are eliminated from consideration
because these materials will not meet specifications for high ADT levels.
The state of Tennessee has an abundance of carbonate aggregates espe-
cially in middle and east Tennessee. However, in the past, Tennessee also had
significant sources of highly polish-resistant bituminous surface aggregates
such as slag in these same areas. Cautious specifications defining what an
approved surface aggregate source is posed no problem when a large supply
of highly polish-resistant aggregates was available. However, when the supply
of these excellent aggregates declined, the cost of producing safe bituminous
pavement surfaces increased. With only a small number of approved aggregate
sources, two problems began to occur. First, as demand increased, the aggre-
gate price increased. Second, transporting aggregates from approved sources to
distant areas increased pavement costs.
In 1992, the Tennessee Dept. of Transportation (TDOT) initiated a project
to pair aggregate performance with the pavement functional needs (based on
ADT) so that all Tennessee aggregate sources could be used efficiently.
Research Objectives
This project was undertaken to achieve the following objectives:
1. Ascertain what laboratory methods are currently available to pre-
evaluate aggregate polish-resistance for bituminous surfaces.
2. Determine the relative effectiveness of the methods through a literature
review.
3. If no suitable methods are found, attempt to develop a test to character-
ize an aggregates ability to retain microtexture over time. The test
must be inexpensive, repeatable, and not operator sensitive.
CROUCH AND BROWN, doi:10.1520/STP104347 3
4. Perform a preliminary evaluation of the new method.

5. Make recommendations for implementation of project findings.
Objectives 1 and 2 were accomplished in phase 1 of the project [1,2]. Infor-
mation on laboratory test methods for characterizing aggregate polish resist-
ance currently available or being developed was obtained. A survey was made
of State Departments of Transportation throughout the country, along with
associated industry and academia. Only three standardized laboratory tests
were commonly used by respondents: the percent insoluble residue [3], petro-
graphic analysis [4], and British polishing wheel/British pendulum [5,6].
Objectives 3, 4 and 5 were part of Phase 2 of the project. The Tennessee tex-
tural retention method (TTRM) was developed and evaluated under phase 3
[7,8]. The TTRM was found to be repeatable, not operator sensitive, and inex-
pensive. However, the TTRM did not reveal an aggregates terminal texture.
Because of this, TDOT sponsored a fourth phase of the project to evaluate
and refine the Tennessee Terminal Textural Condition Method (T3CM) that was
developed to characterize a surface aggregates probable terminal texture condi-
tion [9,10]. In phase 4, 32 bituminous surface projects were selected for evalua-
tion. The projects selected included both test strips and in-service bituminous
surfaces. Further, several projects contained aggregates from the same source.
When possible, the coarse aggregates used in the projects were sampled at the
HMA cold-feed stockpile rather than the aggregate source to ensure that the
exact aggregates placed in the bituminous surface course were tested. The aggre-
gates were subjected to two physical tests: T3CM and British polishing wheel/
British pendulum, as well as three chemical/mineralogical tests: percent silica
ASTM C25 [11], percent silica by x-ray diffraction, and loss-on-ignition (LOI).
Further, the bituminous surface courses containing these aggregates were moni-
tored for pavement distress and skid resistance. Skid resistance data was obtained
by TDOT personnel using the locked-wheel trailer skid tests [12].
By comparing the skid performance versus traffic with laboratory test
results, aggregate performance was categorized for effective use. The informa-
tion generated in the project will be used to help TDOT achieve minimum
HMA resurfacing cycle goals of eight years on interstates and twelve years on
state routes. Further, with all Tennessee bituminous surface aggregates cor-
rectly categorized, TDOT personnel can make informed decisions about the
most economical aggregate to meet the pavements functional requirements.
As a result, the department will save money and extend the supply of approved
bituminous surface aggregates while maintaining a high level of skid resistance
to ensure the motoring publics safety.
Background and Literature Review

There are many factors affecting skid resistance. The three primary factors influ-
encing skid resistance performance of bituminous roadways are pavement
distress, macrotexture, and microtexture. The pavement distress modes of rutting

and bleeding greatly decrease skid resistance [13]. Surface rutting allows water
to pond in the ruts. Bleeding drowns aggregate microtexture, thus decreasing
skid resistance due to loss of tire-pavement interaction. Macrotexture is the result
of the size, shape, and arrangement of aggregate particles in the mix. For a pave-
ment that is not rutted or excessively bleeding, macrotexture controls water film
thickness on the roadway surface [14]. Macrotexture relieves the water pressure,
which is built up in the forward portion of the tire-pavement interface; thus
allowing a large portion of the tire area to remain in contact with the pavement
surface [15]. Macrotexture is a measure of the general coarseness of the pave-
ment. Macrotexture also controls seasonal variations seen in the skid number
(SN) readings taken in the field [16]. Microtexture is the fine texture of the
aggregate that makes it smooth or rough to the touch. Microtexture provides the
adhesion component of skid resistance [15]. Goodwin [17] and Gramling [18]
found that while accumulating field skid resistance research data, the coarse ag-
gregate in bituminous mixtures was more influential in determining skid resist-
ance than other mix constituents. Nitta et al. [19] agreed, indicating that mixture
component changes such as asphalt content, type of fine aggregate and coarse ag-
gregate grading are less important than coarse aggregate type.
Based on field experience, a highly skid-resistant surface course will reach
a terminal skid number during the pavement service life. Research in New Jer-
sey on bituminous pavement showed cyclic skid resistance values with a con-
stant mean value, hereafter referred to as the terminal skid number, may be
expected after approximately 2 106 vehicle passes [20]. After a pavement has
reached its terminal skid number, the pavements skid resistance is primarily
affected by cyclic seasonal effects [20].
Several researchers including Kandhal et al. [21], Nitta et al. [19], and
Skerritt [22] agree that aggregates will reach an ultimate polishing value or ter-
minal textural condition. This behavior is similar to that in the field (measured
with the locked-wheel trailer, AASHTO T242) as the pavement is exposed to
more traffic repetitions. An aggregates terminal textural condition is the mini-
mum surface roughness and particle angularity achieved when the aggregate is
subject to polishing and wear. For bituminous surface courses which are not
rutted or significantly bleeding, the terminal textural condition of the coarse ag-
gregate particles is the controlling factor in determining the terminal skid num-
ber. Therefore, adequate characterization of an aggregates terminal textural
condition is essential in determining its contribution to the skid resistance of
the bituminous surface course.
Dirringer used the British Polishing Wheel and British Pendulum to
achieve a terminal aggregate texture in the laboratory [20]. The specimens
were polished far beyond the standard nine-hour period. Unfortunately, Kula-
kowski et al. [23] found considerable variation in the British Pendulum results.
The cause of the variation was considered to be the complex, cumbersome and
ineffective calibration procedures. Therefore, a new method for evaluating

aggregate polish resistance was needed.
To summarize, the literature review suggests that any new pre-evaluation
method for bituminous surface aggregates should:
Concentrate on the terminal textural condition of aggregates,
Be repeatable, and
Be operator insensitive.
Summary of Tennessee Terminal Textural Condition Method (T3CM)
General
The T3CM Voids Device is based on a modification of the AASHTO T304
Uncompacted Voids Apparatus for Fine Aggregate [24]. The modification con-
sists of a larger reservoir to hold the aggregate and a larger funnel through
which the aggregate flows to fill the reservoir. The reservoir used in the modi-
fied apparatus is a 2832-mL AASHTO T19 mold [25]. The funnel was chosen
to have approximately the same shape as the standard funnel yet have a volume
equal to twice that of the modified cylindrical reservoir. The remainder of the
apparatus was increased in size proportional to the new reservoir (see Fig. 1 for
details).
FIG. 1Schematic T3CM uncompacted voids content apparatus modified

AASHTO T304
Procedure
The complete procedure for the T3CM is described in the second authors dis-
sertation [26]. In general, five 12-kg samples are alternatively aged for 500 rev-
olutions in the Los Angeles Abrasion Machine [27] using no steel spheres and
tested using a modified AASHTO T304 (T3CM) apparatus until a terminal tex-
tural condition (TTC) is reached. To insure that each sample is comprised of a
single size (6.39.5 mm) aggregate, each sample is re-sieved after each 500-
revolution aging cycle to eliminate all finer particles. The samples were contin-
ually aged 500 revolutions at a time and sieved until all specimens had reached
8000 revolutions. Prior to 8000 revolutions, the number of samples may be
reduced depending on the degradation of the aggregate without wasting any
material. No individual sample was allowed to exceed 12 kg. No further
uncompacted voids tests were performed prior to 8000 revolutions. When each
sample has been aged and sieved for 8000 revolutions, all samples are com-
bined and mixed. If the combined sample exceeded 24 kg, it was reduced to 24
kg by wasting. The combined sample was then split into two 12-kg samples.
The two samples at 8000 revolutions were tested using the modified AASHTO
T304 device. Three tests were conducted on each sample for a total of six tests.
Following testing, the samples were continually aged, sieved, and tested every
500 revolutions until each sample had been aged 14 000 revolutions. When
each sample reached 14 000 revolutions, the two samples were combined and
mixed. If the combined sample exceeded 12 kg, it was reduced to 12 kg by
wasting. The sample was then tested six times with the modified AASHTO
T304 device. Following the testing procedure, the sample was continually
aged, sieved, and tested every 500 revolutions until a terminal condition was
reached (see next paragraph) or the sample was exhausted.
Good bituminous surface aggregates achieve a higher terminal textural
condition. That is, after a certain amount of aging (abrasion and wear), the
aggregates angularity and surface texture begin to regenerate and no longer
decline. Not all aggregates have a terminal textural condition. Specifically, an
aggregate is considered to have reached a TTC if the linearly regressed slope
(on a plot of uncompacted voids versus aging revolutions) of the last seven
points has a 0.00003 or greater slope. This value was determined experimen-
tally. The limiting slope was chosen for two reasons. First, the vast majority of
TDOT proven-performing limestones had a slope greater than this value. Sec-
ond, it allowed for some oscillation about the terminal textural condition simi-
lar to the oscillation of field skid numbers. For aggregates in a terminal textural
condition, the Tennessee Terminal Textural Condition Rating is equal to the
average of the last seven average percent air voids tests. Occasionally, the
slope of the curve will temporarily increase above 0.00003 before the aggre-
gate is truly in a terminal condition. To avoid erroneous results, testing should
be continued until TTC has been achieved four consecutive times before
reporting a TTC for any aggregate. The final TTC reported is the mean of the
four TTC values. Poor performing bituminous surface aggregates continue to
polish and do not achieve a terminal textural condition.
Rationale
Many Tennessee limestones are composed of a combination of hard and soft
materials. These materials have differential polishing rates under traffic. Softer
materials in the aggregate particles at the pavement surface polish faster than the
harder component, which regenerates the particle surface texture. The central
question in allowing the use of such particles is the minimum amount of harder
materials in the particles necessary to achieve an acceptable regenerative texture.
In the past, chemical and mineralogical methods have been used in an attempt to
answer this question. However, neither chemical nor mineralogical methods
reveal anything about the distribution of hard constituents in the aggregate par-
ticles. The T3CM attempts to determine regenerative texture physically by estab-
lishing a terminal textural condition (TTC) for an aggregate.
Results
Results of the laboratory tests on the 32 project aggregates are shown in Table 1.
TDOT personnel evaluated the British Pendulum Number after 9 h of polishing
(BPN 9) and percent silica by ASTM C25 of the selected aggregates. Tennessee
Tech researchers conducted percent silica by x-ray diffraction, LOI, and T3CM.
LOI values were obtained by determining the percent weight loss when a 600-g
sample of the aggregate was subjected to 950 C for 8 h in a muffle furnace.
A summary of field test results and related information is shown in Table 2.
TDOT measures SN40 in the left wheel path of the AASHTO A lane with a
ribbed tire at 64 kph (40 mph). No project was considered to have had a terminal
skid number unless more than three skid data points were available. A plot of
project skid data versus time was considered to have a terminal skid number if
the later skid results were stable or increasing. The procedure used to determine
which projects had a terminal skid number was both subjective and optimistic.
However, the author needed at least some terminal skid numbers for comparison
to the laboratory data. It is certainly possible that some terminal skid number
judgments (based on the very limited available data) were in error.
Twelve aggregate sources were subjected to a preliminary laboratory eval-
uation for possible use in Tennessee bituminous surface courses. A summary
of the phase 4 aggregate source evaluations is shown in Table 3.
Two aggregates were selected for random production sampling. Each ag-
gregate source was sampled and tested at approximately three month intervals
for one year. The results of the random production sample testing are shown in
Table 4.
TABLE 1Laboratory test results for 32 project aggregates.
Aggregate Project TTC BPN9 LOI (%) X-ray silica (%) ASTM C25 silica (%)
1 1 44.51 33 17.05 63.59 53.20
1 2 43.40 35 17.08 58.19 46.56
1 3 43.40 31 23.60 55.70 46.44
1 4a 43.89 32 15.91 61.46 53.12
1 4b 43.67 35 16.68 61.69 53.90
2 1 43.71 21 NA NA 19.10
2 2 43.18 23 35.50 21.93 18.80
2 3a 42.71 NA 35.73 23.86 19.84
2 3b 43.03 NA 32.33 21.48 18.60
2 4 42.97 NA 36.80 21.69 15.48
2 5 43.22 22 34.77 21.06 19.44
3 1 43.80 32 20.82 57.92 47.28
3 2 43.54 34 19.29 57.63 45.60
3 3 44.59 30 21.48 41.42 36.36
3 4 44.73 27 12.5 29.57 22.32
4 1 44.71 33 20.15 53.71 43.70
4 2 44.63 31 23.82 46.75 36.60
4 3 44.71 NA 21.77 46.51 36.64
4 4 44.77 32 24.06 43.82 34.60
5 1 46.41 26 23.08 52.44 43.20
5 2 44.29 29 22.43 51.65 42.96
5 3 43.21 38 18.01 58.39 45.40
5 4 44.37 26 20.53 49.39 43.12
6 1 43.03 22.4 36.90 17.15 12.77
6 2 43.37 19 36.98 16.51 12.56
7 1 40.63 34 1.72 97.72 88.10
7 2 40.64 36 1.67 97.82 85.10
8 1 44.46 31 23.42 50.30 40.80
8 2 44.73 30 8.18 52.54 41.60
9 1 44.20 31 22.75 44.91 36.00
10 1 43.08 NA 33.43 21.71 14.76
11 1 45.00 22 0.05 NA 36.88
Analysis of Results
Correlation with Pavement Performance

The current project focused on correlating aggregate polish resistance to field
performance. Figures 27 show correlations between available terminal skid
numbers and TTC, LOI, X-ray percent silica, ASTM C25 percent silica, and
BPN 9 values, respectively.
Table 5 shows correlation coefficients between these laboratory pre-
evaluation methods and available field terminal skid numbers.
TABLE 2Field test results and related information for 32 project aggregates.
Completion ADT Number of SN 40 Terminal

Aggregate Project date provided provided points provided skid number?
1 1 No No 3 No
1 2 No No 0 No
1 3 Yes Yes 7 50
1 4a No No 4 46
1 4b No No 4 46
2 1 No No 1 No
2 2 Yes Yes 5 37
2 3a Yes Yes 9 No
2 3b Yes Yes 9 No
2 4 Yes Yes 3 No
2 5 Yes Yes 4 41
3 1 No No 3 No
3 2 Yes Yes 6 41
3 3 No No 1 No
3 4 Yes No 5 50.5
4 1 No No 2 No
4 2 No No 5 52
4 3 Yes Yes 8 54
4 4 No No 3 No
5 1 No No 1 No
5 2 Yes Yes 7 50
5 3 Yes Yes 9 45
5 4 No No 1 No
6 1 No No 1 No
6 2 No No 0 No
7 1 No No 2 No
7 2 No No 4 No
8 1 No No 1 No
8 2 Yes No 4 54.5
9 1 No No 2 No
10 1 No No 3 No
11 1 Yes Yes 2 No
The T3CM had the highest correlation coefficient by far. The correlation
coefficients for the other laboratory procedures indicate that for the available
data there is no reliable relationship. In the 1970s, AASHTO realized that
ASTM C25 percent silica is a good indicator of aggregates that are likely to be
highly polish susceptible. However, the method should not be used as a princi-
ple means of predicting aggregate polish resistance [15]. Further, percent silica
by X-ray diffraction measures the same thing (see Fig. 7) with only a slightly
better correlation with terminal skid performance. Finally, LOI, which had the
TABLE 3Summary of phase 4 aggregate source evaluations.
Aggregate TTC BPN 9 LOI (%) X-ray silica (%) ASTM C25 silica (%) Recommendation
6 43.35 NA 35.83 15.10 14.68 Already approved
7 40.68 NA NA NA NA Already approved
8 44.41 NA NA 49.62 NA Test strip
9 44.17 30 25.55 42.10 42.08 Test strip
12 42.56 NA 17.98 50.10 37.20 Test strip
13 42.40 NA 1.05 NA 82.64 Already approved
14 42.90 NA 33.20 12.20 14.53 Test strip
15 42.77 23 NA NA 3.57 No test strip
16 42.00 NA NA NA NA None
17 NA NA NA NA 17.88 None
18 42.33 NA 40.66 9.81 NA No test strip
19 43.06 31 27.40 NA 13.12 Test strip
second best correlation coefficient, is simply an improved percent silica test,

easier to perform, more repeatable and possibly more accurate because of a
much larger sample size.
LOI provides the same information as percent silica. Fig. 8 shows the cor-
relation between ASTM C25 percent silica and LOI. In Fig. 9, there appears to
be two outlying points. The possible outlying points are shown encased in rec-
tangles. Correlation coefficients are provided with and without possible
outliers.
Aggregate Variability
Average laboratory properties as well as statistical information such as prop-
erty range and coefficient of variation are important in selecting the correct sur-
face aggregate for a particular bituminous surface course. Table 6 shows
average laboratory test results for project aggregates that had more than three
TABLE 4Random production sample aggregate test results.
Aggregate Date sampled TTC ASTM C25 silica (%) BPN 9

3 2/20/02 44.15 48.92 33
3 5/22/02 43.84 45.90 32
3 8/30/02 44.52 45.90 36
3 11/22/02 43.58 39.2 35
8 1/27/03 44.50 42.4 30
8 4/11/03 44.47 40.0 31
8 7/09/03 44.07 40.8 30
8 10/13/03 44.57 41.8 32
FIG. 2Terminal skid number versus TTC.
FIG. 3Terminal skid number versus LOI.

FIG. 4Terminal skid number versus x-ray percent silica.
FIG. 5Terminal skid number versus ASTM C25 percent silica.

FIG. 6Terminal skid number versus BPN 9.
FIG. 7X-ray percent silica versus ASTM C25 percent silica.
TABLE 5Correlations between laboratory pre-evaluation methods and terminal skid numbers.
Test method Correlation coefficient with available terminal skid numbers

T3CM 0.7051
LOI 0.3662
X-ray diffraction percent silica 0.1157
ASTM C25 percent silica 0.0756
BPN 9 0.0652
FIG. 8ASTM C25 percent silica versus LOI.
field projects. Similarly, Tables 7 and 8 show the range and coefficient of varia-
tion of laboratory test results for the aggregates in Table 6.
Table 9 shows the ranking (1 lowest, 5 highest) of aggregate variability
for each laboratory test method. The cumulative rank is the sum of all
FIG. 9X-ray percent silica versus LOI.

TABLE 6Average laboratory test results of project aggregates with more than three projects.
Number of X-ray ASTM C25

Aggregate projects LOI (%) percent silica percent silica TTC BPN 9
1 5 18.06 60.13 50.64 43.77 33.20
2 6 35.03 22.00 18.54 43.14 22.00
3 4 18.52 46.64 37.89 44.17 30.75
4 4 22.45 47.70 37.89 44.71 32.00
5 4 21.01 52.97 43.67 44.57 29.75
TABLE 7Range of laboratory test results of project aggregates (3 projects).

1 5 7.69 7.89 7.46 1.11 4.00
2 6 4.47 2.80 4.06 1.00 2.00
3 4 8.98 28.35 24.96 1.19 7.00
4 4 3.91 9.89 9.10 0.14 2.00
5 4 5.07 9.00 2.28 3.20 12.00
TABLE 8Coefficient of variation of laboratory test results of project aggregates (3 projects).

1 5 17.33 5.23 7.49 1.05 5.39
2 6 4.78 4.93 8.44 0.77 4.55
3 4 22.23 29.47 30.18 1.32 9.71
4 4 8.22 8.85 10.54 0.13 3.13
5 4 10.83 7.25 2.65 3.00 19.09
TABLE 9Rank of coefficient of variation of laboratory test results of project aggregates with more
than three projects.
X-ray ASTM C25 Cumulative

Aggregate LOI (%) percent silica percent silica TTC BPN 9 rank
1 4 2 2 3 3 14
2 1 1 3 2 2 9
3 5 5 5 4 4 23
4 2 4 4 1 1 12
5 3 3 1 5 5 17
TABLE 10Random production sample statistics.
Aggregate 3 8
TTC average 44.02 44.40
TTC range 0.94 0.50
TTC Coefficient of variation (%) 0.92 0.51
ASTM C25 percent silica average 44.98 41.25
ASTM C25 percent silica range 9.72 2.40
ASTM C25 percent silica coefficient of variation (%) 9.13 2.58
BPN 9 average 34 30.75
BPN 9 range 4 2
BPN 9 coefficient of variation (%) 5.37 3.11
individual laboratory test method rankings for an individual aggregate. The cu-
mulative rank can be used as a measure of overall aggregate variability. Aggre-
gates 2 and 4 had relatively low variability compared to the other aggregates
used in the study. Conversely, aggregates 3 and 5 had relatively high variability
compared to the other aggregates used in the study.
Table 10 shows a statistical analysis of the results of the random produc-
tion sample testing. Aggregate source 8 exhibited lower variability than aggre-
gate source 3 for all laboratory test procedures conducted.
Conclusions
On the basis of the laboratory data, limited terminal skid numbers available
and the preliminary analysis done, the following conclusion can be drawn. The
T3CM appears to be able to better discern the performance of Tennessee bitu-
minous surface aggregates than the British Pendulum and British Wheel
Method, ASTM C25 percent silica, LOI, or X-ray diffraction percent silica.
Correlation coefficients indicating the strength of relation between laboratory
test methods and field skid performance of bituminous surface aggregates
showed T3CM to be vastly superior to the other methods. In summary, the
T3CM appears to be a conservative, reliable surface aggregate pre-qualification
method. T3CM measures what is really important in selecting bituminous sur-
face aggregate, which is the aggregates terminal textural condition.
Acknowledgments
The writers gratefully acknowledge the support of the Tennessee Department
of Transportation and the Federal Highway Administration. The writers also
thank the many aggregate producers in Tennessee, KY, NC, and Arkansas
whose cooperation made the project possible. In addition, the authors thank
Richard Maxwell, Tim Dunn, and Perry Melton for their patience and skill in
fabrication, maintenance, and repair of the equipment. The writers thank Patty
Buchannan, Jason Laxson, Neal Whitten, Jake Williams, Leslie Parker, John
Gravely, John Davis, Brandon Hill, John Dudley, Bart Saucier, Vern Prentice,
Keith Honeycutt, Todd Walker, Jamey Dotson, Adam Ledsinger, Bart
Romano, Kevin Cagle, Nicky Wells, Jamie Waller, Rob Bailey, Kim Couch,
George Zima, Adam Borden, Shane Beasley, Adam Walker, Charles Tucker,
Jason Burgess, Joel Gothard, Greg Shirley and Tony Zhang for their help in the
laboratory. The writers appreciate the administrative and information technol-
ogy support provided by the Tennessee Technological University Center for
Energy Systems Research. Disclaimer: The opinions, findings, and conclusions
expressed here are those of the authors and not necessarily those of the Tennes-
see Dept. of Transportation or the Federal Highway Administration.
References
[1] Crouch, L. K., Gothard, J., Head, G., and Goodwin, W., Evaluation of
Textural Retention of Pavement Surface Aggregates, Transportation
Research Record 1486, TRB National Research Council, Washington,
D.C., 1995.
[2] Gothard, J., 1994, Evaluation of Aggregate Polish-Resistance by the
TTU Textural Retention Method, Master of Science thesis, TN Techno-
logical Univ., Cookeville, TN.
[3] ASTM D3042-97, 2003, Standard Test Method for Insoluble Residue in
Carbonate Aggregates, Annual Book of ASTM Standards, Vol. 04.03,
ASTM International, West Conshohocken, PA, pp. 284286.
[4] ASTM C295-03, 2004, Standard Guide for Petrographic Examination of
Aggregates for Concrete, Annual Book of ASTM Standards, Vol. 04.02,
ASTM International, West Conshohocken, PA, 2004, pp. 189196.
[5] AASHTO T287-90 (1999), 2003, Standard Method of Test for Surface
Frictional Properties Using the British Pendulum Tester, AASHTO
Standard Specifications for Transportation Materials and Methods of
Sampling and Testing, Part 2B Tests, 23rd Ed., Washington, D.C.
[6] AASHTO T279-96 (2001), 2003, Standard Method of Test for Acceler-
ated Polishing of Aggregates Using the British Wheel, AASHTO Stand-
ard Specifications for Transportation Materials and Methods of Sampling
and Testing, Part 2B Tests, 23rd ed., Washington, D.C.
[7] Shirley, G., 1995, Pre-Evaluation of Polishing Resistance of Aggregates
for Tennessee Bituminous Surface Courses, Master of Science thesis,
TN Technological Univ., Cookeville, TN.
[8] Crouch, L. K., Shirley, G., Head, G., and Goodwin, W., Aggregate Pol-
ishing Resistance PreEvaluation, Transportation Research Record
1530, TRB National Research Council, Washington, D.C., 1996.
[9] Tucker, C., 1997, Aggregate Terminal Textural Condition, Master of

Science thesis, TN Technological Univ., Cookeville, TN.
[10] Burgess, J., 1998, Evaluating Bituminous Surface Aggregates with the
T3CM, Master of Science thesis, TN Tech Univ., Cookeville, TN.
[11] ASTM C25-93a, 1994, Standard Test Methods for Chemical Analysis of
Limestone, Quicklime, and Hydrated Lime, Annual Book of ASTM
Standards, Vol. 04.01, ASTM International, Philadelphia, PA, pp. 936.
[12] AASHTO T242-96 (2000), 2003, Standard Method of Test for Frictional
Properties of Paved Surfaces Using a Full-Scale Tire, AASHTO Standard
Specifications for Transportation Materials and Methods of Sampling and
Testing, Part 2B Tests, 23rd ed., Washington, D.C.
[13] Standard Nomenclature and Definitions for Pavement-Components and
Deficiencies, Highway Research Board Special Report 113, Highway
Research Board, National Research Council, Washington, D.C., 1970, p. 6.
[14] Yager, T. and Buhlmann, F., Macrotexture and Drainage Management
on a Variety of Concrete and Asphalt Surfaces, Pavement Surface Char-
acteristics and Materials, ASTM STP 763, C. M. Hayden, Ed., American
Society for Testing and Materials, Philadelphia, 1982, pp. 1630.
[15] The American Association of State Highway and Transportation Officials,
1976, Guidelines for Skid Resistant Pavement Design, Washington, D.C.
[16] Hill, B. and Henry, J., Surface Materials and Properties Related to Sea-
sonal Variations in Skid Resistance, Pavement Surface Characteristics
and Materials, ASTM STP 763, C. M. Hayden, Ed., American Society for
Testing and Materials, Philadelphia, 1982, pp. 4560.
[17] Goodwin, W., Evaluation of Pavement Aggregates For NonSkid
Qualities, Twelfth Annual Symposium on Geology as Applied to Highway
Engineering, Univ. of Tennessee, Knoxville Tennessee, Feb 1011, 1961.
[18] Gramling, W., Effect of Aggregate Mineralogy on Polishing Rate and
Skid Resistance in Pennsylvania, Highway Research Report 341, HRB,
National Research Council, Washington, D.C., 1970, pp. 1821.
[19] Nitta, N., Saito, K., and Isozaki, S., Evaluating the Polishing Properties
of Aggregates and Bituminous Pavement Surfaces by Means of the Penn
State Reciprocating Polishing Machine, Surface Characteristics of Road-
ways: International Research and Technologies, ASTM STP 1031, W. E.
Meyer and J. Reichart, Eds., American Society for Testing and Materials,
Philadelphia, 1990, pp. 113126.
[20] Diringer, K., Aggregates and Skid Resistance, Report No. FHWA/NJ
89-008-7110, New Jersey Department of Transportation, Trenton, NJ
March 1990.
[21] Kandhal, P., Frazier, P., Jr., and Bishara, E., Evaluation of Alabama
Limestone Aggregates for Asphalt Wearing Courses, Transportation
Research Record 1418, TRB, National Research Council, Washington,
D.C., 1993, pp. 1221.
[22] Skerritt, W., Aggregate Type and Traffic Volume as Controlling Factors
in Bituminous Pavement Friction, Transportation Research Report
1418, TRB, National Research Council, Washington, D.C., 1993, pp.
2229.
[23] Kulakowski, B., Henry, J., and Lin, C., A Closed-Loop Calibration
Procedure for a British Pendulum Tester, Surface Characteristics of
Roadways: International Research and Technologies, ASTM STP 1031,
W. E. Meyer and J. Reichert, Eds., American Society for Testing and
Materials, Philadelphia, PA, 1990, pp. 103112.
[24] AASHTO T304-96 (2000), 2003, Standard Method of Test for Uncom-
pacted Void Content of Fine Aggregate, AASHTO Standard Specifica-
tions for Transportation Materials and Methods of Sampling and Testing,
Part 2B, 23rd ed., Washington, D.C.
[25] AASHTO T19/T 19M-00, 2003, Standard Method of Test for Bulk Den-
sity (Unit Weight) and Voids in Aggregate, AASHTO Standard Specifi-
cations for Transportation Materials and Methods of Sampling and
Testing, Part 2A Tests, 23rd ed., Washington, D.C.
[26] Sauter (Brown), H. J., 2001, Identification of Aggregates for Tennessee
Bituminous Surface Courses, Phase IV, Ph.D. dissertation, Tennessee
Technological Univ., Cookeville, TN.
[27] AASHTO T96-02, 2003, Standard Method of Test for Resistance to
Degradation of Small-Size Coarse Aggregate by Abrasion and Impact in
the Los Angeles Machine, AASHTO Standard Specifications for Trans-
portation Materials and Methods of Sampling and Testing, Part 2A Tests,
23rd ed., Washington, D.C.
STP 1555, 2012
DOI:10.1520/STP104447
Edgar de Leon Izeppi,1 Gerardo Flintsch,2 and Kevin McGhee3
Effect of Water, Speed, and Grade on

Continuous Friction Measurement
Equipment (CFME)
REFERENCE: de Leon Izeppi, Edgar, Flintsch, Gerardo, and McGhee,

Kevin, Effect of Water, Speed, and Grade on Continuous Friction Measure-
ment Equipment (CFME), Pavement Performance: Current Trends, Advan-
ces, and Challenges on December 5, 2011 in Tampa, FL; STP 1555, B.
Choubane, Editor, pp. 2039, doi:10.1520/STP104447, ASTM International,
West Conshohocken, PA 2012.
ABSTRACT: The regional pooled-fund project known as the Pavement
Surface Properties Consortium is a research program focused on enhancing
the level of service provided by the roadway transportation system by opti-
mizing pavement surface texture characteristics. The program was set up in
2006 with support from the Federal Highway Administration (FHWA), and six
departments of transportation. The Consortium provides a practical mecha-
nism to conduct research on pavement surface properties and explore their
relationships with ride quality, friction, and noise. Complementing this effort,
additional research focuses on the review, testing, and evaluation of emerg-
ing technologies. Some of the technologies tested have been Continuous
Friction Measurement Equipment (CFME) operating with fixed-slip ratios. In
2010, it was decided to run comparisons of all the units to validate and estab-
lish a database and perform repeatability and sensitivity evaluations at the
Virginia Smart Road. In this paper, we evaluate CFME equipment with tests
conducted at different speeds and a fixed 0.5-mm water-film-thickness flow
rate. A second comparison was also made in which the water-film thickness
was varied, while the speed was kept constant at 65 km/h (40 mph). All of the
Manuscript received October 21, 2011; accepted for publication June 4, 2012; published online
October 2012.
1
Center for Sustainable Transportation Infrastructure (CSTI), Virginia Tech Transportation
Institute (VTTI), Blacksburg, VA 24061.
2
Director, CSTI, VTTI, Blacksburg, VA 24061; and Dept. of Civil and Environmental Engineering,
Virginia Polytechnical Institute and State Univ., Blacksburg, VA 24061.
3
Virginia Center for Transportation Innovation and Research, Charlottesville, VA 22903-2454.
Copyright V
19428-2959.
20
DE LEON IZEPPI ET AL., doi:10.1520/STP104447 21
devices were tested back to back to eliminate the effect of temperature differ-
ences on the pavement and on the water. Preliminary results show that at
greater speeds, the measured friction coefficients measured are lower. Like-
wise, the greater the water-film thickness used, the lower the measured fric-
tion coefficients. There is also evidence that some of these units might be
affected by the longitudinal grade.
KEYWORDS: continuous friction, pavements, friction measurements
Introduction
The regional pooled-fund project known as the Pavement Surface Properties
Consortium is a research program focused on enhancing the level of service
provided by the roadway transportation system by optimizing pavement sur-
face texture characteristics. The program was set up in 2006 with support from
the Federal Highway Administration (FHWA) and six departments of transpor-
tation (DOTs) (CT, GA, MS, PA, SC, and VA). The Consortium provides a
practical mechanism to conduct research on pavement surface properties and
explore their relationships with ride quality, friction, and noise. Complement-
ing this effort, additional research focuses on the review, testing, and evalua-
tion of emerging technologies.
Some of the technologies tested have been Continuous Friction Measure-
ment Equipment (CFME)specifically, two Findlay Irvine grip testers and a
Dynatest highway friction tester (HFT), both operating with fixed-slip ratios.
Concurrently, the Federal Highway Administration implemented a friction
loan program that allowed practitioners and/or researchers to utilize these
CFMEs, while personnel from the Virginia Tech Transportation Institute
(VTTI) provide on-site training and remote technical support for the period of
the loan at each site.
In 2010, when the last piece of equipment was incorporated into the fric-
tion loan program, it was decided that it would be useful to run comparisons of
all three units to validate and establish a database and perform some repeatabil-
ity and sensitivity evaluations at the Virginia Smart Road. This paper summa-
rizes the findings of these tests.
Objective
The general objective of this paper is to evaluate and compare CFME equipment
utilized to measure friction for devices that are participating in the friction loan
program. The specific objectives for this comparison study were to establish a
database from which quick comparisons can be made between deployments of the
equipment to ensure the quality and validity of the data obtained at each site and
determine the effect of different conditions in the measurements.
To accomplish this, comparison tests were made running the units at 40,
65, and 90 km/h (25, 40, and 55 mph) with a fixed 0.5-mm water-film-thickness
flow rate. A second comparison was made where the water-film thickness was
varied between 0.25, 0.50, and 1.0 mm, while keeping the speed constant at
65 km/h (40 mph). All three units tested back to back to eliminate the effect of
temperature differences on the pavement and on the water.
Background
To compare the results of the measurements for the CFME units, it is first use-
ful to explain how this technology compares with other friction measurement
devices, such as the locked-wheel skid testers. As it has been pointed out
before, CFME friction profiles provide more information than locked-wheeled
testers, but they also suffer from the disadvantage that the slip ratio generally
varies within a small range. This creates larger variability in the results of the
CFMEs friction measurements because CFMEs measure friction with a slip ra-
tio near the friction peak of the slip versus friction curve. Small variations in
the slip near the peak friction affect the friction data collected more than they
do equipment that operates at or near full slip, where the friction curve has bot-
tomed out, as indicated in Fig. 1.
However, it is anticipated that CFMEs will offer a practical alternative for
network-level pavement friction data collection at highway speeds because
they can carry out substantial friction surveys with low water consumption and
they are measuring continuously rather than averaging values over several
FIG. 1Pavement longitudinal friction versus tire slip (after Henry [1]).
hundred feet. These devices are also highly maneuverable and especially
adapted for investigating accident sites, supporting wet-weather accident
reduction programs and identifying localized areas with reduced friction.
Having said that, the fact remains that it is necessary to compare different
CFMEs to assess their accuracy and results by evaluating how different opera-
tional factors (such as water-film thickness, grade, and speed) affect their
results. Traditionally, the approach to comparing the data from CFMEs surveys
has been to use simple averages, as has been the case in tests done at the Smart
Road, Wallops Island, etc. Although this has been the case and averages for
each of the different pavement sections have sufficed, this is only appropriate
because the comparisons have been with locked-wheel devices that typically
report data every 91.4 m (300 feet) or more. However, if this approach is used
in highway testing, the data should be analyzed before it is averaged to rule out
the presence of any slippery segments that might be localized sufficiently to
not be picked up when averaged.
The quest for a method to compare friction profiles is a universal problem
affecting the highway community worldwide. Pavement friction measurements
made by two different devices are not unequivocally correct measurements and
their comparison can only be an assessment of the degree of agreement
between them. Agreement measures the closeness between readings, in this
case between two pavement friction measurements [2].
Limits of Agreement
This paper uses the limits of agreement (LOA) method, as defined by Bland
and Altman in 1986, because it was found to be an appropriate method for
assessing the agreement of pavement friction equipment comparisons because
it is simple and easy to understand [35]. LOA has been used in medical
research for over 20 years with excellent results. In the medical research field,
the use of correlation coefficients for equipment comparisons has been discour-
aged as irrelevant and misleading [6]. It should be noted that this happens
when the real value for the parameter being tested is not known or there is no
accepted reference method to measure it. If this was the case, it would not be
an equipment comparison; it would be a simple calibration exercise.
The complete LOA method is explained elsewhere [7]. In this method, the
combined effect of the variability of two instruments being compared, or sc, is
made up of three components. The first two represent the variability inherent
for each device, from their variances (s21 and s22 ). As also defined by ASTM
C-670 [8], they are the repeatability coefficients for each device, when their
square root is multiplied by 2H2 times or 2.83, which are then labeled as r1
and r2, respectively, in this paper. Because there are more than two replicates, the
values s21 and s22 are obtained from an ANOVA1 analysis for both series of runs for
each device. Inside the radical that defines sc in eq 1, they are multiplied by a factor
equal to (11/m), where m is the number of replicates in each case, to account for
the contribution of each according to the number of data points for each.
The third term, which measures the variability produced by the interaction
between the devices, referred to as sD, is obtained from the standard deviation
of the differences of the averages of both measurements for each series of
measurements.
Therefore, the combined effects standard deviation used by this method
(sc) is:
s

2 1 2 1 2
sc sD 1 s1 1 s: (1)
m1 m2 2
To obtain the limits of agreement, sc is then multiplied by a factor of 1.96 to

obtain 95 % reliability confidence limit, which has a positive and a negative
limit from the mean of both differences in measurements. This mean should be
zero if there is no bias. The results shown in this paper only list the two repeat-
ability factors for each device (r1 and r2) and the absolute value of the LOA
factor, or 1.96 sc.
Testing Facility
The Virginia Smart Road is a facility particularly appropriate for this kind of
testing because it includes a variety of real world flexible and rigid pavement
surfaces and it is operated under controlled traffic conditions. It offers seven
different types of asphalt concrete surfaces, including five different Superpave
mixtures, a 12.5-mm stone mastic asphalt (SMA), and a 12.5-mm open-graded
friction course (OGFC) and two rigid concrete pavement surfaces featuring a
continuously reinforced concrete pavement (CRCP) and a jointed reinforced
concrete pavements (JRCP) section.
The continuously reinforced concrete pavement (CRCP) has a transversely
tined finish, and it includes two 30.5-m (100-ft) sections with epoxy overlays
consisting of a double layer of aggregate bonded by epoxy coatings. The test
road has a maximum longitudinal grade of 6 %. All of these section character-
istics are summarized in Table 1.
Devices Used in this Study

The grip tester is a continuous friction measuring device manufactured by Findlay
Irvine in the U.K. Lower water consumption extends data collection and makes
the device more appropriate for use in a network-level pavement-management
TABLE 1Virginia Smart Road section pavement surfaces.
Section Name Surface Mix Type Length (m (feet)) MPD Downhill (mm) MPD Uphill (mm)
1 Loop SMA 19.0 N/A 1.00 0.80
2 A SM-12.5D 105.8 (347) 0.53 0.89
3 B SM-9.5D 88.1 (289) 0.68 1.01
4 C SM-9.5E 89 (292) 0.71 0.79
5 D SM-9.5A 124 (407) 0.56 0.70
6 E SM-9.5D 81.7 (268) 0.60 0.70
7 F SM-9.5D 92 (302)
8 G SM-9.5D 92.7 (304)
9 H SM-9.5D 89 (292)
10 I SM-9.5A(h) 103 (338) 0.92 0.73
11 J SM-9.5D 85.3 (280) 1.05 0.85
12 K OGFC 92 (302) 1.63 1.80
13 L SMA-12.5D 99.4 (326) 1.00 1.08
14 CRCP Trans. tined 698 (2,290) 0.70 0.80
15 JRCP Grooved/ground 180.1 (591) N/A N/A
Note: MPD mean profile depth, CRCP continuously reinforced concrete pavement, JRCP jointed rein-
forced concrete pavement (the epoxy overlays in the CRCP section have average MPD of Cg 1.85 mm and
EP-5 1.18 mm). Section I has a high (h) compaction finish.
system. The grip tester is a fixed slip (16.4 %) instrument that measures the
drag and load continuously (every 3 ft), while towing at any speed from 5 km
to 130 km/h (3 up to 80 mph). The measuring tires are manufactured under
ASTM 1844. The loan program currently has two of these units.
A special traveling unit has been designed by VTTI to include all the nec-
essary components to allow immediate testing upon arrival to any location. A
full tank, approximately 757 l (200 gal of water), will allow about 32 km (20
miles) of continuous data collection under normal conditions [0.5-mm film
thickness at 65 km/h (40 mph)].
The third unit used is a Dynatest 6875H HFT manufactured by Dynatest to
measure continuously a coefficient of friction on highways. This system con-
sists of a fully instrumented heavy duty pickup truck with a two-axis transducer
mounted on a retractable fifth wheel assembly to provide vertical and horizon-
tal tractive force measurements. When the wheel is lowered, the hydraulic sys-
tem causes the test wheel to rotate with a 14 % slip ratio. An 1893-l (500-gal)
water tank allows about 40 km (25 miles) of continuous data collection under
the same normal conditions as stated above for the grip tester [0.5-mm water-
film thickness 65 km/h (40 mph)]. Figure 2 shows both the grip tester and the
Dynatest HFT.
All three of these units are available to state DOTs and other government
agencies on a first-come first-served basis, with training and technical support
provided for by VTTI for the basic period for each loan. All of the shipping
FIG. 2CFME units available from the friction loan program.
arrangements are made by VTTI. The agency is responsible for personnel to

operate it, and a brief report documenting their findings.
Tests
Two sets of tests were made by varying the water-film thickness and also by
varying the speed. Both downhill and uphill runs were made on the same wheel
path (eastbound lane left wheel-path) as can be seen in Fig. 3. The two sets of
tests were: (a) runs at 40, 65, and 90 km/h (25, 40, and 55 mph) with a fixed
0.5-mm water-film-thickness flow rate, and (b) runs varying water-film thick-
nesses between 0.25, 0.50, and 1.0 mm, keeping the speed constant at 65 km/h
(40 mph). Both tests were run five times downhill and five times uphill for
each setting with an extra run to pre-wet the pavement surfaces. Figure 4 is a
FIG. 3CFME units driving uphill on the eastbound lane.
plot of the averaged results of five runs for the speed test at 65 km/h (40 mph)
with a 0.5-mm water-film thickness for each unit running downhill.
The legend shows the results of the first unit labeled as 40440d, which
represents the measurements of grip tester 404, running at 65 km/h (40 mph),
on the downhill direction. In the same way, 40540d and DYN-40d represent
the measurements of grip tester 405 and the Dynatest HFT running at 65 km/h
(40 mph) downhill, respectively.
The vertical lines represent the boundaries between the different sections
of the Smart Road identified in capital letters as referred in Table 1 above.
Between sections D and E, there is a bridge that is marked to point out how the
equipment measurements tend to spike when going over joints or similar pave-
ment features that cause the units to jump. Sections E through H are not
marked individually because their surface mix is the same.
FIG. 4Example results of three CFME units on the Virginia Smart Road
downhill.
The total distance tested in each direction in every test was about 1555 ms
(5100 ft). After the data collection was completed, a synchronization problem
was found with the data acquisition of the different units at the beginning and
at the end of each test. To solve this problem, the comparison of the results is
limited to the middle 1417 ms (4650 ft), from section B to the end of the sec-
ond epoxy concrete overlay (EP-5). The corrected results are shown below in
Fig. 5, where it can be seen there is no data in section A or in the last CRCP
section.
Figure 5 shows that both grip testers and the Dynatest HFT friction profiles
follow a similar pattern over the whole length being compared. It can also be
seen that both of the grip tester profiles tend to have higher measurements
throughout the whole length. Although the measurements of the grip testers are
very close, there are minor differences in them. These observations can be seen
by zooming-in; for example, from 2450 to 2550, as shown in Fig. 6.
Figure 6 also shows that the patterns for both grip testers are similar, and
although the one for the Dynatest HFT is not so different, it has a lower mu, on
average, of about 0.05 units. This phenomena, however, is very different for
the uphill plot. To make a direct comparison from the same stretch of road, the
results and the distances associated with them were inverted to directly com-
pare the same road section based on the downhill distance in the road.
FIG. 5Results of three CFME units on the Virginia Smart Road downhill,
corrected.
The Dynatest HFT collects data every 0.30 m (1 ft), whereas the grip tester
does it every 0.91 m (3 ft). For this reason, for the HFT, an additional dotted
line was added in the plot to illustrate the lower interval. The solid line is an
averaged result of three measurements, to equal the interval for the grip tester
units.
FIG. 6Short section 40-mph speed test results of the three CFMEs downhill.
In Fig. 7, it can be seen that the measured mu from the grip testers going
uphill are very similar to the measured mu of the Dynatest unit, although it is
still, for the most part, a bit lower.
On both of these plots, it seemed as if the dotted Dynatest results had
higher variability than those of the grip testers. This impression is the result of
the shorter data collection interval from the Dynatest HFT. However, they are
not as dissimilar if the standard deviation and coefficient-of-variation (COV)
results are compared, as shown in Table 2. The downhill averages confirm the
differences in magnitudes between units (60.05), not so much going up.
Figure 8 is a box plot of these results. It shows the means, quintiles, vari-
ability, and the effect of the outliers, for the three different instruments, in both
directions, for this short segment of road. This plot clearly shows the difference
in the magnitude of the measurements in the downhill and the uphill direction
mentioned above.
The next section discusses the results for the three speeds with their respec-
tive differences between the overall averages of the length of road being com-
pared which is 1417-ms long (4650 ft) roughly, from section B to section EP-5.
Complete Speed Tests Results

Table 3 shows the results for the three speeds, for each device, and the differ-
ence between the total average measurements. Table 3 also has repeatability
factors for each direction (r1 and r2), and the limits of agreement for both direc-
tions. This is the effect of the grade on the results for each unit, at three differ-
ent speeds. If the measurements are further averaged by ten times, at 9.1 and
91.4 ms (30 and 300 ft), the result is an improvement in both repeatability and
LOA.
FIG. 7Short section 40-mph speed test results of the three CFMEs uphill.
TABLE 2Short section (24502550) results for 65 km/h (40 mph) speed test all units.
Statistic 404-40d 405-40d DYN-40d 404-40u 405-40u DYN-40u

Average 0.89 0.86 0.79 0.79 0.79 0.76
Standard deviation 0.02 0.01 0.02 0.02 0.02 0.02
Coefficient of variation 2.1 % 1.4 % 1.9 % 2.1 % 2.0 % 2.8 %
Figure 9 below shows a boxplot with means, variability, outliers, and quin-
tiles for each device, of the three speeds for all three devices for an averaging
interval of 0.91 m (3 ft).
A comparison made between the limits of agreement for each different unit
in the same direction shows the real effect of the speed difference for each
pair, as shown in Table 4. At each averaging interval, the effect in the agree-
ment caused by the effect of each unit is represented by its repeatability factor,
and the overall effect including the interaction is given by their LOA. Notice
how, at higher averaging intervals, the effects of the variability, or the noise of
the profile, is reduced every time. Noticeable also is the effect of the higher
and lower speeds. This result seems to show that 65 km/h (40 mph) is the best
speed to make these device comparisons.
To better understand why there are differences because of the effect of the
grade, Fig. 10 below shows a plot for each unit with the results in each direc-
tion at 65 km/h (40 mph). Similar plots can be made at each speed and they
would all have the same behavior. In these plots, it can be seen that in those
corresponding to the grip tester units, the results of the downhill and the uphill
average runs are very different, whereas the one for the Dynatest HFT unit, the
results are not so different. This was reflected in the results shown in Table 3
and highlighted as Bias.
FIG. 8Short section 40-mph speed test boxplot of the three CFMEs, down
and up.
TABLE 3Effect of grade with varying speed.
Grip tester 404 Grip tester 405 Dynatest HFT

Grade effects
(same unit) 25 mph 40 mph 55 mph 25 mph 40 mph 55 mph 25 mph 40 mph 55 mph
Mean down 0.93 0.84 0.73 0.89 0.82 0.74 0.80 0.75 0.66
Mean up 0.85 0.79 0.64 0.82 0.74 0.66 0.80 0.75 0.64
Bias 0.08 0.05 0.09 0.07 0.08 0.08 0.00 0.00 0.02
Averaging every 3 ft
Repeatability factor 0.10 0.10 0.19 0.14 0.09 0.14 0.12 0.11 0.15
r1: down
r2: up
Limits of agreement 0.14 0.19 0.24 0.17 0.16 0.20 0.19 0.15 0.19
r1: down
r2: up
r1: down
r2: up
FIG. 9Effect of grade with varying speed for the three CFMEs.
TABLE 4Effect of the change of speed between units (same direction).
40 km/h (25 mph) 65 km/h (40 mph) 90 km/h (55 mph)
Speed effects (same direction) GT1-GT2 GT1-HFT GT2-HFT GT1-GT2 GT1-HFT GT2-HFT GT1-GT2 GT1-HFT GT2-HFT
L of A downhill: 3 ft 0.15 0.18 0.20 0.14 0.17 0.16 0.21 0.23 0.22
L of A downhill: 30 ft 0.11 0.14 0.16 0.09 0.12 0.12 0.15 0.16 0.15
L of A downhill: 300 ft 0.07 0.09 0.11 0.05 0.07 0.06 0.10 0.10 0.09
L of A uphill: 3 ft 0.15 0.18 0.19 0.20 0.19 0.19 0.20 0.21 0.21
L of A uphill: 30 ft 0.09 0.13 0.14 0.12 0.14 0.12 0.15 0.15 0.16
L of A uphill: 300 ft 0.06 0.08 0.09 0.05 0.07 0.07 0.08 0.09 0.09
DE LEON IZEPPI ET AL., doi:10.1520/STP104447
33
FIG. 10Individual inverted results for three CFME on the Virginia Smart
Road
TABLE 5Average water-film thickness results for the three units, both directions.
Grip tester 404 Grip tester 405 Dynatest HFT
Grade effects (same unit) 0.25 mm 0.50 mm 1.00 mm 0.25 mm 0.50 mm 1.00 mm 0.25 mm 0.50 mm 1.00 mm
Mean down 0.84 0.80 0.75 0.84 0.79 0.74 0.76 0.72
Mean up 0.76 0.74 0.69 0.77 0.73 0.66 0.78 0.73
Bias 0.08 0.06 0.06 0.07 0.06 0.08 0.02 0.01
Repeatability factor r1: down 0.13 0.14 0.15 0.15 0.11 0.13 0.13 0.13
Repeatability factor r2: up 0.15 0.18 0.16 0.14 0.13 0.14 0.12 0.12
Limits of agreement 0.25 0.27 0.27 0.25 0.27 0.25 0.23 0.21
35
FIG. 11Effect of grade with varying water-film thickness for the three
CFMEs.
Complete Water-Film Tests Results

Similarly, the results for the water-film thickness also show similar behavior,
as shown in Table 5. It is necessary to mention that the Dynatest HFT unit can-
not test at 0.25 mm film thickness; thus, these results are not shown. However,
it can be seen that the greater the water-film thickness, the lower the friction
averages are, which is to be expected.
Figure 11 shows another boxplot with means, variability, and quintiles for
each device, of the three water-film thicknesses tested, for an averaging inter-
val of 0.91 m (3 ft) at a constant speed of 65 km/h (40 mph).
It can also be seen that the effect of the grade is also present in the same
way as was observed with the speed tests, again presenting a larger bias from
the grip tester units. Table 6 below compares each pair of units traveling in the
same direction to isolate the effect of the different water-film thickness and not
the slope of the road. These results are not very consistent.
Tables 36 have shown the effects of the speed, the grade, and the water-
film thickness. In general, it can be observed that: at greater speeds, friction is
reduced, greater water-film thicknesses reduce the measured friction, and
finally, the greater the averaging interval used, the less the variability is inher-
ently present in the operation of CFME equipment.
Conclusions and Findings

These results show the effect that the grade of a road can have on measure-
ments made with some CFME units. The authors believe that because these
units were originally designed to measure the friction of runways, which rarely
have any significant grade, causes a need for reformulation of the computation
TABLE 6Effect of the change of water between units (same direction).
0.25 mm 0.50 mm 1.00 mm
Water effects (same direction) GT1-GT2 GT1-HFT GT2-HFT GT1-GT2 GT1-HFT GT2-HFT GT1-GT2 GT1-HFT GT2-HFT
L of A downhill: 3 ft 0.17 0.17 0.19 0.17 0.17 0.20 0.18
L of A downhill: 30 ft 0.10 0.10 0.13 0.13 0.12 0.14 0.14
L of A downhill: 300 ft 0.05 0.07 0.09 0.09 0.08 0.09 0.10
L of A uphill: 3 ft 0.20 0.26 0.22 0.24 0.20 0.21 0.20
L of A uphill: 30 ft 0.13 0.19 0.16 0.19 0.14 0.15 0.14
L of A uphill: 300 ft 0.06 0.14 0.09 0.14 0.08 0.09 0.09
37
of the coefficient of friction for those devices that probably assume a constant
weight in the denominator of the equation. The fact that most operational expe-
rience with the CFME units has been on airfield pavements may explain why
this effect of grade has not been reported. When considering the use of CFME
units on highways, the results of the measurements should take into considera-
tion the effect that different longitudinal grade might have in the measured fric-
tion. The maximum grade at the Smart Road is 6 %. Most roads, other than
those on the interstate system, have similar or higher grades than this.
Based on these results the following conclusions can be made:
1. The CFME units tested showed results that confirm that they are very
sensitive to different kinds of pavements. The results also present visi-
ble variability, probably caused by small variations in the slip during
their operation.
2. A method to compare friction profiles from different units and obtain
repeatability and a surrogate for reproducibility called the limits of
agreement was introduced and demonstrated in the comparison of these
measurements. It is expected that more research will be done trying to
use this method to further validate its use.
3. All CFME units tested were affected by joints and other disturbances
that create spikes in the data that needs to be further filtered when
interpreting the results to avoid including false readings in road sec-
tions that contain changes in pavements, pavement types, bridges, or
other disturbances.
4. The CFME units results showed that, at greater speeds, the friction
coefficients measured are lower, as expected.
5. The CFME units results showed that the greater the water-film thick-
ness used, the friction coefficients measured are lower, as expected.
6. All CFME units tested were affected by the grade of the road.
7. All CFME units presented similar variability in their obtained limits of
agreement as a method to compare the precision and accuracy of their
measurements, but it was shown that by averaging the measurements
over longer intervals may allow more meaningful comparisons
between different pieces of equipment.
References
[1] Henry, J. J., Evaluation of Pavement Friction Characteristics: A Synthe-

sis of Highway Practice, NCHRP Synthesis 291, Transportation Research
Board, National Research Council, Washington, D.C., 2000.
[2] Barnhart, H., Haber, M., and Lin, L., An Overview on Assessing Agree-
ment with Continuous Measurements, J. Biopharm. Stat., Vol. 17, No. 4,
2007, pp. 529569.
[3] Bland, J., and Altman, D., Statistical Methods for Assessing Agreement
between Two Methods of Clinical Measurement, Lancet, Vol. i, 1986,
pp. 307310.
[4] Bland, J., and Altman, D., Measuring Agreement in Device Comparison
Studies, Stat. Devices Med. Res., Vol. 8, 1999, pp. 135160.
[5] Bland, J., and Altman, D., Agreement between Devices of Measurement
with Multiple Observations per Individual, J. Biopharm. Stat., Vol. 17,
No. 4, 2007, pp. 571582.
[6] Carstensen, B., Simpson, J., and Gurrin, L., Statistical Models for
Assessing Agreement in Device Comparison Studies with Replicate
Measurements, Int. J. Biostat., Vol. 4, No. 1, 2008, pp. 126.
[7] de Leon Izeppi, E., Flintsch, G., and McGhee, K., Limits of Agreement
Method for Comparing Pavement Friction Measurements, Paper No. 12-
1864, 91st Annual Meeting of the Transportation Research Board, Wash-
ington, D.C., 2012.
[8] ASTM Standard C-670, 2010, Practice for Preparing Precision and Bias
Statements for Test Methods for Construction Materials, Annual Book of
ASTM Standards, Vol. 04.02, ASTM International, West Conshohocken,
PA.
STP 1555, 2012
DOI:10.1520/STP104396
Kevin Gaspard1 and Zhongjie Zhang2
Ascertaining Parametric Correlations

between Lock-Wheel Friction-Testing
and Fixed-Slip Friction-Testing Devices
on Louisiana Highways
REFERENCE: Gaspard, Kevin and Zhang, Zhongjie, Ascertaining Paramet-

ric Correlations between Lock-Wheel Friction-Testing and Fixed-Slip Friction-
Testing Devices on Louisiana Highways, Pavement Performance: Current
Trends, Advances, and Challenges on December 5, 2011 in Tampa, FL; STP
1555, B. Choubane, Editor, pp. 4058, doi:10.1520/STP104396, ASTM Inter-
national, West Conshohocken, PA 2012.
ABSTRACT: Pavement friction testing originated in the 1950s and has been
studied internationally. There are four major types of friction-testing methods:
lock-wheel test (LWT), fixed slip (FXSL), variable slip, and side force. The
Federal Highway Administrator solicited the Louisiana Transportation
Research Center (LTRC) to participate in a national study to evaluate fixed-
slip testers that it had purchased from Dynatest International (FXSLD) and
Findlay Irvine (FXSLG). LTRC developed a testing program where the fixed-
slip testers were correlated to its LWT. Eight roadways were assessed with
each device at 80.5 kph within a 2-h interval. Approximately 2286 m were
assessed on each roadway. Testing with the LWT included both the smooth
(LWTS) and ribbed tires (LWTR). The regression analysis between fixed-slip
testers and a lock-wheel tester (ribbed and smooth tire) indicated that statisti-
cally viable correlations exist for the roadway surfaces and friction number
(FN) ranges assessed in this study. More comprehensive testing should be
conducted to validate the trend found in this study prior to establishing equa-
tions to be utilized on a statewide basis. The one-way analysis of variance
(ANOVA) results indicated that a statistical difference exists between the FN
maximum (umax) values obtained by all devices used in this study. In the
Manuscript received October 5, 2011; accepted for publication April 12, 2012; published online
October 2012.
1
Senior Pavement Research Engineer, Louisiana Transportation Research Center, 4101 Gourrier
Ave., Baton Rouge, LA 70808 (Corresponding author), e-mail: kevin.gaspard@la.gov
2
Research Administrator, Ph.D., Louisiana Transportation Research Center, 4101 Gourrier Ave.,
Baton Rouge, LA 70808, e-mail: doc.zhang@la.gov
Copyright V
19428-2959.
40
GASPARD AND ZHANG, doi:10.1520/STP104396 41
authors opinion, this is a significant issue that must be addressed by the

friction-testing community and regulation industry. Though both FXSL devi-
ces were calibrated in accordance with their manufacturers recommenda-
tions, different FN maximums (lmax), as well as precisions (variances), were
discovered. Such a scenario would make it unfeasible to establish investiga-
tory or intervention friction criteria for a roadway network using FXSL testers
under these conditions.
KEYWORDS: friction testing, rado model, lock-wheel friction tester, fixed-
slip friction tester, variable-slip tester, side-friction tester, antilock brakes
Introduction
Pavement friction testing originated in the 1950s when the United Kingdom
commissioned its Transport and Research Laboratory (TRRL) to develop and
test high friction pavement surfaces [1]. The Federal Highway Administration
(FHWA) followed suit in 1989 by commissioning the University of Michigan
to study roadway surface properties and their relationship to truck crashes [2].
Exhaustive efforts have occurred since the 1950s to model and assess friction
characteristics of pavements by numerous groups such as the National Cooper-
ative Highway Research Program (NCHRP), the Permanent International Asso-
ciation of Road Congresses (PIARC), National Aeronautics and Space
Administration (NASA) Wallops Friction workshops, and the Hermes Project
[37].
There are four major types of high-speed friction-testing methods: lock
wheel, fixed slip, variable slip, and side force [35].
Lock-wheel testers are used primarily to obtain the friction number (FN) or
skid number (SN) at 100 % slip (lmin) [8]. However, the friction profile from 0
to 100 % slip is available as presented in Figs. 1 and 2. At 100 % slip, there is
no tire rotation. Slip speed is defined as the percentage slip multiplied by the
speed of the testing vehicle or device [9]. This type of measurement is meant
to simulate the friction that occurs when vehicle brakes completely lock tires,
which is the typical fashion of braking that existed prior to the advent of anti-
lock braking technology. Investigations have shown, as the slip speed
increases, pavement friction rises to a peak and then decreases as shown in
Fig. 1 [3,9,10]. Furthermore, pavement friction is dependent on the test vehicle
speed with friction numbers increasing as test speed decreases, refer to Fig. 2
[3,9,10]. Friction numbers can be obtained with either or both smooth tires or
ribbed tires [11,12]. Ribbed tires are relatively insensitive to macrotexture,
which makes them ideal for investigating both the microtexture quality of the
pavement as well as its high speed friction characteristics [35]. Smooth tires
are very sensitive to macrotexture and provide insight into the macrotexture
quality of the pavement, as well as its low speed friction characteristics.
Macrotexture refers to those pavement surface properties that are controlled by
aggregate gradation, shape, and size in asphalt pavement [35]. In concrete
pavements, macrotexture is further enhanced by methods, such as grooving.
FIG. 1Typical FN curve and Rado model.
Microtexture is primarily controlled by the surface properties of the aggregate

particles themselves [35].
Unlike locked wheel testers, fixed-slip testers are designed to obtain fric-
tion numbers at a fixed-slip percentage [13]. Research has shown that the
FIG. 2Illustration of friction number versus vehicle speed versus % slip.

maximum friction number (lmax) generally occurs between 10 to 20 % slip as

shown in Fig. 1 [35]. This FN is believed to more closely represent the fric-
tion present during breaking on vehicles with antilock brakes and will be dis-
cussed later [3,9].
Variable-slip testers are designed to obtain FN over a range of predeter-
mined slip speeds varying from 0 to 100 % slip [13]. Some manufacturers
design lock-wheel testers so that they may be programmed to record the fric-
tion from free rolling (0 % slip) to the completely wheel-locked position
(100 % slip) [14]. The lock-wheel tester used in this study has that capability.
Figure 3 presents the curve for the friction number versus slip speed obtained
from the LWT from a test site in this project.
Side-force friction testers collect friction data from a friction tire that is
positioned at an angle (yaw angle) to the pavement [15]. Relevant to highway
pavements, it monitors the lateral friction forces experienced in horizontal
curves on roadways, a value particularly important during wet weather, as well
as ice/snow conditions [4].
There are indirect measures available that are used to obtain the macrotex-
ture and microtexture of the pavement. Macrotexture can be estimated using
the sand patch method, circular texture meter, and outflow meter, whereas
microtexture can be estimated with the British portable tester, Japanese
dynamic friction tester, and sideways-force coefficient routine investigation
machine (SCRIM) [3,1619].
FIG. 3Lock-wheel friction tester and best-fit Rado curve.

Individuals who model pavement surface friction generally use either the
Penn State model or Rado model to describe the relationship between friction
and slip speed [10].
The Penn State model uses a exponential function to describe the relation-
ship between friction (l) and slip speed. It was discovered through correlations
that macrotexture related to the speed gradient, whereas microtexture better
represented to the friction properties. This model was used in the development
of the International Friction Index (IFI) [5]. As a result, the following equation
was developed [3,10]:
PNG
l l0 e 100 S (1)
The Rado model was developed to characterize the friction conditions that
manifest as breaking transitions from a free rolling condition (0 % slip) to the
lock-wheel condition (100 % slip) [3,9]. This model produces a curve that
shows that friction increases from zero slip speed to a peak value and then
decreases to the friction value at 100 % slip as presented in Fig. 1. Research
has shown that the peak value is generally between 10 and 20 % slip. This
model also utilizes a characteristic shape factor (C), typically around 2.5, that
indicates the harshness of the texture. The ascending portion of the curve is de-
pendent more upon the tire properties, whereas the descending portion of the
curve is dependent upon the properties of the pavement [3,9]. Antilock braking
systems (ABS) are designed to control vehicle brakes such that wheel locking
(100 % slip) never occurs. It is believed that ABS operate in the friction zone
near the peak portion of the curve as shown in Fig. 1 [3,4,9]:
" #2
s
ln s
peak
c
ls lpeak e (2)
ABS were successfully introduced into the automobile production market by

Mercedes-Benz in 1978 [20]. The first wheel-lock-prevention system can be
traced to Gabriel Voisin in the 1920s, with the first motor vehicle wheel-lock-
prevention patent in 1936 by Bosch [20].
ABS consists of four major components: speed sensors, pump(s), valves,
and a controller [2023]. The speed sensors monitor the speed of the wheels,
while the pump increases or decreases braking pressure through the values, as
the entire process is carefully sequenced through the controller. Just prior to
the wheel locking, rapid deceleration occurs. For instance, it could take up to
5 s to stop a normal passenger vehicle but only 1 s to lock the wheel. The con-
troller is sensitive to wheel decelerations and regulates braking pressure and
value positions up to 15 times per second when excessive wheel decelerations

are experienced [21].
There are three primary types of ABS: four channel, three channel, and one
channel [2123]. Four-channel systems have sensors on all four wheels, and
the braking pressure can be controlled to each wheel, whereas three-channel
systems have sensors and valves for each front wheel, but only one channel
and valve for the rear wheels. One-channel systems are hybrids between ABS
and lock-wheel braking systems. They have only one sensor and valve for the
rear wheels and the front wheels are controlled by lock-wheel braking. Passen-
ger vehicles generally have four-channel systems, and pickup trucks generally
have either three- or one-channel systems.
Purpose
LTRC was approached by FHWA to participate in a nationwide study to assess
the feasibility of using fixed-slip friction testers (FXSL) in place of lock-wheel
friction testers (LWT), which is the more common system utilized to assess
pavement friction characteristics in the U.S. [3,4]. There is both a national and
international interest in assessing pavement friction qualities as they relate to
ABS. FHWA has purchased several fixed-slip friction testers and are encourag-
ing state agencies to utilize the devices and provide feedback, preferably
through publications.
Methodology
Experiment Design
Roadway Site SelectionSeventeen sites were tested in this study. A num-

ber of these sites could not be tested with the grip tester because of equipment
problems. As such, only eight of the 17 sites underwent at least one test with
the friction testers used in this study. For that reason, only the analysis of those
sites will be presented in this paper. The eight sites consisted of the following:
four Superpave Asphaltic Concrete mix sites (three of which were stone
mastic asphalt),
one NovaChip site,
one open graded friction course site (OGFC),
one jointed concrete pavement (JCP) site, and
one continuously reinforced concrete pavement (CRCP) site.
The material designs for each pavement type conformed to LADOTD spec-
ifications [24].
Testing DevicesThree testing devices were used in this study; two of

which were part of an FWHA loan program:
The LWT used in this study is owned by LTRC and conforms to ASTM
E274/E274M-11 standards [8]. It has both a smooth tire (ASTM E 524-
08) and a ribbed tire (ASTM E 501-08) and was manufactured by Inter-
national Cybernetics Corporation (ICC) [11,12]. Friction testing during
this study was conducted in the left wheel path, so tires were rotated dur-
ing the course of testing to ensure that the pavement was tested with
both the ribbed tire (LWTR) and smooth tire (LWTS). The LWT is
equipped with a 32-kHz selcom texturing laser, which made it possible
to assess the mean profile depth on the left wheel path. The estimated tex-
ture depth, hereafter referred to as mean texture depth, was obtained using
the relationship outlined in ASTM E 1845-09 [25]. Pavement surface tem-
perature readings were recorded with a handheld infrared temperature de-
vice, recorded at the beginning and ending of each test period, which is
standard LTRC testing protocol. Figure 3 shows a plot of the output
obtained from the LWT data along with a best fit RADO model curve
[3,9]. The RADO model was fitted to all LWT data so that umax, peak slip
speed, and C values could be catalogued and used in the statistical analy-
sis of the data sets. Testing was conducted at 80.5 kph on all sites.
One of the fixed-slip testers was a Dynatest 6875 H model friction tester
(FXSLD), purchased by FHWA as part of the friction-tester loan pro-
gram [26,27]. It collected friction data utilizing an ASTM E 1551-08
smooth tire with a slip ratio fixed at 15 % (12.7 kph slip speed) [3,4,28].
Collection of mean profile depth could be accomplished using an
onboard texturing laser. Roadway surface and friction tire temperatures
were collected during testing as well. Friction measurements can be pro-
grammed to report at intervals beginning from 0.3 ms.
The remaining fixed-slip device purchased by FHWA for the loan pro-
gram was an ASTM E2340/E2340M-11 grip tester (FXSLG) was manu-
factured by Findlay Irvine [27,29]. It collects friction data in the left
wheel path with an ASTM 1844-08 smooth tire with the slip ratio fixed
at 16.4 % (13.9 kph slip speed) [3,4,30]. It is capable of providing fric-
tion measurements every meter.
Site Assessment Factorial
Effort was made to ensure that testing began and ended at the same locations,
but matching test point to test point exactly was impossible because of the dif-
ferences in testing methods, as well as documenting the location of each test.
All testing was conducted at 80.5 kph on the same day within a 2-h interval.
Testing always began with the LWT, and the texture data from the first trial
was used in the parametric study described later.
LWTThe LWT device can be set to begin testing at a specified distance

using its distance measuring instrument (DMI). However, its tests are
conducted and recorded in the time domain with a 2.5-s interval [8]. For
instance, the device was programmed to begin testing 305 m from a start point.
After this 305 m is traversed, water will be sprayed on the road for 0.75 s, at
which time skidding will begin. Water continues to be sprayed during skidding
for a duration of 1.0 s. After which time skidding ceases. Water continues to be
sprayed for an additional 0.75 s before the cycle stops. There is a short delay
based on computer processing time before the cycle repeats as presented in
Fig. 4. The average friction number (umin) per 1-s interval was used in the
regression analysis described later in this report. Where possible, 30 cycles of
testing were performed on each site and generally 2286 m were traversed. Fac-
tors, such as intersections and major driveway conflicts, on some projects pre-
vented the planned 30 tests per site from being obtained.
FXSLDThe FXSLD is capable of being programmed similar to the LWT

and testing can also be initiated using a laser and cone. Test results were pro-
grammed to report at 0.3-m intervals.
FXSLGThis device must be activated manually and reports friction

measurements at 1-m intervals
Because the FXSLD and FXSLG produced more readings than the LWT
per test zone, the authors decided to average their results per a 76 m interval
that corresponded with the LWT test areas as shown in Fig. 4.
Parametric Study
Regression, one way analysis of variance (ANOVA), and descriptive statistical
methods were employed on this project [3133].
FIG. 4Friction-testing intervals.

Regression methods were used to discover the relationships between the

LWT (ribbed and smooth) tire and the FXSLD, as well as the FXSLG.
The stepwise variable selection procedure using SAS 9.2 was used to deter-
mine the significant variables evaluated at the alpha 0.01 level. The multiple
regression assumptions of multicollinearity, normality of residuals, and homo-
geneity of variance were assessed using the variance inflation factor,
Kolmogorov-Smirnov procedure, and residual plots, respectively. All variables
used in the analysis were independent and outliers were removed with statisti-
cal diagnostic methods. The results of the regression modeling were tabulated
in Table 1, with Table 2 presenting the mean friction values obtained from
each device used in this study for each test site.
The variables used in the evaluation were the friction numbers from each de-
vice, the mean texture depth obtained with the LWT device on the first trial, pave-
ment surface temperature obtained with the handheld device, and speed of testing
using the LWT data. Research has shown that vehicle speed, pavement surface
temperature, and mean texture depth can have significant influences on friction
values [35]. The following describes the relationships that were evaluated:
ln[FN(LWTR)] fln[FN(FXSLD)], ln[MTD], ln[PTEMP], ln[SP(LWT)]g
ln[FN(LWTS)] fln[FN(FXSLD)], ln[MTD], ln[PTEMP], ln[SP(LWT)]g
ln[FN(LWTR)] fln[FN(FXSLG)], ln[MTD], ln[PTEMP], ln[SP(LWT)]g
ln[FN(LWTS)] fln[FN(FXSLG)], ln[MTD], ln[PTEMP], ln[SP(LWT)]g
ln[FN(FXSLG)] fln[FN(FXSLD)], ln[MTD], ln[PTEMP], ln[SP(LWT)]g
where FN is the friction number, ln is the natural log, FXSLD is the fixed-slip
friction tester (Dynatest), FXSLG is the fixed-slip friction tester (grip tester),
LWTR is the lock-wheel tester (ribbed tire), LWTS is the lock-wheel tester
(smooth tire), MTD is the mean texture depth (mm), PTEMP is the surface
temperature of pavement ( C), and SP(LWT) is the vehicle speed of the lock-
wheel tester (kph).
Note: The LWT friction numbers (umin) used in the regression analysis
were at 100 % slip.
The transformed regression model is:
lnYi b0 b1lnX1i b2lnX2i b3lnX3i b4lnX4i,
with
Yi FN (umin) of LWT (ribbed or smooth tire),
X1i FN of fixed-slip tester (Dynatest or grip tester),
X2i mean texture depth (MTD (mm)),
X3i surface temperature of pavement (PTEMP ( C)),
X4i speed of lock-wheel tester vehicle (SP(LWT) (kph)), and
B(04) regression coefficients.
The one-way ANOVA method at alpha 0.01 was employed to obtain or-
thogonal contrasts for LWT and FXSL FN (umax) data using PROC mixed with
Tukey-Kramer and Satterthwaite adjustments in SAS 9.2 [3133]. Statistical
hypothesis testing, FNLWTR FNLWTS FNFXSLD FNFXSLG, was
TABLE 1Multiple regression analysis results.
Device LWTR FXSLD LWTS FXSLD LWTR FXSLG LWTS FXSLG FXSLG FXSLD
Summary statistics FN avg. 41.4 65.2 36.2 65.2 41.4 49.6 36.2 49.6 49.6 65.2
St. dev. 5.3 12.3 8.9 12.3 5.3 7.8 8.9 7.8 7.8 12.3
Cov % 12.8 18.9 24.5 18.9 12.8 15.8 24.5 15.8 15.8 18.9
Max. 53.3 88.8 53.9 88.8 53.3 72.7 53.9 72.7 72.7 88.8
Min. 31.7 26.2 14.1 26.2 31.7 27.5 14.1 27.5 27.5 26.2
n 217 217 217 217 214 214 214 214 214 214
Device(s) LWTR-FXSLD LWTS-FXSLD LWTR-FXSLG LWTS-FXSLG FXSLG-FXSLD
Regression parameters Intercept 2.61605 0.47452 1.70872 1.90547 1.97859
LN(FXSL) 0.37050 0.74215 0.51686 1.40049 0.57588
LN(MTD) 0.20916 N.S. 0.18594 0.29700 N.S.
LN(PTEMP) 0.12322 N.S. N.S. N.S. 0.13612
LN(SPLWT) N.S. N.S. N.S. N.S. N.S.
r-squared 0.5727 0.3192 0.6622 0.6944 0.4737
RMSE 0.0824 0.2168 0.0730 0.1475 0.1136
Regression assumptions Multi-collinearity No No No No No
Normality of residuals Yes Yes Yes Yes Yes
Homogeneous variance Yes No Yes Yes No
GASPARD AND ZHANG, doi:10.1520/STP104396
49
TABLE 2Average FN values for each test site and device.
Device
Pavement Type LWTR (FN) LWTS (FN) FXSLG (FN) FXSLD (FN)
JCP 50 48 63 68
CRCP 42 24 42 54
OGFC 36 37 41 56
Super Pave 37 30 46 58
Nova Chip 48 48 57 74
conducted using the F-test to determine if the groups were different followed
by orthogonal contrasts designed to assess if means differed between selected
groups, as shown in Table 3 [3133]. This testing regime revealed the similar-
ities or differences between maximum friction values obtained from the devi-
ces, as well as types of tires used to obtain pavement friction values.
To obtain the % slip and C valve at umax, 423 curves using the Rado model
were fit to LWTS and LWTR data as presented in Table 4 [35,9]. This served
two purposes. First, the authors could determine if the % slip and C values at
umax were within the ranges reported by others. Second, this information could
be used to establish if the % slip for fixed-slip testers should be adopted for use
by state agencies.
Analysis of Results
Regression Analysis
The statistical analysis revealed the assumptions for the regression analysis
were best met by using linear transforms wherein both the dependant and inde-
pendent variables were expressed in terms of natural log (ln) [33]. Table 1
illustrates the results of the analysis and Table 2 presents the average FN value
for each device at each test site. Table 1 was arranged so the regression coeffi-
cients for the intercept and each variable (X1 to X4) could easily be compared.
LWTR to FXSLD ModelThe regression analysis revealed that the model

was
LN(FNLWTR) 2.61605 0.37050 LN(FNFXSLD) 0.20916 LN(MTD)
0.12322LN(PTEMP)
with an r-squared value of 0.5727 and root-mean-square error (RMSE) of
0.0824. All variables were significant at the alpha 0.01 level except speed.
TABLE 3One-way ANOVA tests.a,b
Difference in Adjusted Statistical Similar

Device LWTR LWTS FXSLG FXSLD averages p value hypothesis test or different
Summary Statistics FN avg. 69 55 50 65
St. dev. 8.4 11.4 7.8 12.3
Cov % 12.1 20.4 15.8 18.9
Max. 87.6 83 72.7 88.8
Min. 51.7 24 27.5 26.2
n 218 205 214 214
Orthogonal Contrasts Test 1 X X X X N/A <0.0001 F Different

Test 2 X X 14 <0.0001 Tukey-Kramer Different
Test 4 X X 4 0.0002 Tukey-Kramer Different
a
X(s) in cell(s) indicates those devices were compared.
b
FN values in this table are the maximum friction (umax) values.
GASPARD AND ZHANG, doi:10.1520/STP104396
51
TABLE 4Descriptive statistics.
Device Parameter Speed (kph) umax Speak (kph) C Percent slip umin
LWTS Average 82.86 55 17.85 2.24 21.54 36
LWTS St. dev. 1.28 11 4.58 0.42 5.52 9
LWTS Cov % 1.54 20 25.65 18.95 25.65 24
LWTS Max. 85.94 83 40.23 3.4 48.92 54
LWTS Min. 79.82 24 9.66 1.05 11.9 14
LWTS n 205 205 205 205 205 205
LWTR Average 82.25 69 16.12 2.26 19.6 41

LWTR St. dev. 0.85 8 2.47 0.24 3.01 5
LWTR Cov % 1.04 12 15.33 10.45 15.34 13
LWTR Max. 83.85 88 28.97 2.9 35.09 53
LWTR Min. 79.02 52 11.27 1.45 13.65 32
LWTR n 218 218 218 218 218 218
LWTS-LWTR Average 82.55 63 16.96 2.25 20.54 39

LWTS-LWTR St. dev. 1.12 12 3.74 0.34 4.51 8
LWTS-LWTR Cov % 1.36 19 22.08 15.13 21.95 20
LWTS-LWTR Max. 85.94 88 40.23 3.4 48.92 54
LWTS-LWTR Min. 79.02 24 9.66 1.05 11.9 14
LWTS-LWTR n 423 423 423 423 423 423
Because all tests were conducted at 80.5 kph, there was no conflict with other
studies where friction testing was conducted at different speeds such as 50.7,
67.6, and 80.5 kph [35]. Had data been collected at various speeds, the speed
variable should have been significant as well.
LWTS TO FXSLD ModelThe regression analysis revealed that the model

was LN(FNLWTS) 0.47452 0.74215 LN(FNFXSLD) with an r-squared
value of 0.3192 and RMSE of 0.2168. The variables of speed, mean texture
depth, and pavement surface temperature were found to be not significant in
this analysis. This was surprising because both devices use smooth tires and
are generally sensitive to macrotexture.
LTWR to FXSLG ModelThe regression analysis revealed that the model

was LN(FNLWTR) 1.70872 0.51686 LN(FNFXSLG)-0.18594 LN(MTD)
with an r-squared value of 0.6622 and RMSE of 0.0730. As with the LWTR-
FXSLD model, speed and pavement surface temperature were found to be not
significant in this analysis.
LWTS to FXSLG ModelThe regression analysis revealed that the model

was LN(FNLWTS) -1.90547 1.40049 LN(FXSLG) 0.29700 LN(MTD)
with an r-squared of 0.6944 and RMSE of 0.1475. Both the speed and pave-
ment surface temperature were found to be insignificant.
FXSLG TO FXSLD ModelThe regression analysis revealed that the model

was LN(FNLWTS) 1.97859 0.57588 LN(FXSLD) 0.13612 LN(PTEMP)
with an r-squared of 0.4737 and RMSE of 0.1136. In this model, speed and
mean texture depth were both found to be insignificant.
Summary of Regression ResultsEach of the models produced intercepts

with the LWTS/FXSLD model having the intercept nearest to zero. This could
have occurred because the range of FN values used to build the models did not
include low FN values (i.e.,<10). Significant relationships between the FN val-
ues from each device were discovered and only the FXSLG-FXSLD and
LWTS-FXSLD models did not have a significant MTD variable. This was sur-
prising to the authors because generally smooth tire friction-testing devices are
sensitive to macrotexture and ribbed tire devices are not. Pavement surface
temperature was significant in the LWTR/FXSLD and FXSLG/FXSLD models
but not the others.
Based upon the results of the test sections evaluated in this study, as well
as the range of friction values used to build the models, reasonable correlations
between the devices were discovered [33]. This infers that lmin could be rea-
sonably predicted from the fixed-slip devices used in this study.
ANOVA Analysis
The ANOVA analysis (umax) revealed that a significant difference exists
between all devices as presented in Table 3. As expected, the mean value for
the LWTR was higher than the other devices because ribbed tires are less sensi-
tive to water than smooth tires [35]. The LWTSs mean was higher (5 FN)
than the FXSLG, but lower (-10 FN) than the FXSLD. The FXSLG and
FXSLD mean values differed by (15 FN) with the FXSLD having the higher
mean. There variances differed as well.
Summary of ANOVA ResultsThe results indicated that a statistical differ-

ence exists between the FXSLD and FXSLG and their variances were
non-homogeneous, as shown in Table 3. The authors feel that this is a signifi-
cant issue that must be addressed by the friction-testing community and regula-
tion industry. Though both devices were calibrated in accordance with their
manufacturers recommendations, different FN maximums (lmax), as well as
precisions (variances), were discovered. Such a scenario would make it unfea-
sible to establish investigatory or intervention friction criteria for a roadway
network using fixed-slip testers. Part of the issue can be attributed to the lack
of an ASTM standard where criteria, such as percentage slip, tire size, and
precision criteria, are set in stone. Furthermore, the fixed-slip testers used in
this study both used smooth tires that are very sensitive to macrotexture,
instead of using ribbed tires that are not. This is contrary to the historical trend
with lock-wheel testers where ribbed tires are generally used to collect friction
data and in many instances the smooth tires were phased out [3,4]. The authors
were not surprised to discover a difference between the LWT and FXSL test-
ers. The peak value was selected from the curve for LWTs whereas the FXSL
testers produced umax values based on fixed parameters.
Descriptive Statistics of % Slip and C Values

The purpose of this analysis was to determine the percentage slip at umax and
the C value from the Rado curve. The analysis revealed that the percentage slip
at umax was 21.5, 19.6, and 20.5, for the LWTS, LWTR, and LWTS-LTWR,
respectively, as presented in Table 4. Research has shown that these values
range from 10 to 20 %, so the data from this research was at the upper end of
that range [35]. The C values discovered in this analysis were 2.24, 2.26, and
2.25 for the LWTS, LWTR, and LWTS-LTWR, respectively. This is near the
typical reported C value of 2.5 [35,9].
Supplemental Discussion
ABS: Friction TestingAntilock braking systems have been modeled as if

they were operating in the range of lmax, which is approximated by fixed-slip
testers. However, the slip ratio cannot be measured directly [21]. ABS is primar-
ily controlled by deceleration rates of the wheel, not the slip speed of the tires
[2123]. The authors postulate that antilock braking systems are operating
between lmax and lmin,, not directly at lmax. If that is the case, a variable-slip
friction tester would better simulate antilock brakes than a fixed-slip friction tes-
ter. Alternatively, a new friction-testing device could be developed utilizing a
deceleration device to control slip speed that monitors friction during that pro-
cess because that would more closely simulate antilock brakes. As shown in Fig.
2, the LWT used in this study has the ability to model the friction curve through-
out its slip speed/friction cycle. From the curve, umax (0.53), umin (0.36), peak
slip speed (16.9), and the C value (2.7) for the Rado model were determined.
International Friction Index (IFI) Though not elaborated on in this pa-

per, scientific efforts have taken place to establish an IFI. Such an index would
create a standard friction number reference internationally, but would be diffi-
cult to truly implement, as research has shown, because of the numerous varia-
bles, such as the myriad testing devices and friction surfaces that exist [35]. It
may be more feasible to establish indices as opposed to an index to
account for device type, surface type, speed, and temperature [35]. Such an
effort is possible and would produce better correlations, as well as repeatability

and reproducibility.
Conclusions
The parametric study between fixed-slip testers and a lock-wheel tester (ribbed
and smooth tire) indicated that statistically significant correlations at the
alpha 0.01 level exist for the roadway surfaces and FN ranges assessed in
this study [33]. More comprehensive testing should be conducted to validate
the trend found in this study prior to establishing equations to be utilized on a
statewide basis. Because it is common practice for state agencies to develop
indices based on local calibration, the need for national- or international-based
indices is minimal, except to serve as a guideline.
In the U.S., most state agencies have historical data based on LWT ribbed
and/or smooth tires. These data are often utilized for investigatory and inter-
vention purposes [3,4]. If a new friction-testing device, such as a fixed-slip fric-
tion tester, were to be implemented, statistically significant correlations, as
established by the governing agency or organization such as ASTM, would
need to be in place and fixed-slip-derived data (lmax) would have to be trans-
formed into lock-wheel friction data (lmin) to continue to monitor the friction
trends on projects [3,4,9,33]. However, based upon the information gained in
this pilot study, it may be more feasible to fit the RADO model to LWT derived
data because that produces the entire friction profile from 0 to 100 % slip as
shown in Fig. 3 [3,4,9]. That could be simplified with the development of post
processing software that would yield both lmax andlmin, as well as their corre-
sponding slip speeds [3,4,9,10].
Acknowledgments
The writers wish to thank LADOTD and LTRC for providing the resources
necessary for this project and FHWA for providing the fixed-slip friction test-
ers. The assistance of Robert Orthomeyer, Frank Holt, Edgar De Leon Izeppit,
Mitch Terrell, Shawn Elisar, and Glen Gore was very helpful during the course
of this study.
References
[1] Nichols, J.C., Trials of High Friction Surfaces for Highways, Transpor-
tation Research Laboratory, England, TRL, Report No.125:22, 1998.
[2] Julian, F. and Moler, S., Gaining Traction in Roadway Safety, Fed.
Highw. Admin., Vol. 72, No. 1, 2008, p. 10.
[3] Henry, J. J., NCHRP Synthesis of Highway Practice 291: Evaluation of
Pavement Friction Characteristics, TRB, National Research Council,
Washington, D.C., 2000.
[4] Hall, J., Smith, K., Titus-Glover, L., Wambold, J., Yager, T., and Rado,
Z., NCHRP Guide for Pavement Friction Web Document 108, TRB,
National Cooperative Highway Research Program (submitted Feb 2009).
[5] Wambold, J. C., Antle, C. E., Henry, J. J., and Rado, Z., International
PIARC Experiment to Compare and Harmonize Texture and Skid Resist-
ance Measurements, Final Report submitted to the Permanent Interna-
tional Association of Road Congresses (PIARC), State College, PA, 1995.
[6] Wambold, J. C., Henry, J. J., and Yager, T., NASA Wallops Tire/ Run-
way Friction Workshops 19942003, Paper presented at the 5th Sympo-
sium on Pavement Surface Characteristics Conference, SURF 2004,
World Road Association, Paris, 2004.
[7] Descornet, G., The HERMES Project, Paper presented at the 5th Sym-
posium on Pavement Surface Characteristics Conference, SURF 2004
(CD-ROM), World Road Association, Paris, 2004.
[8] Standard Test Method for Skid Resistance of Paved Surface Using Full-
Scale Tire, ASTM Standard Test Method E274/E274M-11, ASTM Inter-
national, West Conshohocken, PA, 2011, DOI: 10.1520/E0274_E0274M-
11, www.astm.org.
[9] Rado, Z., Analysis of Texture Profiles, PTI Report No. 9510, PA Trans-
portation Institute, State College, PA, 1994.
[10] Leu, M. and Henry, J., Prediction of Skid Resistance as a Function of
Speed from Pavement Texture, Transportation Research Record 946,
Transportation Research Board, National Research Council, Washington,
D.C., 1983.
[11] Standard Specification for Standard Smooth Tire for Pavement Skid-
Resistance Tests, ASTM Standard Test Method E524-08, Annual Book of
ASTM Standards, Vol. 04.03, ASTM International, West Conshohocken, PA,
2008.
[12] Standard Specification for Standard Rib Tire for Pavement Skid-
Resistance Tests, ASTM Standard Test Method E501-08, Annual Book of
PA, 2008.
[13] Standard Specification for Rubber Sheet Floor Covering Without Back-
ing, ASTM Standard Test Method 1859-10, Annual Book of ASTM Stand-
ards, Vol. 15.04, ASTM International, West Conshohocken, PA, 2010.
[14] Standard Test Method for Determining Longitudinal Peak Braking Coef-
ficient of Paved Surfaces Using Standard Reference Test Tire, ASTM
Standard Test Method E1337-90(2008), ASTM International, West Con-
shohocken, PA, 2008, DOI: 10.1520/E1337-90R08, www.astm.org.
[15] Standard Test Method for Side Force Friction on Paved Surfaces Using
the Mu-Meter, ASTM Standard Test Method E670-09, ASTM Interna-
tional, West Conshohocken, PA, 2009, DOI: 10.1520/E1337-90R08,
www.astm.org.
[16] Standard Test Method for Measuring Pavement Macrotexture Depth

Using a Volumetric Technique, ASTM Standard Test Method E965-96
(2006), ASTM International, West Conshohocken, PA, 2006, DOI:
10.1520/E0965-96R06, www.astm.org.
[17] Standard Test Method for Measuring Pavement Macrotexture Properties
Using the Circular Track Meter, ASTM Standard Test Method E2157-09,
ASTM International, West Conshohocken, PA, 2009, DOI: 10.1520/
E2157-09, www.astm.org.
[18] Standard Test Method for Measuring Surface Frictional Properties Using
the British Pendulum Tester, ASTM Standard Test Method E303-93
(2008), ASTM International, West Conshohocken, PA, 2008, DOI:
10.1520/E0303-93R08, www.astm.org.
[19] Standard Test Method for Measuring Paved Surface Frictional Properties
Using the Dynamic Friction Tester, ASTM Standard Test Method E1911-
09ae1, ASTM International, West Conshohocken, PA, 2009, DOI:
10.1520/E1911-09AE01, www.astm.org.
[20] Johnson, A., Hitting the Brakes, Duke University Press, Durham, NC,
2009.
[21] Miller, F., Vandome, A., and McBrewster, J., Anti-Lock Braking System,
Alphascript Publishing, Beau Bassin, Mauritius, 2010.
[22] Bauer, H., Dietsche, K., Crepin, J., and Dinker, F., Automotive Handbook,
Bouch, Stuttgart, Germany, 2000.
[23] Mills, V., Samuel, B., and Wagner, J., Modeling and Analysis of Auto-
motive Antilock Brake Systems Subject to Vehicle Payload Shifting,
Veh. Syst. Dyn, Vol. 37, No. 4, 2002.
[24] Louisiana Standard Specifications for Roads and Bridges, LA Dept. of
Transportation and Development, 2006.
[25] Standard Practice for Calculating Pavement Macrotexture Mean Profile
Depth, ASTM Standard Test Model E1845-09, Annual Book of ASTM
Standards, Vol. 04.03, ASTM International, West Conshohocken, PA, 2009.
[26] Dynatest International, for more information, see: www.dynatest.com
(Last accessed June 1, 2012).
[27] Standard Test Method for Measuring the Skid Resistance of Pavements
and Other Trafficked Surfaces Using a Continuous Reading, Fixed Slip
Technique, ASTM Standard Model E2340 / 2340M-11, ASTM Interna-
tional, West Conshohocken, PA, 2011, DOI: 10.1520/E2340_E2340M-
11, www.astm.org.
[28] Standard Specification for Special Purpose, Smooth-Tread Tire, Oper-
ated on Fixed Braking Slip Continuous Friction Measuring Equipment,
ASTM Standard Model E1551-08, Annual Book of ASTM Standards, vol.
04.03, ASTM International, West Conshohocken, PA, 2008.
[29] Findlay Irvine, for more information, see: www.findlayirvine.com (Last
accessed June 1, 2012).
[30] Standard Specification for A Size 10 45 Smooth-Tread Friction Test

Tire, ASTM Standard Mode E1844-08, Annual Book of ASTM Standards,
Vol. 04.0, ASTM International, West Conshohocken, PA, 2008.
[31] Statistical Analysis Software (SAS), version 9.2, for more information,
see: www.sas.com (Last accessed June 1, 2012).
[32] Devore, J., Probability and Statistics for Engineering and the Sciences,
7th ed., Duxbury Press, Pacific grove, CA, 2007.
[33] Neter, J., Kutner, M., Nachtsheim, C., and Wasserman, W., Applied Lin-
ear Statistical Models, 4th ed., Irwin Publications, Scarborough, ON,
Canada, 1990.
STP 1555, 2012
DOI:10.1520/STP104426
Shuo Li,1 Samy Noureldin,1 Karen Zhu,1 and Yi Jiang2
Pavement Surface Microtexture: Testing,

Characterization, and Frictional Interpretation
REFERENCE: Li, Shuo, Noureldin, Samy, Zhu, Karen, and Jiang, Yi,
Pavement Surface Microtexture: Testing, Characterization, and Frictional
Interpretation, Pavement Performance: Current Trends, Advances, and
Challenges on December 5, 2011 in Tampa, FL; STP 1555, B. Choubane,
Editor, pp. 5976, doi:10.1520/STP104426, ASTM International, West Con-
shohocken, PA 2012.
ABSTRACT: Some researchers have investigated the evaluation of pave-
ment friction using macrotexture measurements and found that the relation-
ship between friction and macrotexture is extremely weak. Textures that
affect surface friction include both macrotexture and microtexture. While
macrotexture can be readily measured at highway speeds currently, micro-
texture is evaluated by the friction from a surrogate device at low speeds.
Microtexture depends mainly on the surface properties of the aggregates
and plays an important role in developing friction force. The evaluation of
pavement friction solely from texture measurements will be undermined with-
out considering microtexture. This paper presents a study conducted to
examine the use of laser-based sensors in measuring microtexture. The
requirement of frequency was first established for choosing lasers for meas-
uring pavement textures. Three variables, including the mean profile depth
the slope variance and the root-mean-square were evaluated and utilized to
characterize the microtexture profiles. These three variables varied with the
baseline length and produced useful information for evaluating pavement fric-
tion. The correlation between surface friction and macrotexture and microtex-
ture was also examined.
KEYWORDS: pavement friction, microtexture, laser sampling rate, laser
sampling frequency, speed, wavelength, baseline length, mean profile depth,
slope variance, root-mean-square
Manuscript received October 17, 2011; accepted for publication March 26, 2012; published
1
Division of Research and Development, Indiana Dept. of Transportation, 1205 Montgomery St.,
West Lafayette, IN 47906.
2
Dept. of Building and Construction Management, Purdue Univ., West Lafayette, IN 47907.
Copyright V
19428-2959.
59
Introduction
The Indiana Dept. of Transportation (INDOT) has been conducting annual net-
work inventory friction testing in accordance with ASTM E274 [1] to pinpoint
locations with potential low friction numbers over the past decades. Low friction
numbers were submitted to the districts for necessary action. However, issues
have arisen on the test procedures and the credibility of test results. The current
inventory friction testing is conducted at the mile markers and cannot provide a
continuous coverage of the network pavements. Some slippery pavement loca-
tions may be missed. Also, it is difficult to measure pavement friction at the
standard speed due to the real world traffic conditions. The conversion of friction
numbers at other speeds to those at the standard speed requires the so-called
speed gradients. However, speed gradients vary with pavement surface features
and the type of test tire. It consumes much time and labor to establish the speed
gradients for both rib and smooth tires [2,3]. Moreover, safety concerns always
arise for field testing, in particular on high-speed highway facilities.
Pavement surface friction is a function of pavement surface texture charac-
teristics. Compared to surface friction, surface texture is a unique and physical
feature, and does not vary with test speed and condition. In addition, the texture
testing using a laser sensor may provide a seamless, longitudinal texture profile
and is capable of capturing all slippery pavement locations. Some researchers
have made efforts to evaluate pavement surface friction from pavement surface
texture measured using laser-sensors [4,5]. However, it was found that the rela-
tionship between friction and macrotexture is extremely weak [4]. The textures
that affect friction include both macrotexture and microtexture [6]. Macrotex-
tures have wavelengths of 0.5 mm to 50 mm and can be readily measured in
the field [7]. Nevertheless, microtextures have wavelengths of 0.001 mm to 0.5
mm and are currently evaluated by the friction at low speeds using a surrogate
device. Microtexture depends mainly on the surface properties of the aggre-
gates [8] and plays an important role in developing friction force, particularly
in the long term. In addition, the pavement friction measured using the ASTM
E274 locked wheel trailer is very sensitive to surface microtexture. Therefore,
any effort to establish the relationship between pavement friction and texture
could be undermined without taking into account the microtexture.
The Joint Transportation Research program (JTRP) funded a pilot study to
investigate the use of laser-based sensors in measuring pavement surface
microtexture. This study first investigated the theoretical requirements for
lasers used to measure texture profiles at different test speeds, in particular the
laser frequency at highway speeds. Subsequently, this study conducted pave-
ment texture testing on three typical pavement surfaces in the INDOT friction
test track and examined the characteristics of microtexture profiles plotted
from the test data. Finally, this pilot study evaluated the possible variables that
may be employed to characterize the microtexture profile and the potential use
LI ET AL., doi:10.1520/STP104426 61
of pavement surface microtexture in evaluating pavement surface friction. The

current paper reports the main results obtained from this pilot study. In the
authors opinion, such information is not only useful for pavement engineers to
better understand the characteristics of pavement surface microtexture, but also
beneficial to bring pavement engineers a closer step to establishing procedures
and variables for correlating surface friction and texture properties.
Requirement of Frequency for Microtexture Measuring Lasers
Pavement Texture Frequency

Pavement surface macrotexture is commonly measured in terms of the so-
called macrotexture profile depth (MPD). A texture profile consists of a series
of periodic textures. To accurately characterize the texture profile, the sampling
interval must be small enough so that a true texture profile can be established.
In other words, the laser sampling rate or frequency should be fast enough to
capture the variations of texture profile, in particular while testing at highway
speeds. Pavement textures are physical, stationary surface features, and are cur-
rently characterized in terms of wavelength and amplitude. No frequency has
been identified while measuring texture profiles. In reality, the laser system is
usually mounted on a test vehicle. During testing, the test vehicle travels on the
pavement at a certain speed and the laser scans the pavement surface at the test
vehicle speed. This implies that the pavement texture profile can be considered
as a series of traveling waves passing the laser at the speed of test vehicle.
Once the texture profile frequency is identified, the frequency requirement
for the laser can be estimated from the fundamental physics. Frequency refers
to as the number of a repeating or periodic event, such as oscillations, vibra-
tions, or waves occurring per unit of time. In general, frequency is related to
wavelength and velocity as follows:

f (1)
k
where:
f frequency,
velocity, and
k wavelength.
It should be pointed out that the frequency defined in Eq 1 is also referred
to as temporal frequency in light of a unit of hertz (Hz) and its reciprocal is the
period. Therefore, replacing , i.e., the velocity in Eq 1 with the speed of test
vehicle yields the frequencies of textures with typical wavelengths as shown in
Table 1.
TABLE 1Texture frequencies by wavelength at different speeds.
Frequency by wavelength (Hz)
Test speed (mph) 0.5 mm 0.03 mm 0.001 mm

10 8939 148981 4469444
20 17878 297963 8938889
30 26817 446944 13408333
40 35756 595926 17877778
50 44694 744907 22347222
60 53633 893889 26816667
Requirements for Lasers

During field testing at highway speeds, the sampling frequency is an important
indicator of the texture measuring capability of the laser system. One published
study indicates that the laser sampling frequency should be great enough to
handle the signal variations caused by the background fluctuations [9]. Some
researchers indicated that the sampling frequency of the laser should meet the
requirement by the Nyquist sampling theorem [10]. Fundamentally, the
Nyquist sampling theorem states that the sampling frequency should be at least
twice the highest frequency contained in the signal to avoid aliasing [11].
Notice that the real world pavement texture is complex and consists of many
wavelength components that produce different frequencies. Thus, the lasers
sampling frequency should be at least twice the highest frequency according to
the Nyquist sampling theorem.
The required sampling frequencies for lasers capable of measuring micro-
textures can be determined from Eq 1 according to the traveling speed of the
test vehicle and are presented in Table 2. It is shown that as the test speed
increases, the required frequency increases. The increase rate is much greater
for the 0.001-mm wavelength than those for 0.03-mm and 0.5-mm wave-
lengths. For example, the laser should have a sampling frequency not less than
54 kHz to capture all macrotexture peaks and troughs at 30 mph. To be capable
of capturing all microtextures at highway speeds, however, a laser with a high
frequency (HF) to a very high frequency (VHF) should be utilized. Neverthe-
less, the lasers with frequencies greater than 70 kHz are currently very expen-
sive and not commercially available. As illustrated in Eq 1, frequency is
inversely proportional to wavelength, but proportional to velocity. This implies
that when choosing a laser for microtexture testing, a compromise should be
made between the test speed and the laser frequency.
Notice that when choosing a laser, there are also several other important
factors that must be considered, including laser spot size, sampling resolution,
and sampling spacing. After weighing the study goals, the accuracy require-
ment, and the availability of commercial lasers, a 1-kHz laser scanner was
LI ET AL., doi:10.1520/STP104426 63
TABLE 2Required laser frequencies for microtexture testing.
Required frequency by texture wavelength (kHz)
Test speed (mph) 0.5 mm 0.025 mm 0.001 mm

0.1 0.2 3.0 89.4
1 1.8 29.8 893.9
2 3.6 59.6 1787.8
3 5.4 89.4 2681.7
4 7.2 119.2 3575.6
5 8.9 149.0 4469.4
10 17.9 298.0 8938.9
20 35.8 595.9 17877.8
30 53.6 893.9 26816.7
40 71.5 1191.9 35755.6
50 89.4 1489.9 44694.4
60 107.3 1787.8 53633.3
chosen in the microtexture testing. Currently, the 1-kHz laser scanner is prob-
ably the one with best resolution and capability. The 1-kHz laser scanner is a
stand-alone system and has a very low scanning speed of approximately 15
mm=s (0.033 mph). The 1-kHz laser scanner has a laser spot size of 0.050 mm,
vertical resolution of 0.015 mm, sampling spacing of 0.015 mm, and wave-
length range of 0.03 mm to 50 mm. While this laser system is not capable of
capturing microtextures with wavelengths less than 0.03 mm, the microtexture
acquired using this laser scanner may play an important role in wet pavement
friction development. It is the authors opinion that there may exist a lower
bound of wavelength. The effect of the microtextures with wavelengths less
than the lower bound may be negligible, particularly when the pavement sur-
face is wet.
Development of Microtexture Profile
Texture Testing
Pavement texture testing was collected on the cores taken from the three pave-
ments in the INDOT friction test track, including slick concrete pavement, hot
mix asphalt (HMA) pavement, and tined concrete pavement. These three surfa-
ces were specially prepared to represent the possible magnitude of pavement
surface friction on the real world highways. The HMA pavement was con-
structed with 9.5-mm mix. Both the slick and tined concrete pavements were
constructed with regular cement concrete mix. However, the slick concrete
pavement was finished to provide a smooth surface that is usually observed on
highways experiencing very poor friction. The tined concrete surface was
FIG. 1Photos of cores taken INDOT friction test track.
finished with 3-mm-deep, 3-mm-wide transverse grooves at a spacing of 1820

mm. In the HMA mix, the coarse aggregates consist of 27 % steel slag and 27
% dolomite aggregates. As shown in Fig. 1 are the photos of three cores taken
from the three pavements, respectively.
Summarized in Table 3 are the surface friction properties of these three
pavements, including the mean profile depth (MPD) values of macrotexture,
dynamic friction tester (DFT) numbers measured from the dynamic friction tes-
ter [12], and the locked wheel friction numbers. During the texture testing
using the 1-kHz scan laser, ten parallel scan lines were taken for each core
sample. The original scan data was first filtered with a bandpass range of 0.25
mm to 0.5 mm. While filtering the scan data, the sample spacing was set up to
be 0.084 mm that is one third of the lowpass. The filtered scan data was then
used to plot microtexture profiles that were utilized in data analysis and calcu-
lation of the microtexture properties.
Microtexture Profiles
Figure 2 shows three typical microtexture profiles measured on the three surfa-
ces, respectively. The microtexture profile on the slick concrete surface exhib-
its uniform occurrences of texture waves. On the HMA and tined concrete
surfaces, however, the microtexture profiles demonstrate dense occurrences at
some locations and sparse occurrences at some other locations. To examine the
variation of microtexture wavelength, the authors manually measured the
TABLE 3Friction properties of the friction test track surfaces.
Dynamic friction tester
Pavement surface MPDa (mm) DFT Locked wheel FN at 40 mph

Slick concrete 0.08 0.374 8.8
9.5-mm HMA 0.59 0.671 42.5
Tined concrete 1.45 0.816 59.0
a
Macrotexture MPD.
LI ET AL., doi:10.1520/STP104426 65
FIG. 2Typical microtexture profiles on three different surfaces.
wavelengths from the microtexture profile plots as shown in Fig. 3. Careful

inspection of Fig. 3 indicates that the microtexture profile commonly consists
of two categories of waves, i.e., large-scale waves and small-scale waves. The
former indicates the overall shape of the profile (the broken line in Fig. 3), and
the latter indicates the local roughness of the profile.
FIG. 3Example plot for manual calculation of wavelength.
Table 4 shows the results of the wavelengths computed for the three pave-
ment surfaces. For large-scale waves, the slick concrete surface demonstrates
the smallest average wavelength, and the HMA surface the greatest average
wavelength. For small-scale waves, the wavelengths are almost the same for
the three surfaces in light of both the average wavelength and deviation. The
differences between the microtetxure profiles of different pavement surfaces
are mainly associated with the large-scale waves. This may be extended to
make a tentative statement that the long-term friction performance may mainly
be contributed to by the small-scale waves that probably depend to a large
extent on the aggregate surface features. If both the large- and small-scale
waves are combined together, the statistic results of the wavelengths can be
also computed and are presented in Table 4. The average wavelengths are very
close. The main difference is associated with the maximum wavelengths that
are 0.395, 0.290, and 0.475 for the slick concrete, HMA, and tined concrete
surfaces, respectively. The slick concrete surface shows the smallest
TABLE 4Summary of microtexture wavelength.
Slick concrete HMA Tined concrete
Wavelength Large-scale Small-scale Large-scale Small-scale Large-scale Small-scale

Ave (mm) 0.131 0.046 0.172 0.046 0.151 0.043
SD (mm) 0.062 0.013 0.053 0.013 0.073 0.013
Combined
Ave (mm) 0.082 0.083 0.080
SD (mm) 0.059 0.065 0.067
N 157 156 154
Range (mm) 0.0250.395 0.0200.290 0.020.475
LI ET AL., doi:10.1520/STP104426 67
TABLE 5Summary of microtexture depth distributions.
Microtexture depth (mm)
Percentile Slick concrete HMA Tined concrete

50th 0.009 0.010 0.014
70th 0.014 0.016 0.022
85th 0.019 0.023 0.030
90th 0.021 0.027 0.034
95th 0.025 0.034 0.040
100th 0.049 0.050 0.075
Ave. depth 0.010 0.014 0.017
SD 0.008 0.021 0.013
COV (%) 75.5 148.7 75.9
wavelength standard deviation, and the tined concrete surface demonstrates the
greatest wavelength variation range. However, it was confusing that the greater
maximum wavelength on the slick concrete surface is greater than that on the
asphalt surface.
Presented in Table 5 are the texture depths for different percentiles. The
percentile is defined as the total percentage of the texture depths less than a cer-
tain value. It is shown that most microtexture depths (more than 95 %) are less
than 0.05 mm. Also, summarized in Table 5 is the information on the distribu-
tions of microtexture depths, including the average texture depth (ave. depth)
values, the standard deviation (SD) values, and the coefficient of variation
(COV) values for the three pavement surfaces. The tined concrete surface
exhibits the greatest average texture depth, and the slick concrete surface
exhibits the smallest average texture depth. However, the slick and tined con-
crete surfaces demonstrate the similar degree of variations (see the COV val-
ues). The HMA surface demonstrates the greatest deviation. This is probably
because of the use of 27 % of steel slag, and 27 % of dolomite in the HMA
coarse aggregates.
Characterization of Microtexture Profiles
Microtexture Profile Properties

Currently, there is no technical specification or standard available to guide the
evaluation of the microtexture profile properties. It should be kept in mind that
in nature, macrotexture and microtexture are only two different texture catego-
ries of the same family. The testing and calculation of the microtexture profiles
should be consistent with those in the macrotexture profile testing. Apparently,
it is reasonable to adopt the MPD to evaluate the microtexture profile
properties. Notice that microtexture is the texture that produces harsh or asper-
ity surface to provide good friction. The use of a sole variable such as the MPD
may not be enough to characterize the surface asperities. Therefore, two sup-
plemental variables were also employed by the authors to evaluate the micro-
texture profile properties.
The first supplemental variable is the slope variance (SV) that measures
the sharpness of asperities [13] and is calculated as follows:
sX
SVi2
SV (2)
n
where:
yi1 yi Dyi
SVi i 0; 1; 2; n 1
xi1 xi Dxi
xi the ith measured x coordinate of the microtexture profile,

yi the ith measured y coordinate of the microtexture profile, and
n the number of segments in a sample length or a baseline.
The second supplemental variable is the root mean square (rms) of the
microtexture profile. RMS is a measure of the magnitude of the asperities of
the microtexture profile through erasing the signs of the negative peak values
and is computed as follows:
s
X
2
Peaki
RMS (3)
n
where:
Peaki jyi Yi j i 1; 2; n;
yi the ith measured y coordinate of the microtexture profile,

Yi the ith predicted y coordinate of the microtexture profile, and
n the number of segments in a sample length or a baseline.
Effect of Baseline Length

The calculation of MPD from a macrotexture profile is based on a baseline of
100 mm that is twice the maximum macrotexture wavelength. Because the base-
line length affects the reference or regression line, different baseline lengths will
yield different MPD values. The authors computed the microtexture MPD, SV,
LI ET AL., doi:10.1520/STP104426 69
and rms in light of various baseline lengths, including 1 mm, 12.75 mm (0.5 in.),
25.4 mm (1 in.), 50.8 mm (2 in.), and 100 mm (4 in.), and the results are plotted
in Fig. 4. In general, MPD, SV, and rms increase as the baseline length increases,
regardless of the type of pavement surface. The increase rate decreases as the
baseline length increases. However, all curves demonstrate a turning point with
respect to the same baseline length of 12.75 mm. The MPD, SV, and rms for the
HMA surface increase more dramatically than those for the slick and tined con-
crete surface. The increases are negligible for the slick and tined concrete surfa-
ces when the baseline length is greater than 12.75 mm.
Mathematically speaking, increasing the baseline length will invariably
increase the MPD because a baseline of greater length provides greater possibil-
ity for greater texture depths to present according to the computation procedures.
However, the SV and rms tend to remain more or less constant after the baseline
length reaches 12.75 mm. This implies that a baseline of 12.75 mm may be a rea-
sonable option in the computation of microtexture profile properties. To be con-
sistent with the calculation of MPD for a macrotexture profile, the authors first
divided each 100-mm microtexture profile sample into eight 12.75-mm seg-
ments, and then computed the MPDi, SVi, and rmsi for each 12.75-mm segment
(i 1, 2, , and 8). The MPD, SV, and rms are the average of MPDi, SVi, and
rmsi of the eight segments (see Table 6), respectively. The HMA surface demon-
strated the greatest MPD, probably because of the fact that the HMA surface
experienced the greatest depth variations. The tined concrete surface produced
the greatest SV, which implies that the microtexture on the tined concrete surface
produced greater sharpness than that on the slick concrete or HMA surface. The
HMA and tined concrete surfaces demonstrated the similar rms values. It appears
that the variations of SV and rms followed a similar trend to the variation of sur-
face friction (see Table 3). As SV and rms increased, pavement friction
increased. This indicates that SV and rms may provide information to better
characterize the properties of pavement microtexture.
Use of Microtexture Measurements in Estimating Friction
Pavement Friction Components

It was pointed out that the pavement surface friction force can basically be
determined by its two main components, including adhesion force and hystere-
sis force [14]. The former depends on the interface shear strength and the con-
tact area, and the latter depends on the damping losses within the rubber. On a
dry pavement, the friction force is dominated by the adhesion force if the sur-
face is smooth and by the hysteresis force when the surface is rough. When
water is present and pavement surface becomes lubricated, the adhesion force
decreases and therefore the friction force decreases. Presented in Table 7 are
FIG. 4Variations of Microtexture MPD, SV, and rms with baseline length.
LI ET AL., doi:10.1520/STP104426 71
TABLE 6Microtexture MPD, SV, and rms based on 12.75-mm baseline.
Surface Slick concrete HMA Tined concrete

MPD (mm) 0.039 0.079 0.061
SV (mm) 0.700 1.000 1.016
rms (mm) 0.013 0.021 0.021
the friction numbers measured on an in-service asphalt pavement. The pave-

ment surface was smooth and experienced no surface distresses but polished
aggregates. The wet surface indicates that water was applied to the pavement
surface and the dry surface indicates that no water was applied to the pavement
in the entire process of testing.
Three observations were made from the data shown in Table 7. First, under
the dry condition, the smooth tire generated a friction number greater than that
generated by the ribbed tire. This is because on the dry pavement surface, the
dominant friction force component is the adhesion force that depends to a great
extent on the contact area. The standard smooth tire commonly generates a
greater contact area than the standard ribbed tire. Second, the friction numbers
decreased dramatically when the pavement surface became wet. It decreased
by approximately 71 points for the smooth tire and 41 points for the ribbed tire.
In other words, the adhesion force might be 71 for the smooth tire and 41 for
the ribbed tire in terms of friction number. Because the hysteresis force became
dominant when the surface was wet, the hysteresis force might be 37 for the
ribbed tire and 27 for the smooth tire.
Comparison of Texture and Friction

Presented in Fig. 5 are the variations of friction number (FN) with macrotex-
ture MPD and microtexture MPD, respectively. The variations with macrotex-
ture MPD are very limited in dry pavement. The wet pavement friction
increases as the macrotexture MPD increases. The friction difference or
(FNdryFNwet) decreases as macrotexture increases. In the authors opinion, the
friction difference is the result of the combined effect of macro- and microtex-
tures. Similar to macrotexture, the microtexture MPD demonstrates limited
effect on the dry pavement friction. However, as the microtexture MPD
TABLE 7Friction numbers measured on wet and dry pavement.
Surface condition Test tire Test speed Friction number

Wet Smooth 40 mph 27
Dry Smooth 40 mph 98
Wet Ribbed 40 mph 37
Dry Ribbed 40 mph 78
FIG. 5Variations of surface friction with texture depth.
increases, the wet pavement FN decreases and then increases. Consequently,

the friction differences decrease to a certain value and then increase as the
microtexture MPD increases. If the observation is valid, there may exist a turn-
ing point for the microtexture MPD. Figure 6 shows the variations of FN with
microtexture SV and rms, respectively. The variation of FN with rms follows a
trend very similar to that with rms. This indicates that to estimate friction from
microtexture measurements, the use of SV is as effective as the use of rms. It is
not necessary to include both SV and rms when developing relationship
between friction and microtexture measurements. Also, the wet pavement FN
is proportional to the microtexture SV. As SV increases, the FN difference
decreases.
LI ET AL., doi:10.1520/STP104426 73
FIG. 6Variations of FN with microtexture SV and rms.
Presented in Table 8 are the Pearson correlation coefficients between the

friction and texture measurements on the three surfaces. Apparently, the wet
pavement friction has a positive relationship with all three texture measurements,
including macrotexture MPD, microtexture MPD, and microtexture SV. While
the greatest degree of dependence occurred between the wet pavement FN and
the macrotexture MPD, the microtexture SV demonstrated an equivalent effect
of the wet pavement FN. The correlation coefficient for the microtexture SV is
TABLE 8Pearson correlations between friction and texture measurements.
Texture measurement
Friction measurement Macrotexture MPD Microtexture MPD Microtexture SV

Wet pavement FN 0.9722 0.6026 0.9159
Dry pavement FN 0.4645 0.9211 0.6131
much greater than that for the microtexture MPD. This indicates that the wet
pavement friction is more sensitive to the microtexture SV than to the microtex-
ture MPD. On the dry pavement, the friction number has a negative relationship
with all three texture measurements. The greatest degree of negative dependence
arises between the FN and the microtexture MPD, and the least degree of nega-
tive dependence between the FN and the macrotexture MPD. This indicates that
dry pavement friction is not as sensitive to macrotexture as to microtexture.
Notice that only the microtextures with wavelengths 0.03 mm greater were
considered because of the limitation of the equipment.
Conclusions
Pavement texture profile may be considered as a series of waves passing the
laser mounted on a test vehicle travelling at a specific speed. Therefore, the fre-
quencies of a pavement texture profile are decided by the classic frequency-ve-
locity-wavelength relationship in light of the texture wavelength and test
speed. The required sampling frequencies for lasers capable of measuring
microtextures may also be determined from the frequency-velocity-wavelength
relationship according to the Nyquist sampling theorem. To be capable of cap-
turing all microtextures at highway speeds, a laser with a high frequency (HF)
to a very high frequency (VHF) should be utilized. Currently, the HF and VHF
lasers are not commercially available. A compromise must be made between
the test speed and the laser frequency.
The microtexture data on three typical surfaces indicated that a microtex-
ture profile consists of two categories of waves, i.e., large- and small-scale
waves. The former indicates the overall shape of the profile and the latter the
local roughness of the profile. For large-scale waves, the slick concrete surface
demonstrated the smallest average wavelength, and the HMA surface the great-
est average wavelength. For small-scale waves, the wavelengths are almost the
same for the three surfaces in light of the average wavelength and deviation.
The differences between the microtexture profiles of different pavements arose
mainly associated with the large-scale waves. The tined concrete surface exhib-
ited the greatest average texture depth, and the slick concrete surface the small-
est average texture depth. The slick and tined concrete surfaces demonstrated a
similar degree of variations in texture depths. The HMA surface exhibited the
greatest deviation, due probably to the use of steel slag and dolomite in
aggregates.
In addition to the MPD, SV and rms may be adopted to evaluate the prop-
erties of microtexture. The length of baseline will affect the MPD, SV, and rms
values. However, it appears that a baseline greater than 12.75 mm (0.5 in.) will
yield stable SV and rms values. As SV and rms values increase, pavement fric-
tion increases. The tined concrete surface demonstrated the greatest SV, and
LI ET AL., doi:10.1520/STP104426 75
the slick concrete surface the smallest SV. It appears that the introduction of
SV and rms may provide information to better characterize the properties of
microtexture profiles. The effect of microtexture SV on wet pavement friction
is equivalent to that of microtexture MPD. The wet pavement friction is more
sensitive to the microtexture SV than to the microtexture MPD.
Acknowledgments
This research project is sponsored by INDOT in cooperation with the Federal
Highway Administration through the Joint Transportation Research Program
(JTRP). The writers thank Mark Leichty of Ames Engineering for his valuable
assistance. The contents of this paper reflect the views of the authors who are
responsible for the facts and the accuracy of the data presented herein. The
contents do not necessarily reflect the official views or policies of INDOT and
the Federal Highway Administration. This paper does not constitute a standard,
specification, or regulation.
References
[1] ASTM E274-97, 2004, Standard Test Method for Skid Resistance of
Paved Surfaces Using a Full-Scale Tire, Annual Book of ASTM Stand-
ards, Vol. 04.03, ASTM International, West Conshohocken, PA.
[2] ASTM E501-94, 2004, Standard Specification for Standard Rib Tire for
Pavement Skid-Resistance Tests, Annual Book of ASTM Standards, Vol.
04.03, ASTM International, West Conshohocken, PA,
[3] ASTM E524-88, 2004, Standard Specification for Standard Smooth Tire
for Pavement Skid-Resistance Tests, Annual Book of ASTM Standards,
Vol. 04.03, ASTM International, West Conshohocken, PA,
[4] Jackson, N. M., Choubane, B., Holzschuher, C., and Gokhale, S.,
Measuring Pavement Friction Characteristics at Variable Speeds for
Added Safety, J. ASTM Int., Vol. 2, 2005, pp. 5972.
[5] Kevin, K., McGhee, K. K., and Flintsch, G. W., High-Speed Texture Mea-
surement of Pavements, Virginia Transportation Research Council, Char-
lottesville, VA, 2003.
[6] PIARC Technical Committee on Surface Characteristics, Optimization
of Pavement Surface Characteristics, Report to the XVIIIth World Rd.
Congress, Brussels, Belgium, 1987.
[7] ASTM E1845-01, 2004, Standard Practice for Calculating Pavement
Macrotexture Mean Profile Depth, Annual Book of ASTM Standards,
Vol. 04.03, ASTM International, West Conshohocken, PA.
[8] Hall, J. W., Smith, K. L., Titus-Glover, L., Wambold, J. C., Yager, T. J.,
and Rado, Z., Guide for Pavement Friction, NCHRP Web-Only
Document 108, Transportation Research Board, The National Academies,

[9] Mu, X. Y., Li, L., Tang, N., Liu, C., and Jiang, X. H., Laser-Based Sys-
tem for Highway Pavement Texture Measurement, Proceedings of the
2003 IEEE International Conference on Intelligent Transportation Sys-
tem, Vol. 2, 2003, pp. 15591562.
[10] Xie, J. B., Liu, R., and Michalk, B., Automatic Skid Number Evaluation
Using Texture Laser Measurements, IEEE International Conference on
Networking, Sensing, and Control, Sanya, Peoples Republic of China,
April 68, 2008, pp. 3742.
[11] Smith, S. W., The Scientist and Engineers Guide to Digital Signal Proc-
essing, CA Technical Publishing, San Diego, CA, 1997.
[12] ASTM E-98, 2004, Standard Test Method for Measuring Paved Surface
Frictional Properties Using the Dynamic Friction Tester, Annual Book of
PA,
[13] Schmidt, B., Laser Texture Measurements of Asphalt Concrete, Nordic
Rd. and Transport Research, No. 1, Danish Rd. Directorate, Hedehusene,
Denmark, 2000.
[14] Kummer, H. W., and Meyer, W. E., Tentative Skid-Resistance Require-
ments for Main Rural Highways, Report No. NCHRP-37, Highway
Research Board, National Research Council, 1967.
STP 1555, 2012
DOI:10.1520/STP104543
Ulf Sandberg1
Measuring Rolling Resistance of Typical

Swedish Pavements
REFERENCE: Sandberg, Ulf, Measuring Rolling Resistance of Typical

Swedish Pavements, Pavement Performance: Current Trends, Advances,
and Challenges on December 5, 2011 in Tampa, FL; STP 1555, B. Chou-
bane, Editor, pp. 7795, doi:10.1520/STP104543, ASTM International, West
Conshohocken, PA 2012.
ABSTRACT: Rolling resistance properties of a large number of Swedish
pavements and a few Danish pavements have been measured in recent
years by means of a special trailer produced and owned by the Technical Uni-
versity of Gdansk in Poland. Three different automobile and light truck tires
have been used as reference tires, two of them representing automobile tires
and one representing truck tires (with regard to pavement properties). There
are several problems related to the measurement methods, which are briefly
dealt with in this paper. The main part of the paper reports the rolling
resistance coefficients (RRCs) that have been measured over a range of
pavements common in Sweden and over a range of conditions. It appears
that there is a very good and repeatable correlation between RRC and mean
profile depth (MPD) of the pavement macrotexture, explaining approximately
80 % of the total variation. The variation in RRC between the worst and the
best pavement is 45 %, which would correspond to approximately 5 % of fuel
consumption and CO2 variation in highway driving. Looking at the most com-
mon pavements in Sweden, the range is approximately 20 % 25 %
between the best and the worst. The conclusion is that there is a potential for
reducing fuel consumption of road traffic by using pavements with low rolling
resistance properties.
KEYWORDS: rolling resistance, pavement, pavement characteristics, fuel
consumption, mean profile depth, texture, macrotexture, megatexture, mea-
surement method, MPD, trailer, reference tires
Manuscript received November 14, 2011; accepted for publication April 19, 2012; published
1
Swedish National Rd. and Transport Research Institute (VTI), SE-58195 Linkoping, Sweden,
e-mail: ulf.sandberg@vti.se
Copyright V
19428-2959.
77
Introduction
Rolling resistance is a form of energy loss caused by the interaction between a
rolling tire and the pavement. This functional property of pavements and tires
is one of the most important from both an economical and environmental point
of view, something that is acknowledged widely with regard to tires, but which
is most often totally neglected with regard to pavements. There seems to be the
misconception that rolling resistance is a property of tires alone, rather than an
interaction, just like skid resistance or tire/pavement noise, in which both com-
ponents are equally guilty.
Pavements are traditionally selected essentially based on properties, such
as skid resistance, durability, and cost; sometimes they are selected for ride
comfort and tire/pavement noise emission. Rolling resistance is never
considered.
One reason why rolling resistance of pavements is hardly ever considered
an interesting property is the lack of practical measurement methods. Rolling
resistance of tires is measured on laboratory drums using ISO and SAE meth-
ods, but to take these methods out on the road is virtually impossible. The lack
of proper measurements has resulted in ignorance about the effect of pavement
on rolling resistance.
This article presents recent measurements of rolling resistance, made possi-
ble by means of a specially designed trailer, on a number of common pavement
types in Sweden, supplemented with a few pavements in Denmark. The overall
purpose is to study the potential for saving energy and CO2 emissions by add-
ing rolling resistance data in pavement management systems.
Pavement Features Contributing to Rolling Resistance

There are many mechanisms that contribute to the rolling resistance energy
loss [1], but this will not be treated in this article. However, with respect to the
pavement, the following features are potentially influential [1]:
pavement texture (microtexture, macrotexture, and megatexture),
unevenness,
rutting, and
stiffness.
Road surface macro- and megatexture creates local deflections in the tire,
which causes extra energy losses if the tire rubber is not free of hysteresis.
Microtexture affects the frictional properties and will thus influence the
stickslip movements in the tire/pavement interface, which are connected with
energy losses. Unevenness may cause losses in the vehicle suspension and
some in the tire itself. Rutting may mean that the tire will be exposed to correc-
tional side forces. Some rolling losses are even consumed in the road or pave-
ment rather than in the tire. Stones in the road surface may move a little
SANDBERG, doi:10.1520/STP104543 79
tangentially under the forces of the tire and the surface may deflect somewhat
vertically (or radially) under the tire load, depending on pavement stiffness;
these are two types of deflection that consume some energy if the materials
involved are not totally elastic.
Definition of Rolling Resistance and Rolling Resistance Coefficient (RRC)

Rolling resistance may be defined as the energy required to move a rolling tire
a certain distance in the desired direction. A practical definition of rolling re-
sistance must consider what kind of physical measure that can be used to mea-
sure this variable. Energy does not appear to be a practical physical measure to
use in a rolling resistance measurement standard. For example, measuring heat
dissipation in a tire, although not impossible, seems impractical.
The most relevant measure, also practical to use, seems to be something
based on the force (force of rolling resistance (FRR)) that is required to move
the rolling tire in the desired direction. However, it is clear that this force will
depend on the vertical load that is applied to the tire. Studies have found an
approximately linear relation between rolling resistance and tire load, repre-
sented by a vertical force Fz (equal to the mass m of the load times the gravita-
tional constant g). A rolling resistance coefficient, RRC or Cr (both terms are
commonly used; ISO 28580 uses the notion Cr), has been created to represent
the characteristic of tire/road rolling resistance: the dimensionless ratio of roll-
ing resistance to tire load:
Rolling resistance coefficient RRC Cr FRR=Fz
Typical values of RRC for tires in new condition are in the range 0.006 to
0.015 for automobile tires, and 0.004 to 0.012 for heavy truck tires.
Earlier Studies of Pavement-Related Rolling Resistance

Coast-down measurements were done as early as in the 1920s to investigate
the influence of different road pavements on rolling resistance [2], see Fig. 1.
Some groundbreaking work on rolling resistance and fuel consumption
related to road surface properties were made in the 1980s. Using a special
rolling resistance trailer, as well as profilometers for the texture and the
unevenness ranges, Descornet at the Belgian Road Research Centre (BRRC)
analyzed the relation between the rolling resistance coefficient (RRC) and
unevenness, megatexture, and macrotexture. He found that the most sensitive
range was the megatexture range [3], see Fig. 2. In the beginning of the 1980s,
Sandberg made corresponding measurements and analyses, but the big
difference was that he measured fuel consumption with a full Volvo 240 car
and not rolling resistance; see results in Fig. 3 [4].
These two diagrams look quite different. However, the difference is logi-
cal, because when making fuel consumption measurements, in contrast to
trailer measurements, the suspension losses are present and they should peak in
the area where Sandbergs data peak. It may be noticed that in the megatexture
and macrotexture areas Descornets and Sandbergs data are not very different.
The value of these data is that they suggest that one shall measure both the
tire rolling resistance and the suspension losses, and that not only macrotexture
but also megatexture is important for tire rolling resistance. Nevertheless, in
the newer studies, megatexture has been neglected; albeit VTI and some others
regularly measure it. One reason is that it is very difficult to setup a measure-
ment program in which macro- and megatexture are not well correlated,
FIG. 1Ames Space-time Recorder for measuring retardation (speed)

during coastdown, as used in Ref 2.
FIG. 2Correlation between RRC and road roughness/texture level as a func-

tion of texture wavelength [3].
FIG. 3Correlation between fuel consumption per km and road roughness/

texture level as a function of texture wavelength [4].
implying that their effects cannot be separated. Another is that, as mega- and
macrotexture normally are well correlated, there is less need to determine
which one is the most influential, and one can choose the measure that is most
commonly collected, which is the mean profile depth (MPD).
After these measurements in the first half of the 1980s (published in 1990)
and some others (see the review in Ref 1), nothing of importance happened
with regard to pavement-related rolling resistance until the Technical Univer-
sity of Gdansk (TUG) in Poland, under the leadership of Prof. Jerzy A.
Ejsmont, between 2003 and 2006, produced an improved version of the
Belgian trailer used by Descornet. This author, who had tried without success
to raise the interest in the matter in Sweden since his study in the 1980s, and
had made substantial studies on tire rolling resistance, see e.g., [5], ordered the
first measurements in 2005 with the TUG trailer, while it still was not vali-
dated, and used the results to increase interest in the subject at the Swedish
Road Administration. Also some large international projects on energy use in
the road sector helped in this task; most of all was the Energy Conservation in
Road Pavement Design (ECRPD) project [6].
After it became evident that Swedish roads have very unfavorable textures
for rolling resistance, and thus create extra fuel consumption and CO2 emis-
sions from traffic, the interest in the subject has been steadily increasing in
Sweden. The intention is that models for the selection of pavements and their
maintenance shall include a rolling resistance parameter in the near future.
Since 2010, several countries in northern and middle Europe, as well
as California, have cooperated in a pool-funded rolling resistance project
named MIRIAM (Models for Rolling Resistance in Road Infrastructure
Asset Management Systems) [7]. For example, this has resulted in a special
session on rolling resistance being organized at the annual meeting of TRB in
2012 [8].
Rolling resistance trailers in Germany, Poland, and Belgium produced in
the previous decade or earlier [1] have helped to raise the interest in pavement-
related rolling resistance. What has made this study possible is the access to an
appropriate and reasonably accurate and practical measurement instrument, the
Polish R2 trailer.
Measurement Methods in General

The measurement methods for rolling resistance, which are available, can be
grouped into four general categories:
Laboratory drum (DR) method: Laboratory measurements made with
test tires rotating on drums. The drums may be equipped with sandpaper
or replica road surfaces, apart from the steel surface of the drum. This is
the method used in tire testing, with variants specified in SAE J1269 and
SAE J2452, as well as in ISO 18164:2005 and ISO 28580:2009.
Trailer (TR) method: Measurements with test tires rolling in a special

towed trailer. The trailer may be designed either for passenger automo-
bile tires or for heavy truck tires. Used on the road or on test tracks.
Coast-down (CD) method: Coast-down measurements using any type of
vehicle, measuring deceleration and a number of other parameters, from
which rolling resistance may be calculated. The test vehicle is coasted
(gearbox in neutral) from one higher to one lower speed. Used on the
road or on test tracks.
Fuel consumption (FC) method: Fuel consumption measurements using
specially instrumented (normal) vehicles, from which rolling resistance
may be calculated by means of a fuel consumption/rolling resistance
model. A variant of this would be when using a vehicle not powered by
fuel, such as an electric vehicle. Then the method should be called
energy consumption (EC) method. Note that driver influence might be
high in this method.
Measurement Method and Equipment in This Study

As already mentioned, in this study, a trailer from TUG in Poland has been
used. This trailer is designed to be towed by a reasonably powerful passenger
car. The construction of the trailer is self-supporting (three-wheeled). The front
wheels that stabilize the trailer have self-aligning properties and assure good
stability of the trailer. An independent front suspension, based on double trans-
verse arms, was constructed as an adaptation of passenger car suspension.
However, during tests the suspension is blocked by removable bars, which
ascertains proper levelling of the trailer. There is a single test wheel at the rear,
which is supported by a vertical arm (4)2 that is an important element of the
force-measuring system (Fig. 4).
The front and rear suspension are connected to a horizontal arm (1). The
rotation axis (3) is placed directly in the geometrical centre of the rotation axis
of the front wheels. Tire load is provided by an arm (2) that has a common
rotation axis with the horizontal arm (1). The load block (6) is resting on an
arm (2).
Suspension element (7) carries the load from an arm (2) to the horizontal
arm (1). The rear end of the horizontal arm (1) is connected to the vertical arm
(4), which is equipped with the test wheel hub. Undesirable vibrations of the
vertical arm (4) that may be induced during tests are suppressed by a Foucault
currents electromagnetic brake (not shown in Fig. 4). Inflation pressure in the
test tire is maintained by a remote-controlled release valve and pressure sensor.
2
Numbers in parentheses (x) refer to parts of the trailer shown in Fig. 4.
FIG. 4The measurement principle of the R2 trailer from TUG in Poland.
During tests, the rolling resistance force acting on the test tire pulls
(deflects) the vertical arm (4). The deflection rate is measured by the laser sen-
sor installed on the horizontal arm (1) and sending the laser beam toward the
vertical arm (4). The rolling resistance coefficient is defined as a ratio of
rolling resistance force Pf (same as FRR mentioned earlier) and vertical load
Fz. The trailer is equipped with a patented compensation device that eliminates
the influence of factors, such as road inclination and longitudinal acceleration,
that otherwise very substantially would disturb the measurements.
The appearance of the R2 trailer is illustrated in Fig. 5.
Measurements are made as follows: After approximately 20 min of running
to warm-up the test tire, the trailer is towed over the test section at as constant
a speed as possible (usually at 50 and 80 km/h), over which the RRC is meas-
ured as a function of position. The measurements are repeated in the opposite
direction if possible, and one or more extra runs (loops) are made. If the test
sections are at least 400 m long, it is usually enough with two runs (loops). It is
possible to measure sections as short as 40 m, although then one must make
several repetitions.
In addition to rolling resistance, measurements were made also of noise
properties and texture on most pavements, and, in some cases, skid resistance.
The noise and skid resistance measurements are not presented here, but the tex-
ture measurements will be used.
Texture measurements were made with VTIs mobile laser profilometer
running at 3640 km/h and measuring texture in the same path as the rolling
resistance measurements. The performance of this profilometer makes possible
measurements from 500 mm down to 2 mm texture wavelengths. Although full
profile curves and texture spectra were recorded, in this article, only the mean
FIG. 5The tire/road rolling resistance measurement trailer from TUG in the
shape and condition of 2010 (before 2009 the test tire was not enclosed).
profile depth (MPD) values are used. These were measured following the ISO
13473-1 standard from 1997. A similar standard is the ASTM E1845-09.
Reference Tires
As it is very impractical to measure rolling resistance using a great number of
tires representing the tire market when classifying or ranking pavement proper-
ties for rolling resistance, it is practical if not necessary to use reference tires.
The purpose of these is to be representative of the category of tires that they
are intended to represent and to provide stable and repeatable conditions. A
common reference tire concept is that one tire or set of tires shall represent the
fleet of automobile tires on the roads (tire category C1), and another tire or set
of tires shall represent the fleet of heavy truck tires (C3). One might also want
to have a tire representing the middle range, van tires (C2). Reference tires
must be available for a long time. ASTM has defined a set of standard refer-
ence test tires (SRTTs) for C1, C2, and C3 tires, of which the one for C1 has
been employed in the studies reported here (ASTM F2493-06).
This reference tire concept is already implemented in the drafts ISO/DIS
11819-2 and ISO/TS 11819-3, which are two documents specifying the so-
called close-proximity (CPX) method for classification of noise properties of
pavements. A draft for an ASTM method for a Standard Test Method for
Measurement of Tire-Pavement Noise Using the On-Board Sound Intensity
(OBSI) Method specifies the use of one reference tire (the SRTT of ASTM
2493-06). The tires used in the CPX method by ISO are shown on the left
(SRTT) and in the middle (AAV4) in Fig. 6. The tire on the right is an extra
FIG. 6Reference tires used in the tests reported in this article. Refer to the
text for more information.
tire used by TUG from the time they started to make rolling resistance meas-
urements, and has been kept since then for the purpose of providing a link to
the old measurements.
The SRTT is a tire specified in ASTM F2393-06 as a reference tire for var-
ious purposes. The Avon AV4 tire (AAV4) is a tire tested and found to classify
pavements (for noise) in roughly the same way as a selection of regular heavy
truck tires does. It is, in fact, a light truck tire, but, as the smallest dimension
for this series of tires is used, the AAV4 fits on large passenger cars, as does
the SRTT.
The SRTT and the AAV4 are tires considered in the MIRIAM project to
become reference tires also for rolling resistance, and have been tested for this
purpose in MIRIAM.
As it is a reference tire specified by ASTM, the SRTT is likely to be avail-
able for several decades in the future. The AAV4 tire will not be manufactured
in the future unless the users of CPX tires order a full batch of 100 or more tires
simultaneously, which is indeed the plan.
Test Pavements
The tested pavements include the following numbers and types:
11 dense asphalt concrete, maximum aggregate sizes: 6, 8, 11, and 16 mm;
9 stone matrix asphalt (SMA), maximum aggregate sizes: 6, 8, 11, and 16 mm;
1 hot-rolled asphalt (HRA), U.K.-type, maximum aggregate size: 16 mm;
3 dense-graded asphalt rubber (Arizona-type adapted to Sweden), maxi-
mum aggregate sizes: 11 and 16 mm;
1 open-graded asphalt rubber (Arizona-type adapted to Sweden), maximum
aggregate size: 11 mm;
3 porous asphalt concrete, single layer, maximum aggregate sizes in top

layer: 8 and 11 mm;
3 porous asphalt concrete, double layer, maximum aggregate sizes in top
layer: 8 and 11 mm;
2 chip seals (surface dressings), single layer, maximum aggregate size: 11 mm;
6 thin asphalt layers (dense), maximum aggregate sizes: 6, 8, and 16 mm;
1 exposed aggregate cement concrete, max aggregate size: 16 mm; and
1 SMA, medium texture but very uneven, maximum aggregate size: 16 mm.
These have been tested in the various conditions that each one was in,
including both new pavements and a few near the end of nominal life.
Measurement Results
In this article, the results will not be listed in tables for each test condition, tire,
and pavement. It would be a huge table, difficult to evaluate. Instead, the RRC
data will be plotted against the MPD data in regression diagrams.
The first diagram shows results measured in 2005 and 2007, at a time when
the test tire was not yet protected from the air flow by an enclosure. Figure 7
shows results corrected for the air-flow effect (by testing at various speeds and
deducting the speed effect), where RRC at 80 km/h has been plotted against
MPD values for the same pavements. These measurements were made by using
FIG. 7Rolling resistance coefficient plotted against macrotexture (MPD) for

measurements in 2005 and 2007. The round symbols are for dense asphalt,
SMA pavements, and an extremely rough-textured chip seal (the highest point),
whereas the single square symbol is for an EACC (cement concrete). Note that
these measurements were made with four tires (RRC averaged), of which one
was the SRTT; the others were regular market automobile tires.
four automobile tires, including one SRTT. The round symbols are for dense
asphalt and SMA pavements having maximum aggregate sizes 8, 11, and 16
mm, and an extremely rough-textured chip seal at the top; whereas the square
symbol is for an exposed aggregate cement concrete (EACC) with maximum
aggregate size 16 mm.
Apart from a correlation so high that it is probably just by chance, it may
be noted from this diagram that the EACC pavement follows the same RRC-
MPD trend as the asphaltic pavements. It is claimed by the cement concrete
industry in North America that such rigid pavements have lower RRCs than
flexible pavements but our measurements could not verify this. Nevertheless,
things may be different for larger tires, higher loads, and hot weather. Note
also the very large range of RRC: over the measured MPD range, the highest
RRC on the regression line is 45 % higher than the lowest RRC on the same
regression line. One may say that this MPD range covers the macrotexture
range of Swedish paved roads, if one excludes pavements with obvious defects,
such as bleeding asphalt and pavements with lots of ravelling and potholes.
When measurements were made after 2008, TUG had fitted an enclosure
over the test tire, more or less eliminating the air-flow resistance around the
tire, adding to the rolling resistance. It then appeared that the RRC data meas-
ured at 50 km/h correlated almost perfectly to those at 80 km/h. This shows
that the measurements were repeatable and subject to only small disturbances.
But because of this, in this article for the 20092010 data, only the results for
80 km/h are displayed.
Figure 8 shows the results of the latest measurement campaigns, using the
TUG trailer with an enclosure to eliminate the air-flow resistance. In this case,
all measurements in 20092010 are put into the same diagram without distin-
guishing between different series of measurements (i.e., different weeks). One
can see that the correlation is very low; R2, which shows the variance explained
by the RRC-MPD regression as part of the total variance in this set of data, is
only 0.26. This is statistically significant on the 95 % level, but it is very disap-
pointing compared to Fig. 7. One may, therefore, suspect that there is one vari-
able (or more) that is out of control in this scenario and which ruins the
correlation.
The next figure (Fig. 9) reveals, at least, part of the problem. In this dia-
gram, data have been analyzed and plotted separately for each measurement se-
ries. A measurement series is a set of measurements at similar temperatures
covering one, two, or possibly three consecutive days. In such cases, a single
calibration is assumed to be valid for all measurements within the series. Note
that the two points in the upper right corner are for a chip seal (in two different
tracks), which had some potholes and were measured at temperatures near the
freezing point.
It appears that there is an excellent and reproducible relation between
RRC and MPD, i.e., that the rolling resistance pavement effect is largely

measurements in 2009 and 2010. All data for 80 km/h and averaged for all
three test tires (Fig. 6) have been put together without distinction.
caused by macrotexture, and/or perhaps something very closely correlated with

macrotexture. The latter may be megatexture. Note that the slope coefficient in
the regression line equation is rather consistent (0.0170.022).
Another observation is that the measurements are influenced by a biasing
factor with high influence related to something that is unique for the series of
measurements. One such parameter may be ambient temperature (or pavement
or tire temperatures, but they have been shown to be highly correlated with
each other). And there are some temperature differences here that can explain
some of the bias, but not all of it. These results are not shown here, as it is yet
not known what the appropriate RRC-temperature relations might be, and not
all data have been analyzed in this way yet. Future analyses and planned meas-
urements will hopefully shed some light on this.
Another potentially biasing effect might be calibration; possibly there is a
drift with time in the system, which is unknown and not controlled by the
equipment operator. If so, this drift seems to have some random or, at least,
inconsistent relation with time, as it is impossible to see a relation between this
bias and time when the series is measured. This bias problem is a matter that
must be studied much more in the near future, and will be within project
MIRIAM.

measurements in 2009 and 2010. As the previous figure, but here a distinction
is made between different measurement series (i.e., different regressions). The
most essential regression data, equation for regression line and correlation
coefficient squared, are indicated for each series of measurement.
Discussion
In the first part of this article, it was stated that megatexture seems to have
even better correlation with rolling resistance than macrotexture. This is veri-
fied by a study made in 2011, within project MIRIAM [9], but only provided
that both mega- and macrotexture data are based on values linearly or logarith-
mically related to the profile curve. If macrotexture is represented by MPD,
which is more sensitive to top than bottom parts of the profile, then MPD is
better correlated to RRC than megatexture is.
Because megatexture (measured in accordance with ISO 13473-5) is well
correlated with RRC, why not try megatexture as a parameter instead of the
MPD, or as a supplementary texture parameter? One reason is that the present
set of tested pavements has not been selected with the possibility of testing
megatexture level versus MPD, because these two parameters are mostly very
closely correlated. A second reason is that MPD is a variable that is sensitive to
the direction of the texture, namely, that its features upwards (peaks) in the pro-
file are much more influential than its features downwards (valleys). This
should be favorable to represent rolling resistance sensitivity to texture. So far,
there is no similar factor that can modify megatexture values in a similar way,
and it is not sure that it would be relevant. A third reason is that the correlation
between RRC and MPD seems to be so good that it explains something like 80
% of the total variation.
Acknowledging all the potential errors in these measurements, both with
respect to rolling resistance and texture, it is unlikely that much higher correla-
tions are possible with present methods and equipment. An exemption is a
study in MIRIAM where it was noted that if the profile curve was modified to
represent the enveloping features of the tires, even higher correlations are
achieved, but still MPD is the best parameter [9]. Therefore, megatexture has
not yet been tried as a replacement or supplement to MPD, but it lies in the
pipeline for further studies. It is probable that including megatexture might
give only a slight improvement, but maybe most important is that when pave-
ments deteriorate, it is mostly megatexture that changes, and this should be
picked up by a proper model.
Another thing is that unevenness, represented by the IRI (the International
Roughness Index), should also have an influence. So far, only a few of the test
sections have been measured with regard to unevenness, so such analysis has
not yet been made, although an uneven section was indeed compared once to a
more even section and some logical differences were observed, but more stud-
ies are needed here.
It shall be noted that several effects other than by MPD, for example, IRI,
have been studied in a comprehensive model study recently made at VTI [10].
This study is based not only on the same measurements as presented here, but
also, and moreover, on a set of advanced coast-down measurements using an
automobile and a heavy truck. In this study, approximately the same influence
of MPD on RRC has been found, but it has also been found that IRI has a sig-
nificant influence. For automobiles the IRI influence seems to be lower than the
MPD influence (for an unevenness range typical of European roads), but for
trucks the IRI influence seems to be relatively more important than for automo-
biles. Whether the IRI or MPD influence is the greatest for trucks is uncertain
[11]. In this case, one must note that IRI (unevenness) should indeed have an
extra effect when testing a full vehicle (including its normal suspension)
instead of when testing with a trailer, in which suspension has different and
probably less influential characteristics (this issue is presently under study at
TUG).
Another thing not yet performed in full, and thus not possible to include
here, is to distinguish between the three test tires used. It is obvious from the
raw data (not shown here) that the three tires give regressions between RRC
and MPD that are different from each other. Especially the AAV4 tire, sup-
posed (or rather hoped) to be a suitable proxy for heavy truck tires, has a roll-
ing resistance, which is substantially less sensitive to texture. This is most
probably very desirable, as it is known that heavy truck tires have much lower
sensitivity to texture with regard to noise emission [12], and, at the low fre-
quencies of interest here, it is likely that noise emission and rolling resistance
have a significant correlation. The tire influence on these results will also be
studied much more in planned near-future projects.
It is a little surprising how consistent the coefficients in the regression
line are for the relation RRC-MPD. An average seems to be close to 0.0020
and the variation around this is only about 615 %. However, one must be
aware of that if more variables were taken on board in the regression, as, for
example, megatexture and IRI, the coefficient for MPD could be reduced.
Essentially, all these results were confirmed in a MIRIAM study made in
2011 [8].
The variation in rolling resistance over the range of pavements tested here
is comparable to that of tires, i.e., pavements may affect the rolling resistance
coefficient equally as much as tires.
Estimated Effects on Fuel Consumption

Finally, some words about how one can estimate the effects on fuel consump-
tion of vehicles from these rolling resistance data. It is often assumed in the lit-
erature that the return factor, which is the change in fuel consumption for a
change in rolling resistance, is between 5 % and 30 %. This large range is dis-
cussed in Ref 1, where it is concluded:
Based on these US and Swedish data, it seems that rounded rule-of-
thumb values for the Return Factor of cars could be 10 % for urban driving
and 20 % for highway driving, as well as 15 % for trucks in urban driving and
30 % for trucks in highway driving. These stereotype values are not recom-
mended to be used in serious technical calculations, but may be useful to
remember by heart to get a general feeling for the case.
However, more recent calculations at VTI, for project MIRIAM, have indi-
cated much lower return factors [13]. A calculation of potential fuel
consumption (FC) savings for the entire Swedish national highway system
indicated an FC savings of 0.9 % for a rolling resistance reduction of 8.6 % for
light vehicles and 1.2 % FC savings for 12.5 % of rolling resistance reduction
for heavy vehicles. This implies a return factor of approximately 10 % for both
vehicle categories. This would be for mainly highway and motorway driving.
The calculation is based on an advanced model and assumes that the rolling
resistance reduction is achieved by an MPD reduction of 0.5 mm and
unchanged IRI of 1.02.0 [13]. It also assumes that speeds are not influenced
by the change in MPD.
For example, when this article has indicated a range of RRC between com-
mon pavements in Sweden and Denmark of 20 % 25 % (up to 45 % if includ-
ing chip seal pavements), using a 10 % return factor means a fuel consumption
influence of 2.0 % 2.5 %. This may be achieved for those roads that have a
very rough texture. However, what is realistic on a national scale may be a

reduction of MPD from the present national average, which is 0.9 to 1.0 mm
according to an estimation based on Ref 14, down to a future 0.5 mm, and this
would generate fuel consumption savings of 1 % according to the calculation
in Ref 13. In terms of national consumption in the road transport sector, this
corresponds to substantial cost savings, which are possible if the present rather
rough-textured pavements are replaced by fine-textured ones.
Conclusions
The results presented in this article show the following:
Macrotexture, represented by the parameter MPD, is a major factor influencing
rolling resistance. It is not known if it is the most important parameter of the
pavement, but it is so well correlated with rolling resistance that it is difficult
to find a better single variable for the purpose of quantifying the pavement
influence on rolling resistance.
The relation is rather consistent measured in different and independent measure-
ment series: around a coefficient X of 0.0020 in an equation of
RRC X MPD Y, where Y is the basic RRC as measured on a hypotheti-
cally absolutely smooth pavement.
There is a substantial bias between various series of measurements, the source
of which is not yet known, but it is believed that temperature is part of the
solution and that uncertain calibration might be another part of the solution.
The range in rolling resistance between the best and worst Swedish pavements
is at least 45 % (the worst has an RRC 45 % higher than the best), although
the more common pavements exposed to high traffic flows show a range of
20 %25 % in rolling resistance. It is estimated that Swedish fuel
consumption savings may amount to approximately 1 % if MPD values
could be reduced by an average of 0.5 mm.
Rolling resistance is not only a property of tires, but is also a property of the
pavement, which is of high importance for the energy consumption in the
road transport sector and must be systematically considered along with other
functional properties in pavement management systems.
The work with this pavement property has only just started and a lot more
research is needed in the near future, not the least with regard to measure-
ment methods.
Acknowledgments
The writer is grateful to the Swedish Transport Administration for sponsoring
most of this study in the project Rolling Resistance and Noise Properties of
Nordic Roads and Its Relation to Pavement Texture. Part of the study is also
sponsored through a Nordic project called Road Surface Texture for Low
Noise and Low Rolling Resistance. Some of the information in this article has
been collected within the MIRIAM project [7]. The writer is also grateful to
the Technical University of Gdansk (TUG) in Poland and its staff under the
direction of Prof. Jerzy A. Ejsmont for assisting VTI with all the rolling
resistance measurements made with their R2 trailer.
References
[1] Sandberg, U., ed., 2011, Rolling ResistanceBasic Information and

State-of-the-Art on Measurement Methods, Report MIRIAM_SP1_01,
project MIRIAM, http://www.transguide.org/VTI%20publ/MIRIAM-
SoA-Report-Final-110601.pdf (Last accessed April 16, 2012).
[2] Agg, T. R., Tractive Resistance of Automobiles and Coefficients of Fric-
tion of Pneumatic Tires, Iowa State College of Agriculture and Mechanic
Arts Official Publication, Bulletin 88, Vol. XXVI, No. 70, May 2, 1928.
[3] Descornet, G., Road Surface Influence on Tire Rolling Resistance, Sur-
face Characteristics of Roadways: International Research and Technolo-
gies, ASTM STP 1031, W. E. Meyer and J. Reichert, Eds., ASTM
International, West Conshohocken, PA, 1990.
[4] Sandberg, U., Road Macro- and Megatexture Influence on Fuel Con-
sumption, Surface Characteristics of Roadways: International Research
and Technologies, ASTM STP 1031, W. E. Meyer and J. Reichert, Eds.,
ASTM International, West Conshohocken, PA, 1990.
[5] Sandberg, U. and Ejsmont, J. A., Noise Emission, Friction and Rolling
Resistance of Car TiresSummary of an Experimental Study, Proc. of
NOISE-CON 2000, Newport Beach, CA, 2000.
[6] ECRPD, 2010, Energy Conservation in Road Pavement Design, Mainte-
nance and Utilisation, ECRPD Final Publishable Report, Grant Agreement:
EIE/06/039/SI2.448265, Intelligent Energy Executive Agency (IEEA), http://
www.transguide.org/eu/ecrpd/20091001_D5(a)%20ECRPD%20Road%20
surface%20effects%20on%20rolling%20resistance%20090416.pdf (Last
accessed April 16, 2012).
[7] MIRIAM, http://www.miriam-co2.net (Last accessed April 16, 2012).
[8] Special Sessions 265 and 325, 91st Annual Meeting of the Transporta-
tion Research Board, Washington, D.C., 2012, http://www.miriam-co2.
net/Publications/Publications.htm (Last accessed April 16, 2012).
[9] Bergiers, A. L., Goubert, L., Anfosso-Ledee, F., Dujardin, N., Ejsmont,
J. A., Sandberg, U., and Zoller, M., Comparison of Rolling Resistance
Measuring EquipmentPilot Study, MIRIAM SP1 Deliverable No. 3,
Final Version 2011-12-31, http://www.miriam-co2.net/Publications/Pub-
lications.htm (Last accessed April 16, 2012).
[10] Karlsson, R., Hammarstrom, U., Sorenssen, H., and Eriksson, O.,
Road Surface Influence on Rolling ResistanceCoastdown
Measurements for a Car and an HGV, Report VTI Notat 24A-2011,

Swedish National Road and Transport Research Institute (VTI), Link-
oping, Sweden, 2011.
[11] Karlsson, R., Project MIRIAM Report, Swedish National Road and Trans-
port Research Institute (VTI), Linkoping, Sweden (unpublished).
[12] Sandberg, U. and Ejsmont, J. A., 2002, Tyre/Road Noise Reference
Book, Informex, SE-59039 Kisa, Sweden, http://www.informex.info
(Last accessed April 16, 2012).
[13] Hammarstrom, U., Eriksson, J., Karlsson, R., and Yahya, M.-R., A Roll-
ing Resistance Model, a Fuel Consumption Model and the Traffic Energy
Saving Potential from Changed Road Surface Conditions, VTI Report
748A, Swedish National Road and Transport Research Institute (VTI),
Linkoping, Sweden, 2012.
[14] Gustafsson, M. and Lundberg, T., Prognosmodeller for tillstandsmatt i
Trafikverkets Pavement Management SystemMakrotextur, MPD, VTI
Report 7XX, Swedish National Road and Transport Research Institute
(VTI), Linkoping, Sweden, 2012.
STP 1555, 2012
DOI:10.1520/STP104446
Robert Otto Rasmussen1 and Richard C. Sohaney2
Measurement and Geometric and Functional

Characterization of Pavement Texture
REFERENCE: Rasmussen, Robert Otto and Sohaney, Richard C.,

Measurement and Geometric and Functional Characterization of Pavement
Texture, Pavement Performance: Current Trends, Advances, and Chal-
lenges on December 5, 2011 in Tampa, FL; STP 1555, B. Choubane, Editor,
pp. 96109, doi:10.1520/STP104446, ASTM International, West Consho-
hocken, PA 2012.
ABSTRACT: Pavement texture is defined by the irregularities on a pavement
surface that deviate from an ideal, perfectly flat surface. Texture influences
many aspects of road safety and comfort, including friction, smoothness,
splash and spray, rolling resistance, and numerous others. The focus of this
paper is on techniques to measure and analyze texture, particularly that
which influences tire-pavement noise and friction. Standards for characteriz-
ing texture geometrically are described, and techniques that describe texture
in terms of its influence on functional performance are explored.
KEYWORDS: pavement, texture, profile, tire-pavement noise, friction,
smoothness, splash and spray, rolling resistance
Introduction
Pavement texture is defined by the irregularities on a pavement surface that
deviate from an ideal, perfectly flat surface [1]. As shown in Fig. 1, the World
Road Association (PIARC) has established standard categories of texture,
classified by wavelength. These categories include microtexture (wavelengths
up to 0.5 mm), macrotexture (0.5 mm to 50 mm), megatexture (50 mm to
October 2012.
1
The Transtec Group, Inc., Austin, TX, 78731 (Corresponding author), e-mail: robotto@
thetranstecgroup.com.
2
The Transtec Group, Inc., Austin, TX, 78731
Copyright V
19428-2959.
96
RASMUSSEN AND SOHANEY, doi:10.1520/STP104446 97
FIG. 1World Road Association (PIARC) texture definitions and their influence
on pavement surface characteristics.
500 mm), and roughness (wavelengths larger than 500 mm). Texture influences
many aspects of road safety and comfort, including friction, smoothness, splash
and spray, rolling resistance, and numerous others [2,3]. Figure 1 also illus-
trates the ranges of texture that influences these pavement surface characteris-
tics, with a positive influence shown in green and deleterious effect shown in
red. The focus of this paper is on techniques to measure and analyze texture,
particularly that which influences tire-pavement noise and friction.
In modern concrete pavement construction, texture is deliberately imparted
in a reasonably controlled manner. The nominal geometry of the texture is typi-
cally prescribed, either directly or indirectly. For example, as illustrated in
Fig. 2, the spacing and depth of the directional striations produced by tining
are often described in a specification. Diamond ground textures will be defined
in large part by the geometry of the grinding head (blade widths and spacers).
Even the texture of a drag surface will be defined to some degree by the spe-
cific mechanical means used to impart these surfaces.
While it would be ideal for the as-constructed texture to match the as-
designed (nominal) geometry, this is not usually the case. Numerous factors
will affect the final texture, including variability in materials and construction
technique. This is illustrated in Fig. 3, which shows variability of pavement
texture on two areas of a project that were constructed within minutes of each
other. In addition to initial variability, texture will change over time due to
both traffic and the environment.
With texture affecting so many different surface characteristics, it is an
ideal parameter to be evaluated as part of construction quality control.
FIG. 2Typical texture specifications.
Furthermore, because texture is simply geometry, it is a fundamental aspect of

a pavement surface to measure. This paper discusses pavement texture mea-
surement, with specific discussion of the methods used in the Concrete Pave-
ment Surface Characteristics Program (CPSCP) conducted by the National
Concrete Pavement Technology Center (CP Tech Center). The methods and
metrics used to quantify texture are also described herein.
Measuring Texture
Laser-based profilometry is the most commonly used technique to collect pave-
ment texture data. Various standards have been developed to describe the req-
uisite equipment in such a way that is relevant to the type of texture to be
FIG. 3Material and construction variability affecting the as-constructed

texture.
measured. Of these, ISO 13473 [47] is the most commonly cited. While other
texture measurement methods have been advanced, they often possess techni-
cal or practical limitations. Laser-based texture profilometry uses technology
that is similar to road profilers that evaluate pavement smoothness. However,
when measuring texture, a greater degree of precision and accuracy is needed
with respect to small features that are normally filtered out when evaluating
road roughness.
Most profilers in use today measure texture using a single point laser which
results in a 2-dimensional (2-D) texture profile; with distance along the pave-
ment surface as one dimension, and the texture elevation as the second. Con-
crete pavement texture is complex though; it is anisotropic, meaning that the
texture varies depending on the direction of the measurementlongitudinal or
transverse. Measuring a 2-D profile therefore fails to completely describe char-
acteristics of the texture that are important to many surface characteristics. In
addition, measuring with a single point laser can introduce artifacts in the
measurement that distort the profile, and thus introduce error in predicting the
surface characteristic of interest.
To overcome the limitation of using a spot sensor, a line laser can be used
instead. Traditionally, line lasers are found in machine vision applications. In
recent years, vendors of these sensors have worked with the highway industry
to develop variants of the sensors that are suitable for texture profile applica-
tions. Figure 4 illustrates the measurement principle of a line laser. Measuring
with a line laser results in a 3-dimensional (3-D) texture profile; with distance
along the pavement surface as one dimension, distance along the laser line as
the second dimension, and texture elevation as the third.
FIG. 4Measurement principle of LMI-Selcom RoLine line laser height

sensor.
RoboTex
A line laser sensor alone cannot measure pavement texture properly. A plat-
form with a means to mobilize the sensor is required. In 2005, a prototype of
an innovative system for texture measurement was developed: a robotic-based
texture measurement system (RoboTex) [8,9]. As shown to the left of Fig. 5,
the system is built on a robotic chassis for remote control of the speed and
direction of travel of the measurement platform.
RoboTex includes customized software that simultaneously collects and
stores data from a number of sensors. As illustrated to the right of Fig. 5, these
include the following:
Line laser sensorfor height data;
Accelerometerto establish an inertial reference elevation;
Wheel encoderto determine the precise position of the robot;
GPSto establish a global position of the robot for reference;
Timeto determine speed and for global reference; and
Digital imaging systemfor a visual record of the surface.
The result is a concept device that provides a better tradeoff between reso-
lution and efficiency in texture data collection. RoboTex is capable of sampling
100 or more texture elevation points across a 100-mm wide laser line at 1000
Hz as it travels down the road under its own power at approximately 0.5 m/s.
This yields a pavement texture measurement with a spatial resolution of about
0.4 mm2, and a height resolution of 0.01 mm. More importantly, the result is a
3-dimensional texture profile along a 100-mm wide swath of pavement surface,
as illustrated in Fig. 6.
Geometric Texture Metrics

While illustrations of the measurements as shown in Fig. 6 are useful, they do
not quantify texture. However, there are a multitude of ways to characterize
FIG. 5RoboTex texture measurement system photo (left), and schematic of

components (right).
FIG. 6Sample RoboTex data from concrete pavement textures.
texture. The selection of a proper metric should begin with an understanding of

the fundamental mechanisms that describe the surface characteristics of inter-
est. Ideally, a texture metric should describe the texture in a way that relates its
interaction with a tire.
Quite often, texture depth is used as a texture metric. However, there are
numerous ways that texture depth can be calculated. To illustrate this, refer to
a short (100 mm) sample of tined concrete pavement texture, as illustrated in
Fig. 7.
A straightforward way to calculate texture depth is to calculate an average
(rectified) deviation of the texture from the mean depth. This is illustrated to
the top of Fig. 8, where the sum of the shaded areas both above and below the
mean profile elevation is divided by the length of the profile (100 mm). The
result is in an average texture depth (Ra) of 0.42 mm. Alternatively, the texture
profile shown in Fig. 7 can be squared, as illustrated to the bottom of Fig. 8.
Taking this area, dividing by the length of the profile, and then taking the
square root, results in the root-mean-square or rms texture depth (Rq). In this
example, the rms is 0.63 mm. Mathematically, these calculations are made as
follows using measured texture data:
v
uN
PN uP 2
jzn j u zn
n1
t
Ra and Rq n1
N N
where:
N number of measured samples of texture profile, and
zn texture profile elevation data (starting with n 1, and ending with
n N) after detrending.
FIG. 7Sample texture profile.
One disadvantage of most texture depth metrics is the lack of sensitivity to

directionality of the texture. For example, the same Ra and Rq values are calcu-
lated for the texture profile in Fig. 7 even if the profile is flipped over (upside
down). In the flipped over case, the tined grooves become sharp spikes and
clearly, a tire would interact with this profile very differently than the right-
side up profile. To help overcome this limitation, a directionality metric termed
texture skewness (Rsk), is calculated as follows:
P
N
z3n
n1
Rsk
R3q N
For textures like that shown in Fig. 7, the skew is negative (in this
example, 1.7). If the profile is flipped over the texture has a positive skew.
Historically, pavement texture depth has been measured using the volumet-
ric (sand) patch technique and reported as a mean texture depth (MTD) [10].
The test involves carefully spreading a known volume of small glass beads into
a circle on the pavement surface; the MTD is simply the volume of beads di-
vided by the area of the circle. As laser-based profilers began to be used, the
mean profile depth (MPD) metric was developed as a means to estimate the
MTD [4,11]. As shown in Fig. 9, calculating MPD involves dividing the 100
mm texture profile into two 50-mm segment halves. A peak profile point in
each segment half is determined, then averaged, and then the mean profile
FIG. 8Calculation of average texture height, Ra (top) and rms texture height,
Rq (bottom).
elevation (in this case, zero) is subtracted. The result in this example is 0.69
mm. It should be noted that the currently proposed revision to the ISO 13473-1
specification refers to the calculation within a 100-mm sample as a mean seg-
ment depth (MSD); the MPD is then an average of numerous (minimum of
five) MSD values.
FIG. 9Calculation of mean profile (mean segment) depth.
Functional Texture Metrics

While geometric metrics such as depth and skewness are intuitive and straight-
forward to calculate, by themselves they do not describe the interaction of a
tire and the pavement texture. Functional texture metrics are more useful in
this regard. ISO specification 13565-2 describes a process of transforming a
texture profile into a profile bearing ratio curve [12]. As illustrated in Fig. 10,
this begins with the recognition of three distinct regions of texture. The peaks
are the first to be contacted by a tire, and the most rapid to wear. The second
region is the core, where most of the texture typically lies, and thus describes
the longevity of the texture. The lowest regions are termed valleys, and are
where air and water can possibly reside that are not necessarily in contact with
the tire. Calculation of the heights of these three regions first involves convert-
ing the texture profile to a Profile Bearing Length Ratio (a type of cumulative
distribution of texture elevations). From this, various chord and area calcula-
tions derive the parameters of reduced peak height (Rpk), core roughness depth
(Rk), and reduced valley depth (Rvk).
Spectral Texture Metrics

Spectral analysis of pavement texture is used to identify the presence or signifi-
cance of various repeating features. To illustrate this, Fig. 11 shows a spectral
analysis of the same tined pavement shown in Fig. 7. More on the technique
used for this type of analysis can be found in ISO 13473-4 [7]. From this plot,
FIG. 10Calculation of profile bearing ratio from a texture profile.
a characteristic peak at 25 mm is visible, which corresponds to the tine spacing.

Other less obvious characteristics of the texture can sometimes be derived
from spectral analyses, particularly when the texture wavelength is plotted in
narrow band (as opposed to third-octave, illustrated here). Furthermore, there
is often a correlation between some pavement surface characteristics and the
texture content within select spatial frequency bands (or texture wavelengths).
It should also be noted that in Fig. 11, a transformation of the texture elevation
has been made to a texture level using a decibel scale and a reference rms tex-
ture of 1 lm. This should not be confused with the decibel scale that is used for
noise measurement and reporting.
Bridging Filter
Ideally, the texture metrics that are calculated should correlate well to the pave-
ment surface characteristics of interest, however this is not always the case.
Tire-pavement noise, in particular, is a difficult response to predict given its
complexity. One technique that has successfully improved this correlation is
the application of a bridging (or envelopment) filter. The filter is applied to tex-
ture profiles prior to calculating key texture metrics, most notably spectral
analysis.
The specifics of the bridging filter are governed by the response of interest.
For example, in the case of tire-pavement noise, the filter geometry is on the
same scale as that of the tire tread. As illustrated in Fig. 12, the filter simulates
FIG. 11Spectral analysis of texture profile.

FIG. 12Principles of tire tread bridging filter.
how a tire tread would impregnate the texture. The elevation of the filtered pro-
file is calculated as a balance of the downward force on the tire, and the upward
reaction due to the elasticity of the tire. This occurs when an equivalent uni-
form tire rubber displacement (e.g., 1 mm) is achieved [13]. At this point of
equilibrium, a single filtered elevation is calculated and stored as the bridged
elevation, and the process is repeated again after incrementally moving the fil-
ter forward. Mathematically, the filtering process can be further simplified by
sorting the texture profile elevations. This is illustrated in Fig. 13.
Conclusions
Texture is a vitally important pavement property because it relates to most of
what defines functional performance vis-a-vis pavement surface characteristics.
It has been shown that measuring the texture of concrete pavements in an accu-
rate and relevant manner requires 3-dimensional data, which in the CPSCP was
collected using an innovative device called RoboTex. The data from RoboTex
was used to calculate numerous texture metrics including those described
FIG. 13Calculation of bridged elevations using tire tread bridging filter.
herein. As a result, the as-constructed texture could be related to surface char-

acteristics such as tire-pavement noise and friction, and better practices devel-
oped for the design and construction of concrete pavement surfaces.
Acknowledgments
The writers would acknowledge the support of the National Concrete Pavement
Technology Center including Tom Cackler, Dale Harrington, and Paul Wie-
gand, as well as the sponsorship of Transportation Pooled Fund TPF-5 (139)
including contributions from FHWA and the State DOTs of CA, IA, MN, NY,
TX, WA, and WI. The writers would also like to acknowledge ISO TC 43/SC
1/WG 39, and in particular, their convener Dr. Ulf Sandberg. Finally, the
authors wish to acknowledge the assistance and support of LMI-Selcom
throughout this effort.
References
[1] ASME B46.1, 2009, Surface Texture (Surface Roughness, Waviness,

and Lay), ASME, New York.
[2] ASTM, Surface Characteristics of Roadways: International Research
and Technologies, ASTM Spec. Tech. Publ., Vol. 1031, 1990, pp. 5527.
[3] ASTM, Vehicle, Tire, Pavement Interface, ASTM Spec. Tech. Publ.,
Vol. 1164, 1992.
[4] ISO/WD 13473-1, 2009, Characterization of pavement texture by use of
surface profilesPart 1: Determination of Mean Profile Depth, ISO/TC
43/SC 1/ N 1xx, International Organization for Standardization, Geneva,
Switzerland.
[5] ISO 13472-2, 2002, Characterization of Pavement Texture by Use of

Surface ProfilesPart 2: Terminology and Basic Requirements Related
to Pavement Texture Profile Analysis, International Organization for
Standardization, Geneva, Switzerland.
[6] ISO 13473-3, 2002, Characterization of Pavement Texture by Use of
Surface ProfilesPart 3: Specification and Classification of
Profilometers, International Organization for Standardization, Geneva,
Switzerland.
[7] ISO/TS 13473-4, 2008, Characterization of Pavement Texture by Use of
Surface ProfilesPart 4: Spectral Analysis of Surface Profiles, Interna-
tional Organization for Standardization, Geneva, Switzerland.
[8] Rasmussen, R. O., et al., The Little Book of Quieter Pavements, Report
No. FHWA-IF-08-004, Federal Highway Administration, Washington,
DC, 2007.
[9] Rasmussen, R. O., Sohaney, R., Wiegand, P., and Harrington, D.,
Measuring and Analyzing Pavement Texture, Concrete Pavement Sur-
face Characteristics Program Tech Brief, National Concrete Pavement
Technology Center, Ames, IA, 2011.
[10] ASTM E 965-01, 2001, Standard Test Method for Measuring Pavement
Macrotexture Depth Using a Volumetric Technique, Annual Book of
PA.
[11] ASTM E 1845-96, 1996, Standard Test Method for Calculating Pave-
ment Macrotexture Mean Profile Depth, Annual Book of ASTM Stand-
ards, Vol. 04.03, ASTM International, West Conshohocken, PA.
[12] ISO 13565-2, 1996, Geometrical Product Specifications (GPS)Surface
Texture: Profile Method; Surfaces Having Stratified Functional Proper-
tiesPart 2: Height Characterization Using the Linear Material Ratio
Curve, International Organization for Standardization, Geneva,
Switzerland.
[13] Karamihas, S. M., Critical Profiler Accuracy Requirements, University
of Michigan Research Report UMTRI-2005-24, University of Michigan,
Ann Arbor, MI, 2005.
STP 1555, 2012
DOI:10.1520/STP104436
Bernard Izevbekhai1 and Alexandra Akkari1
Performance of Exposed Aggregate Surface

of Composite Pavements at MnROAD
REFERENCE: Izevbekhai, Bernard and Akkari, Alexandra, Performance of

Exposed Aggregate Surface of Composite Pavements at MnROAD, Pave-
ment Performance: Current Trends, Advances, and Challenges on December
5, 2011 in Tampa, FL; STP 1555, B. Choubane, Editor, pp. 110134,
doi:10.1520/STP104436, ASTM International, West Conshohocken, PA
2012.
ABSTRACT: This paper summarizes the construction and early performance
assessment of an exposed aggregate surface on a composite pavement test
cell (cell 72) at MnROAD. Although exposed aggregate surfaces and com-
posite pavement systems have become extremely popular in Europe, they
are a relatively new concept in the United States. The performance of the
exposed aggregate surface is compared to that in the two other composite
pavement cells at MnROAD: cell 70, hot mix asphalt (HMA) over a recycled
aggregate concrete, and cell 71, diamond grind concrete over recycled ag-
gregate concrete. The strength, tire pavement noise (on board sound inten-
sity), sound absorption, friction, texture, and international roughness index
were tested. The results suggest that the exposed aggregate concrete sur-
face does not provide significant noise reduction. Innovative diamond grind-
ing of composite pavements might be beneficial for noise reduction. Exposed
aggregate surfacing can provide more than adequate friction for driver safety,
but it does not show any improvement relative to typical HMA or diamond
ground surfaces. Exposed aggregate surfaces have a texture (or mean pro-
file depth) similar to that of traditional diamond ground surfaces but might be
detrimental to the ride quality. Continued monitoring of these test cells will aid
the development of a wide understanding of exposed aggregate surfaces and
composite pavements needed for more effective design and accurate service
life models.
KEYWORDS: exposed aggregate surfaces, composite pavements, tire
pavement interaction noise, sound absorption, ride quality
Manuscript received October 17, 2011; accepted for publication May 3, 2012; published online
October 2012.
1
Minnesota Dept. of Transportation, Office of Materials and Road Research, 1400 Gervais Ave.,
Maplewood, MN 55109.
Copyright V
19428-2959.
110
IZEVBEKHAI AND AKKARI, doi:10.1520/STP104436 111
Introduction
The functional requirements of pavements often dictate the time of repair,

rehabilitation, or replacement, whether or not the structural requirements are
satisfied. The process of selecting a surface texturing type consequently
becomes indispensible, as it ultimately influences the friction, ride quality, and
acoustic properties of the pavement. At the MnROAD pavement testing facil-
ity, 18 different surface texture types, including 7 different concrete texture
types, have been constructed for research and implementation. Exposed aggre-
gate concrete (EAC) surfaces are commonly used as the surface layer in com-
posite pavements. EAC utilizes very high quality aggregate. To expose the
aggregate to the surface, one applies a set retardant to the wet concrete surface
and subsequently brushes off the surface mortar. Studies of different EACs in
Europe found that the aggregate size can vary from less than 1 mm to up to 22
mm, but these studies almost always found the best performance in pavements
utilizing gap graded aggregate with a maximum size of less than 8 mm [1].
Exposed aggregate pavements are occasionally used in Europe for their noise
reduction capability and increased durability and friction relative to conven-
tional pavement surfacing techniques. However, although various international
scan tours [1,2] reported that exposed aggregate in Europe provided quiet pave-
ment surfaces, earlier attempts made in the United States at exposed aggregate
finishes were not very quiet and did not provide sufficient skid resistance.
In order to address this concern, a test section finished with exposed aggre-
gate was built in 2010 at the MnROAD facility and was subjected to high vol-
ume traffic, as well as snow and ice operations. In order to achieve the
requirements of an exposed aggregate finish, a high quality concrete mix was
desired. Therefore, to minimize the overall cost, it was found adequate to place
the high quality mix layer on a recycled aggregate concrete layer. Skid resist-
ance, acoustic properties, and ride quality were measured before and after tex-
turing, followed by seasonal monitoring for comparison with pre-existing
transverse tined textures. The performance of the exposed aggregate was also
compared to that of a contiguous bituminous cell, contiguous diamond ground
lanes, and a neighboring transverse tined texture.
Europe and Canada have been experimenting with the construction of new
pavements as composite pavement systems [2], a technique that can provide
several benefits over single layer construction. These new composite pavement
systems consist of two distinct layers. New composite pavement systems can
be very economically effective because of the use of low cost materials, such
as recycled or moderate quality aggregate, in the lower layer. The surface layer
can then provide better performance characteristics, such as low noise charac-
teristics and improved friction and ride characteristics, along with minimal
maintenance, with the utilization of new, innovative pavement techniques. The
lower layer of these composite systems is generally much thicker than the
upper layer, so costs are further reduced because less of the more expensive
and durable materials are required. Composite pavements can be made using
asphalt concrete over new concrete, along with construction of two new con-
crete layers. In summary, composite pavement designs have the ability to pro-
vide long-lasting, economical, and environmentally sustainable transportation.
Many transportation departments have become interested in recycling and
reuse as means to reduce pavement costs and achieve better surface character-
istics and durability, and they have consequently started experimenting with
two lift construction and exposed aggregate surfaces. Despite this recent effort,
the performance of composite pavements and EAC is not as clearly understood.
More performance data are needed for effective design, along with better con-
struction specifications and guidelines.
Project Overview and Design
Overview
Cells 70, 71, and 72 at the MnROAD research facility were constructed as part
of the SHRP 2-R21 project to evaluate two different types of composite pave-
ments: an asphalt layer over a Portland cement concrete (PCC) layer, and a
high surface quality PCC layer over a lower quality PCC layer. An exposed ag-
gregate surface was used in cell 72. MnROAD consists of two distinct seg-
ments of roadway: the Mainline, a segment of Interstate 94, and the Low
Volume Road. There are a total of 56 test cells between the Mainline and the
Low Volume Road, each with a distinct pavement type and design. The three
composite pavement test cells discussed in this paper are all located on the
Mainline. This paper serves the following main objectives:
Discuss important material parameters of pavement layers and exposed
aggregate surfaces and identify their behavior and performance charac-

teristics in relation to durability, ride, and noise.
Determine recommendations for construction specifications, techniques,
and other quality management procedures in relation to composite pave-

ments and exposed aggregate surfaces [3].
Cell Design
The top layer of cell 70 is high quality hot mix asphalt (HMA). An EAC mix is
used as the top layer of cells 71 and 72. The bottom layer of cells 70 and 71 is
concrete utilizing 50 % recycled concrete aggregate (RAC). The bottom layer
of cell 72 consists of an economical low cost mix (CHP) that utilizes a
relaxed aggregate gradation. Figure 1 shows a cross section of the three com-
posite cells along with the aggregate base designations. All cells were con-
structed with 15 ft panels. Cells 71 and 72 have 1 in. dowels throughout both
lanes. The material properties and cost information are provided in Table 1.
FIG. 1Cell composition.
Figure 2 illustrates the unique gradation of the exposed aggregate mixture

designed to produce a surface texture with good friction characteristics and
reduced noise generation.
The EAC mix is an extremely rich mixture with more than 600 lb of
cement and 725 lb of total cementation material per cubic yard. However, the
low cost and recycled aggregate mixes contain a total cementitious content of
only 600 lb/yd3. The difference between the two mixes is the fly ash content.
The low cost mix uses a high class F fly ash content of 360 lb/yd3, whereas the
recycled mix only has 240 lb/yd3. The sources of fine aggregate, cement, and
fly ash are the same for all three mixes.
Although both cells 71 and 72 were constructed using the high quality
exposed aggregate mix in the top layer, three different surface treatments were
used to evaluate a broader range of pavement types. Cell 71 was finished with
diamond grinding, with an innovative diamond grind in the driving lane and
the traditional diamond grind in the passing lane. Cell 72 was treated using
common practices for exposed aggregate finishes involving the application of a
set retarder to the surface. The set retarder used was a water based compound.
This chemical provides etch retention and consistency in temperatures of up to
130 F for up to 16 h [4]. The retardant is both odorless and non-flammable. It
does not require the use of plastic covers, which minimizes the required
TABLE 1Mix designs.
Recycled Low Cost Exposed Aggregate

Mix designation CHP RAC EAC
Pavement location Lower Lower Upper wear
Fly ash, % 60 40 15
Aggregate Relaxed gradation 50% CA from RCA 98 % passing 3/8 in.
Quantity 255.0 yd3 (4323 yd2) 466.0 yd3 (4323 yd2) 255.0 yd3 (3059 yd2)
Cost $140/yd3 or $30/yd2 $135/yd3 or $30/yd2 $175/yd3 or $48/yd2
Water, lb/yd3 234 172 283
Cement, lb/yd3 360 (Type I/II) (SG 3.15) 240 (Type I/II) 616 (Type I/II)
(SG 3.15) (SG 3.15)
Fly ash, lb/yd3 240 (Class C/F) (SG 2.5) 360 (Class C/F) 109 (Class C/F)
(SG 2.5) (SG 2.5)
Water/cement 0.39 0.29 0.39
C. sand (OD lb/yd3) 1200 (SG 2.63) 1263 (SG 2.63) 843 (SG 2.63)
(Abs 0.90 %) (Abs 0.90 %) (Abs 0.90 %)
CA1 (OD lb/yd3) 825 (size 1.5 in., 787 (size 1.5 in., 1133 (size 0.5 in.
SG 2.75, Abs 0.90 %) SG 2.75, W. chips, SG 2.72,
Abs 0.90 %) Abs 0.40 %)
CA2 (OD lb/yd3) 920 (recycled, SG 2.49, 1102 (size 0.75 843(size 0.375 in.
Abs 2.93 %) in., SG 2.69, W. chips, SG 2.72,
Abs 1.30 %) Abs 0.40 %)
Max slump, in. 3 3 3
Percent air 7 7 7
Multi-air 25, oz/yd3 2 to 15 2 to 15 2 to 15
Adm 686, oz/cwt 1 to 7 1 to 5 1 to 5
Admixture 3, oz/cwt 0 to 30 (non-chloride 0 to 30 (non- 0 to 5 oz/cwt admix-
accelerator) chloride accelerator) ture as needed for
slump retentions
application labor. This compound allows the unhardened surface mortar to be

removed via brushing, revealing the aggregate to the surface.
Construction Sequence
Mainline Pre-construction and Demo Slab Paving

Construction of the three composite cells at MnROAD began in April 2010. In-
place bituminous shoulders were milled to be used as reclaimed asphalt pave-
ment by C.S. McCrossan, Inc (CSM). The existing concrete pavement in cells
70, 71, and 72 was broken, removed, and hauled back to CSM to be crushed
and used in the new first layer of composite cells 70 and 71.
Before the new composite cells were paved, a practice demo slab was con-
structed in the stockpile area of the MnROAD test facility. The bottom layer
FIG. 2Aggregate gradations.
for the first 100 ft of the demo slab was paved using the low cost PCC mix, and
the remaining 100 ft was done using the recycled concrete mix. It was placed
on the existing granular grade, with 15 ft panels and 1 in. dowels throughout.
Crews experienced issues with the workability and consistency of the recycled
PCC mix. The slowed delivery was expected to influence the degree of bonding
between the two layers. However, the issues seemed to be resolved near the
end of the slab. The contractor initially experimented with the low cost mix.
The first few batches were sent away for being too wet, but adjustments were
made in subsequent batches so as to achieve the desired consistency. The top
layer of the entire slab was paved using the exposed aggregate mix. This mix
seemed to achieve the necessary properties for placement and finishing.
Figure 3 shows the construction of the demo slab.
Mainline Paving Construction and Texturing

Paving of the composite cells on the Mainline of MnROAD began on May 5,
2010, with the bottom layer of cell 70. The RAC was placed on May 5. The
first few batches of RAC had some problems with consistency and air content
FIG. 3Demo slab.

TABLE 2Fresh concrete testing.
Mix Time Slump, in. Air Content, %

Cell 70
RAC 7:30 a.m. 0.75 7
RAC 7:50 a.m. 1.5 10.8
RAC 8:15 a.m. ... 10.6
RAC 9:07 a.m. 3.25 8.5
RAC 9:43 a.m. 1.5 6.8
RAC 10:55 a.m. 1.75 6.8
Cell 71
RAC 8:10 a.m. 1.25 6.6
EAC 9:10 a.m. 2.25 5
EAC 9:20 a.m. 2.25 6.2
Cell 72
CHP 8:23 a.m. 2.75 7.2
CHP 9:33 a.m. 2.5 6.4
EAC 10:14 a.m. 3.25 5.7
EAC 10:55 a.m. 2 9
EAC 11:12 a.m. 2.5 7.5
compared the RAC used in the demo slab, which had a more desirable work-
ability and a slump near 1.5 to 1.75 in., allowing foot traffic after only 45 min.
Table 2 shows the results from fresh concrete testing of all three cells.
Following the placement of the bottom layer of cell 70, paving of cell 71
began. There was a shortage of RAC due to a 40 % loss of the old PCC during
washing and the rejection of two loads of RAC for high slump. Consequently,
cell 71 reached only 266 ft in length. The time between the two lifts varied
between approximately 30 min and 1 h while paving. The RAC achieved better
slump and consistency than the demo slab.
Cell 72 was constructed after cell 71. It was 681 ft long, accounting for the
208 ft that were not paved in cell 71. Multiple obstacles were encountered
throughout the day during paving. Three loads of concrete were rejected, two
by the contractor and one by the Minnesota Department of Transportation (Mn/
DOT). Similar to cell 71, the time between the paving of the top and bottom
lifts varied from 30 min to over 1 h as paving progressed. It began raining 15
min after the last load had arrived. Figure 4 shows the construction of cell 72.
The surface was treated with a water-based top surface retarder intended for
EAC. After the cell was brushed to expose the aggregate surface, the slab was
covered with poly. Because of the complications from rain and low tempera-
ture, there were some difficulties in the brushing of the surface.
On the morning of the day after the paving of cell 72, the surfaces of the
remaining sections were brushed. This is shown in Fig. 5. This effort was much
FIG. 4MnROAD Mainline cell 72 construction.
more successful at achieving the desired texture than that of the previous day.
The poly was removed from the cell. It rained on site after paving had been
completed.
Later, the shoulders were prepared for a class 2 aggregate, the bituminous
tack coat was placed on the concrete surface, and the paving of the HMA layer
of cell 70 was completed. This bituminous paving was done in two lifts. Over
the next week, the second lift of shoulders was placed, joints were sawed and
sealed in the HMA in cell 70, and the aggregate shoulders were placed and
tack sealed. Thereafter, the passing lane in cell 71 was ground using the con-
ventional diamond grind. The innovative diamond grind of the driving lane in
FIG. 5MnROAD Mainline cell 72 surface treatment.

cell 72 was completed on May 27. The final sweeping of the project was done
on Friday, May 28.
Cells 70, 71, and 72 on the Mainline were officially opened to traffic the
week after paving had been completed and initial testing and evaluation were
completed. Subsequently, periodic lane closures were used to monitor the per-
formance of the composite cells. The results of this evaluation are discussed in
the next section.
Time Lag between Lifts
The time between paving the first lift and the second lift of wet-on-wet com-
posite pavements is an important factor influencing the bond between the two
layers. Although data are not available for the actual time lag between the two
pavers during the construction of cells 71 and 72, the truck arrival time taken
from the contractors batch tickets can be used as a reasonable estimate for the
time when paving began. Each truck of bottom lift and top lift concrete deliv-
ered approximately 8 yd3 of material. Because each cell is 27 ft wide (a 14 ft
driving lane and 13 ft passing lane), each load can pave approximately 16 lon-
gitudinal feet of bottom lift concrete and 32 longitudinal feet of top lift con-
crete. Using this information, the time lags between lifts have been estimated
and are plotted in Figs. 6(a) and 6(b). The rejected truck loads were not
included in the calculations for the amount of material delivered.
These plots clearly illustrate the differences in lag time between the two
cells and within the cells themselves. Cell 72, however, does seem to generally
have a much longer delay in paving between the two lifts than cell 71. This
might be because construction of the top lift of cell 71 began much sooner after
the bottom lift than in cell 72. However, paving of the top lift in cell 71 seemed
to slow down as time went on, with the major drop in lag time being due to the
rejection of two loads of lower lift concrete. The lag between lifts in 72 seems
to be more erratic, with more frequent jumps and dips. Figure 6(b) shows the
progression of each lift separately in order to illustrate which layer was experi-
encing delays. A larger gap between the two lines corresponds to a longer lag
between paving of the two lifts.
Early Performance Evaluation

This section provides the test methods and results of the early performance
assessment of cell 70, cell 71, and cell 72. Strength, ride, noise, and surface
characteristics are discussed in this section.
Compressive, Flexural, and Bond Strength

Compressive and flexural strength specimens were made at the construction
site at the time of paving using the three different concrete mixes used in cells
FIG. 6Time lag between lifts: (a) cell 71 and 72 comparison; (b) cell 72 detail.
TABLE 3Compressive, flexural, and bond strength.

Compressive Strength
7 Day 14 Day 28 Day

EAC 5044 5315 5601
RCA 3599 4117 4509
CHP 3773 4364 5003
Flexural Strength
7 Day 28 Day
EAC 739 846
RCA 578 658
CHP 468 575
EAC/RAC 795 910
CHP/EAC 860
Bond Strength
7 Day 14 Day
EAC/CHP demo slab 3470 4790
EAC/RAC demo slab 3080 3720
EAC/RAC cell 71 5610 6950
EAC/CHP cell 72 5670* 6560
70, 71, and 72. Composite beams with two different mix layers were made
using different combinations of the three mixes. The specimens were trans-
ported back to the Mn/DOT Office of Materials and Road Research after initial
curing. They were tested by the concrete laboratory staff using ASTM C39 [5]
and ASTM C78 [6]. The results are shown in Table 3.
As expected, the EAC achieved higher flexural and compressive strength
than the low cost mix or the recycled mix. This difference in strength illustrates
the change in quality between the two layers of the composites. Interestingly,
however, the two composites achieved slightly higher strength than the fully
exposed aggregate concrete beams. When the composites were tested, they
were placed in the third point bending machine such that the EAC layer was in
tension and the lower quality (RAC or CHP) mix was in compression. Because
of this load orientation, the measured flexural strength was essentially the
strength of the EAC. This explains why exposed aggregate beams and compos-
ite beams achieved comparable flexural strengths. In the future, composite
beams should be tested in a multitude of different orientations, with the bond
plain both vertical and horizontal, in order to avoid this phenomenon.
The bond strength between the two layers of the composite pavements is
another very important property that might influence the long-term
TABLE 4Performance data.
Cell 70 Cell 70 Cell 71 Cell 71 Cell 72 Cell 72

Driving Passing Driving Passing Driving Passing
Property Date Lane Lane Lane Lane Lane Lane
Surface type HMA HMA Innovative Traditional EAC EAC
Diamond Grind Diamond Grind
Mean profile June 1, 2010 0.20 0.40 1.11 0.88 0.72 0.83
depth, mm Oct. 20, 2010 0.25 0.44 0.99 0.83 0.78 0.85
Friction num- Sept. 20, 2010 52.5 53.8 44.9 49.2 44.2 48.8
ber (ribbed)
Friction num- Sept. 20, 2010 28.0 41.2 47.9 47.9 39.3 40.8
ber (smooth)
IRI, m/km June 7, 2010 1.06 1.11 2.39 0.79 1.79 1.56
Sept. 21, 2010 1.32 1.40 2.51 0.97 1.85 1.71
Nov. 16, 2010 1.51 1.67 2.50 1.10 1.92 1.83
SA at 1000 June 7, 2010 0.053 0.151 0.048 0.010 0.031 0.034
Hz Oct. 18, 2010 0.049 0.081 0.064 0.051 0.036 0.039
OBSI, dBA July 28, 2010 100.4 100.7 96.9 100.2 101.6 101.6
Sept. 17, 2010 101.1 101.6 98.8 101.5 103.0 103.0
Nov. 17, 2010 103.0 103.3 100.9 103.4 104.5 104.4
performance. Instead, slant-shear cylindrical specimens were made to test the

bond strength in accordance with ASTM C882 [7]. The specimens were made
at the paving site by bonding two layers of concrete, EAC over CHP or EAC
over RAC, at an angled plane in a cylinder. Table 3 shows the measured bond
strengths of EAC over RAC from cell 71 and EAC over CHP from cell 72.
Specimens were also tested using the concrete from the demo slab.
All of the specimens tested achieved a bond strength well over 1000 psi,
the minimum required in order for a bond between concretes to be considered
adequate in pavement applications. It is interesting to note the reduced bond
strength in the specimens made from the demo slab concrete. Also, the bond
between EAC and CHP seems to be stronger than that between EAC and RAC
in the demo slab concrete, but the opposite trend was observed in the speci-
mens made from cells 71 and 72.
Performance Data
Table 4 provides the performance data collected from the first two years of
service for cells 70, 71, and 72. The resulting data for mean profile depth, fric-
tion number, international roughness index (IRI), sound absorption (SA), and
on board sound intensity (OBSI) are discussed in the subsequent sections.
On Board Sound IntensityThe OBSI test measures the noise generated

from the tirepavement interaction. One advantage of using the OBSI method
to measure sound generation is that it allows the noise generated by the

pavementtire interaction to be isolated from noise from other sources, such as
engine noise and the surrounding landscape noise, making it favorable over the
traditional Statistical Pass By Method. OBSI testing is done according to the
AASHTO TP 76-08 (11) procedure [8]. The process analyzes data recorded
with microphones located close to the tirepavement contact. This setup is
shown in Fig. 9. The dominant noise generation source becomes the tire
pavement interaction when cars travel at freeway speeds. Therefore, the test is
performed with a vehicle moving at 60 mph over a 440 ft stretch of pavement
in order to adequately capture the desired noise source.
The innovative diamond ground EAC in cell 71 had the lowest initial
OBSI. The traditional diamond grind had an OBSI similar to that of the HMA.
More significantly, the exposed aggregate finish consistently had the highest
OBSI of the four surfaces compared. Pavements are generally considered quiet
when they achieve an OBSI of less than 100 dBA; as a result, the data suggest that
the diamond grind in cell 71 is the only composite pavement to be considered
quiet. In a survey of EAC pavements in Europe conducted by the National Con-
crete Pavement Technology Center, OBSI values were found to range from 101 to
106 dBA, which is similar to the results obtained for cell 72 [9]. The 1/3 octave
sound intensity spectrums used to calculate the OBSI values are show in Figs. 7(a)
and 7(b). All cells show a similar shape in their sound intensity spectrums.
Sound AbsorptionThe SA coefficient is the ratio of the absorbed sound

energy to the transmitted sound energy sent to the pavement surface. The ratio
is measured when a white noise with a frequency ranging from 315 to 1800 Hz
is projected into the pavement within an impedance tube placed normal to the
pavement surface. A set a sensitive microphones then captures the incident and
reflective waves. Software then windows the reflected waves and converts the
data to the third octave SA coefficient at 315, 400, 500, 750, 1000, 1250, and
1650 Hz. The setup for measuring SA using a BSWA 425 device is shown in
Fig. 9.
SA coefficients range between 1 and 0; a value of 1 would mean that all of
the sound is being absorbed. Figures 8(a) and 8(b) show the measured SA coef-
ficients. The negative SA coefficients in the plots imply that the microphones
are measuring more sound than is being emitted from the noise source. This
suggests that there was contamination during testing and the tube-base com-
bination did not adequately block sound from outside noise sources. To enable
a more useful comparison between the SA in different cells, the absorption
coefficients at 1000 Hz are given in Table 4. The 1000 Hz frequency is chosen
because it is commonly considered the frequency at which the tirepavement
interaction noise is the highest [10].
The HMA shows the highest SA of all the composite cells. The diamond
grind in cell 71 generally has higher absorption than the exposed aggregate in
FIG. 71/3 octave sound intensity spectrum: (a) cell 72 on different dates; (b)
Nov. 17, 2010, in different cells.
FIG. 8SA coefficient spectrum: (a) June 7, 2010 and (b) Oct. 17, 2010.
cell 72. As expected, the innovative diamond grind has higher SA than the tra-
ditional diamond grind.
FrictionThe standard method for testing friction at MnROAD utilizes

the KJ Law Friction Trailer to perform skid testing of the pavement surface.
Friction testing is done in accordance with the following three ASTM stand-
ards: ASTM E274 [11], ASTM E501 [12], and ASTM E524 [13]. Both ribbed
and smooth tires are used because they each measure adhesion and hysteresis
differently. Because the ribbed tire removes the water from the surface more
efficiently than the smooth tire, it generally examines the pavement friction on
the micro-texture portion. The smooth tire, however, is more affected by the
macro-texture. If the macro-texture does not adequately drain water on the
pavement surface, the smooth tire will hydroplane, resulting in lower friction
values. The locked-wheel skid trailer and truck setup is shown in Fig. 9.
The test generates friction numbers ranging from 0 to 100, with higher
numbers indicating higher friction. A pavement with a friction number of 25
from a smooth tire is considered a safe pavement with adequate skid resistance,
whereas a pavement with a friction number of less than 15 would require reha-
bilitation in order to achieve sufficient skid resistance [14].
Table 4 shows that all three test cells achieved adequate skid resistance for
driver safety. The HMA in cell 70 achieved the highest friction number in test-
ing with the ribbed tire, but it achieved the lowest friction number in testing
with the smooth tire. The diamond grind in cell 71 consistently achieved better
friction that the exposed aggregate in cell 72. However, this difference is sig-
nificant only in tests done with the smooth tire. When testing with the ribbed
tire, the diamond grind and exposed aggregate achieved comparable results.
TextureThe circular track meter (CTM) shown in Fig. 9 is a laser-

equipped device that scans the surface of a pavement in accordance with
ASTM E2157 [16]. The CTM is equipped with a charged coupled device laser
displacement sensor that sweeps the pavement surface in a circle 11.2 in. in di-
ameter and 35 in. in circumference. The displacement sensor for this instru-
ment is mounted on an arm that rotates 3 in. (80 mm) above the surface at a
tangential velocity of 6 m/min. The data are segmented into eight 111.5 mm
arcs of 128 samples each. From each segment, the computer software computes
the mean profile depth (MPD), the root-mean-square texture depth (RMS) of
each segment, and the average of all eight segments. A plot of the eight seg-
ments of MPD is also produced. This plot is difficult to decipher unless the
sweeping action of the laser in each sector is visualized. Effectively, only two
of the eight segments mimic the texture. The MPD generated from the CTM on
two different occasions is provided in Table 4.
The diamond grind in cell 71 and the exposed aggregate in cell 72 have
significantly higher MPDs than the HMA in cell 70. There is a consistent
FIG. 9Test equipment.

increase in MPD from the driving lane to the passing lane for cells 70 and 72,
which suggests that traffic loading can influence pavement texture. The innova-
tive diamond grind in cell 71 was consistently higher than both the traditional
grind and the EAC. There was no trend in the change in MPD from summer to
winter months.
International Roughness IndexThe IRI is a standard measure of ride

quality. IRI is mathematical property of a two-dimensional road profile. It is
the sum of the deviations from an assumed plane surface based on the vertical
acceleration of the quarter car in response to the surface profile. The IRI algo-
rithm is therefore subject to the resonant frequency of the quarter car and pro-
vides different multipliers for different frequencies to mimic the human
comfort level. Mn/DOT measures the IRI with a lightweight inertial surface an-
alyzer (LISA). The LISA is a profile device that measures the amount of verti-
cal rise over a horizontal distance. This is done with two separate laser sources
on the side of the vehicle. One laser takes continuous profile measurements
over a 4 in. path while the other measures three discrete profiles across the 4 in.
path. The raw data are then used to calculate the IRI, with a higher IRI corre-
sponding to rougher pavement.
The Federal Highway Administration (FHWA) roughness categories desig-
nate pavements with IRI values of less than 1.5 as being in good condition,
with pavements with an IRI of less than 2.6 being considered unacceptable.
All test cells did achieve acceptable IRIs, but only the HMA and traditional
grind were categorized as good pavement. The exposed aggregate had higher
IRI values than all surfaces except for the innovative diamond grind.
Analysis of Texture Orientation of Surface Configuration

One of the tire pavement noise explanatory variables (8) is the texture orienta-
tion or spikiness of the surface. It is defined as the skewness of a profile, a pa-
rameter that describes the shape of the amplitude distribution function (ADF)
and not the shape of the configuration. Skewness is a simple measure of the
asymmetry of the ADF, or, equivalently, it measures the symmetry of the ADF
of a profile about its mean line. The universal definition of the skewness (SKU)
of a profile is the quotient of the mean cube value of the ordinate values Y(x)
and the cube of the root mean square value of the ordinate values Y(x) through
a sampled length L.
L
1 1
SKU Y 3 xdx (1)
Rq3 l 0
where
L
1
Rq Y 2 xdx (2)
l 0
However, this equation is discretized to accommodate statistical texture ordi-

nate data that are not necessarily continuous. Thus the simplified form
becomes
N 3
R Yi Y
i1
SKU (3)
N 1s3
where:
Y 0 at the bottom of the groove,
Y mean,
s standard deviation, and
N number of data points.
It must be noted that in Eq 3, the SKU of the probability density function is
indicated, not the SKU of the profile asperity ordinates. Typically, using this
method, conventional grinds are mildly positively oriented, innovative grinds
are negatively oriented, and transverse tined are negatively oriented. A nega-
tive texture is clearly less than 0 SKU (<0.1), and a neutral texture has an
SKU between 0.1 and 0.1, whereas in positive textures the SKU is typically
>0.1. Ideally, drag textures are sometimes neutral textures, but because of their
proximity to the sine wave, the accuracy of the discrete method of slices might
not result in zero SKU. As a consequence of the above expression, a reliable
method of texture spikiness quantification is as follows: Sign SKU of
ADF Texture Spikiness.
The exposed aggregate was expected to have negative texture; however, it
turned out to have positive texture, which partly explains its noisiness. Izevbe-
khai [15] found texture orientation to be one of the most powerful predictors of
tire pavement noise, surpassed only by texture direction. Figure 10(a) shows a
CTM scan from cell 72.
Texture Characteristics and Acoustics over Time

Figures 10(b) and 10(c) plot the MPD and OBSI trend from the time of con-
struction to the most current measurements, and Table 5 provides the raw data.
The MPD increased with time in the driving lane, where, it can be hypothe-
sized, the higher traffic volume resulted in more paste loss, which in turn
resulted in deeper inter-asperity features. In comparison, there was no corre-
sponding increase in the texture orientation as the SKU increased to a neutral
texture from the first test to the second and fell at the third test back to the
FIG. 10(a) CTM cell 72. (b) MPD driving. (c) MPD passing.
TABLE 5Texture and noise performance over time.
72x1RR (Driving) 72x1LL (Passing)
June 1, 2010 Oct. 1, 2010 June 1, 2011 June 1, 2010 Oct. 1, 2010 June 1, 2011
MPD, mm 0.36 0.49 0.54 0.45 0.4 0.37
RMS 0.16 0.24 0.26 0.24 0.21 0.18
Skewness 0.24 0.01 0.32 0.33 0.06 0.62
Cumulative ESAL 13 301 500 13 523 900 14 367 100 3 369 200 3 431 900 3 640 900
OBSI 101.3 102.7 103.0 101.7 102.9 103.2
original negative texture measured on the same spot in three consecutive tests.
It changed from 0.24 in the summer of 2010 to 0.01 in the fall of 2010, and
to 0.32 in the summer of 2011. Results for the driving lane were similar to
the observations made in the passing lane, although the equivalent single axle
loads (ESALs) in the former were four times that of the latter. The passing lane
swung from a more negative texture (0.33) to a more positive texture (0.06)
from summer to fall of 2010. It fell to a very negative value of 0.62 in the
summer of 2011. There was an increase in OBSI through the one-year period,
but the increase appeared to be more rapid between the first two readings in
both lanes than between the second and third readings. In Table 5, the OBSI-
101 value has been used in order to enable the detection of small changes.
A continuous drop in MPD was observed in the passing lane with lesser
traffic, in contrast to the driving lanes, where a continuous increase in MPD
was observed. It appears based on this observation that the traffic effect at the
early stage of the service period results not in the polishing of aggregate but
instead in the removal of more paste, thus increasing the MPD while increasing
the texture spikiness up to a point beyond which further paste removal is asso-
ciated with further spike removal and the net effect is negative texture. Never-
theless, one-year performance shows to some degree some non-descript
features that might become more ordered and defined after many years of mon-
itoring. It must be noted, however, that texture depth changes do not necessar-
ily result in increased noise, nor are they necessarily associated with increased
MPD in the EAC surface observed.
Conclusion
This paper summarizes the construction and early performance assessment of
three composite test cells at the MnROAD (cell 70, HMA over an RAC; cell
71, diamond grind concrete over RAC; and cell 72, EAC over a low cost con-
crete) and emphasizes the performance of the exposed aggregate finish. The
construction of cells 70, 71, and 72 was part of a composite pavement project
of the Strategic Highway Research Program. The following conclusions were
made with regard to the different composites and surfaces tested and the
exposed aggregate surface performance in relation to other contiguous test cells
with familiar surface types.
Early performance assessment of the three test cells suggests that the
EAC surface did not provide significant noise reduction, as it had a

higher OBSI than either the HMA or the traditional diamond grind
surfaces tested.
Innovative diamond grinding of composite pavements might be beneficial
for noise reduction, as it resulted in a lower OBSI than the HMA, EAC,
or traditional grind surfaces, and it also had greater SA than the EAC.
Exposed aggregate surfacing provided more than adequate friction for
driver safety but did not show any advantage over typical HMA or dia-
mond ground surfaces.
The exposed aggregate surface tested has a texture (or MPD) similar to
that of traditional diamond ground surfaces, but it might have reduced

ride quality as laser induced IRI values due to isotropy.
The acoustic performance is highly dependent on the aggregate content.
Consequently, it is recommended to use crushed rock to reduce the texture ori-
entation to a more likely negative value, blended with rounded aggregate to
enhance quietness. However, the use of 100 % rounded aggregates in a gap
gradation is not encouraged, as it might compromise friction.
References
[1] Tompkins, D., Khazanovich, L., and Darter, M. I., 2008 Survey of Euro-
pean Composite Pavements, Transportation Research Board, Washington,
D.C., 2010.
[2] Akkari, A. and Izevbekhai, B. I., Composite Pavements at MnROAD,
Minnesota Department of Transportation, St. Paul, MN, 2010.
[3] SHRP 2-R21, 2011, Composite Pavement Systems, Transportation
Research Board, Washington, D.C., http://apps.trb.org/cmsfeed/TRBNet
ProjectDisplay.asp?ProjectID=2173 (Last accessed October 7, 2011).
[4] Products and Innovations, 2011, MBT Reveal, BASF Construction
Chemicals, Cleveland, OH, http://www.basf-admixtures.com/en/products/
retarders_release/mbt_reveal/Pages/default.aspx (Last accessed October
10, 2011).
[5] ASTM C39, 2011, Standard Test Method for Compressive Strength of
Cylindrical Concrete Specimens, Annual Book of ASTM Standards, Vol.
04.02, ASTM International, West Conshohocken, PA, pp. 2329.
[6] ASTM 257, 2011, Standard Test Method for Flexural Strength of Con-
crete Using a Simple Beam with Third-point Loading, Annual Book of

PA, pp. 4144.
[7] ASTM C882, 2011, Standard Test Method for Bond Strength of Epoxy-
Resin Systems Used with Concrete by Slant Shear, Annual Book of
PA, pp. 487490.
[8] AASHTO TP 76-08, Standard Method of Test for Measurement of Tire/
Pavement Noise Using the On-Board Sound Intensity (OBSI) Method,
American Association of State and Highway Transportation Officials,
Washington, D.C., August 1, 2008, pp. 111.
[9] Rasmussen, R., 2011, Hearing is Believing, General Blog, Concrete Pave-
ment Surface Characteristics Program, http://www.surfacecharacteristics.
com/index.php?q=blog (Last accessed October 10, 2011).
[10] Ongel, A., Harvey, J. T., Kohler, E., Lu, Q., and Steven, B. D., Investiga-
tion of Noise, Durability, Permeability, and Friction Performance Trends
for Asphaltic Pavement Surface Types: First- and Second-Year Results,
University of California Pavement Research Center, Berkley, CA, 2008.
[11] ASTM E274, 2012, Standard Test Method for Skid Resistance of Paved
Surfaces, Annual Book of ASTM Standards, Vol. 04.03, ASTM Interna-
tional, West Conshohocken, PA, pp. 10831088.
[12] ASTM E501, 2012, Skid Testing Using a Standard Ribbed Tire, Annual
Book of ASTM Standards, Vol. 04.03, ASTM International, West Consho-
hocken, PA, pp. 10981102.
[13] ASTM E524, 2012, Skid Testing Using a Smooth Tire, Annual Book of
PA, pp. 11031105.
[14] Eller, A., Drainable Pavements at MnROAD: Pervious Concrete and Po-
rous Concrete Overlay Cells 39, 85, and 89, Minnesota Department of
Transportation, St. Paul, MN, 2010.
[15] Izevbekhai, B. I., 2012, Tire Pavement Interaction Noise of Concrete
Pavements, Ph.D. dissertation, University of Minnesota, Minneapolis,
MN.
[16] ASTM E2157, 2012, Standard Test Method for Measuring Pavement
Macrotexture Properties Using the Circular Track Meter, Annual Book
of ASTM Standards, Vol. 04.03, ASTM International, West Consho-
hocken, PA, pp. 13261330.
STP 1555, 2012
DOI:10.1520/STP104524
John Hencken,1 Edwin H. Haas,2 and Thomas Bennert3
Analyzing Spectral Repeatability of Different

Asphalt Pavements Using the OBSI Method in
New Jersey and Massachusetts
REFERENCE: Hencken, John, Haas, Edwin H., and Bennert, Thomas,

Analyzing Spectral Repeatability of Different Asphalt Pavements Using the
OBSI Method in New Jersey and Massachusetts, Pavement Performance:
Current Trends, Advances, and Challenges on December 5, 2011 in Tampa,
FL; STP 1555, B. Choubane, Editor, pp. 135149, doi:10.1520/STP104524,
ASTM International, West Conshohocken, PA 2012.
ABSTRACT: For the past two years, the Center for Advanced Infrastructure
and Transportation has used the On-Board Sound Intensity Method (OBSI)
to examine tire/pavement noise on in-service pavements throughout NJ and
the northeast United States. The research initiative originated as a result of
the NJ Department of Transportations (NJDOTs) pavement resource pro-
gram, which monitors the conditions of NJ roadways and provides
engineering-based research about mechanistic pavement design, pavement
material characterization, and pavement evaluation. NJDOT is interested in a
long-term evaluation of the effects of the age and condition of asphalt pave-
ment on the noise generated at the tire-pavement interface. Of the vast num-
ber of asphalt pavement design types found on interstates in NJ, Superpave-
based dense graded asphalt (DGA) mixes and open-graded friction course
(OGFC) mixes constitute the majority. Throughout the analysis of the OBSI
levels measured, patterns that would be beneficial to the industry were illus-
trated in the one-third octave band profile for each pavement type. This paper
provides a brief background of the OBSI method utilized in NJ and MA, an
explanation of the analysis utilized to evaluate these pavements, and an
Manuscript received November 8, 2011; accepted for publication April 12, 2012; published
1
Research Engineer, Rutgers Univ., Center for Advanced Infrastructure and Transportation, 100
Brett Rd., Piscataway, NJ 08854 (Corresponding author), e-mail: jhenck@gmail.com
2
Graduate Student/Research Assistant, Rutgers Univ., Dept. of Civil and Environmental
Engineering, 623 Bowser Rd., Piscataway, NJ 08854, e-mail: edhaas829@gmail.com
3
Ph.D., Senior Research Engineer, Rutgers Univ., Center for Advanced Infrastructure and
Transportation, 100 Brett Rd., Piscataway, NJ 08854, e-mail: bennert@rci.rutgers.edu
Copyright V
19428-2959.
135
explanation of the different spectral signatures attributed to DGA and OGFC

pavements.
KEYWORDS: OBSI, on-board sound intensity, tire-pavement interface
noise, sound intensity, spectral signature
Background
The mitigation of transportation noise is a major concern for departments of
transportation (DOTs) nationwide. From a responsibility standpoint, once a
noise source leaves the boundary of a property, control of the noise falls under
the jurisdiction of the DOT. In the past, mitigation by the DOT was typically
accomplished through the erection of a sound barrier. If there was more noise
or louder noise, a larger wall was used. As the construction of these walls is
prohibitively expensive, DOTs are increasingly interested in mitigating noise
at the source [13]. It is known that on highways, the controlling noise generat-
ing mechanism is the tire-pavement interface [1,3]. The NJ Department of
Transportation (NJDOT) and the NJ Turnpike Authority have received numer-
ous noise complaints from residents who are located near interstates and state
highways throughout NJ. In order to begin to understand how to mitigate noise
at the source, the first step was to gather information about the noise generated
on in-service pavements throughout the state.
Noise reduction is an important benefit of functional overlays. It is impor-
tant to consider the noise reduction capabilities of the overlay in addition to
splash and spray reduction, improved skid resistance, improved friction, and
improved smoothness [4,5]. In order to effectively analyze the noise reduction
benefits of functional overlays, the need arose to catalog pavement noise gener-
ated on in-service pavements throughout the state. Two complications are in-
herent: firstly, asphalt is viscoelastic and therefore changes with time and
temperature, and secondly, tire-pavement interface noise testing variables fluc-
tuate during testing. The method chosen in order to overcome this by the Cen-
ter for Advanced Infrastructure and Transportation and NJDOT was to monitor
pavement noise characteristics seasonally over the service life for each pave-
ment type.
Prior to this long-term pavement noise characterization, a significant
amount of research had been conducted both in NJ [6] and in CA, CO, and
elsewhere throughout the United States [79]. Portland cement concrete pave-
ments were shown to typically produce higher noise levels from the tire-
pavement interface when compared to different asphalt-based surfaces [3].
Studies looking into the effectiveness of quiet pavements have compared typi-
cal dense graded hot mix asphalt (DGA) pavements to porous and semi-porous
asphalt pavements in order to identify potential noise mitigation properties
associated with each design type [10]. Prior work on porous pavements was
conducted so as to measure the durability and effective design life of the noise
mitigating properties of open graded friction courses (OGFCs) and other
HENCKEN ET AL., doi:10.1520/STP104524 137
functional overlays [6]. In order to utilize OGFC surface treatments for noise
reduction at the tire-pavement interface, the structural properties of the mixes
need to be analyzed with respect to regionally specific needs [11]. For instance,
it has been shown that OGFC overlays should be used cautiously in areas
where the use of studded tires or snow chains during the winter season is preva-
lent, because studded tires and snow chains significantly reduced the life span
of an OGFC [12]. In 2005, NJ focused its research needs on the use of func-
tional thin lift overlays, specifically, OGFC overlays and their practicality as a
pavement wearing surface. The conclusion of this initiative was that open
graded asphalt pavements are a viable option for the NJDOT and can aid in
noise mitigation [13]. As a result of prior research, both NJDOT and the NJ
Turnpike Authority have widely adopted the use of asphalt rubber open graded
friction course (AROGFC) pavements as surface wearing courses, but they still
hold reservations regarding the longevity of the functional benefits over the
service life of these special pavements [6]. When the application of an
AROGFC is not possible because of environmental constraints, the state typi-
cally utilizes stone-mastic-asphalt pavements or standard Superpave mixes.
Each of these pavements varies according to the region in which it is being
applied in, based on different aggregate qualities and availabilities, even
throughout NJ. The different mix designs vary in terms of several factors,
including, but not limited to, binder type, binder percentage, binder stiffness,
aggregate angularity, design density, splash and spray reduction, and effective
film thickness. These factors affect the functional properties of the asphalt,
including the acoustic properties. In order to determine the effects of pavement
types throughout NJ, several test areas have been studied at the tire-pavement
interface, covering the full range of in-service pavements in NJ, utilizing the
On-Board Sound Intensity Method (OBSI).
Introduction
The objective of this paper is to review and confirm the idea that different
asphalt pavement mix designs yield distinct acoustical properties. Researchers
in CA, CO, and several other states throughout the United States have
described the one-third octave band frequency spectrum for various concrete
and asphalt pavement surfaces [79]. This paper works through a method of
analysis to decipher the difference between one in-service pavements distinct
acoustical properties and anothers. In addition, the method helps to verify that
similar mix types exhibit similar acoustical properties, and it provides pave-
ment noise results that are less variable through normalization and statistical
measures while utilizing a quick and practical approach. The pavement indus-
try will benefit from opportunities to understand how tire/pavement noise
changes for a particular pavement over a long period of time.
Testing Procedures
The testing was conducted in the right wheel path of the right lane at 60 mph 6 1
mph (96.6 km/h 6 1.6 km/h). Two technicians collected data in every instance.
The driver was required to keep the right wheels within the standard wheel path
within the lane, verify speed control throughout each test, monitor the equipment
in the right side mirror, and assess the roadway for anything that could change the
noise quality in each test (e.g., a large truck passing by). Figure 1 displays the
G.R.A.S phase-matched microphones and preamplifiers used for the leading and
trailing edge intensity probes, which were utilized to measure the tire/pavement
noise on the outside of the vehicle. A technician with a laptop began each test and
watched the test section to make sure each test began at the same point during
each run, as exemplified in Fig. 2. The equipment utilized to acquire and process
the data was a Bruel & Kjaer PULSE system. The technician running the laptop
was required to monitor the coherence and Pressure-Intensity Index (PI) spectrum
during each test in order to ensure the validity of each measurement. At a mini-
mum of one location along each different pavement material, the ambient air tem-
perature, pavement temperature, tire temperature, wind speed, barometric
pressure, and tire pressure were recorded. Microphone calibration was completed
with an acoustic calibrator, and the resultant data were recorded and validated
according to AASHTO TP 76-10 [14] at the beginning and end of each test period.
The standard frequency analysis of the measured sound intensity was performed
using one-third octave band resolution as stipulated by AASHTO TP 76-10 [14].
Each measurement was recorded over 440 ft (134.1 m) for a 5.0 s measure-
ment period. The signals were A-weighted prior to digitization, and all of the
levels reported within this paper reflect the A-weighting. The sound intensity
levels of the two probes were energy averaged for each run. The averages were
then averaged together arithmetically. A minimum of three runs were
FIG. 1Sound intensity testing apparatus.

FIG. 2Test setup from inside the vehicle.
completed for each test section analyzed in this report. AASHTO TP 76-10
[14] requires two conforming runs, but the authors preferred three to allow for
better statistical comparison. The test site selection was made on location by
the sound testing crew after they completed a dry run over the material in order
to determine appropriate sites to test. A good site explicitly had a contiguous
section of pavement with no material changes, joints, or bridge decks. Typi-
cally, one measurement was recorded every half-mile.
The analysis of the measurements taken was completed in several separate
processes using the following methods. Following the test section selection pro-
cess and the testing procedure set forth by the OBSI method referencing
AASHTO TP 76-10 [14], all test sites were averaged together in order to get a
representation for each material. Overall bar charts and one-third octave band
frequency spectrum graphs were created for each site and then averaged to repre-
sent each material. The frequency graphs show the measured sound intensity lev-
els along the one-third octave band spectrum, which is the typical frequency
band used to show sound measurements for OBSI. The different materials were
then compared in order to determine differences in the way the tire generated
noise will be perceived by a receiver for each. A-weighted measurements are uti-
lized because it is the weighting scheme that resembles the average human per-
ception of sound; the 400 Hz to 5000 Hz frequencies are depicted as per
AASHTO TP 76-10 [14].
Discussion
Data Normalization
Run-to-Run VariabilityThe first step required for each set of data is to

compile any measurements from each testing location that conform to
AASHTO TP 76-10 [14]. A minimum of two measurements that are within 0.6
dB(A) are required, but very often more are acquired. An example of this can
be seen in Fig. 3. If a 5 mile section is tested, there will be roughly 20 measure-
ment sections to compile.
The graph represents five different runs from testing that ensued in March
2010; all runs started at milepost 11.0 East on a 14 mile test section on I-78.
The ambient temperature for the day was recorded as 92.6 F. This section was
set up so as to evaluate an AROGFC that was paved in the summer of 2009.
This graph shows how small the run-to-run variability was for that milepost.
Each time the test vehicle circled around the 14 mile test section, the vehicle
was positioned well within the wheel path and the laptop technician was able
to start the measurement at the same spot, and the equipment positioning and
stability remained constant throughout.
Moderate Roadway VariabilityThe best case and most desirable sam-

pling would consist of even measurements in which no variables changed from
run to run. Unfortunately, this is not always the case. Even if the environmental
factors remain exactly the same for the entire testing period, the reality of
asphalt pavement is that it is variable over a distance. Whether this variability
is derived from slight variations in the mix from the plant, variations in com-
paction temperature, variations in compaction densities, or a range of other
FIG. 3I78 Run-to-run variability.

factors, the more sections that are tested of a single material along the same
roadway, the more variability that will be shown through the OBSI measure-
ments. For the example above, every milepost had similar run-to-run variabili-
ty, in which there was minimal to no change over the course of one testing day.
Figure 4 shows the compilation of the testing on I-78 for the same 14 mile sec-
tion depicted in Fig. 3.
With such a long section, it was expected that we would see a wide range
throughout the one-third octave band frequency levels when all of the mile posts
were compared. With a closer look at Fig. 4, though, it is apparent that a major-
ity of the measurements at each milepost are very similar, with only a few out-
liers. All of the mile posts have been averaged to create a representative
pavement profile for that AROGFC pavement. This average can be seen in
Fig. 5. Shown as one line, the shape of the curve still closely resembles that seen
in Fig. 4. Added to this average, the 1s standard deviation is graphically depicted
in order to show the variability within the average. The outliers from Fig. 4
cause the standard deviation to be higher across the entire frequency spectrum
for the average shown in Fig. 5, and the plot shows what the typical frequency
spectrum looked like on average for that AROGFC pavement on I-78 that day.
Low Roadway VariabilityAnother example shows how it is possible for a

5 mile section to exhibit almost no variation on any stand-alone section or as a
compilation of all of the sections. Figure 6 shows a compilation of each mile
marker tested on Rt. 202 in April 2010.
FIG. 4Moderate roadway variability.

FIG. 5I-78 road average.
The ambient temperature on the day of testing was recorded as 83.4 F. The
pavement tested was a 9.5 mm Superpave mix with 76-22 binder. Figure 6
shows minimal variations across the depicted one-third octave band frequency
spectrum. The average of these materials is shown in Fig. 7; it very closely
resembles what is seen in Fig. 6.
Pavement Type Comparisons
Similar PavementsOnce each pavement is compiled into an average, it is

interesting to look at how similar pavement types compare to each other in dif-
ferent sections on different roadways. The averaged plot makes it easier to
compare the same section with future measurements as well, in order to deter-
mine whether the effects of temperature, age, or some other factor changed any
noise property generated at the tire-pavement interface. Figure 8 shows three
differently aged DGA averages obtained at different times of year and with dif-
ferent recorded ambient temperatures.
The three pavements shown are from Rt. 202, I-78, and the Garden State
Parkway (GSP). The Rt. 202 section is the 9.5 mm Superpave mix represented
in Fig. 7. The tested portion of I-78 was a 3 mile section of Superpave
12.5M76 that was tested in May 2010 at a recorded ambient temperature of
60.5 F. The third pavement was the GSP Superpave 12.5H76, tested in April
2010 at a recorded ambient temperature of 62.8 F. The relationship that we
observed here is one in which each DGA has a curve that is very similar to the
FIG. 6Low roadway variability.

others. Even though the 9.5 mm mix was designed for heavy loading (greater
than 30 106 ESALs) and the Rt. 202 12.5 mm mix was designed for medium
loading (10 106 to 30 106 ESALs), their curves look the same. Generally,
the standard deviation bars overlapped for all of the pavements. This is also a
sign that the pavements, from a statistical point of view, are similar.
FIG. 7Rt. 202 road average.

FIG. 8DGA averages.

Because this relationship worked for the DGAs, three OGFC pavements
were plotted next to each other in order to determine whether they looked simi-
lar to each other. Figure 9 shows three different OGFC pavements side by side.
The first is the I-78 AROGFC average from Fig. 5. The second is an
AROGFC that was tested 1 week after it was paved on I-195 in October 2010.
FIG. 9OGFC averages.

The ambient temperature during testing on I-195 was recorded as 62.2 F. The
third pavement is an OGFC with a polymer modified binder (MOGFC) that
was tested on the GSP in June 2010. The MOGFC was paved in 2005. The am-
bient temperature recorded while testing the MOGFC on the GSP was 86.0 F.
This spectrum analysis showed us that the OGFC pavements, similar to the
DGAs, had very similar curves when they were plotted on the one-third octave
band spectrum. As with the DGA pavements, the OGFC pavements that are
shown in Fig. 9 have standard deviation bars that overlap significantly. If the
testing location or road information was not known for any of the pavements
shown here, one could assume that they were similar.
Different PavementsGiven that the three examples of different DGA

pavements shown in Fig. 8 exhibited similarly shaped one-third octave band
spectra and, separately, the three examples of different OGFC pavements
shown in Fig. 9 exhibited similarly shaped one-third octave band spectra, the
next logical step was to plot a DGA and an OGFC together in order to see how
different they looked. Figure 10 shows that the two mix designs have different
acoustical properties, as each has statistically different spectra.
The two pavements shown in Fig. 10 demonstrate an obvious visual differ-
ence when overlaid on the same chart. Because the analysis method utilized
throughout the paper thus far has been capable of differentiating between two
different mix types in NJ, verification of geographical differences in mix types
was necessary. To accomplish this, data acquired in MA were processed using
FIG. 10DGA versus OGFC.

the same method in order to compare similar pavement types paved in different
regions.
New Jersey versus Massachusetts
Example 1: Dense Graded Hot Mix AsphaltsThe first example of the ver-
ification, shown in Fig. 11, is comparing the NJ DGA from the GSP, which is
shown in Fig. 8, to a DGA tested in MA on I-495.
The MA DGA was tested in October 2010, and the ambient temperature
recorded during testing was 73.3 F. In Fig. 11, both pavements seem to have
very similar frequency curves, and the MA DGA looks very similar to the other
DGA examples discussed throughout the paper.
Example 2: Open Graded Friction CoursesFor the final example, the NJ

I-78 AROGFC shown in Fig. 5 and Fig. 9 is overlaid by an OGFC that was
tested on I-290 in MA. This example is shown in Fig. 12.
The MA I-290 OGFC in Fig. 12 was tested in October 2010, with a
recorded ambient temperature of 76.2 F. Both curves appear to be very similar,
and the MA OGFC, as a pavement type, looks very similar to the NJ OGFC
examples shown throughout the paper. This regional comparison shows that
although pavements have inherent variation, the spectral curves represented by
similar pavement types are distinctly unique to that pavement type.
FIG. 11NJ versus MA DGA comparison.

FIG. 12NJ versus MA OGFC comparison.
Summary
The OBSI has been recognized as an effective method for quickly and effec-
tively evaluating tire-pavement interface noise generated on in service pave-
ments [6]. The method is designed to standardize tire-pavement interface
measurements as much as possible [9]. Two years of data collection and analy-
sis have revealed noticeable patterns in the one-third octave band spectrum
analysis plots for different pavement types. These patterns followed other
OBSI research in CA and CO [7,8].
From a pavement materials perspective, deciphering the patterns that
emerged logically became the first step in understanding the acoustical proper-
ties of different asphalt mixes. It was found that in the one-third octave band,
all the pavements tested conform within the range of 65 dB(A) to 110 dB(A)
throughout 400 Hz to 5000 Hz; however, within that range, different pave-
ments types began to display unique signatures. The OGFC pavements, exem-
plified in Fig. 9, exhibited a curve that is most easily described by two
parabolic curves, one in the 400 Hz to 1600 Hz center frequency range and the
second in the 1600 Hz to 5000 Hz center frequency range. The maximum
within the 400 Hz to 1600 Hz range was about 95 dB(A) between the 630 Hz
and the 1000 Hz center frequency. The considerably lower maximum within
the 1600 Hz to 5000 Hz range was about 84 dB(A) around the 2000 Hz center
frequency. The lower levels in the 1600 Hz to 5000 Hz range are typically
associated with absorption resulting from the higher air void percentage
inherent in an OGFC mix type [7,8]. In contrast, the curve exhibited by the
DGA spectra could most easily be described as a trapezoid. The DGA curve
had two peaks occurring at the 1000 Hz center frequency around 94 dB(A) and
at the 2000 Hz center frequency around 92 dB(A), respectively. These signa-
tures reoccurred when different roadways were evaluated, and this helped to
validate the idea that a standard pavement type designed using a mechanistic
pavement design should yield a standard tire/pavement noise spectral signa-
ture. In order to confirm that this idea was not regionally valid only in NJ,
results recorded in MA were compared to the findings in NJ. It appears that the
patterns discovered were universal, as they held in both regions. This is useful
to know, because if noise testing is required in a particular area but the pave-
ment type is not known, it is possible to go out in the field and determine
roughly what type of pavement it is, independent of region, provided a long
enough section is measured and an appropriate average is taken. The examples
of slight differences between two similar pavement types that are displayed in
Figs. 8 and 9 stem from any number of unknown environmental factors, pave-
ment age discrepancies, or human induced error during testing. Even with these
variations, the same basic spectral signatures are represented for each, but at
slightly higher or lower decibel levels, respectively.
Because the purpose of this study was to describe a method of analysis
with which to determine the difference between one in-service pavements dis-
tinct acoustical properties and anothers, the analysis provided herein has iden-
tified, defined, and verified distinct patterns. This research will aid in
discovering and tracking changes over the duration of the long term NJDOT
pavement resource program study.
Acknowledgments
The writers would like to acknowledge the NJDOT for providing funding to
perform OBSI testing in NJ and Susan Gresavage of the NJDOT for overseeing
the work performed.
References
[1] Rasmussen, R., Bernhard, R., Sandberg, U., and Mun, E., 2007, The
Little Book of Quieter Pavement, FHWA-IF-08-004, Federal Highway
Administration, U.S. Department of Transportation, Washington, D.C.
[2] Sandberg, U., Tyre/Road NoiseMyths and Realities, Proceedings of
the 2001 International Congress and Exhibition on Noise Control Engi-
neering, Inter-Noise, The Hague, The Netherlands, 2001, pp. 2730.
[3] Donovan, P. and Lodico, D., NCHRP Report 630: Measuring Tire-
Pavement Noise at the Source, Transportation Research Board of the
National Academies, Washington, D.C., 2009.
[4] Kandahl, P. and Lockett, L., Construction and Performance of Ultrathin

Asphalt Friction Course, Report No. 97-05, National Center for Asphalt
Technology, Auburn, AL, 1997.
[5] Liu, K., Alvarez, A., Martin, A., Dossey, T., Smit, A., and Estakhri, C.,
Synthesis of Current Research on Permeable Friction Courses: Perform-
ance, Design, Construction, and Maintenance, Report No. 0-5836-1,
Texas Transportation Institute, College Station, TX, 2010.
[6] Bennert, T., Hanson, D., Maher, A., and Vitillo, N., Influence of Pave-
ment Surface Type on Tire/Pavement Generated Noise, J. Test. Eval.,
Vol. 33, No. 2, 2005, pp. 94100.
[7] Donavan, P., Eight Year Evaluation of the Noise Performance of the Cal-
trans Aspahlt Research Pavements on LA 138, Illingworth & Rodkin,
Inc., Sacramento, CA, 2011.
[8] Hanson, D. and James, R., Colorado DOT Tire/Pavement Noise Study,
Report No. CDOT-DTD-R-2004-5, National Center for Asphalt Technol-
ogy, Denver, CO, 2004.
[9] Bernhard, R. and Wayson, R., An Introduction to Tire/Pavement Noise
of Asphalt Pavement, Publication SQDH 2005-1, The Institute of Safe,
Quiet, and Durable Highways, Purdue University, West Lafayette, IN,
2005.
[10] McDaniel, R. Thornton, W., and Dominguez, J., Field Evaluation of Po-
rous Asphalt Pavement, Report No. SQDH 2004-3, The Institute for
Safe, Quiet, and Durable Highways, West Lafayette, IN, 2004.
[11] Smit, A. and Waller, B., Sound Pressure and Intensity Evaluations of
Low Noise Pavement Structures with Open Graded Asphalt Mixtures,
Report No. 07-02, National Center for Asphalt Technology, Auburn, AL,
2007.
[12] Russel, M., Uhlmeyer, J., Sexton, T., Weston, J., and Baker, T.,
Evaluation of Long-term Pavement Performance and Noise Characteris-
tics of Open-graded Friction CoursesProject 3, Publication WA-RD
749.1, Washington State Department of Transportation, Olympia, WA,
2010.
[13] Bennert, T., Fee, F., Sheehy, E., Jurnikis, A., and Sauber, R.,
Comparison of Thin-lift Hot-mix Asphalt Surface Course Mixes in New
Jersey, Transportation Research Record.1929, Transportation Research
Board, Washington, D.C., 2005, pp. 5968.
[14] AASHTO TP 76-10, 2010, Provisional Standard Test Method for Mea-
surement of Tire/Pavement Noise Using the On-Board Sound Intensity
(OBSI) Method, AASHTO, Washington, D.C.
STP 1555, 2012
DOI:10.1520/STP104439
Qing Lu1
Estimating Acoustic Performance Trends

of Various Asphalt-Surface Mixtures Based
on Field Measurement
REFERENCE: Lu, Qing, Estimating Acoustic Performance Trends of Various

Asphalt-Surface Mixtures Based on Field Measurement, Pavement Perform-
ance: Current Trends, Advances, and Challenges on December 5, 2011 in
Tampa, FL; STP 1555, B. Choubane, Editor, pp. 150159, doi:10.1520/
STP104439, ASTM International, West Conshohocken, PA 2012.
ABSTRACT: As one measure to reduce traffic noise, designing a noise-
reducing (quiet) pavement surface has received increasing attention. For
asphalt pavements, some existing asphalt mixtures have shown better noise-
reducing capabilities than conventional dense-graded asphalt concrete
(DGAC) pavements. To design a new asphalt mixture optimized for noise
reduction, the acoustic performance of existing asphaltsurface mixtures
needs to be evaluated and understood. For this purpose, a study was con-
ducted to measure tire/pavement noise on a variety of asphalt pavements
for four consecutive years. The tire/pavement noise was measured with an
on-board sound-intensity (OBSI) method that was continuously improved
during the study. Regression analysis was applied to determine the levels
and increase rates of tire/pavement noise on four major asphalt mixtures,
and the effects of mix design variables on tire/pavement noise. Results show
that tire/pavement noise generally increases with pavement age on all types
of asphalt pavements. For newly placed or overlaid pavements, open-graded
or rubberized gap-graded mixtures exhibit lower noise levels than conven-
tional dense-graded asphalt mixture. Open-graded asphalt mixtures can
retain noise-reduction benefit for a longer period than gap-graded mixtures.
Using rubberized binders in the open-graded mixtures can further reduce the
noise-increase rate. Pavement surface permeability, macrotexture, and exist-
ing surface distresses may affect the tire/pavement noise level.
KEYWORDS: tire/pavement noise, asphalt mixture, OBSI (on-board sound
intensity), pavement
1
Dept. of Civil and Environmental Engineering, Univ. of South Florida, Tampa, Florida 33620.
Copyright V
19428-2959.
150
LU, doi:10.1520/STP104439 151
Introduction
Highway noise from moving vehicles consists of three components: aerodynamic

noise, power-unit noise, and tire/pavement noise, among which the tire/pavement
noise dominates when vehicle speed exceeds 50 km/h [1]. As one of the effective
measures to reduce highway noise, designing a noise-reducing (quiet) pave-
ment surface can decrease the tire/pavement noise at the source, which may elim-
inate the need of constructing costly sound-barrier walls and, therefore, save cost
and improve roadside aesthetics. For asphalt pavements, some existing asphalt
mixtures (typically, open-graded mixtures) have shown noise-reducing capabil-
ities compared to conventional dense-graded asphalt concrete (DGAC) pave-
ments. These mixtures, however, are originally designed for improving roadway
safety during wet weather instead of reducing tire/pavement noise. To design
new asphalt mixtures that are optimized for noise reduction, while meeting other
pavement surface performance requirements, the acoustic performance of exist-
ing asphalt-surface mixtures needs to be evaluated and understood.
Tire/pavement noise is generated by various types of mechanisms, includ-
ing impact between tire tread and pavement texture, air pumping (rapid com-
pression and squeezing out of air trapped in gaps between tire tread and
pavement texture), and stickslip and sticksnap of tread blocks interacting
with a pavement surface [2]. Previous studies show that air-pumping tends to
produce a significant level of noise over a frequency of 1000 Hz when a pave-
ment surface is nonporous and smooth [3], whereas tread vibration is mostly
responsible for low-frequency noise. Therefore, variables that may increase
tire vibration (e.g., increased macrotexture, roughness, and aggregate size)
would be expected to increase low-frequency noise; whereas variables that mit-
igate the air-pumping mechanism, such as increased air voids, would be
expected to decrease high-frequency noise. Overall noise levels are likely influ-
enced by the combined effects at various frequencies. To reduce the tire/pave-
ment noise, from a pavement engineers perspective, pavement surface needs
to be designed appropriately following the noise-generation mechanisms.
In 2005, a multi-year research project was initiated by the California Dept.
of Transportation (Caltrans) to investigate the surface performance of various
asphalt mixes, including noise, durability, permeability, and friction trends [4].
One primary objective of the study was to determine the level and trend of tire/
pavement noise on various asphalt pavements. From 2006 to 2009, field data
were collected in 4 consecutive years from about 80 asphalt-pavement sections
covering a variety of asphalt mix types, pavement ages, traffic, and environ-
mental conditions. This paper presents the analysis and modeling results of the
tire/pavement noise data collected from four major asphalt-surface mixes: con-
ventional dense-graded asphalt concrete (DGAC), open-graded asphalt con-
crete (OGAC), rubberized open-graded asphalt concrete (RAC-O), and
rubberized gap-graded asphalt concrete (RAC-G). These four mixes are widely
used for asphalt-pavement-surface layers not only in California, but also in

other states and countries.
The objectives of the analysis are to determine the level and trend of tire/
pavement noise on different surface mixes and to identify mix design variables
that may significantly affect tire/pavement noise.
Data Collection
Tire/pavement noise was measured on about 80 pavement sections selected
from a range of climate and traffic conditions and surface mix ages (<1 yr, 1 to
4 yr, or 4 to 8 yr), using one version of the on-board sound-intensity (OBSI)
method that was developed and continuously updated in California. The test
procedure, currently standardized in AASHTO TP 76 Standard Method of Test
for Measurement of Tire/Pavement Noise Using the On-Board Sound Intensity
(OBSI) Method [5], uses two sound-intensity probes installed on a test vehicle
and positioned near the interface of pavement and a standard reference test tire
(SRTT) to measure the tire/pavement noise. Signals from the probe micro-
phones are processed by a two-channel, real-time analyzer and output as
on-board sound-intensity (OBSI) values to represent noise levels. On each
pavement section, OBSI measurement was repeated three times. The standard
test vehicle speed is 96 km/h. On some pavement sections where there were
constraints from roadway geometry or traffic condition, however, the test vehi-
cle speed was reduced to 48 km/h. Details of the noise-data collection proce-
dure can be found in the literature [4].
Before analysis, the OBSI data obtained under various test conditions were
converted to equivalent values at the same set of test conditions (i.e., test vehicle
speed of 96 km/h, air density of 1.21 kg/m3, and a standard reference test tire
[SRTT]) using a set of conversion equations developed during the study [6].
OBSI is presented in terms of spectral content in one-third octave bands, covering
a frequency range from 500 Hz to 5000 Hz. An overall A-weighted sound-inten-
sity level is calculated by summing sound-intensity levels at each frequency.
Results and Analysis

The average overall OBSI value from three repeated measurements is presented
in Fig. 1 for each pavement section of the four mix types. In the figure, each con-
nected line represents multi-year data from one pavement section. Generally there
are four data points in each connected line, representing data collected in the 4
survey years. Because of pavement overlay during the survey or measurement
errors, data are only available for 2 or 3 years for certain pavement sections.
It can be seen from Fig. 1 that the overall OBSI shows a general increase
trend with pavement age for all mixes. For new pavements (age between 0 and
2 yr), the tire/pavement noise levels measured on DGAC pavements are higher
LU, doi:10.1520/STP104439 153
FIG. 1Trend of overall OBSI over four survey years for each pavement section.
than those measured on the other three types of mixes. After the pavements
were exposed to traffic, the overall sound-intensity measured on RAC-G pave-
ments quickly approached the average value measured on DGAC pavements of
same ages. The overall sound intensity measured on the OGAC pavements
appears to only increase slightly for about 5 years and then increase quickly
with pavement age. With a few exceptions, the overall sound intensity measured
on the RAC-O pavements appears not to change significantly for about 7 years.
Based on these observations, the rank of the four mix types in terms of noise is
RAC-O > OGAC > RAC-G > DGAC. These observations based on the 4-yr
measurements are consistent with the findings based on the first 3-yr measure-
ment [7]. There are a few pavement sections on which the measured sound
intensity changed significantly in the later survey years. This was because of
either measurement errors or development of significant surface distresses and
corresponding pavement maintenance activities in between survey years [7].
These data were treated as outliers and excluded from further analysis.
Figure 2 shows the box plots of overall OBSI over four survey years for
different mix types for the three original age categories. As the plots show,
sound intensity generally increases with pavement age for the same group of
pavement sections. Overall, the increase rate of sound intensity is the lowest on
RAC-O pavements, which means that RAC-O pavements remain quieter than
DGAC pavements longer than do OGAC pavements.
To model the tire/pavement noise trend and to determine the effects of mix
design, traffic, and climate variables on this noise, a multiple regression analy-
sis was performed. Two separated models were estimated for the effects of mix
type and mix properties, respectively.
In the first model, only the mix type variable is included in the explanatory
variables, whereas mix property variables (e.g., permeability and aggregate
gradation) are all excluded. The regression model appears below as Eq 1
OOBSI b0 b1 T b2 IO b3 IRG b4 IRO b5 H b6

N b7 A b8 IRV b9 IRT b10 T IO b11 T
IRG b12 T IRO 1
where:
OOBSI overall on-board sound intensity, dB(A);
T pavement age, year;
IO indicator variable, 1 if mix is OGAC, 0 otherwise;
IRG indicator variable, 1 if mix is RAC-G, 0 otherwise;
IRO indicator variable, 1 if mix is RAC-O, 0 otherwise;
H surface layer thickness, mm;
N number of degree days over 30 C;
A annual average daily traffic in the measurement lane;
LU, doi:10.1520/STP104439 155
FIG. 2Box plots of overall OBSI values for different mix types and different
initial age categories in 4 survey years.
IRV indicator variable, 1 if raveling distress is present on the pavement,

0 otherwise;
IRT indicator variable, 1 if rutting distress is present on the pavement, 0
otherwise;
bi coefficients to be estimated, i 0, 1,,12.
The estimated values and P-values of the coefficients are shown in Table
1, which are similar to the estimation results based on the first 3 years data [7].
A P-value less than 0.05 indicates that the corresponding coefficient is sig-
nificant at a 95 % confidence level. From Table 1, it can be seen that pavement
age (b1) is statistically significant in affecting tire/pavement noise, and a posi-
tive b1 value suggests tire/pavement noise increases with pavement age. The
significance and negative estimated values of coefficients b2, b3, and b4 indi-
cate that new pavements with surface mixes OGAC, RAC-G, and RAC-O pro-
duce lower tire/pavement noise than new pavements placed with DGAC mix.
The average noise reductions (compared to DGAC pavements) for newly
placed OGAC, RAC-G, and RAC-O pavements are about 2.7, 1.7, and 2.8
dB(A), respectively. Pavement surface layer thickness is significant is affecting
tire/pavement noise, and the estimated coefficient (b5) suggests that a higher
thickness leads to a lower noise level. The presence of surface distresses of
TABLE 1Estimation results for first linear regression model.
Coefficient Estimated Value P-Value

b0 102.5020 <0.0001
b1 0.1217 0.0246
b2 2.7261 <0.0001
b3 1.6732 0.0003
b4 2.7697 <0.0001
b5 0.0210 0.0003
b6 0.003262 0.1030
b7 0.000002430 0.8667
b8 0.7070 0.0010
b9 0.7468 0.0170
b10 0.1318 0.1093
b11 0.0878 0.2668
b12 0.0481 0.5306
Note: Residual standard error: 1.160 on 187 degrees of freedom; multiple R2: 0.58.
raveling and rutting also significantly affect tire/pavement noise. The two esti-
mated coefficients (b8, b9) indicate that the presence of raveling or rutting
increases tire/pavement noise by about 0.7 dB(A). The interaction terms
between age and mix type (b10, b11, b12) are statistically insignificant, indicat-
ing that the growth rate of overall sound intensity is not statistically different
among the four pavement types. This is different from the direct observations
from Fig. 1. This is mostly because of the constraints applied by the multiple
regression analysis that assumes a linear increase of noise with age for all
mixes. Use of different growth function forms for different mixes in the same
regression model significantly increases the complexity of parameter estima-
tion and result interpretation, so it is adopted in this model. Instead, in estimat-
ing separate acoustic models for each surface mix type, various noise increase
rates may be estimated.
In the second model, the effects of mix property variables instead of mix
type were estimated, and the model was estimated separately for each mix
type. The regression model is shown in Eq 2
OOBSI b0 b1 lnP b2 T b3 F b4 M b5 H b6 N
b7 A (2)
where:
P pavement surface permeability, cm/s;
F fineness modulus of all aggregates in the mix;
M mean profile depth, lm; and
bi coefficients to be estimated, i 0, 1,,7.
The model was estimated separately for each mix type, and the results are
shown in Table 2.
LU, doi:10.1520/STP104439 157
TABLE 2Estimation results for second linear regression model.
Estimated Value
Coefficient DGAC OGAC RAC-G RAC-O

a a a
b0 99.3934 101.4530 95.1318 107.6118a
b1 0.3004a 0.1780a 0.2960a 0.03748
b2 0.03512 0.3478a 0.2130a 0.20822a
b3 0.7708 0.8991 0.05397 1.01597
b4 0.004036a 0.00035 0.00159a 0.00080
b5 0.001539 0.00936 0.00805 0.06961a
b6 0.001804 0.00357 0.01327a 0.00113
b7 0.00005800 0.000104a 0.000028 0.0000008
R2 0.67 0.73 0.68 0.44
a
Represents the coefficient is statistically significant at a 95 % confidence level.
It can be seen from Table 2 that, except for the RAC-O mix, the R2 for the
individual mix model is higher than that of the first model that includes all mix
types, indicating that generally the separate models for each mix type fit the
measurement data better than the first model. The results show that at a 95 %
confidence level, pavement age is statistically significant in affecting tire/pave-
ment noise on all types of mixes except DGAC. The estimated values for coef-
ficient (b2) show that for the two open-graded mixes, noise increase rate is
higher on OGAC pavements than on RAC-O pavements. Pavement surface
permeability is a significant variable for all mixes except RAC-O. The negative
value of coefficient b1 suggests that higher permeability leads to lower noise
levels. Pavement mean profile depth is a significant variable for DGAC and
RAC-G pavements, and higher profile depths (b4) correspond to higher noise
levels. For OGAC and RAC-O pavements, mean profile depth does not have a
significant influence on noise level. The surface layer thickness (b5) is signifi-
cant only for RAC-O, possibly because of the fact that for other mix types the
thicknesses were typically very similar.
For all mix types, the aggregate gradation variable, fineness modulus (b3)
does not seem to significantly affect tire/pavement noise. Traffic volume (b7) is
a significant variable that increases tire/pavement noise for OGAC mixes. For
RAC-G mixes, high temperature duration is significant and the estimated coef-
ficient (b6) indicates that tire/pavement noise increases when the number of
high temperature days increases.
Conclusions
The study analyzed and modeled tire/pavement noise levels and trends on four
types of asphalt-pavement surfaces based on on-board sound-intensity data col-
lected from about 80 pavement sections in 4 consecutive years. The following
findings were obtained:
1. Overall tire/pavement noise generally increases with pavement age.

For newly paved overlays, the overall sound intensities measured on
OGAC, RAC-G, and RAC-O pavements are lower than the values
measured on the DGAC pavements. The average noise reductions
(compared to DGAC pavements) for newly paved OGAC, RAC-G, and
RAC-O mixes are about 2.7, 1.8, and 2.8 dB(A), respectively. After the
pavements are exposed to traffic, the overall sound intensity measured
on RAC-G pavements quickly approaches the typical value measured
on DGAC pavements of similar ages. The overall sound intensity
measured on the OGAC pavements does not change much for about
5 yr and then increases quickly with pavement age. With a few excep-
tions, the overall sound intensity measured on the RAC-O pavements
does not change much for about 7 yr and then increases quickly with
pavement age. The ranking (from best to worst) of the four mix types
in terms of noise reduction is RAC-O, OGAC, RAC-G, and DGAC.
2. Multiple regression analysis on all mixes shows that overall sound in-
tensity increases with the presence of raveling or rutting damage, and
decreases with increased surface layer thickness. Multiple regression
analysis on individual mix types shows that pavement surface perme-
ability is a significant factor on all mixes except RAC-O, and higher
permeability leads to a lower noise level. For all four mix types, the
aggregate gradation variable (fineness modulus) does not seem to sig-
nificantly affect tire/pavement noise. Pavement surface macrotexture
as represented by mean profile depth is a significant factor for DGAC
and RAC-G pavements, and higher mean profile depths correspond to
higher noise levels.
Acknowledgments
This paper describes research activities that were sponsored by the California
Department of Transportation (Caltrans), Division of Research and Innovation.
Caltrans sponsorship is gratefully acknowledged. The contents of this paper
reflect the views of the author and do not reflect the official views or policies of
the State of California or the Federal Highway Administration.
References
[1] Nelson, P. M. and Phillips, S. M., Quieter Rd. Surfaces, TRL Annual
Review, Transportation Research Laboratories, United Kingdom, 1997.
[2] Robert, O. R., Robert, J. B., Ulf, S., and Eric, P. M., The Little Book of
Quieter Pavements, Report No. FHWA-IF-08-004, Federal Highway
Administration, 2007.
[3] Sandberg, U. and Descornet, G., Rd. Surface Influence on Tire/Rd.
Noise: Part 1, Proceedings of Inter-Noise 80, Miami, FL, 1980.
LU, doi:10.1520/STP104439 159
[4] Ongel, A., Experimental Analysis of Open-graded Asphalt Concrete

Mixes in Terms of Safety, Durability, and Noise, Ph.D. thesis, University
of California, Davis, 2007.
[5] American Association of State Highway and Transportation Officials,
Standard Method of Test for Measurement of Tire/Pavement Noise using
the On-Board Sound Intensity (OBSI) Method, AASHTO Specification
TP 76-10, 2010.
[6] Lu, Q., Harvey, J. T., and Wu, R. Z., Investigation of Noise and
Durability Performance Trends for Asphaltic Pavement Surface Types:
Four-Year Results, UCPRC-RR-2010-05, University of California,
Davis, 2010.
[7] Lu, Q., Kohler, E., and Harvey, J., Field Investigation of Acoustic
Performance of Various Asphalt Surface Mixes, Proceedings of the 10th
International Conference of Chinese Transportation Professionals,
Beijing, China, 2010.
STP 1555, 2012
DOI:10.1520/STP104444
Robert Otto Rasmussen1 and Richard C. Sohaney2
Measuring and Visualizing TirePavement

Noise Using On-Board Sound Intensity
REFERENCE: Rasmussen, Robert Otto and Sohaney, Richard C.,

Measuring and Visualizing TirePavement Noise Using On-Board Sound
Intensity, Pavement Performance: Current Trends, Advances, and Chal-
lenges on December 5, 2011 in Tampa, FL; STP 1555, B. Choubane, Editor,
pp. 160169, doi:10.1520/STP104444, ASTM International, West Consho-
hocken, PA 2012.
ABSTRACT: The Concrete Pavement Surface Characteristics Program
(CPSCP) has recently developed guidelines and practices for reducing tire
pavement noise. The primary means for doing this is via measurements
using the on-board sound intensity (OBSI) technique. OBSI measures tire-
pavement noise at the source utilizing specialized microphones mounted to
the outside of a vehicle, near the tirepavement interface. Measurements are
performed while the vehicle drives across the pavement being evaluated. To
facilitate innovation, specifications have been developed based on an end-
result approach. In this paper, a means for how to measure OBSI is provided,
along with a means to interpret the as-constructed quality through visualiza-
tion techniques. With it, the spatial variability of tire-pavement noise, as
measured using OBSI, can be readily determined.
KEYWORDS: pavement, tirepavement noise, texture, noise, OBSI,
on-board sound intensity, concrete
Introduction
In December 2004, a coalition was formed between the National Concrete
Pavement Technology Center (CP Tech Center), Federal Highway Administra-
tion (FHWA), American Concrete Pavement Association (ACPA), and the
International Grooving and Grinding Association (IGGA). The mission was to
October 2012.
1
Ph.D., The Transtec Group, Inc., Austin, TX 78731.
2
The Transtec Group, Inc., Austin, TX 78731.
Copyright V
19428-2959.
160
help optimize concrete pavement surface characteristics, and, more specifi-

cally, to find quieter pavement solutions that will not compromise safety, dura-
bility, or cost effectiveness. The Concrete Pavement Surface Characteristics
Program (CPSCP) emerged from this coalition, with a focus on identifying spe-
cific guidance to properly design and construct quieter concrete pavements.
The key recommendations developed out of the CPSCP are largely based
on tirepavement noise tests conducted worldwide using the on-board sound
intensity (OBSI) method. This test method measures tirepavement noise at
the source utilizing microphones in a sound intensity probe configuration
mounted to the outside of a vehicle, near the tirepavement interface. Measure-
ments are performed while the test vehicle drives across the pavement of inter-
est. In this paper, the OBSI test method is described, including some of the
history behind its development, the test setup, important factors that must be
controlled during the test, the general test procedure, and how the results can
be reported and interpreted with specific application to constructing more uni-
form, quieter concrete pavement surfaces.
Sound Intensity
Sound is caused by small and rapid pressure changes in the air. Our ears
respond to these changes in sound pressure, and they can similarly be measured
with a microphone. When a sound level is reported (e.g., 90 dBA), it is most
often a measure of the amplitude of these sound pressure changes. Sound inten-
sity is different from sound pressure in that it has both amplitude and an associ-
ated direction [7]. Intensity is related to the flow of acoustic energy, and more
specifically, it is the acoustic power per unit area (units of W/m2). Because it
includes direction, it is not as simple to measure as sound pressure, but one
common technique is to use a probe consisting of two microphones spaced
apart by specified distance. Measuring both the amplitude and phase of sound
at the two microphones gives the probe directional characteristics and allows
for the direction of sound sources to be determined. Figure 1 shows an example
of a sound intensity probe used to measure tirepavement noise. In this case,
the microphones are in a side-by-side configuration.
There are several advantages of using sound intensity instead of sound
pressure for measuring tirepavement noise at the source [5]. First, the direc-
tional characteristic of the probe makes it better suited for measuring a specific
noise source while attenuating sounds from other sources in other directions
(e.g., engine or exhaust noise). Second, sound intensity is much less contami-
nated by random noise, such as wind noise generated as the vehicle is mov-
ing. Third, because sound intensity measures the acoustic energy propagating
away from the source to the roadside, it correlates well with sound measured at
the roadside (known as pass-by or wayside measurements).
FIG. 1Sound intensity probe used for OBSI testing showing two 12.7-mm
(1/2-in.) microphones.
History of OBSI
Sound intensity has always been part of acoustic theory, but until the 1970s,
was not readily measured because of its computational demand [7]. As com-
puters became smaller, faster, and less expensive, instrumentation capable of
measuring sound intensity was developed, and, more specifically, a two-
microphone technique [2]. In the 1980s, General Motors pioneered the applica-
tion of the two-microphone sound intensity method to automotive noise
measurements, and P. Donavan, in particular, applied the method to measuring
tire noise [4]. This work resulted in the General Motors Road Tire Noise Eval-
uation Procedure, Test Procedure GMN7079TP, 2004.
In the early 2000s, while at Illingworth and Rodkin of Petaluma, CA,
Donavan continued development of what is now the OBSI method for meas-
uring tirepavement noise [5]. With the assistance of Caltrans (including B.
Rymer) and Arizona DOT (L. Scofield), numerous refinements to the method
were made in the early years of its use. The CPSCP adopted the OBSI technol-
ogy in 2005, and since that time has also contributed to its evolution. The
OBSI technique employed today uses a dual probe fixture configuration and a
standard reference test tire described by the ASTM International Standard
F2493 [3]. Work has been conducted in parallel under NCHRP Study 1-44 to
document the more critical aspects of the OBSI test protocol; this work was
published as NCHRP Report 630 [6].
In recent years, standardization efforts were initiated, and the acronym
OBSI was adopted. The OBSI method was first standardized for the highway
community by AASHTO in 2008, and since has undergone annual updates as
provisional standard TP76, Standard Method of Test for Measurement of
Tire/Pavement Noise Using the On-Board Sound Intensity (OBSI) Method
[1]. In addition to other industry experts, members of the CPSCP team have
been integral to the development of the AASHTO standard by way of an
FHWA Technical Working Group on tirepavement noise.
OBSI Testing
Research has shown that tirepavement noise can be well described by meas-
uring at two principal locations near the tirepavement interface [6]. These
locations are defined as the leading and trailing edge of the contact patch. As
illustrated in Fig. 2, the OBSI test procedure specifies that the probes be located
close to these spots, more specifically, 101.6 mm (4 in.) horizontally from the
tire side wall, 76.2 mm (3 in.) vertically above the pavement, and 104.8 mm
(4.125 in.) in front and behind the axle centerline [1].
A special fixture is used to mount to the probes at the specified locations.
Early versions of the test procedure used a fixture that positioned only a single
FIG. 2Sound intensity probe positions with respect to the tirepavement

interface.
FIG. 3OBSI configurations with a single sound intensity probe (left) and
dual probe (right).
sound intensity probe (Fig. 3, left). With this single-probe fixture, two runs
were necessary to measure both noise sources; a first run with the probe posi-
tioned at the leading edge was done, and a second with the probe positioned at
the trailing edge. Driven by demand for increased test efficiency, most OBSI
systems in use today employ a dual-probe configuration (Fig. 3, right) where
these measurements are collected simultaneously. Both configurations have
been used over the course of the CPSCP measurements.
Microphone windscreens must be used to reduce the noise caused by air
flowing around the microphones while driving during the test. For OBSI meas-
urements, windscreens are special foam spheres, nominally 89 mm (3.5 in.) in
diameter [1].
Experience and research has shown there are several factors that must be
controlled during the OBSI test. Among the more significant are the test tire,
vehicle speed, and vehicle noises.
Test Tire
The current OBSI standard [1] specifies use of an ASTM F2493 Standard Ref-
erence Test Tire (SRTT) (P225/60R16) [3]. However, at the beginning of the
CPSCP project (through May 2006), OBSI testing was conducted using the de
facto standard tire, a Goodyear Aquatred 3 tire (P205/70R15). Differences in
tire construction include the tread pattern (see Fig. 4) and rubber compounds;
both of these affect the noise measurements from these two tires. The differ-
ence in tirepavement noise generated by these two tires was important to char-
acterize so that data collected early in the CPSPC could be compared to that
collected in recent years. This relationship is not a simple one though. In terms
FIG. 4Standard tires used in CPSCP OBSI Testing: Goodyear Aquatred 3

(left) and ASTM F2493 SRTT (right).
of overall sound intensity level, the SRTT is quieter than the Aquatred by 1 to
3 dBA (decibels, A-weighting). The specific differences, however, are better
determined by frequency, and, furthermore, by the nominal pavement texture
being evaluated.
Vehicle Speed
The OBSI standard specifies a vehicle speed of 96.6 km/h (60 mph) [1]. In sit-
uations where safety does not permit this speed, the variance must be clearly
noted, because the resulting measurement will be affected. For small variations
in speed (i.e., 64.8 km/h (63 mph)), a correction factor can be applied during
post processing. For larger variations, however, error can be introduced if a sin-
gle (generic) correction is applied. The relationship between speed and sound
level varies significantly depending on the specific combination of tire and
pavement surface.
Vehicle Noise
The test vehicle must not make any abnormal noise that could contaminate the
tirepavement noise measurement. Examples include noise caused by foreign
matter on the tire tread, suspension/shock squeak, wheel bearing squeal, and
brake noises. Good practice is for the test operator to listen to the microphone
signals during the data acquisition to identify if these or other abnormal noises
are occurring.
OBSI Test Sections

The AASHTO TP76 OBSI standard requires test sections that are 134.1 m
(440 ft) long. At 96.6 km/h (60 mph), it takes the test vehicle 5 s to traverse a
FIG. 5Vehicle with sound intensity probes mounted on the right rear wheel
and traversing a test section.
test section, and the results are an average over this 5-s interval [1]. Whereas
the CPSCP testing is in general conformance with the AASHTO standard,
test sections are of varying length. The CPSCP testing was initiated years
before the standard was developed, but, more importantly, test section boun-
daries are determined based on a number of criteria. For example, where
there are multiple surface types back-to-back, test section limits are defined
in part by these transitions. The CPSCP test sections have all been thoroughly
documented, though, so that the exact same sections can be retested in subse-
quent years.
As part of the CPSCP testing, reflective cones are placed on the shoulder at
the start and end of the test sections. An optical sensor mounted on the test ve-
hicle detects the cones as the vehicle passes by, and that signal is integrated
into the acoustic recording system and used as a trigger. Figure 5 illustrates
this process.
Reporting and Visualizing OBSI Data

For an individual test section, a single OBSI level (units of dBA) is often cited.
This value is a combination of the sound intensity measured at the leading and
trailing edges of the tirepavement contact patch, A-weighted, averaged over
the length of the test section, and limited to the frequency range covered by the
400 to 5000 Hz one-third octave bands. For concrete pavements, this level typi-
cally ranges from 99 dBA for quieter pavements to 108 dBA or more for louder
pavements [8,9].
When the frequency content of tirepavement noise is of interest, the
sound intensity spectrum may also be reported. As illustrated at the top of
Fig. 6, it is common to report one-third octave bands from 400 to 5000 Hz.
FIG. 6Results from OBSI processing: one-third octave spectral analysis

(top) and OBSI level versus distance showing variability over the test section
(bottom).
To study the variability of the tirepavement noise response over the length
of a test section, A-weighted OBSI levels can be calculated using short inter-
val averages, and plotted as a function of distance. This is shown at the bot-
tom of Fig. 6.
Finally, to help clearly visualize OBSI data, a tool has been developed that
allows the results to be viewed in Google Earth. The tool also has the ability to
view photographs and listen to the tirepavement noise via an embedded audio
player. This is illustrated in Fig. 7.
FIG. 7Google Earth reporting of OBSI results.
Conclusions
The OBSI technique has proven an invaluable tool during the course of the
CPSCP. Even casual observations of OBSI data using visualization tools, such
as Google Earth, can be important. Using OBSI, the location of the quieter
and louder areas of pavement allow for a more thorough evaluation of the
materials and construction elements at strategically selected locations. From
this, better practices and specifications have been developed that serve to
reduce the sound generated on concrete pavements without compromising
safety, durability, or cost. These documents are publicly available via the
website of the National Concrete Pavement Technology Center

(www.CPTechCenter.org).
Acknowledgments
The writers acknowledge the support of the National Concrete Pavement Tech-
nology Center including Tom Cackler, Dale Harrington, and Paul Wiegand, as
well as the sponsorship of Transportation Pooled Fund TPF-5 [139], including
contributions from U.S. DOT, FHWA, and the U.S. State DOTs of CA, IA,
MN, NY, TX, WA, and WI. The writers also acknowledge Dr. Paul Donavan
and the numerous individuals that have contributed to the standardization of
OBSI as part of the FHWA Tire-Pavement Noise Technical Working Group.
References
[1] AASHTO TP76-11, 2011, Standard Method of Test for Measurement of

Tire/Pavement Noise Using the On-Board Sound Intensity (OBSI) Method,
American Association of State Highway and Transportation Officials,
Washington, D.C.
[2] ANSI S1.9-1996, 2006, Instruments for the Measurement of Sound
Intensity, American National Standards Institute, Washington, D.C.
[3] ASTM F2493, 2008, Standard Specification for P225/60R16 97S Radial
Standard Reference Test Tire, Annual Book of ASTM Standards, Vol.
09.02, ASTM International, West Conshohocken, PA.
[4] Chung, J.-Y., Sound Intensity Meter, U.S. Patent No. 4,236,040 (Nov
25, 1980).
[5] Donavan, P. and Rymer, B., Quantification of Tire/Pavement Noise:
Application of the Sound Intensity Method, Proceedings of Internoise
2004, Prague, the Czech Republic, 2004.
[6] Donavan, P. and Lodico, D., Measuring Tire-Pavement Noise at the
Source, Report No. 630, National Cooperative Highway Research Pro-
gram, Washington, D.C., 2009.
[7] Fahy, F. J., Sound Intensity, 2nd ed., E & FN Spon, London, 1995.
[8] Rasmussen, R. O., Bernhard, R. J., Sandberg, U., and Mun, E. P., The
Little Book of Quieter Pavements, Report No. FHWA-IF-08-004,
USDOT Federal Highway Administration, Washington, D.C., 2007.
[9] Rasmussen, R. O., Whirledge, R. P., Turner, D. J., Light, R. J., Ferragut,
T. R., and Wiegand, P. D., Identifying Quieter Concrete Pavements
Using On-Board Sound Intensity, Transportation Research Board An-
nual Meeting, Washington, D.C., Jan 1317, 2008.
STP 1555, 2012
DOI:10.1520/STP104473
Regis Carvalho,1 Richard Stubstad,2 Larry Wiser,3 and

Olga Selezneva4
Simplified Deflection-based Probabilistic

Technique for Network-level Pavement
Analysis
REFERENCE: Carvalho, Regis, Stubstad, Richard, Wiser, Larry, and Selez-

neva, Olga, Simplified Deflection-based Probabilistic Technique for Net-
work-level Pavement Analysis, Pavement Performance: Current Trends,
Advances, and Challenges on December 5, 2011 in Tampa, FL; STP 1555,
B. Choubane, Editor, pp. 170191, doi:10.1520/STP104473, ASTM Interna-
tional, West Conshohocken, PA 2012.
ABSTRACT: The concept of a pavement management system (PMS) was
developed following the successful AASHTO Road Test, in the 1950s, when
pavement performance was characterized as the necessary information to
compare design alternatives and expectations over the service life of both
flexible and rigid pavement structures. A link was established between per-
formance and the activities of planning, designing, constructing, and main-
taining pavements, and the concept of a PMS was defined accordingly.
Currently, surface distress and roughness measurements are the main com-
ponents in determining a pavements existing condition in a PMS. The struc-
tural condition, although capable of providing important information on how
the pavement will perform in the future (rate of deterioration, types and sever-
ity of expected distresses, etc.), is not widely used. Reasons for this lack of
presence in a system-wide PMS are mostly related to time-consuming analy-
ses and the need for an experienced analyst. This paper presents the devel-
opment of simplified, deflection-based analytical techniques that provide
rapid, automated screening of pavements structural capacity for inclusion in
a network-level PMS. The approach is based on stochastic-logistic modeling
commonly used in reliability engineering. Data from the Long-Term Pavement
October 2012.
1
Dynatest Consulting, Upper Marlboro, MD 20774.
2
Dynatest Consulting, Ventura, CA 93001.
3
Federal Highway Administration, McLean, VA 22101.
4
Applied Research Associates, Elkridge, MD 21075.
Copyright V
19428-2959.
170
CARVALHO ET AL., doi:10.1520/STP104473 171
Performance database was used to develop appropriate models, which relate

the expected future performance to deflection-based parameters extracted
analytically from falling weight deflectometer deflection basins and site-
specific conditions such as climate and traffic. It is of paramount importance
to consider the existing structural capacity of pavements together with pave-
ment distresses in order to predict future pavement conditions more accu-
rately and thus successfully fulfill the overall goals of the original concept of
pavement management.
KEYWORDS: pavement deflection, falling weight deflectometer, pavement
management system, network-level analysis, pavement performance
Introduction
Deflection data analysis provides both a qualitative and a quantitative assess-
ment of the structural integrity and bearing capacity of a pavement. Pavements
with poor structural quality are more likely to develop distresses prematurely.
It is commonly expected that the rate of deterioration will increase as the struc-
tural condition worsens. Identifying pavements with poor structural condition
is very important for preventing the early and rapid development of load-
related distresses.
Unfortunately, the use of deflection data for network-level pavement analy-
ses is not common. This lack of practice is mainly due to costs associated with
testing and the difficulties associated with it, which often involve lane closures
and safety concerns to road users and technicians performing the tests alike.
However, this obstacle can be overcome with efficient falling weight deflectome-
ter (FWD) test point spacings and the identification of an appropriate frequency
over time for network-level testing. For optimum results, it is not necessary to
test at project-level intervals, nor is it necessary to test every year [1].
Another main issue for limited deflection data applications for network-level
analysis is the apparent lack of an ability to relate the deflections directly to a
structural index that can be applied to the entire network and which does not
need expert opinion or traditional backcalculation. Currently, deflection data are
often used to backcalculate layer moduli in order to evaluate remaining life and
overlay requirements on a project-level basis. Backcalculation is a technique that
in most cases requires technical expertise. For this reason, its application in
network-level analyses contradicts the prevailing premise of a pavement man-
agement system (PMS): semi-automated pavement analysis for budget and per-
formance optimization. If simple parameters derived directly from the deflection
basin are used, then the question becomes one of how to interpret these data in a
systematic way that both fits the purpose of a PMS and eliminates the need for
engineering expertise to oversee the entire optimization process.
The purpose of this research was to identify deflection-based parameters
that could be used in a simplified technique developed for network-level appli-
cations. Although the technique has been shown throughout the course of the
research to be equally well suited for both flexible and rigid pavement types,
the emphasis of this paper is on flexible pavements.
Overview of Deflection Data Usage for Network-level Applications

Currently, there are only a few agencies that are using deflection data or have
plans to incorporate it directly into their PMS. The AK Department of Transpor-
tation uses a multi-year cycle for deflection readings based on ongoing pre- and
post-rehabilitation projects; their PMS approach using FWD data was only
recently initiated [2]. Although interesting, it cannot be considered a simple
process because it currently uses backcalculation to determine the remaining life.
The TX Department of Transportation currently uses deflection data in its
PMS for flexible pavements only (district optional) [36]. Basin parameters
were developed and used in lieu of a backcalculation approach because of the
lack of pavement structural data (number of layers, layer thicknesses, and ma-
terial types). From these parameters, a structural strength index is determined
based on the thickness of the asphalt layer. This approach allows the user to
determine where the inadequacies lie, whether within the subgrade or some-
where within the pavement structural section itself above the subgrade.
The VA Department of Transportation recently conducted a research study
to evaluate the applicability of the use of deflection data in their PMS [7]. The
approach is simple. Flexible pavements are analyzed by calculating the sub-
grade resilient modulus, the effective pavement modulus, and the effective
structural number. Rigid and composite pavements are analyzed by calculating
the area under the deflection basin and the static modulus of the subgrade reac-
tion (k-value). In both cases, a closed form solution is used, which eliminates
backcalculation and the need for expert oversight. The study was carried out
using a limited number of sections, and based on the results obtained it was
recommended that the process could be expanded to cover the entire primary
pavement network of VA.
The Republic of South Africa uses FWD testing at 200 m (1/8 mile)
intervals to ascertain the bearing capacity of three roadway zones or layers
(surface, base, and subgrade). The methodology relies on direct deflection ba-
sin parameters, not backcalculation [8]. A so-called benchmarking technique
was developed to relate direct deflection values and basin parameters to the
structural condition of the pavement at both project and network levels.
The above-mentioned applications have in common the need for the devel-
opment of threshold levels on a case-by-case basis, often relying on experience
and historical data. Although this approach might work for a more detailed
analysis such as a project-level design, it is not ideally suited for network-level
applications, unless a thorough investigation is conducted prior to the imple-
mentation of the recommended techniques. On such occasions, deflection data,
combined with a history of rehabilitation and performance, are used to create a
rating system that yields a qualitative assessment of the structural condition.

This assessment is used to determine prospective maintenance and rehabilita-
tion (M&R) treatments or, if no such treatments are needed immediately, the
remaining service life of the pavement.
Based on these premises, a probabilistic approach was developed to deter-
mine the likelihood of premature failure using load-deflection techniques. Pre-
mature failure is defined by the presence of excessive distresses (i.e., levels of
distress higher than the design threshold) occurring prior to the end of the
design life of the pavement section.
The probabilistic approach proposed in this research can be used to estab-
lish thresholds for deflection data based on performance measures and assist
with the creation of the network-level structural decision matrix. The benefits
of using a deflection-based probabilistic model sensitive to structural pavement
performance are as follows:
It provides a direct link to pavement performance by estimating the sta-
tistical likelihood of the development of load-related distresses in early

stages of the pavements service life. The intent is to obtain not a well-
defined predictive model based on deflection data, but rather a simplified
procedure with which to identify critical sections with a high likelihood
of developing structural distresses prematurely and to enable planning
for these premature failures on a network-level basis, not job by job or
section by section.
Rating criteria can be created in the same a manner as a functional deci-
sion matrix. Different tiers of structural quality can be defined based on

deflection-based thresholds and the likelihood of acceptable performance.
The structural decision matrix can then follow rating criteria similar to a
functional decision matrix, which in turn facilitates implementation
within existing PMS algorithms and databases.
The probabilistic models can be locally calibrated to meet network char-
acteristics (e.g., typical surface distresses observed, interval between

M&R projects, and acceptable threshold limits for the various distress
types).
The probabilistic models can be recalibrated on an ongoing basis as new
load-deflection and surface distress survey data are collected over time.
Binary Logistic Model
Logistic regression is useful for situations in which one wants to predict the
likely outcome of an event based on a set of predictor variables (numerical and
categorical). Logistic models are best suited for situations in which the depend-
ent variable is dichotomous (i.e., a variable that assumes binary values, such as
acceptable and not acceptable). Binary (logistic) regression is a powerful anal-
ysis tool that can be used to model complex problems in which the number of
variables affecting the outcome is large [9].
Formulation and Calibration

The logistic regression models general equation is provided in Eq 1. In this
case, P is the probability of the occurrence of an event, and b is the exponent
parameter in the logistic distribution, which is a linear function of the predictor
variables as given in general form in Eq 2. The event is defined as the per-
formance of the pavement sectionacceptable for sections outlasting the
design life and not acceptable for sections exhibiting premature failure. The
predictor variables, also known as explanatory variables, are selected so as to
best quantify the behavior of the event, such as pavement cracking, rutting, or
ride quality. In this research, the predictor variables are deflections and site
characteristics such as climate, traffic, and pavement structure. A generalized
logistic model probability function of a predictor variable X is shown in Fig. 1
1
Pevent (1)
1 eb
b a0 a1 b1 a2 b2 an bn (2)
Predictor variables can be continuous (e.g., traffic volume) or discrete (e.g.,

pavement subgrade type assigned as coarse- or fine-grained, or non-cohesive or
cohesive). The dichotomous event is modeled using binary numbers: 1 for ac-
ceptable, and 0 for not acceptable. The model coefficients are computed using
optimization algorithms. The most widely used of these algorithms is the
FIG. 1Typical logistic model probability function of a predictor variable X.

maximum likelihood estimate. In this algorithm, the coefficients of the pre-

dictor variables are obtained so as to maximize the probability of the events
occurring. The maximum likelihood estimate is well suited for problems in
which the event occurs in the majority of cases. This is the case with pavement
performance, as the majority of the networks pavement sections are expected
to perform well.
There is no practical limitation to the number of predictor variables that
can be used. In fact, it is part of the process to eliminate variables with no sig-
nificant relevance to the probabilistic function. This characteristic greatly helps
with the analysis, as there is no need to refine the selection of variables, and all
possible candidates can be considered. The selection of significant variables is
achieved through a stepwise regression. Initially, all predictor variable candi-
dates are used in the calibration process. The predictor coefficients are calcu-
lated using the optimization algorithm, and the significance of each predictor
with regard to the probability function is computed. One by one in each step,
the predictors with the least significance to the problem are eliminated [10].
The process is concluded when either all remaining variables have significance
levels higher than 95 % or the maximum number of iterations is reached, in
which case the analyst has the option of reanalyzing the model with a higher
number of iterations or accepting the model at the level of accuracy
determined.
Measurement of Accuracy
Conventional statistics cannot be used to measure the goodness of fit of binary
logistic models. Instead, so-called receiver operating characteristics (ROC)
curves are used to evaluate the accuracy of the model. ROC curves have been
used in signal detection theory to depict the tradeoff between hit rates and false
alarm (false negative) rates of classifiers. They are also widely used in medical
decision making [11]. In recent years, ROC curves have also been used in
research associated with data mining techniques. One important characteristic
is that ROC analysis allows the selection of possibly optimal models and the
option to discard suboptimal ones independent of the class distribution of the
predictor values.
ROC curves are useful for organizing predictors and visualizing their per-
formance. The ROC curve is drawn using only the true positive rate (TPR) and
the false positive rate (FPR). The TPR is determined by the number of true pos-
itive predictions, normalized to the total number of positive observed values.
The FPR defines the number of incorrect positive results occurring among all
negative samples available during the stochastic analysis.
The ROC space is defined by the TPR and FPR, represented by the x-axis
and the y-axis, respectively, and this space depicts the relative trade-off
between true positives and false positives. The TPR is often described as the
sensitivity of the model, whereas (1 FPR) is often seen as the true negative
rate or specificity. Therefore, the ROC curve is also referred to as the sensitiv-
ity versus (1 specificity) curve.
Each case consisting of a prediction and observed values represents one
point in the ROC space. Given the discrete nature of the problem (i.e., the pre-
dicted probability is associated with a discrete definition of pavement perform-
ance [acceptable or not acceptable]), algorithms are used to create the ROC
curve [11].
An example of an ROC curve is shown in Fig. 2. Several points in the
ROC space are important to note. The best possible prediction would yield a
point in the upper left corner or coordinate (0,1) of the ROC space, represent-
ing 100 % sensitivity (no false negatives) and 100 % specificity (no false posi-
tives). The (0,1) point is also called a perfect classification. A completely
random guess (e.g., determined by flipping a coin) would give a point along a
diagonal line, also referred to as the line of no discrimination, which consists
of points (0,0) and (1,1). Points above the diagonal line indicate good classifi-
cation results. In comparative terms, one point in ROC space is better than
another if it is to the northwest (either the TPR is higher or the FPR is lower, or
both) of the first point.
The area under the ROC curve defines the goodness of fit of the predic-
tions. In an analogy with measures of linear optimization techniques, the area
under the ROC curve can be viewed as the equivalent of an R-squared value in
linear regression. By definition, the area under the ROC curve cannot be lower
than 0.5, which corresponds to the reference line, shown in Fig. 2. The highest
value is determined by the boundary of the chart, which represents the perfect
classification with an area under the ROC curve equal to unity, or 1.
FIG. 2Example of an ROC curve.

In addition to the goodness of fit, the ROC curve is used to define the cut-
off value of the model. This is a very important parameter that defines how the
results from the probability density function can be interpreted in a binary sys-
tem (i.e., the threshold used to separate probabilities that are referred to as ac-
ceptable or not acceptable). The cut-off is defined as the probability value that
corresponds to the point in the ROC curve closest to the perfect classification
point (0,1). When the cut-off is used to convert the predicted probabilities into
binary outcomes, the model yields the highest level of accuracy (i.e., the high-
est TPRs and true negative rates).
Stochastic Approach for Evaluating Load-deflection Techniques for

Network-level Applications
The objective of pavement design is to provide a structure that will exhibit
adequate performance throughout its design life. In this case, performance is
defined by threshold levels of distress that can affect the ride quality, plus
cracking and rutting, of the pavement. Therefore, it is defined that a pavement
has an acceptable level of performance if distress levels do not exceed the
threshold levels during a period of time (e.g., fatigue cracking not to exceed
25 % of the total area within 15 years of service for flexible pavements).
Because the logistic model describes the likelihood of an events being accept-
able or not acceptable as a function of one or more predictor variables, it can
be used to model the probability of acceptable performance that includes meas-
urements of deflections or deflection parameters directly computed from the
deflections themselves, not via backcalculation, which, as stated previously,
usually requires an engineering expert interface.
The goal of the stochastic approach proposed is to use the probability den-
sity function to obtain insightful information about the pavements structural
condition required in order to achieve the desired performance during a pre-
defined period of time. This new knowledge can then be used to define strategic
treatments (maintenance, preservation, or rehabilitation) to be used in the PMS
optimization process.
A typical logistic model probability function is shown in Fig. 3. It
describes the likelihood that the pavement section will exhibit acceptable per-
formance in a particular distress as a function of a given deflection parameter
(in this case center deflection). In this example, the cut-off value was deter-
mined from the ROC curve and is plotted in the figure. It should be noted here
that a different probability function is regressively calculated for each uniform
pavement section as a function of various important co-input design variables,
such as traffic, pavement thickness, etc. However, the cut-off value of the
deflection parameter X is optimized so as to apply to all pavement sections in
the network. The use of site characteristics makes this approach site-specific,
thus enhancing its applicability throughout the pavement network.
FIG. 3Example of a logistic model probability function using center deflec-

tion as the deflection parameter.
In the proposed stochastic approach, the probability density function is
used to determine the thresholds that define the structural condition. By defini-
tion, the cut-off value in a logistic model determines how one will convert the
continuous probability prediction as a dichotomous outcome (i.e., predicted
probabilities above the threshold are defined as acceptable, and others are
defined as not acceptable). In addition to the cut-off threshold, other intermedi-
ate cut-off values can be used to further subdivide the structural condition. As
an example, Fig. 3 describes a second threshold arbitrarily assigned at a 0.4
probability level that can be used to differentiate sections with a fair struc-
tural condition from those with a poor one. (Note that this definition is arbi-
trary and can be adjusted to better fit an agencys rehabilitation practices or
network characteristics.)
Once thresholds and their respective deflection parameters are identified, a
structural decision matrix (also referred to as structural rehabilitation matrix)
can be generated. Table 1 shows an example of a structural decision matrix
obtained from the probability density function shown in Fig. 3.
The advantages of using the stochastic process to evaluate the structural
condition of the pavement for network-level analyses are as follows:
The structural condition is related to the performance through the use of
a probability density function.
TABLE 1Example of structural decision matrix.
Structural Condition Deflection Parameter

Good <200
Fair 200350
Poor >350
The probability density function is based on qualitative performance

measures, i.e., acceptable and not acceptable.
The stochastic model is site-specific (i.e., additional variables are incor-
porated that reflect important site characteristics, such as traffic, pave-
ment structure, etc.).
Different structural decision matrices can be defined based on different
types of distress. Therefore, the final evaluation can be made based on
the most critical or the most typical distress type.
The calculations are simple and do not require an interactive process.
Therefore, the logistic model can be automated and incorporated in vir-
tually any PMS in which deflection data are or will soon become
available.
Logistic models can be calibrated locally in order to enhance their good-
ness of fit and the accuracy of their predictions. Depending on the histor-
ical availability of deflection data and pavement distress types, local
calibration can begin immediately if both datasets are available. If not,
several years of gathering ongoing deflection and performance data will
be needed before local calibration can take place.
Despite the preceding point, even incomplete deflection data obtained
from project-level analysis or quality control (Q/C) after construction
can be used in the local calibration process.
If local calibration cannot take place, the relationships derived from
the Long-Term Pavement Performance (LTPP) nationwide database
during this research should be used as soon as deflection data (even
limited data) are available, along with the other requisite input
variables.
Model Calibration
The model calibration was performed using the nationwide LTPP database.
The evaluation of applicable deflection techniques was based on the quality
and accuracy of the corresponding logistic model developed. A selection pro-
cess was established to evaluate each model within each distress group and
pavement type (e.g., roughness in flexible pavements). The evaluation criteria
considered the accuracy of the logistic model, its consistency with engineering
principles, and errors in the predictions (error types I and II). The entire analy-
sis process used is based on the national database for this project and is sum-
marized by the following steps:
1. Data preparation:
a. Compile data including performance measurements, deflections,
pavement structure, climate, and traffic for each eligible site in
the LTPP database.
b. Compute deflection techniques as selected.
c. Rate performance over time as acceptable or not acceptable

according to load-related distresses.
d. Prepare subset of the data for model validation.
2. Evaluation of selected techniques based on observed performance:
a. Conduct a preliminary screening to sort out the best deflection
techniques as potential candidates for logistic model development.
b. Develop and calibrate a logistic model using the extensive
LTPP analysis database. (One model was developed for each
load-related distress type using the best predictive deflection-
based parameter for each critical distress type.)
3. Sensitivity analysis of model parameters
4. Validation analysis
The following sections describe in greater detail the steps used in the anal-
ysis process. Three models were developed for flexible pavements: roughness,
rutting, and fatigue cracking. The example provided refers to the development
of the flexible pavement roughness model.
Data Preparation
The data selected for this study were retrieved from the LTPP database. Basic
statistics on the final dataset used are presented in Table 2. The dichotomous
performance variable was determined based on the expected service life of
flexible and rigid pavements, arbitrarily defined as 15 years. The average
length of service when the first deflection measurement was taken was 2 years.
The LTPP sections examined for this study were still in service and were at an
average service life of 11.3 years when the last performance measurement sur-
veys were conducted. As a requirement of the logistic models assumption for
acceptable performance, all pavement sections had to be evaluated at the end
of the chosen expected service life. Therefore, minor extrapolations of the per-
formance curves were computed in order to achieve this requirement. Extrapo-
lations were computed on a section-by-section basis by fitting non-linear
regression models to the historical data.
The threshold for acceptable performance was defined based on the Mecha-
nistic-Empirical Pavement Design Guide (ME-PDG) pavement performance
threshold values for acceptable terminal distress levels, described as follows:
Roughness: 172 in./mile
Rutting: 0.5 in.
Fatigue cracking: 25 % of total lane area
The acceptable level for each distress can also be defined based on the net-
work characteristics and the agencys practice for defining rehabilitation and
maintenance timing, which would provide the agency with greater flexibility to
adjust the technique based on its network characteristics and preferred M&R
timing. One section would be classified as exhibiting acceptable performance
TABLE 2Input LTPP data summary for flexible pavement sections.
Average Standard Deviation Minimum Maximum Median

Service life used at 1.5 1.4 0.0 10.5 1.0
the beginning of the
analysis period
Annual 843.9 442.2 73.2 2153.3 876.9
precipitation, mm
Annual minimum 6.4 5.9 7.0 20.8 6.7
temperature, C
Annual maximum 19.6 6.2 3.8 31.3 20.3
temperature, C
Annual mean 13.0 6.0 1.6 25.6 13.3
temperature, C
Average annual daily 723.8 728.7 7.3 3955.0 454.7
truck traffic (AADTT)
Total thickness, in. 22.1 9.4 6.3 85.5 20.3
AC thickness, in. 7.3 3.5 0.9 23.2 7.1
Class 5 (volume) 146.8 141.7 2 1770.0 106.3
Class 9 (volume) 401.7 554.6 1 2838.0 175.9
Fine Grain Soil (1) Coarse Grain Soil (0)

Subgrade type 32.6 % 67.4 %
Bound Base Layer (1) Unbound Base Layer (0)

Pavement type 42.3 % 57.7 %
for a given distress if the deterioration were lower than or equal to the threshold
at 15 years for flexible pavements and 20 years for rigid pavements.
The final step in data preparation was the selection of a subset of data for
model validation. The selection was based on a stratified sampling technique.
The technique consisted of selecting a total of 12 % of the original dataset in
an automated random process designed to pick sections distributed throughout
a variety of site characteristics that would adequately represent the original
dataset. Climate conditions, traffic volume, pavement type (in terms of base
layer type), and subgrade type were used.
Evaluation of Selected Techniques Based on Observed Performance

A total of 51 deflection parameters were computed based on the literature
review findings. The first deflection survey available after construction/rehabil-
itation was used to compute these parameters. The latest performance measure-
ments available were used. This process was repeated for all load-related
distresses and for both pavement types evaluated.
The evaluation of the selected deflection techniques was based on the
logistic model approach and the notion that the best deflection techniques yield
the most accurate models. Because there were many deflection parameters to
be evaluated, a first screening was conducted in order to eliminate poor correla-

tions. A simple linear Pearson regression was used to test the level of correla-
tion between the measured performance and the deflection parameter in
question. The Pearsons R-squared and the significance level were used to iden-
tify good candidates for the full binary logistic analysis.
A statistical package, now called PASW Statistics (formerly known as
SPSS), was used to develop the logistic models. The process of developing each
model followed a backward elimination approach. All variables described in
Table 2 were used in the first step, along with the deflection parameter and the
independent performance variable (defined as acceptable or not acceptable).
The models exponent b was calibrated so as to minimize the FPRs and
false negative rates in the probability analysis. The significance of each vari-
able was computed, and the less significant variables were removed from the
calculation in the next step, thereby eliminating the variable with the poorest
relationship to performance. This process was continued until all remaining
variables had significance levels higher than 95 % or until the maximum num-
ber of iterations was achieved. With each step, the number of variables
decreases and the overall level of significance increases, indicating improve-
ment in the models capability of predicting the likelihood of acceptable per-
formance. This process was repeated for all deflection variables, resulting in a
single model for each performance measure.
Variables that resulted in unreasonable trends were also eliminated. An
example of this situation is a model in which pavement thickness had an
inverse trend with performancethicker pavements would have yielded a
lower probability of acceptable performance, which is contrary to expected
pavement performance behavior.
Sensitivity of Model to Deflection Parameter
A sensitivity analysis was necessary in order to evaluate whether the risk of
premature failure associated with the probability of acceptable (or not) pave-
ment performance would yield deflection parameter values that were reasona-
ble from an engineering standpoint. Two examples of sensitivityone
reasonable and one notare described below.
Figure 4 describes the sensitivity curve of the I2 model based on roughness
performance. I2 is a parameter calculated as the inverse of the deflection D2,
measured at 200 mm (8 in.) from the center of the load. In this case, the cut-
off was set at 0.812, which corresponds to I2 0.0043 1/lm (or D2 232 lm),
which is a very reasonable value for a well-designed pavement structure.
Figure 4 indicates that flexible pavements with D2 < 232 lm have a greater
than 80 % chance of exhibiting acceptable roughness performance for the next
12 years and therefore can be classified as in good structural condition.
In contrast, Fig. 5 shows the sensitivity curve of the subgrade model based
on roughness. In this case, the subgrade modulus was calculated using the
FIG. 4Sensitivity of roughness to acceptable probability levels based on

deflection parameter I2 for flexible pavements.
Hogg (subgrade stiffness) model. It can be observed in this figure that a cut-off
value of 0.802 yields a Hogg modulus value of 65 000 psi, which is very high
and unreasonable for most subgrade types. The Hogg model in this case is bi-
ased toward probabilities under the cut-off value, and consequently toward low
probabilities of acceptable performance.
Final Models Selected

The logistic model based on the deflection parameter I2 (Eq 3) was the one that
resulted in the best accuracy in predicting the probability of acceptable
FIG. 5Sensitivity of roughness to acceptable probability levels based on the

Hogg (subgrade stiffness) deflection parameter for flexible pavements.
TABLE 3Variables in the I2 final model based on roughness performance.
Variables (bi) ai p-value

I2 239.849 0.000005706
Current life 0.189 0.01174
Class 9 (truck) volume 0.0006781 0.0001037
Constant 0.8375 0.01071
Cut-off: 0.812
roughness performance for flexible pavements. Table 3 shows the final model
variables that fitted Eqs 1 and 2 with their respective p-values (i.e., significance
of the variable to the performance variable) and the corresponding cut-off
value that resulted in the highest accuracy.
1
I2 (3)
D2
in which D2 is the deflection 200 mm (8 in.) from the center of the load.
Roughness performance depends on the strength of the entire pavement
structure. For example, it is expected that a robust pavement section will tend
to be smoother for longer periods of time than a weak section, assuming both
have the same climate and traffic conditions and the same subgrade type.
Therefore, the predominance of the deflection variable in the model is justified.
It also makes technical sense that current life and traffic play a significant role
in estimating the likelihood of premature failure due to roughness
performance.
The logistic model based on the deflection parameter CI3 (Eq 4) resulted in
the best accuracy in predicting the probability of acceptable flexible pavement
rutting performance. Table 4 shows the model variables with their respective
p-values (i.e., significance of the variable to the performance variable) and the
corresponding cut-off value that resulted in the highest accuracy.
CI3 D3 D4 (4)
TABLE 4Variables in the CI3 final model based on rutting performance.

CI3 0.01146 0.00001594
Precipitation, mm 0.0005259 0.04018
Class 9 volume 0.0007688 0.00005421
Constant 2.6586 1.0136 1013
Cut-off: 0.792
where:
D3 deflection 300 mm (12 in.) from the center of the load, and
D4 deflection 450 mm (18 in.) from the center of the load.
The rutting model developed was based on deflections located near the
middle of the measured deflection basin. The influences of the base and sub-
base were primarily reflected by the deflections measured at these offset posi-
tions. The strength of the underlying layers is critical in the development of
rutting in the asphalt concrete layer, as well as in the underlying layers. One
important aspect of the rutting phenomenon is the influence of the subgrade. It
has been well documented that the subgrade usually contributes to the overall
magnitude of rutting and the rate of the development of rut depth over time.
However, the influence of the subgrade is not completely captured by the
deflections used in this model. During the model development, the type of sub-
grade (fine versus coarse, cohesive versus non-cohesive) was included as a
variable, but it was excluded during the calibration process. However, the influ-
ence of precipitation was apparent because it directly affected the apparent sub-
grade strength (or stiffness) more than that of any other unbound layer in the
pavement structure. Further improvements to this model at the local calibration
level should include the subgrade type and, if possible, the subgrade modulus
based on a non-linear model.
The logistic model based on the inverse of the center deflection (I1) (Eq 5)
was the one showing the best accuracy in predicting the probability of accepta-
ble fatigue cracking performance. Table 5 shows the model variables with their
respective p-values (i.e., significance of the variable to the performance vari-
able) and the corresponding cut-off value that resulted in the highest accuracy.
1
I1 (5)
D1
in which D1 is the deflection at the center of the load.

Fatigue cracking performance is influenced by the stiffness of the asphalt
concrete layer, the number of load repetitions, and the type/strength of the
underlying layers. Stiff asphalt concrete layers are more susceptible to cracking
TABLE 5Variables in the I1 final model based on fatigue cracking performance.

I1 154.764 0.0006015
Average annual daily truck traffic (AADTT) 0.0005073 0.0007847
Pavement type 0.3774 0.07598
Constant 0.2202 0.4422
Cut-off: 0.605
TABLE 6Statistics of calibrated models based on performance for flexible pavements.
Model Based on Roughness Rutting Fatigue Cracking

Deflection parameter I2 CI3 I1
Area ROC 0.716 0.659 0.620
HosmerLemeshow significance test 0.995 0.267 0.284
Calibration, correct cases, % 64.1 66.5 60.7
Validation, correct cases, % 69.4 66.1 52.5
Error type I (false acceptable), % 6.6 8.6 13.8
Error type II (false not acceptable), % 29.3 24.8 25.5
True positive rate (acceptable), % 64.0 68.5 56.2
True negative rate (not acceptable), % 64.4 59.5 66.9
because the binder hardens and becomes brittle, which in turn reduces its flex-
ural strength. The overall stiffness of the pavement can be captured in part by
the inverse of the maximum deflection, represented in this model as the vari-
able I1. The mechanism of crack initiation and propagation, regardless of where
it starts (bottom or surface), depends on the growing damage due to load repeti-
tions. The truck volume in the model adequately addresses this issue. Finally,
the type of the pavements base layer (unbound or bound) is related to the stiff-
ness of the underlying layers. Based on the data in the LTPP database, pave-
ments with bound base layers are less likely to develop fatigue cracking than
pavements with unbound base layers.
Table 6 provides a summary of all descriptive statistics calculated for each
model. The area under the ROC curve represents the accuracy of each model.
The HosmerLemeshow test is another way to evaluate the significance of the
model. This test is commonly used to evaluate binary logistic models when the
conventional Pearsons R-squared calculation is not possible. Error type I
(false acceptable) is the worst error type for the problem at hand. In such a
case, the model incorrectly shows that the pavement section has a good struc-
tural condition, when in fact the structural condition is just the opposite. From
an agencys perspective, this would be the worst possible outcome from the
model prediction.
The model developed based on roughness performance was the best over-
all. Some of the reasons for this outcome are that (a) there were more data
available for the roughness analysis than for the other two distresses consid-
ered; (b) the variability in roughness data measurement is significantly smaller
than for measurements of rutting and fatigue cracking; and (c) data were col-
lected from all pavements across the United States, with accompanying varia-
tions in climate and subgrade type, which affect rutting and fatigue cracking
performance more than roughness. In the case of a regional or statewide data-
set, four examples of local calibration using the LTPP database from four
different states (TX, MN, CA, and AL) were attempted, and the results showed
significant improvements in the accuracy of the flexible pavement model based

on fatigue cracking [1] in all four cases. In other words, once local calibration
can be carried out, the results obtained by any given department of transporta-
tion or relatively large-scale PMS program will be significantly improved
upon. The other distress types are just as likely to show similar (in fact, in
many cases spectacular) improvements in efficacy.
Validation Analysis
The validation analysis was performed on a subset of the database that was set
aside prior to model calibration. Twelve percent of the total sections available
for calibration were removed from the database. Becasue the LTPP database
covers a wide range of pavement sections, traffic levels, and environmental
conditions, a stratified random sampling technique was employed to retrieve
the sections.
The stratified random sampling technique consists of grouping all sections
in all possible combinations of the characteristics defined for the selection.
These characteristics were climate region, traffic, base layer type, and subgrade
type. After the groups are formed, the target percentage of samples to be
selected is randomly selected proportionally to the size of each group and its
participation in the entire dataset.
Table 7 provides the results of the validation analysis for each of the mod-
els developed. The results indicate that good accuracy was also observed in the
validation dataset, which suggests that the final model is capable of satisfacto-
rily identifying the structural condition of the pavement based on the expected
roughness and rutting performance, but the outcome is not as good for the fa-
tigue cracking model.
Comparison With ME-PDG Predicted Performance

The ME-PDG performance prediction models were used to compare the pre-
dicted distress performance at the end of the design life with the estimated
structural condition obtained from the logistic structural model. This task was
accomplished by applying the ME-PDG models with the mechanistic responses
calculated using Kenpave [12].
TABLE 7Validation results for the structural logistic model based on performance for flexible
pavements.
Model Based on Roughness Rutting Fatigue Cracking

Calibration, correct cases, % 64.1 66.5 60.7
Validation, correct cases, % 69.4 66.1 52.5
FIG. 6Sensitivity of the I1 structural logistic model based on fatigue crack-

ing performance to the ME-PDG predicted fatigue cracking for flexible
pavements.
Two models were considered for flexible pavements: rutting and fatigue
cracking. At first, the roughness model was also included in the scope. However,
this particular model requires additional site factors that often dominate the anal-
ysis. Therefore, a comparison of the outcomes of the structural logistic model
based on roughness and predicted roughness performance was not included in
this exercise. The calculations were performed in Microsoft Excel and not within
the ME-PDG software because of the computational time required in order to
run all scenarios. Given that the objective is the comparison and trend evaluation,
this assumption has no impact on the outcome or conclusions.
FIG. 7Sensitivity of the CI3 structural logistic model based on rutting per-
formance to the ME-PDG predicted rutting for flexible pavements.
As an example, Fig. 6 describes the sensitivity of the I1 structural logistic

model based on fatigue cracking to the ME-PDG predicted fatigue cracking
performance for flexible pavements with two different subgrade stiffnesses. It
can be seen that the trends agree with expectations: high probabilities of an
adequate structural condition are associated with lower levels of fatigue crack-
ing. As the probability decreases, the predicted pavement area with fatigue
cracking increases, as expected.
Figure 7 describes the sensitivity of the CI3 structural logistic model based
on rutting to the ME-PDG predicted asphalt concrete (AC) rutting performance
for flexible pavements with three different subgrade stiffnesses. The trends
agree with expectations: high probabilities of adequate structural condition are
associated with smaller permanent deformations. As the probability decreases,
the predicted AC rutting increases, as expected.
Conclusions
The primary objectives of this study were to develop and verify simplified
deflection-based analytical techniques suitable for the rapid automated screening of
pavement structural capacity for inclusion in a network-level analysis, typically
referred to as a pavement management system (PMS). The Long-Term Pavement
Performance (LTPP) database was the primary source of information for this study.
A probabilistic approach was developed with which to determine the likeli-
hood of premature pavement failure using simplified and easy-to-apply load-
deflection analysis techniques. Premature failure was defined by the presence
of excessive distress occurring prior to the end of the design life of the pave-
ment section (i.e., levels of distress higher than the design threshold). This was
achieved by adopting binary logistic models that utilized deflection techniques
derived from each FWD deflection basin coupled with various site-specific pa-
rameters. Models were created based on performance measured in terms of
roughness, rutting, and fatigue cracking.
The advantages of the stochastic approach for evaluating the structural
condition of the pavement for network-level analysis can be summarized as
follows:
It provides a direct link to pavement performance by estimating the like-
lihood of the development of load-related distresses in the early stages

of the pavements service life. The intent was to have not a regression-
based predictive model based on deflection data, but rather a simplified
procedure for identifying critical sections with a high (stochastic) likeli-
hood of developing structural distresses prematurely.
The probability density function is based on qualitative performance
measures (acceptable and not acceptable).

Rating criteria can be created in a manner equivalent to the functional
decision matrix. Different tiers of structural quality can be defined based

on deflection thresholds and the likelihood of acceptable performance.

The structural decision matrix can then follow rating criteria similar to
those of the functional decision matrix, which facilitates implementation
in existing PMS algorithms.
The stochastic model is site-specific (i.e., additional variables may be
incorporated that reflect site characteristics, such as traffic, pavement

structure, and climatic conditions).
Different structural decision matrices can be defined based on different
types of distresses. Therefore, the final evaluation can be made based on

the most critical or most typical distress(es) observed in the pavement
network.
The calculations are simple and do not require an interactive, labor-
intensive process. Therefore, these simplified data can be automated and

incorporated into virtually any PMS in which deflection data are already
or can eventually become available.
Deflection data obtained from project-level analyses or Q/C after con-
struction can also be used as an agencys initial input data so that it can
begin taking advantage of the addition of a structural component within
the current PMS.
The probabilistic models can be locally calibrated to reflect an agencys
network characteristics (e.g., typical surface distresses observed, the

interval between M&R projects, and the threshold of acceptable limits
for distresses). In addition, the models can be recalibrated every time a
new FWD testing campaign is completed and new load-deflection data
become available.
Deflection data analysis provides both qualitative and quantitative assess-
ments of the structural integrity and bearing capacity of a pavement. Pavements
with poor structural quality are more likely to develop distresses prematurely.
Identifying pavements with inadequate structural condition is very important
for preventing the premature and rapid development of load-related distresses.
The results from this study can be readily implemented in national, state, and
local municipality systems alike.
Acknowledgments
The writers would like to acknowledge the FHWA LTPP program for sponsor-
ing this study, which was conducted under the LTPP Data Analysis project.
References
[1] Carvalho, R. L., Stubstad, R. N., Briggs, R., Selezneva, O., Mustafa, E.,
and Ramachandran, A., Simplified Techniques for Evaluation and Inter-
pretation of Pavement Deflections for Network-Level Analysis, Final
Report Submitted to the Federal Highway Administration (FHWA),

[2] Briggs, R., Project Report, Report Prepared for the Alaska Department
of Transportation by Dynatest Consulting, Inc., Personal Communication,
2011.
[3] Scullion, T., Incorporating a Structural Strength Index into the Texas
Pavement Evaluation System, Final Report No. FHWA/TX-88/409-3F,
Federal Highway Administration, Washington, D.C., 1988.
[4] Perrone, E., Dossey, T., and Hudson, W. R., Network Level Deflection
Data Collection for Rigid Pavement, Interim Report, Texas Department
of Transportation, Austin, TX, 1994.
[5] Zhang, Z., Claros, G., Manuel, L., and Damnjanovic, I., Development of
Structural Condition Index to Support Pavement Maintenance and Reha-
bilitation Decisions at Network Level, Transportation Research Record.
1827, Transportation Research Board, Washington, D.C., 2003, pp.
1017.
[6] Zhang, Z., Manuel, L., Damnjanovic, I., and Li, Z., Development of a
New Methodology for Characterizing Pavement Structural Condition for
Network-Level Applications, Final Report No. FHWA/TX-04/0-4322-1,
Federal Highway Administration, Washington, D.C., 2003.
[7] Diefenderfer, B. K., Network-Level Pavement Evaluation of Virginias
Interstate System Using the Falling Weight Deflectometer, Report VTRC
08-R18, Virginia Transportation Research Council, Charlottesville, VA,
2008.
[8] Horak, E., Benchmarking the Structural Condition of Flexible Pave-
ments With Deflection Bowl Parameters, Journal of the South African
Institution of Civil Engineering, Vol. 50, No. 2, 2008, pp. 29.
[9] Airport Cooperative Research Program (ACRP), ACRP Report 3Anal-
ysis of Aircraft Overruns and Undershoots for Runway Safety Areas,
ACRP, Washington, DC, 2008.
[10] Hosmer, D. and Lemeshow, S., Applied Logistic Regression, John Wiley
& Sons, Hoboken, NJ, 2000.
[11] Fawcett, T., An Introduction to ROC Analysis, Pattern Recogn. Lett.,
Vol. 27, 2006, pp. 861874.
[12] Huang, Y. H., Pavement Design and Analysis, 2nd ed., Prentice-Hall,
Upper Saddle River, NJ, 2003.
STP 1555, 2012
DOI:10.1520/STP104106
Daba S. Gedafa,1 Mustaque Hossain,2 Richard Miller,3

and Douglas A. Steele4
Surface Deflections of Perpetual Pavement

Sections
REFERENCE: Gedafa, Daba S., Hossain, Mustaque, Miller, Richard, and

Steele, Douglas A., Surface Deflections of Perpetual Pavement Sections,
Pavement Performance: Current Trends, Advances, and Challenges on De-
cember 5, 2011 in Tampa, FL; STP 1555, B. Choubane, Editor, pp. 192204,
2012.
ABSTRACT: Structural evaluation provides valuable information at the net-
work level of pavement management for project evaluation and prioritization
purposes. However, due to the expenses involved in data collection and anal-
ysis, structural capacity is usually not evaluated at the network level. A
rolling-wheel deflectometer (RWD), which measures surface deflections at
highway speed, can be an alternate, faster method of pavement deflection
evaluation for network-level data collection. This study was initiated in order
to assess the feasibility of using an RWD for network-level pavement deflec-
tion measurements. RWD deflection data were collected under an 80 kN axle
load and at a highway speed of about 88 km/h on four perpetual pavement
sections in Kansas. Falling-weight deflectometer (FWD) data were also col-
lected concurrently. The FWD first-sensor deflections were normalized to a
load of 40 kN and then corrected to a temperature of 20 C. The RWD deflec-
tions were also corrected for temperature. FWD and RWD center deflections
Manuscript received June 3, 2011; accepted for publication March 13, 2012; published online
October 2012.
1
Ph.D., P.E., Assistant Professor, Dept. of Civil Engineering, Univ. of North Dakota, 243
Centennial Dr., Stop 8115, Upson II Room 260K, Grand Forks, ND 58202-8115,
e-mail: daba.gedafa@engr.und.edu
2
Ph.D., P.E., Professor, Dept. of Civil Engineering, 2118 Fiedler Hall, Kansas State Univ.,
Manhattan, KS 66506, e-mail: mustak@ksu.edu
3
P.E., Assistant Geotechnical Engineer, Kansas Dept. of Transportation, Materials & Research
Center, 2300 Van Buren, Topeka, KS 66611, e-mail: rick@ksdot.org
4
P.E., Applied Research Associates, Inc., 100 Trade Centre Dr., Suite 100, Champaign, IL
61820, e-mail: dsteele@ara.com
Copyright V
19428-2959.
192
GEDAFA ET AL., doi:10.1520/STP104106 193
were then compared statistically. The results show that the deflections meas-
ured by RWD and the center (first sensor) deflections from the FWD are stat-
istically similar. Furthermore, there is a linear relationship between FWD and
RWD center deflections for all sections.
KEYWORDS: perpetual pavement, falling-weight deflectometer (FWD),
rolling-wheel deflectometer (RWD), pavement temperature
Introduction
Increasing traffic volumes and loads, as well as the public expectation of a
longer-lasting transportation infrastructure, have necessitated the design of
flexible pavements with longer lives, sometimes up to 50 years. Perpetual pave-
ments are expected to meet the need for such a longer-lasting pavement. Thick
and stiff pavement layers in perpetual pavements reduce the strains and stresses
at the bottom of the asphalt concrete layer, thereby reducing the potential for
cracking to be initiated at the bottom [1].
Surface deflection testing is now routinely used for pavement structural
evaluation. Although widely used at the project-level, deflection testing at the
network-level can identify the beginning and end of management sections and
group pavement sections with similar structural capacities for condition predic-
tion and remaining service life estimation. However, due to the expenses
involved in data collection and analysis, deflection testing is not routinely done
at the network-level. Haas et al. [2] argue that structural capacity information,
even derived from less intensive sampling than used for project-level purposes,
can be very useful at the network-level for project prioritization purposes.
Problem Statement
Currently, the falling-weight deflectometer (FWD) is the most popular device
for project-level deflection testing. During the past decade, many highway
agencies adopted the FWD as a tool for assessing the structural adequacy of
pavements at the network-level [3]. The Kansas Department of Transportation
(KDOT) owns and operates two FWDs. Currently, each unit is capable of
testing up to 32 lane-km in a ten-hour day during a deflection testing period
that runs from April through October. At this production level, in order to test
the entire network (17 898 lane-km) annually, 200 days of testing (excluding
travel time) would be necessary for the whole network. An alternate, faster
method of deflection testing that can be used on the whole network or on a
representative sample of the network is needed. The rolling-wheel deflectom-
eter (RWD), which measures surface deflections at highway speed, appears to
be very promising for this purpose. Thus, this study was initiated in order to
assess the feasibility of using an RWD for deflection measurements at the net-
work-level.
Study Objectives
The main objectives of this study were as follows:
1. compare FWD and RWD center deflections, and
2. investigate the relationship between FWD and RWD center deflections.
Test Sections
Four sections of the perpetual asphalt pavement project on US-75 near Sabetha,
KS, were selected as test sections. Table 1 shows the pavement structures. The
experiment involved the construction of four thick-pavement structures with
layer thicknesses close to those recommended by KDOTs structural design
method for flexible pavements based on the 1993 American Association of State
Highway and Transportation Officials (AASHTO) Design Guide [4]. The Kan-
sas Asphalt Pavement Association (KAPA) provided the thickness designs for
sections 1, 2, and 3. Section 4 was designed by KDOT. Sections 1 and 3 have
the same thickness, but a softer binder was used in the construction of the base
asphalt mix (PG 64-22 instead of PG 70-22) of section 3, and a richer and more
ductile Superpave mix (SM) was used in the bottom lift of the base layer in that
section. Section 2 is a thin section (287.5 mm) with a predicted fatigue life of
30 106 equivalent single-axle loads per lane. This corresponds to a reliability
level of 85 %, and thus section 2 is referred to as the high-reliability section.
Data Collection
Deflection Data
Falling-Weight Deflectometer Deflection DataFWD testing has been

established worldwide as an effective tool for measuring surface deflections in
the pavement evaluation process. The FWD is a trailer-mounted device that
transmits a load to the pavement surface through a 300 mm diameter circular
TABLE 1Configuration of experimental sections [4].
Section 1 2 3 4
Acronym KAPA (Standard) High Reliability KAPA 2 (Modified) KDOT
Surface course 40 mm, SM 9.5 A (PG70-28)
Binder course 60 mm, SM 19 A (PG70-28)
Base course 225 mm, 187.5 mm, 225 mm, 300 mm,
SM 19A SM 19A SM 19A SM 19A
(PG70-22) (PG64-22) (PG64-22)a (PG64-22)
Stabilized subgrade 150 mm, 6 % hydrated lime mixed with natural soil
Natural subgrade High plasticity clay (A-7-6)
a
Bottom 3 in. designed at 3 % air voids for a binder-rich layer (Pb 6.0 %, design air voids 3 % 6 2 %;
VFA (voids filled with asphalt) 77 %).
plate. A mass is dropped onto the plate with a rubber pad, generating an
impulse load on the pavement that is similar to the stress pulse generated by
moving trucks. The magnitude of the FWD force on the pavement can be var-
ied by altering either the mass of the drop weight or the drop height. The peak
force and maximum deflections at various points along the surface are respec-
tively measured by load cells and velocity transducers, commonly known as
geophones. A single analog integration of a signal generates a deflection-time
trace. Deflection measurements are recorded by a data acquisition system, typi-
cally located in the tow vehicle. FWD deflection data used in this study were
collected with a Dynatest 8000 FWD. The tests were done at 15 m intervals.
Rolling-Wheel Deflectometer Deflection DataAn RWD is an innovative

device for the efficient, fast determination of pavement surface deflection at
highway speed. The current prototype was developed jointly by the Federal
Highway Administration Office of Asset Management and Applied Research
Associates (ARA) [5]. An RWD consists of a set of four triangulation lasers
attached to an aluminum beam mounted beneath a custom-designed 16 m long
trailer. The lasers have an accuracy of 0.2 % or better. The trailer is sufficiently
long to isolate the deflection basin produced by the RWD trailers 80 kN dual-
tire single-axle from deflections produced by the RWD tractor.
An RWD uses a spatially coincident methodology for measuring pave-
ment deflections. Three lasers placed in front of the loaded axle are used to
define the unloaded pavement surface profile, and a fourth laser (D) placed
between the dual tires measures the deflected pavement surface. Deflection is
calculated by comparing the undeflected pavement surface profile with the
deflected pavement profile at the same location. This technique was originally
developed by the Transportation and Road Research Laboratory and was furth-
ered by Harr and Ng-A-Qui [6].
In this study, RWD deflection data were collected under an 80 kN axle
load and at a highway speed of about 88 km/h. Data were collected continu-
ously over entire test sections, but they have been averaged and reported so
that they match FWD deflection data collected at 15 m intervals. The RWD
made 16 passes (15 northbound and 1 southbound) over an approximately 2.4
km long test section. The RWD tests were done on the same day as the FWD.
Gedafa et al. [7] have shown that RWD yields repeatable deflection data based
on an analysis of data from 16 passes.
Temperature Data
Pavement surface temperatures, in conjunction with the air temperature and
asphalt concrete (AC) layer thickness, are needed in order to adjust in situ
deflections to a standard temperature of 20 C. The pavement surface tempera-
ture was measured with an FWD infrared sensor.
The pavement mid-depth temperature was measured on all sections except

section 3. The thermocouple of a temperature gauge was lowered into holes
drilled in the AC layers and filled with oil to measure the temperature at the
mid-depth of each AC layer.
Data Analysis
Pavement Temperature and Deflection Correction

The most important environmental factor affecting surface deflections is the
temperature of the AC layer [810]. Because FWD and RWD testing are per-
formed at different temperatures, deflections need to be adjusted to a standard
temperature to allow for a direct comparison.
In order to determine the corrected deflection, a two-step correction proce-
dure is needed. Typically, the first step consists of measuring/estimating the
effective temperature of the AC layer, and the second step consists of adjusting
the measured deflection to a reference temperature using a correction factor.
This two-step procedure was followed in this study.
Asphalt Concrete Layer TemperatureThree different pavement tempera-

ture prediction equationsBELLS3, Leland, and Watsonwere used to iden-
tify the equation that gives a pavement temperature close to the measured
temperature for all sections. After the best equation was identified, it was used
to estimate the mid-depth pavement temperature for section 3.
Using measured pavement temperatures from the Long Term Pavement
Performance database, the BELLS equation (so named for the first letters of
the last names of the authors who developed the equation: Baltzer, Ertman-
Larson, Lukanen, and Stubstad) was developed for predicting the one-third-
depth temperature [11]. A third model, BELLS3, was subsequently developed
for use during routine FWD testing when the pavement surface is typically
shaded for less than a minute. The BELLS3 model has been used in this study
to calculate the mid-depth pavement temperature [12].
Td 0:95 0:892Ts log d 1:25

A B
1:83 sin 2p 0:448Ts 0:621Tavg 0:042Ts sin 2p
18 18
(1)
where:
Td pavement temperature at layer mid-depth, C,
Ts infrared surface temperature, C,
Tavg average of high and low air temperatures on the day before testing,

C, and
d layer mid-depth, mm.
A and B are computed as follows:
8 8
< td 9:5 if 0 td < 5 < td 9:5 if 0 td < 3
A 4:5 if 5 td < 11 ; B 4:5 if 3 td < 9
: :
td 15:5 if 11 td < 24 td 13:5 if 9 td < 24
where td time of day in decimal hours.

The last two variables are used as arguments to a pair of sine functions
with 18-hour periods and 15.5-hour and 13.5-hour phase lags, respectively.
One cycle per day is allowed. During the other six hours of the day, A and B
are set equal to 4.5 so that the sine functions return a value of 1.
Leland et al. [13] developed Eq 2 using the quasi-Newton method of nu-
merical optimization SYSTAT, a statistical computer program, based on the
pavement surface temperature, pavement depth, and time of FWD testing as
obtained from the Michigan Department of Transportation database.
Tz Tsurf 0:345 0:0432z2 0:00196z3 sin6:3252t 5:0967 (2)
where:
Tz AC pavement temperature at depth z, C,
Tsurf AC pavement temperature at the surface, C,
z depth at which temperature is to be determined, cm, and
t time at which AC surface temperature was measured, days [0 < T < 1
(e.g., 1:30 P.M. 13.5/24 0.5625 days)].
Watson et al. [14] developed an equation using data from the Strategic
Highway Research Program temperature database. The equation allows the
estimation of pavement temperature at any depth based on surface temperature
data.
Td Tsurf 1 0:063d 0:007d 2 0:0004d 3 (3)
where:
Td pavement temperature, F,
Tsurf surface temperature, F, and
d pavement depth, in.
Deflection CorrectionAASHTO [15] recommends a linear relationship

between the temperature and the temperature-correction factor. The relation-
ship is presented in a nomograph form: the slopes of the lines depend on the
thickness of the asphalt layer. The guide recommends two different nomo-
graphs, one for pavements with granular and asphalt-treated bases and another
for pavements with cement and pozzolan-treated bases.
In this study, the FWD first sensor deflection values were normalized to the
40 kN load level and then corrected to a temperature of 20 C using the
AASHTO method [15]. RWD deflection data were also corrected to the same
reference temperature using the same method.
Significant Difference Test

Paired t-tests were done to determine statistical differences in this study. Paired
data arise when two dependent samples are observed using two measurement
methods on a subject, such as the FWD and RWD tests on a pavement. The hy-
pothesis test is performed on the difference of the means of the two samples.
The default value is zero, which results in a test in which the two population
means are equal. Statistical analysis software (SAS) was used to do paired
t-tests for mean differences at a 5 % level of significance.
Linear Regression
Linear regression analysis was conducted for center deflection to determine
whether FWD center deflection can be predicted from RWD center deflection,
and vice versa. RWD deflection was taken as the dependent variable in this
analysis. SAS was used to perform linear regression analysis at a 5 % level of
significance.
Results and Discussion
Comparison of Measured and Estimated Pavement Temperature

The pavement temperature was not measured on section 3, and a comparison
was made only for estimated pavement temperatures. The mid-depth pavement
temperature was estimated for surface, binder, and base layers using BELLS3,
Leland, and Watson equations for all sections. A weighted-average type equa-
tion was used to estimate the mid-depth temperature for the total AC layer
based on the measured and/or estimated pavement temperature in each layer
(Eq 4)
P
n
di Ti
i1
Tavg (4)
Pn
di
i1
where:
di thickness (depth) of layer i,
Ti temperature of layer i, and
n number of layers (three layers: surface, binder, and base).
The lowest to the highest mid-depth pavement temperatures were obtained
from Watson, Leland, and BELLS3 and measured for all sections as shown in
Fig. 1. The BELLS3 equation gives the mid-depth pavement temperature that
is closest to the measured one, and it is recommended that this equation be
used when there is no measured mid-depth pavement temperature. The mid-
depth temperature estimated using BELLS3 was used to correct the deflection
to 20 C for section 3, and measured mid-depth pavement temperatures were
used for other sections.
Significant Difference Test

A significant difference test was done for FWD and RWD center deflections
for all sections. The average RWD deflection is slightly higher than the average
center deflection of the FWD in all test sections except section 3, as shown in
Fig. 2. The minimum FWD center deflection is slightly lower for sections 1
and 4 and higher for sections 2 and 3 than the minimum RWD center deflec-
tion. The maximum FWD center deflection is slightly higher than the maxi-
mum RWD center deflection for all sections. This difference is not significant
from a practical point of view.
FIG. 1Comparison of estimated and measured mid-depth pavement

temperatures.
FIG. 2Comparison of FWD and RWD center deflections.
Table 2 shows significant difference test results. There is no significant dif-

ference between FWD and RWD center deflections, as the p-value is higher
than 0.05 for all sections. The mean difference between FWD and RWD
deflections is positive only for section 3, which shows that the RWD center
deflection is slightly higher than the FWD center deflection for other sections.
The number of data points (n) is 20 for all sections except section 1, which has
only six data points. These results tend to indicate that an RWD can be a valua-
ble tool for collecting deflection data at the network-level.
Linear Regression
RWD deflection was taken as a dependent variable in this analysis. SAS was
used for linear regression at a 5 % significance level. The RWD center deflec-
tion was averaged to match FWD center deflections measured at 15 m inter-
vals. Figure 3 shows the scatter plot for all sections. The equation in each
figure shows the linear regression equation with the intercept, slope, and
TABLE 2Significant difference test for FWD and RWD center deflection.
Average d0 Minimum d0 Maximum d0 t-test (FWD-RWD)
Section FWD RWD FWD RWD FWD RWD Mean Pr > jtj Similar? n
1 0.146 0.151 0.111 0.140 0.189 0.174 0.006 0.49 Yes 6
2 0.207 0.214 0.160 0.154 0.272 0.267 0.008 0.11 Yes 20
3 0.126 0.125 0.104 0.096 0.216 0.231 0.002 0.80 Yes 20
4 0.099 0.102 0.089 0.090 0.119 0.116 0.003 0.08 Yes 20
FIG. 3Scatter plots for FWD and RWD center deflection.
TABLE 3FWD and RWD center deflections (mm).
Section 1 Section 2 Section 3 Section 4
Station, m FWD RWD FWD RWD FWD RWD FWD RWD

0 0.117 0.140 0.233 0.267 0.141 0.110 0.095 0.116
15 0.111 0.140 0.200 0.237 0.112 0.109 0.095 0.097
30 0.156 0.143 0.208 0.217 0.104 0.096 0.094 0.098
45 0.144 0.152 0.204 0.216 0.106 0.100 0.105 0.098
60 0.159 0.160 0.215 0.250 0.110 0.104 0.092 0.091
75 0.189 0.174 0.272 0.249 0.115 0.100 0.100 0.096
90 ... ... 0.224 0.219 0.114 0.102 0.098 0.104
105 ... ... 0.187 0.191 0.109 0.120 0.104 0.101
120 ... ... 0.189 0.221 0.137 0.107 0.099 0.090
135 ... ... 0.213 0.230 0.112 0.103 0.090 0.096
150 ... ... 0.247 0.247 0.113 0.112 0.096 0.100
165 ... ... 0.224 0.223 0.132 0.098 0.101 0.100
180 ... ... 0.190 0.192 0.110 0.118 0.102 0.107
195 ... ... 0.208 0.212 0.117 0.129 0.095 0.113
210 ... ... 0.187 0.217 0.148 0.122 0.097 0.100
225 ... ... 0.184 0.206 0.133 0.115 0.099 0.112
240 ... ... 0.219 0.197 0.128 0.123 0.119 0.115
255 ... ... 0.190 0.182 0.119 0.205 0.108 0.115
270 ... ... 0.183 0.154 0.156 0.231 0.107 0.105
285 ... ... 0.160 0.160 0.216 0.190 0.089 0.091
coefficient of determination (R2). There is no significant scatter for section 1,

as shown in Fig. 3(a), and the coefficient of determination (R2 0.78) is also
relatively high. Figure 3(b) shows the FWD and RWD deflections for section
2, which shows some scatter. Figures 3(c) and 3(d) show some scatter and out-
liers for sections 3 and 4, respectively. The increase in variation might be due
to a high test temperature, and FWD deflection data collection points might not
exactly match the points at which RWD deflections were averaged.
Table 3 shows raw FWD center deflections (normalized and temperature
corrected) collected at 15 m intervals and RWD center deflections (temperature
TABLE 4Linear regression of FWD and RWD center deflection.
Sections Intercept Slope Pr > jtj Linear Relation? R2 n

1 0.09 0.42 0.02 Yes 0.78 6
2 0.04 0.85 0.00 Yes 0.55 20
3 0.02 0.85 0.01 Yes 0.33 20
4 0.04 0.62 0.02 Yes 0.27 20
corrected) averaged at 15 m intervals to match the FWD center deflection.

These data were used for scatter plots (Fig. 3) and linear regression (Table 4).
There is a linear relationship between FWD and RWD center deflections
for all sections, and the p-value is smaller than 0.05, as shown in Table 4. The
slope and coefficient of determination (R2) values vary from 0.42 to 0.85 and
0.27 to 0.78, respectively. Section 1 has the smallest slope, but it is compen-
sated for by a high intercept. The increase in variation for sections 3 and 4
resulted in lower R2 values. These results indicate that in general, RWD deflec-
tion can be predicted from FWD deflection data, and vice versa.
Conclusions
Based on this study, the following conclusions can be made:
The BELLS3 equation yields a mid-depth pavement temperature that is
close to the measured value. It is recommended that the BELLS3 equa-

tion be used to estimate the mid-depth pavement temperature when there
is no measurement available.
There is no statistically significant difference between mean center
deflections from RWD and FWD deflection data. Thus, an RWD can be
used to do deflection surveys at the network-level.
There is a linear relationship between FWD and RWD center deflec-
tions. RWD deflection can be predicted from the FWD deflection data,
and vice versa.
Acknowledgments
The writers would like to thank KDOT for providing the FWD data needed for
this study and ARA for making RWD deflection measurements. The writers
gratefully acknowledge the sponsorship of this study by the Federal Highway
Administration (FHWA).
References
[1] Transportation Research Board, Transportation Research Circular 503:

Perpetual Bituminous Pavements, Transportation Research Board of the
National Academies, Washington, D.C., 2001.
[2] Haas, R., Hudson, W. R., and Zaniewski, J. P., Modern Pavement Man-
agement, Krieger, Malabar, FL, 1994.
[3] Damnjanovic, I. and Zhang, Z., Determination of Required Falling
Weight Deflectometer Testing Frequency for Pavement Structural Evalua-
tion at the Network Level, J. Transp. Eng., Vol. 132, No. 1, 2006, pp.
7685.
[4] Romanoschi, S. A., Gisi, A. J., Portillo, M., and Dumitru, C., First Find-
ings from the Kansas Perpetual Pavements Experiment, Transportation
Research Record. 2068, Transportation Research Board, Washington,
D.C., 2008, pp. 4148.
[5] Applied Research Associates, Rolling-wheel Deflectometer (RWD)
Demonstration for the Kansas Department of Transportation (KDOT),
Final Report, Kansas Department of Transportation, Topeka, KS, 2007.
[6] Harr, M. E. and Ng-A-Qui, N., Noncontact Nondestructive Determina-
tion of Pavement Deflection under Moving Loads, Report No. FAA-RD-
77-127, Federal Aviation Administration, Washington, D.C., 1977.
[7] Gedafa, D. S., Hossain, M., Miller, R., and Steele, D. A., Network-level
Testing for Pavement Structural Evaluation Using a Rolling Wheel
Deflectometer, J. Test. Eval., Vol. 38, No. 4, 2010, pp. 439448.
[8] Kim, Y. R. and Lee, Y.-C., Interrelationships Among Stiffnesses of
Asphalt-Aggregate Mixtures, J. Assoc. Asph. Paving Technol., Vol. 64,
1995, pp. 575609.
[9] Shao, L., Park, S. W., and Kim, Y. R., Simplified Procedure for Predic-
tion of Asphalt Pavement Subsurface Temperatures Based on Heat Trans-
fer Theories, Transportation Research Record. 1568, Transportation
Research Board, Washington, D.C., 1997, pp. 114123.
[10] Park, H. M., Kim, Y. R., and Park, S., Temperature Correction of
Multiload-level, Falling-weight Deflectometer Deflections, Transporta-
tion Research Record. 1806, Transportation Research Board, Washington,
D.C., 2002, pp. 38.
[11] Baltzer, S., Ertman-Larson, H. J., Lukanen, E. O., and Stubstad, R. N.,
Prediction of AC Mat Temperature for Routine Load/Deflection Meas-
urements, Proceedings of the 4th International Conference on the Bear-
ing Capacity of Roads and Airfields, Vol. 1, University of Minnesota,
Minneapolis, MN, 1994, pp. 401412.
[12] Federal Highway Administration, Temperature Predictions and Adjust-
ment Factors for Asphalt Pavement, Publication No. FHWA-RD-98-085,
Federal Highway Administration, Office of Research and Development,
McLean, VA, 2000.
[13] Leland, W., Hill, M., Welna, J. P., and Birkenbeuel, G. K., SYSTAT for
Windows: Statistics, Version 5 (1992). SYSTAT, Inc., Evanston, IL.
[14] Watson, D. E., Zhang, J., and Powell, R. B., Analysis of Temperature
Data for the National Center for Asphalt Technology Test Track, Trans-
portation Research Record. 1891, Transportation Research Board, Wash-
ington, D.C., 2004, pp. 6875.
[15] American Association of State Highway and Transportation Officials
(AASHTO), AASHTO Guide for Design of Pavement Structures,
AASHTO, Washington, D.C., 1993.
STP 1555, 2012
DOI:10.1520/STP104445
Linda M. Pierce,1 Kurt D. Smith,2 Jim E. Bruinsma,3

and Nadarajah Sivaneswaran4
Case Studies Using Falling Weight

Deflectometer Data with
Mechanistic-empirical Design
and Analysis
REFERENCE: Pierce, Linda M., Smith, Kurt D., Bruinsma, Jim E., and Siva-
neswaran, Nadarajah, Case Studies Using Falling Weight Deflectometer Data
with Mechanistic-empirical Design and Analysis, Pavement Performance:
Current Trends, Advances, and Challenges on December 5, 2011 in Tampa,
FL; STP 1555, B. Choubane, Editor, pp. 205218, doi:10.1520/STP104445,
ABSTRACT: The need to accurately characterize the structural condition of
existing pavements has increased with the recent development, release, and
ongoing implementation of the Mechanistic-Empirical Pavement Design Guide
(MEPDG), developed under National Cooperative Highway Research Program
Project No. 1-37A. A number of different material inputs are required in the pro-
cedure, and it is important that these be adequately characterized and defined
so that competent structural designs can be developed. The analysis of deflec-
tion data collected with a falling weight deflectometer (FWD) provides a quick
and reliable way of characterizing many of the parameters of the existing pave-
ment layers. This paper summarizes how deflection data are incorporated into
the MEPDG and describes two case studies, one with a flexible pavement and
one with a rigid pavement. Significant findings and recommendations from the
evaluated flexible pavement case study include the following: surface-down
cracking is critical in the design of the hot mix asphalt (HMA) overlay, correc-
tion factors should be used for adjusting backcalculated layer moduli to labora-
tory determined values, and an FWD testing frequency of 30 Hz should be
1
Applied Pavement Technology, Santa Fe, NM 87508.
2
Applied Pavement Technology, Urbana, IL 61801.
3
Applied Pavement Technology, Plainfield, IL 60544.
4
Federal Highway Administration, McLean, VA 22102.
Copyright V
19428-2959.
205
used for estimating the existing HMA modulus. For rigid pavements, the case
study found that the thinnest overlay produced from the MEPDG was a bonded
Portland cement concrete (PCC) overlay, whereas the HMA overlay was
unreasonably thick. Within the design procedure, the manually entered k-value
is used for unbonded and bonded jointed plain concrete pavements but
does not appear to be used by the program in the HMA overlay design. The
backcalculated dynamic (or static) elastic modulus should be used for the
PCC layer, and the dynamic k-value should be used for the supporting layers.
The backcalculated k-value representing the composite stiffness of all layers
beneath the slab does not appear to have a significant influence on the design
thickness for the pavement structure analyzed.
KEYWORDS: mechanistic-empirical pavement design, material properties,
falling weight deflectometer, rehabilitation
Introduction
Pavement deflection testing is a relatively quick and easy way to assess the
structural condition of existing pavement. The resultant deflection data can be
used in a number of ways, including in the design of structural overlays, in the
appraisal of seasonal variations in pavement response, in the assessment of the
structural variability along the project length, and in the characterization of
pavement layer parameters and subgrade support conditions. In addition, for
rigid pavements, deflection data can be used for assessing load transfer across
joints and cracks and to determine the presence and extent of underlying voids.
A number of different testing devices are available for obtaining pavement
deflection measurements; however, the falling weight deflectometer (FWD)
has become the primary worldwide standard.
The need to accurately characterize existing pavement structures has
increased with the development of the Mechanistic-Empirical Pavement
Design Guide (MEPDG), prepared under NCHRP Project No. 1-37A [1].
The MEPDG predicts pavement performance by simulating the expected
accumulated damage on a monthly or semi-monthly basis over the selected
design period. The amount of incremental damage is dependent on the
prevailing environmental conditions, changes in material properties, and
effects of traffic loading. The incremental damage is then converted to phys-
ical pavement distress and projected roughness levels using calibrated
models [1].
In this process, FWD deflection data can be used to characterize pavement
layer parameters using either forward- or backcalculation methodologies.
Though the MEPDG does not incorporate any deflection analysis routines, the
backcalculation results can be included in the material-characterization portion
of the MEPDG. With the release of the MEPDG, the need arose to determine
how FWD testing and analysis results are integrated, as well as how they influ-
ence the rehabilitation design process.
PIERCE ET AL., doi:10.1520/STP104445 207
Use of FWD Deflection Data in the MEPDG
Deflection-based input data are incorporated into the MEPDG for characterizing
the existing pavement layers for only the Level 1 rehabilitation module. Specifi-
cally, backcalculated results are used for the following pavement layers:
Flexible pavement
- Dynamic modulus of hot mix asphalt (HMA) layers

- Resilient modulus of chemically stabilized layers
- Resilient modulus of unbound base, subbase, and subgrade layers
- Elastic modulus of the bedrock layer (when present)
Rigid pavement
- Elastic modulus of the concrete and base layers

- Subgrade k-value
- Concrete flexural strength (determined from the backcalculated con-
crete elastic modulus)
In addition, the flexible pavement backcalculation results require the appli-
cation of an adjustment factor in order for the backcalculated modulus values
(which are considered dynamic values) to correspond to the laboratory-based
values (which are considered static values) that were used in the development
and calibration of the MEPDG performance models. The MEPDG allows for
the consideration of non-linear unbound layers; however, this functionality is
not recommended until the performance models have been calibrated for non-
linear material properties.
The MEPDG determines the existing HMA layer dynamic modulus by
means of the same procedures as used for a new layer, except that it accounts
for the damage incurred in the layer over the life of the pavement. From this, a
field damaged dynamic modulus master curve is developed and is deter-
mined as follows [1]:
1. FWD testing should be conducted in the outer wheel path. The HMA
layer temperature and the equivalent frequency for the FWD pulse
should be recorded at the time of FWD testing. The layer thickness of
the existing pavement structure should be determined along the project
length from coring or ground penetrating radar testing. The mean back-
calculated HMA modulus (Edam) should be calculated via the combining
of layers with similar materials (including cracked and uncracked
pavements).
2. Determine the mix volumetric parameters (air void content, asphalt con-
tent, and gradation) and asphalt viscosity parameters (regression inter-
cept [A] and regression slope of the viscosity temperature susceptibility)
from cores by following the same procedures as used for determining
binder viscosity-temperature properties for new or reconstruction design.
3. Develop an undamaged dynamic modulus (E*) master curve using the
modified Witczak equation and the data obtained from step 2. The
FIG. 1Illustration of HMA modulus correction [1].
undamaged master curve should be determined using the temperature

recorded in the field and at an equivalent frequency corresponding to
the FWD pulse duration (see Fig. 1).
4. Estimate the fatigue damage in the HMA layer (dac) by using the values
of Edam obtained in step 1 and E* obtained in step 3 in Eq 1
(d regression parameter)
10d E 10d
Edam (1)
1 e0:35 logdac
5. Calculate a0 (1 dac)a, in which a is a function of mix gradation

parameters.
6. Determine the field-damaged dynamic modulus master curve using a0
instead of a in the modified Witczak equation.
Chemically stabilized layers are characterized in Level 1 via standard
backcalculation procedures. In an overlay design, the backcalculated modulus
of the chemically stabilized layer is used as the layer modulus at the time when
the overlay is placed and reduced based on the current condition of the stabi-
lized layer and the pavement surface (see Fig. 2). The percentage of reflected
cracks and the amount of alligator cracking in both the chemically stabilized
layer and the pavement surface layer influence the condition of the chemically
stabilized layer.
Similarly, unbound layers are also characterized in Level 1 via standard
backcalculation procedures, which are adjusted to correspond with laboratory-
obtained values. MEPDG current guidance for adjustment factors includes 0.67
for granular bases and subbase and 0.40 for subgrade soils. Additional MEPDG
modulus ratio recommendations are shown in Table 1.
FIG. 2Illustration of damage estimation for chemically stabilized layers [1].
However, the MEPDG (version 1.003) does not allow the base modulus
to be changed in the Level 1 rehabilitation module; i.e., the base modulus
reverts back to the MEPDG material default value. For example, the default
modulus for A-1-b material is 262 MPa (38 000 lb/in.2), and estimated back-
calculated moduli of 72 MPa (10 400 lb/in.2) were entered into the software
computations; however, the default value of 262 MPa (38 000 lb/in.2) is used
in the analysis.
TABLE 1Backcalculated-to-laboratory modulus ratios [2].
Layer Type Layer Location Mean MR/ER Ratioa

Unbound granular base and Granular base/subbase between two 1.43
subbase layers stabilized layers (cementitious or asphalt
stabilized materials)
Granular base/subbase under a PCC layer 1.32
Granular base/subbase under an HMA 0.62
surface or base layer
Embankment and subgrade Embankment or subgrade soil below a 0.75
soils stabilized subbase layer or stabilized soil
Embankment or subgrade soil below a 0.52
flexible or rigid pavement without a
granular base/subbase layer
Embankment or subgrade soil below a 0.35
flexible or rigid pavement with a granular
base or subbase layer
a
MR, elastic modulus of the in-place materials determined from laboratory repeated load resilient modulus
tests; ER, backcalculated elastic modulus.
In order to illustrate how backcalculated layer parameters impact the

results of the MEPDG analysis, six case studies were evaluated as part of the
research study and examined the following pavement types [3]:
HMA pavement,
HMA pavement on rubblized concrete pavement,
concrete pavement on a granular base,
concrete pavement on a stabilized base,
concrete pavement on an existing HMA pavement, and
composite pavement (HMA overlay on concrete pavement).
The remainder of this paper summarizes only the HMA pavement and the
concrete pavement on granular base case studies in order to illustrate the use of
backcalculated layer parameters in the rehabilitation portion of the MEPDG.
Case Studies
Data from the Long-Term Pavement Performance (LTPP) program experimen-
tal sections were extracted and reviewed for use in each of the case studies.
Deflection data from the project sites were used to backcalculate layer parame-
ters and develop material inputs for the MEPDG design procedure. Details of
the case studies are provided in Volume 2 of the project report [4].
MEPDG version 1.003 was used in all evaluations, as were the nationally
calibrated performance models (i.e., no local calibration was conducted).
Hot Mix Asphalt Pavement Case Study

The HMA pavement section used for this case study included LTPP Section
30-0100. Details of the LTPP section include the following:
Climate zone: Dry-freeze (vicinity of Great Falls, MT)
Functional class: Rural principal interstate (Interstate 15)
Original construction: 1997
Pavement section: 102 mm (4 in.) HMA
- 203 mm (8 in.) aggregate base

- No (or deep) rigid layer
Subgrade classification: Sand (AASHTO A-2-6)
Rehabilitation and repair work: Crack sealing (2003)
- Aggregate seal coat (2004)

Distress: HMA rutting6.4 mm (0.25 in.)
- Granular base rutting: 5.1 mm (0.20 in.)

- Subgrade rutting: 3.8 mm (0.15 in.)
Overall pavement condition: Fair [for Level 3 rehabilitation inputs, the
user is asked to include an overall assessment of the pavement condition

based on an excellent, good, fair, poor, or very poor rating (see Table
2.5.16 of Ref 1)]
TABLE 2Summary of analysis parameter inputs [4].
Variable Value
Initial International Roughness Index (IRI), in./mile 63
Terminal IRI, in./mile 172
HMA surface down cracking, longitudinal cracking, ft/mile 2000
HMA bottom-up cracking, alligator cracking, % 25
HMA thermal fatigue cracking, ft/mile 1000
Chemically stabilized layer fatigue fracture, % N/A
Permanent deformation. total pavement, in. 0.75
Permanent deformation. HMA only, in. 0.25
Reliability, % 90
Design life, years 20
New HMA overlay material properties
Asphalt grading PG 70-28
Asphalt content, % 12.5
Air voids 4.0
Total unit weight, lb/ft3 148
Aggregate gradation
0.75 in. 100
0.375 in. 78
No. 4 50
No. 200 6.5
An HMA overlay was selected as the rehabilitation type for this case study.
For this analysis, milling of the upper 19 mm (0.75 in.) of the existing HMA
layer was assumed to occur prior to placement of the HMA overlay. A sum-
mary of the MEPDG analysis inputs for the new HMA overlay material is
shown in Table 2.
Three backcalculation programs (MODTAG, MICHBACK, and EVER-
CALC) were used to analyze the FWD deflection data for this LTPP site. These
programs were selected because they are widely used, readily available, and
non-proprietary. As the intent of this paper is not to report on the results of the
FWD testing or a comparison of backcalculation programs but to demonstrate
the role of deflection data in the MEPDG, the respective backcalculation results
are not presented here, but they can be found in the project report [4].
In order to illustrate the influence of unbound and bound backcalculated
layer inputs on the MEPDG results, six analysis runs were evaluated (see
Table 3).
For each backcalculation scenario, the HMA overlay thickness required in
order to meet a 90 % reliability level was determined, and the change in dis-
tress predictions was compared. Based on this analysis, the required HMA
overlay thickness for all scenarios was 76 mm (3 in.), except for Scenario D,
which required an overlay thickness of 89 mm (3.5 in.). The primary distress
TABLE 3Layer moduli used in MEPDG analysis.
Layer Moduli, MPa (lb/in.2)
Scenario HMA Base Subgrade

A MEPDG default 262 (38 000) 100 (14 500)
B 4195 (608 500) 115 (16 700) 148 (21 500)
C 4195 (608 500) 72 (10 400) 52 (7500)
D 4116 (597 000) 52 (7500) 62 (9000)
E 2413 (350 000) 91 (13 200) 27 (3900)
F 4075 (591 000) 84 (12 200) 43 (6300)a
57 (8300)b
Notes: A: Default layer properties; HMA modulus determined internally based on material properties.
B: Uncorrected backcalculation values based on MODTAG results for a three-layer system.
C: Corrected backcalculation values based on MODTAG results for a three-layer system.
D: Corrected backcalculation values based on EVERCALC and MICHBACK for a three-layer system.
E: Based on LTPP laboratory testing results.
F: Corrected backcalculation values based on MODTAG results for a four-layer system.
a
Represents a subgrade depth of 610 mm (24 in.).
b
Represents an infinite subgrade depth.
that controlled the required HMA overlay thickness for each scenario was top-
down cracking, with other distress predictions having minimal influence (see
Table 4).
Scenario D has the greatest predicted top-down cracking, though it also has
the weakest base layer modulus with one of the higher subgrade moduli. The
existing HMA modulus for Scenario D is also nearly the same as in Scenarios
B and C. Even though Scenario E has the lowest subgrade modulus, the top-
down cracking is lower than in Scenarios B and C (uncorrected and corrected
backcalculation-based, respectively). However, the permanent deformation for
this analysis increases compared to that of Scenario C, a finding that is partly
attributed to the lower HMA modulus. Scenario F (four-layer system) is com-
parable to the three-layer scenarios.
Additional findings from the MEPDG analysis of the six scenarios include
the following:
Until ongoing studies are completed, the correction factors for backcal-
culated layer moduli that are to be compared to laboratory determined

values, shown in Table 1, should be applied to the backcalculation
results.
The procedure needs to be verified, but for the time being, an equivalent
frequency of 30 Hz should be used. This is calculated as 1/(FWD pulse

duration) 1/0.033 s 30 Hz. Although this formula is technically
incorrect, it is compatible with the equivalent frequency used for calcu-
lating E* in the MEPDG.
TABLE 4Summary of condition results for each scenario.
Scenarioa
Predicted Distress A B C D E F
Terminal IRI, in./mile 109 (98) 97 (100) 97 (100) 97 (100) 101 (99) 98 (100)
Asphalt Concrete (AC) 5.2 (100) 62.7 (92) 69.5 (91) 165.0 (86) 49.4 (93) 50.4 (93)
surface down cracking
(longitudinal cracking), ft/mile
AC bottom-up cracking 0 (100) 0 (100) 0 (100) 0 (100) 0.3 (100) 0 (100)
(alligator cracking), %
AC thermal fracture 1 (100) 1 (100) 1 (100) 1 (100) 1 (100) 1 (100)
(transverse cracking), ft/mile
Chemically stabilized layer NA NA NA NA NA NA
(fatigue fracture)
Permanent deformation 0.13 (99) 0.08 (100) 0.07 (100) 0.08 (100) 0.12 (100) 0.08 (100)
(AC only), in.
Permanent deformation 0.38 (100) 0.08 (100) 0.08 (100) 0.16 (100) 0.12 (100)
(total pavement), in.
Required HMA thickness, in. 3 3 3.5 3 3 3
a
Values include estimated distress quantity and reliability in parentheses.
Concrete Pavement Case Study

The concrete pavement section used for this case study included LTPP Section
32-0200. Details of the LTPP section include the following:
Climate zone: Dry-freeze (Lander County, NV)
Functional class: Rural principal interstate (Interstate 80)
Original construction: 1993
Rehabilitation and repair work: Crack sealing (1995 and 1997)
Pavement section
- 295 mm (11.6 in.) concrete

- 145 mm (5.7 in.) dense-graded aggregate base
- 513 mm (20 in.) granular subbase
- 305 mm (12 in.) lime treated subgrade
Subgrade classification: Sandy silt (AASHTO A-2-6)
Distress:
- Transverse cracking (100 % of slabs)

- Faulting greater than 0.25 mm (0.1 in.)
Overall pavement condition: Poor (see Table 2.5.15 of Ref 1)
For this case study, an HMA overlay, an unbonded jointed plain concrete
pavement (JPCP) concrete overlay, and a bonded JPCP were selected as the
rehabilitation options. A summary of the MEPDG analysis inputs is shown in
Tables 57.
TABLE 5Summary of analysis parameter inputs [4].
HMA Unbonded Bonded

Variable Overlay Overlay Overlay
Initial IRI, in./mile (m/km) 63 (1.0) 63 (1.0) 63 (1.0)
Terminal IRI, in./mile (m/km) 172 (2.7) 172 (2.7) 172 (2.7)
Transverse cracking, % slabs cracked 15 25 25
HMA longitudinal cracking, 2000 (610) NA NA
ft/mile (m/km)
HMA alligator cracking, % 25 NA NA
HMA thermal fatigue cracking, 1000 (305) NA NA
ft/mile (m/km)
Mean joint faulting, in. (mm) NA 0.12 (3) 0.12 (3)
Chemically stabilized layer fatigue 25 NA NA
fracture, %
Permanent deformation. total 0.75 (19) NA NA
pavement, in. (mm)
Permanent deformation. HMA 0.25 (6) NA NA
only, in. (mm)
Reliability, % 90 90 90
Design life, years 20 20 20
Existing concrete elastic modulus, 3 100 000 (21 374) 3 100 000 (21 374) 3 100 000 (21 374)
lb/in.2 (MPa)
Existing concrete modulus of rupture, 550 (3.8) 550 (3.8) 550 (3.8)
lb/in.2 (MPa)
In order to validate the reliability of using FWD backcalculation results in

MEPDG designs, the following three design alternatives were conducted for
each type of overlay (the resulting input values are shown in Table 8):
Alternative 1: The laboratory determined material properties are used as
inputs in the MEPDG.

Alternative 2: The backcalculated values, in terms of the dynamic elastic
modulus for the existing Portland cement concrete (PCC) layer and the
TABLE 6Summary of HMA overlay material inputs [4].
Variable Value
Asphalt grading PG 58-22
Asphalt content, % 5.6
Air voids 4.0
Aggregate gradation (% passing)
0.75 in. (19 mm) 100
0.375 in. (9.5 mm) 82
No. 4 (4.76 mm) 67
No. 200 (0.074 mm) 4
TABLE 7Summary of concrete overlay material inputs [4].
Variable Value
3 3
Unit weight, lb/ft (kg/m ) 138 (2211)
Poissons ratio 0.12
Coefficient of thermal expansion, / F (/ C) 5.3 106 (9.5 106)
Modulus of rupture, lb/in.2 (MPa) 570 (3.9)
dynamic k-value for the supporting layers, are used as inputs, assuming
that the k-value reflects the stiffness of all the layers beneath the base.
Alternative 3: The same inputs as those used in the second alternative
are used, except that the dynamic k-value reflects the composite stiffness
of all layers beneath the slab.
For the rigid pavement backcalculation, the AREA method [5] was used
because it is an established process, includes closed-form equations, and is eas-
ily implemented in a spreadsheet. The average laboratory measured static PCC
modulus, 19 292 000 kPa (2 800 000 lb/in.2), presents a good correlation with
the average backcalculated static PCC elastic modulus of 21 359 000 kPa
(3 100 000 lb/in.2); however, there was significant variation of the backcalcu-
lated values along the section (22 %, which is substantially higher than the
15 % typically assumed acceptable).
The MEPDG results for this case study are summarized in Table 9. The
thinnest design thickness was obtained by using a bonded JPCP overlay,
because it utilizes the remaining structural capacity of the existing slab. The
design thickness of the unbonded overlay was determined considering that the
unbonded JPCP overlays work independently and thus some restraints in thick-
ness have to be provided in order to guarantee its structural capacity. The
MEPDG provides an unreasonably thick HMA overlay design (an overlay
thickness of 150 to 200 mm [6 to 8 in.] would be typical), with the critical per-
formance parameter being surface rutting. Even when various modifications
were made to the HMA mix design, the resulting HMA overlay thickness was
considered unreasonably thick. If surface rutting is addressed through mainte-
nance at some intermediate year (less than 20 years), then a thinner HMA over-
lay would be appropriate.
No difference in terms of the design thickness was found among the three
design alternatives (laboratory/material default values, adjusted
TABLE 8Summary of concrete material inputs [4].
Alternative PCC Elastic Modulus, lb/in.2 (MPa) k-value, lb/in.2/in. (kPa/mm)

1. Laboratory determined 2 783 000 (19 188) NA
2. Backcalculated PCC 3 100 000 (21 359) 310 (84)
3. Composite stiffness 3 100 000 (21 359) 375 (101)
TABLE 9Summary of MEPDG overlay thickness results.
HMA Overlay, Unbonded PCC Bonded PCC

Alternative in. (mm) Overlay, in. (mm) Overlay, in. (mm)
1. Laboratory determined 12 (305) 7 (178) 4 (100)
2. Backcalculated PCC 12 (305) 7 (178) 4 (100)
3. Composite stiffness 12 (305) 7 (178) 4 (100)
backcalculated k-value and backcalculated PCC elastic modulus, and backcal-

culated PCC elastic modulus and k-value), indicating the reliability of using
both the backcalculated dynamic elastic modulus for the PCC layer and the
dynamic k-value for the supporting layers in the MEPDG design.
The backcalculated k-value that represents the composite stiffness of all
layers beneath the slab can be directly entered into the MEPDG without having
a significant influence on the design thickness for the pavement structure ana-
lyzed. However, this does not definitively mean that the MEPDG takes the
stiffness of the base layer into account in the k-value. This could either be due
to the insensitivity of the design thickness on the input k-value or be because
the granular base contributes little to the composite stiffness of all layers
beneath the slab. Other observations made regarding the calculation of the
k-value within the MEPDG program include the following:
With the HMA overlay design, varying the subbase stiffness had very
little influence on the determined k-value, which suggests that the stiff-
ness of the subbase layer is not taken into account in the calculated
k-value. Additionally, the reported k-values in the output file are identi-
cal for cases with varying base layer stiffnesses, indicating that the stiff-
ness of the base is likely not included in the k-value calculation.
Furthermore, the MEPDG appears to ignore the entered dynamic k-value
and calculate the k-values based on the entered layer moduli, as the sum-
marized values were the same regardless of what dynamic k-value was
entered or when a dynamic k-value was not entered.
With the unbonded PCC overlay, additional design runs indicate that the
stiffnesses of the interlayer and the existing PCC are not considered in
the calculation of the k-value. The base layer is taken into account
because there is a difference in k-values when using a stiff versus a
weak base layer. Additionally, the k-values agree well with the entered
dynamic k-value, suggesting that the MEPDG utilizes the entered value
for unbonded PCC overlay designs.
The modulus of the base layer is considered in the calculation of the k-
value for bonded JPCP overlay designs, which agrees with the assump-
tions made for bonded PCC overlays. The calculated k-values also
match the entered dynamic k-value.
Summary
Case study investigations were performed for both HMA and concrete pave-
ments in order to evaluate the use of deflection data in the MEPDG. Significant
findings from this investigation are presented below.
For HMA pavements, backcalculated layer moduli should be adjusted to
laboratory determined values using the correction factors developed by Von
Quintus and Killingsworth [2]. In addition, until the procedure is verified, an
equivalent frequency of 30 Hz should be used for calculating E*.
For concrete pavements over a granular base, no explicit conclusion can be
drawn with respect to defining the layers used in the composite effective
dynamic k-values calculated within the MEPDG for each type of overlay
design, due to the conflicting results obtained. However, it can be concluded
that the manually entered k-value is used for unbonded JPCP and bonded JPCP
overlay designs but not for the HMA overlay design. No appreciable difference
in terms of the design thickness was found among the three design alternatives,
indicating the reliability of using the backcalculated dynamic (or static) elastic
modulus for the PCC layer and the dynamic k-value for the supporting layers
in the MEPDG design. Furthermore, it is found that the backcalculated k-value
that represents the composite stiffness of all layers beneath the slab can be
directly entered into the MEPDG without a significant influence on the design
thickness for the pavement structure analyzed. However, this does not defini-
tively mean that the MEPDG takes the stiffness of the base layer into account
in the k-value. This could be either due to the insensitivity of the design thick-
ness to the input k-value or because the unstabilized granular contributes very
little to the composite stiffness of all layers beneath the slab.
Acknowledgments
The work described in this paper was funded by the Federal Highway Adminis-
tration under Contract No. DTFH61-06-C-00046.
References
[1] National Cooperative Highway Research Program (NCHRP), 2004, Guide

for Mechanistic-Empirical Design of New and Rehabilitated Pavement
Structures, Transportation Research Board, Washington, D.C.
[2] Von Quintus, H. and Killingsworth, B., 1998, Analyses Relating to Pave-
ment Material Characterizations and Their Effects on Pavement Perform-
ance, FHWA-RD-97-085, Federal Highway Administration, McLean,
VA.
[3] Smith, K. D., Bruinsma, J. E., Wade, M. J., Chatti, K., Vandenbossche, J.
M., and Yu, H. T., 2011, Using Falling Weight Deflectometer Data with
Mechanistic-Empirical Design and Analysis, Volume 1: Final Report,

Federal Highway Administration, McLean, VA.
[4] Bruinsma, J. E., Vandenbossche, J. M., Chatti, K., and Smith, K. D.,
2011, Using Falling Weight Deflectometer Data with Mechanistic-
Empirical Design and Analysis, Volume 2: Case Study Reports, Federal
Highway Administration, McLean, VA.
[5] Ioannides, A. M., Barenberg, E. J., and Lary, J. A., Interpretation of Fall-
ing Weight Deflectometer Results Using Principles of Dimensional Ana-
lysis, Proceedings of the 4th International Conference on Concrete
Pavement Design and Rehabilitation, Purdue University, West Lafayette,
IN, 1989.
STP 1555, 2012
DOI:10.1520/STP104438
George K. Chang,1 James Watkins,2 and Robert Orthmeyer3
Practical Implementation of Automated Fault

Measurement Based on Pavement Profiles
REFERENCE: Chang, George K., Watkins, James, and Orthmeyer, Robert,

Practical Implementation of Automated Fault Measurement Based on Pave-
ment Profiles, Pavement Performance: Current Trends, Advances, and
Challenges on December 5, 2011 in Tampa, FL; STP 1555, B. Choubane,
Editor, pp. 219237, doi:10.1520/STP104438, ASTM International, West
Conshohocken, PA 2012.
ABSTRACT: The current U.S. Highway Performance Monitoring System
(HPMS) requires state highway agencies (SHA) to report to the U.S. Federal
Highway Administration (FHWA) a set of pavement-condition data, which
includes faulting data on jointed concrete pavements, among other data
items. To address this pressing need for an automated fault-measurement
method based on pavement profile data to reduce the risk associated with
manual faulting measurements, the authors developed an automated faulting
measurements (AFM) module in the profile viewing and analysis tool
(ProVAL), to allow practical application of this method for state department of
transportation (DOT) users. This paper describes the development of the
AFM algorithm (including joint finding and joint fault computation), the associ-
ated field validation program, recommended profile measurement methodol-
ogy, and usage/limitation of the AFM tool on the ProVAL platform. Through
the peer reviews and field validation efforts, the AFM is shown to be an effec-
tive tool for implementation of automated fault measurements. Therefore, the
AFM is served as the basis for AASHTO R 3612 Standard Practice for Esti-
mating Faulting of Concrete Pavements.
KEYWORDS: pavement, profiling, jointed, concrete, faulting, automation,
software
October 2012.
1
Ph.D., Transtec Group, Austin, TX 78731.
2
Mississippi Department of Transpotation, Jackson, MS 39201.
3
Federal Highway Administration, Matteson, IL 60443.
Copyright V
19428-2959.
219
Introduction
The current Highway Performance Monitoring System (HPMS) requires state
highway agencies (SHA) to report to the U.S. Federal Highway Administra-
tion (FHWA) a set of pavement-condition data, which includes faulting data
on jointed concrete pavements (JCP), among other data items. Therefore, the
U.S. FHWA funded a project to include an automated faulting measurements
(AFM) module in ProVAL with the purpose to reduce the risk associated
with manual faulting measurements. ProVAL is a profile viewing and
analysis software tool sponsored by the U.S. FHWA, Long Term Pavement
Performance (LTPP) program, and the FHWA Profile Pooled Fund Study
since 2000.
The ProVAL research team conducted meetings with FHWA, MS depart-
ment of transportation (MDOT), and Florida department of transportation
(FDOT) to determine the direction for the ProVAL AFM during spring and
summer 2010. The conclusion from these meetings was that the ProVAL AFM
should meet the analysis needs at all projects levels for automated fault meas-
urements based on pavement profiles. The ProVAL AFM should also produce
a list of joint locations identified through an automated process. Then, the users
can inspect and modify joint locations, as necessary, then strive to achieve as
close to 100 % success rates of joint identification using the implemented
methods in ProVAL. The subsequent fault computation should also be compat-
ible with the methods recommended in ASHTO R3612, Standard Practice for
Estimating Faulting of Concrete Pavements [1].
The authors conducted a review of the existing faulting algorithms that
are based on profile data collected by high-speed inertial profilers. These
algorithms have been developed and are available in the public domain.
These algorithms are capable of locating joints and cracks on a concrete
pavement and determining the extent of pavement joint/crack faulting. The
following describes the methods for joint identification and fault computation
based on profile data, and the implemented, practical functionality of the
ProVAL AFM.
Profile Data Collection for Automated Fault Measurements

The automated fault-measurement method consists of two consecutive steps:
(1) joint identification, and (2) fault estimates. If the first step fails, so does the
second step. Joint identification is to locate the joints within pavement profiles
surveyed on jointed concrete pavements (JCP) within a desired precision. Fault
estimates are used to compute faulting based on the profile slices on both sides
of joints.
Automated algorithm for joint identification is often challenged by the fol-
lowing issues.
CHANG ET AL., doi:10.1520/STP104438 221
Profile Issues
Profilers are not certified by the AASHTO R56 procedure [2] for repeat-
ability and accuracy tests,
Sampling intervals may be too wide to include joint locations,
Most profiling has a lack of repeated runs to allow the use of advanced
techniques, such as pattern searches for joint identification, and
There are often distance-measurement-instrument (DMI) drifts, espe-
cially for relatively long profiles.
Pavements Issues
Concrete joints are sometimes filled with sealants or incompressible
materials that prevent it from being detectable by the automated meth-
ods that rely on downward spikes at joints,
Concrete joints are sometimes closed up because of slab thermal expan-
sion; therefore, such joints may not be detectable by methods that rely
on downward spikes at joints,
Cases with skewed joints would force detecting joint locations for both
left and right wheel tracks be conducted separately; therefore, such joint
locations cannot be detected joints using pattern-search methods,
Cases with lack of curl/warp features would make the joint detection dif-
ficult using pattern-search methods,

Slabs with cracks (especially non-transverse cracks) would make it diffi-
cult to detect joint-spacing patterns, and

Uneven joint-spacing patterns (e.g., when joint spacings are not con-
stant/exact or without a fixed pattern, e.g., 1820/1820 ft.) would make

it difficult for joint detection.
Combination of the above factors would pose an even greater challenge for
joint identification. Therefore, the measurements should comply with the fol-
lowing best practices for automated fault measurements using inertial
profilers:
The repeatability and accuracy scores based on cross correlation under
the AASHTO R56 standard need to be equal or above 92 % and 90 %,

respectively.
Profile measurements should be verified with a physical device.
For project-level survey, the sampling interval needs to be 19 mm (0.75
in.) or less. At least three repeat runs are needed during two diurnal peri-
ods of extreme air temperatures. If the profiles exhibit repeatability
above 92 % cross correlation, any of the profile from each diurnal run
can be used for automated fault computation. If the mean fault values
from both diurnal runs are noticeably different (e.g., 5 mm or 0.2 in.),
both values and test conditions should be recorded. No digital filtering
in post-processing data should be allowed. Automated triggering is
recommended to locate the start and end of survey sections with high
precision.
For network-level survey, the sampling interval needs to be 38 mm (1.5
in.) or less. No digital filtering during post-processing data should be
allowed.
Profile data should be collected for both left and right wheel tracks.
Observation should be recorded for profiler sensor footprint, aggressive
surface textures, tinning, slope/grade, spalling, curl/warp, skewed joints,
and sealant-filled joints.
Joint Identification
The review of existing joint-identification algorithms includes the following:
MDOT methodby the Mississippi DOT,
FDOT methodby the Florida DOT, and
FHWA Curl-Warp project methodby the FHWA Curl-Warp research
team.
MDOT Method
The MDOT joint/crack-location-identification algorithm is based on a brute-
force method by detecting whether the elevation differential between adjacent
samples is greater than 2.03 mm (0.08 in.) [3]. The method was developed
based on the MDOT profiler that uses 12.7 mm (1/2 in.) sampling interval.
Based on MDOTs experiences, this brute-force method appears to work well,
especially when only one repeat of profile data is available.
It is uncertain whether the MDOT brute-force method is effective with pro-
file sampling intervals other than [1/2]" (used in the MDOT profilers) or with
joint openings of various sizes. Though it is the simplest method reviewed so
far, the MDOT method can be used effectively to report faulting for the
network-type analysis without the need to identify all joints.
FDOT Method
The FDOT joint/crack-location-identification method is based on detecting
peaks/valleys followed by spacing checks [4,5]. The FDOT algorithm merges
the joint identification and fault computation.
The FDOT method starts with a seed value for the shape factor (SF) of
0.18. It also requires an input for transverse joint spacing (typical 15 to 20 ft in
FL) to compute a maximum number of joints in a given pavement profile. Dur-
ing the iterations to find joint locations, the FDOT method would adjust the SF
value (based on binary search) by comparing the joints found and the maxi-
mum number of joints. For example, the SF seed value, 0.18, would be
deducted by 0.9 for the next iteration if the joints found are less than the
maximum number of joints; the SF seed value would be increased by 0.9 if oth-
erwise. Binary search will loop nine times to find at least one joint location. Af-
ter that, the binary search will loop five times to find a better SF. If the binary
search does not find a better solution, it will stop and take last best found
solution.
The FDOT method has been demonstrated to be effective on their test
cases and the related results have been published in two TRB papers [4,5]. The
FDOT method was also tested on MDOT data and shown to be successful with
the joint detection rates ranging from 100 % to 55.6 %. The lower success rate
corresponds to the profiles with aggressive grades (e.g., Carroll County) or
sealed joints (e.g., WA).
The FDOT method is dependent on a valley-detection technique that may
be challenged when pavement surfaces are flat with joints that are either closed
or filled. Fortunately, the above cases are relative rare. Based on the current
validation results, the success rates for joint identification using the FDOT
method for common site conditions are very satisfactory.
FHWA Curl-Warp Method

The FHWA Curl-Warp method for joint/crack-location identification was
developed by Steve Karamihas under the FHWA Curl-Warp project led by the
primary author [6]. This method is based on a four-step process.
Step 1Filter the profiles,
Step 2Search for the deepest dips,
Step 3Assemble the dip count across the data set, and
Step 4Extract the joint locations.
The details of this method are provided below because its relevancy to the
implementation of the ProVAL AFM and the associated FHWA report is not
published. The procedure described here depends on the recording interval and
height sensor footprint used to make the measurements. Any change in the
measurement procedures would require modification of the settings. However,
the basic strategy is portable to other measurement systems.
This method was demonstrated with a raw profile measurement from the
left wheel track of Section 010A under the FHWA Curl-Warp project [6]. Nar-
row dips protrude from the profile at a regular interval. The dips appear at a
spacing of 4.87 m (15.98 ft), with few missing dips. About a third of the test
FHWA Curl-Warp sections shared this level of clarity in the search for joint
locations. In another third, the joint locations were even more obvious. This
section describes the steps in the joint finding procedure, and discusses the pro-
cess using Section 010A as an example.
Step 1: Filter the ProfilesApply a moving average anti-smoothing filter

with a short base length to all of the profiles. Note that the filtering is used only
to identify joint locations, whereas raw profiles without filtering are used to
compute faults. This eliminates much of the profile content, but does not elimi-
nate the narrow dips. Normalize the filtered profiles by the root-mean-square
value. This facilitates an automated search for the deepest narrow dips. The
outcome of this process is called a spike profile after applying a anti-
smoothing filter with a base length of 250 mm (9.84 in.).
For this measurement of Section 010A, several narrow dips appear with a
magnitude much larger than the root-mean-square value. Over most of the sec-
tion, the dips appear at a regular interval, and few dips appear that do not fol-
low the pattern. However, some locations within the pattern are missing. This
was typical, and it prompted the use of multiple repeat measurements for iden-
tifying joint locations.
Step 2: Search for the Deepest DipsStep through each spike profile at a
regular distance interval and search for deep dips. Within each interval of
distance, register a hit if a value appears in the filtered and normalized pro-
file below a given threshold value. The threshold values depended on the
section.
For Section 010A, the search recorded any value in the spike profile less
than 4. This was typical on test sections with deep narrow dips at the joints.
The high threshold value helped exclude dips caused by features other than
joints. For this example, 71 hits were shown at dips beneath the threshold value
of 4. This includes hits at 61 of the 62 joints and 10 hits that do not correspond
to joints. Some sections with shallower dips required a lower threshold value.
This helped find more of the joints in each repeat measurement, but it also pro-
vided greater opportunity for false readings at transverse cracks and heavy
chatter.
For this process, the search for spikes stepped through the profile at an
interval of 10 mm (0.39 in.). If every profile used the same recording interval,
a new interval would not be needed. However, these profiles had undergone a
linear adjustment to their recording interval to fine-tune their synchronization,
so the recording interval was not consistent.
Step 3: Assemble the Dip Count across the Data SetCount the number of
repeat measurements that registered a hit (i.e., a qualifying dip) at each loca-
tion from all repeated profiles. To do this, sum up the number of hits found
in Step 2 at each location across all of the measurements of a test section.
Express the count as a percentage of the repeat measurements in which a hit
was found. The outcome of this process is a spike incidence profile that
indicates how often a narrow dip appeared over each interval of distance
10 mm wide.
This can be demonstrated by the same example that includes the spike inci-
dence profile for the 29 measurements of Section 010A [6]. A clear pattern is
visible. The locations where more than 20 % of the repeat measurements

included a hit within just 10 mm of distance appear with a regular spacing of
4.87 m. Although other locations show hits in one to three of the repeat meas-
urements, the joint locations are easy to distinguish from the rest.
Many joints were detected in more than 50 % of the repeat measurements
within a window of just 10 mm. Note that downward spikes at the joint often
appeared over a width of more than 10 mm, so the hits were often spread
out over several 10-mm-long intervals. Further, the joints did not always
appear in the same precise location, even with rigorous synchronization. For
example, the first joint on Section 010A registered a spike incidence level of
34.5 % at a location 4.34 m (14.24 ft) from the start of the section. However,
the incidence was 27.6 % at a location 4.35 m from the start of the section
and 31.0 % at a location 4.36 m from the start of the section. Overall, most
of the spikes appeared within five of the 10-mm-wide intervals. The overall
joint width (i.e., tolerance), within which the joints appeared for each section,
is expressed as a joint window. For Section 010A, the joint window is 50
mm. This is an important parameter, because it indicates the amount of pro-
file that must be masked at the slab ends when performing slab-by-slab
analyses.
The clear pattern in Section 010A was typical of about a third of the test
sections. In another third, a pattern just as clear was found, with even fewer
spurious hits away from the joints.
Step 4: Extract the Joint LocationsFirst, weed false hits from the spike
incidence profile that appear near locations with a high concentration of hits.
To do this, seek the location with the highest percentage of hits within the
entire profile, then zero out all values within a give range of that location. Find
the next highest value, clear the values near it, and so on, until only the highest
non-zero values remain.
Second, list only those remaining spike locations that occur in a high
enough percentage of the profiles (defined by the incidence threshold) and
appear with the highest incidence within any width of the pavement equal to an
estimated slab length (defined by the assumed slab length).
Step 4 simply automates the process that a person may use to extract joint
locations from the spike incidence profile by inspection. The outcome is a list
of joint locations. On most test sections, Step 4 produced a list of joint loca-
tions with a clear pattern.
The FHWA JCP Curl-Warp method for joint/crack-location identification
was proved to be very effective for multiple profile runs that make use of small
sampling intervals and small laser footprints (such as dot lasers) for cases that
have clear downward spikes at joints. Note that this method failed at three
FHWA Curl-Warp test sites that consist of flatter slabs with filled joints. The
lack of multiple profile runs for routine surveys makes this technique difficult
to function.
Joint-Identification Methodologies in ProVAL AFM

After detailed, critical review of the above methods, the ProVAL team decided
to adopt the MDOT method and adapt the FHWA Curl-Warp method for joint
identification in ProVAL AFM. A curled-edge detection method was also
developed to handle cases with curled slabs. The FDOT was not included in
the implementation of ProVAL AFM as the above three selected methods can
handle the majority of the real-world cases. For the ProVAL AFM, three meth-
ods for joint identification are implemented:
Downward-spike-detection method based on portion of the FHWA
Curl-Warp method,
Step-detection method based on the MDOT method, and
Curled-edge detection method.
The implemented ProVAL AFM performs joint identification for multiple

profiles at a time. Users are required to provide inputs including: nominal joint
spacing (or the smallest joint-spacing value if multiple joint spacings are pres-
ent), joint window, and whether skewed joints are present. The following pro-
vides the detailed procedures of the above three methods for joint detection
(Fig.1).
Downward-Spike-Detection Method
Perform anti-smoothing filtering using moving average filter at a cutoff
of 250 mm or 9.84 in.,
Normalize the filtered profile with its root mean square (rms) and pro-
duce the spike profile, i.e., making the spike profile unitless,
Detect the locations where the spike profile values exceed a threshold
value (the starting threshold is -4.0), but avoid multiple hits within a
clearance width (0.5 m or 1.64 ft), and
Screen the above locations to differentiate joints from cracks.
Step-Detection Method
Deduct profile elevations between consecutive data points (resulting ele-
vation differences).
Detect the locations where the absolute values of the elevation differen-
ces exceed a threshold value (the starting threshold value is 2.032 mm
or 0.08 in.) but avoid multiple hits within a clearance width (0.91 m or
3 ft).
FIG. 1Identification of joint and crack locations that allow for users modifications in ProVAL AFM.
Curled-Edge-Detection Method
Perform bandpass filtering using moving average filter with short cutoff
at 250 mm or 9.84 in. and long cutoff wavelength at 50 m or 150 ft.
Simulate a rolling straightedge response (with base length of 3 m or
9.8 ft).
Detect the locations where the simulated rolling straightedge responses
exceed a threshold value (the starting threshold is 3 mm or 0.12") but

avoid multiple hits within a clearance width (0.5 m or 1.64 ft).
The downward-spike-detection method performs well with profiles that are

sampled at shorter intervals with downward spikes at joint locations. When
downward spikes are not present in profiles but there are apparent joint faults,
the step-detection method would perform better. When there are no downward
spikes or apparent faults, the curled-edge-detection method performs well
when there are apparent curled up or down slab shapes.
The numerical implementation of joint identification in ProVAL AFM
starts with a default threshold value for a selected method. Iterations of the
joint detections would continue by modifying the threshold until sufficient
joints (currently set at 80 % of all joints) are identified according to the nomi-
nal joint spacing from the user inputs. Therefore, a correct input of the nominal
joint spacing would facilitate the joint finding.
ProVAL AFM presents joint-crack-identification results in tables
and charts. Because none of the existing methodologies can always ensure the
100 % successful rate, ProVAL AFM allows users to examine the joint-
identification results and make corrections if needed (e.g., modify joint/crack
locations, disable joints/cracks, add joints/cracks, and toggle between joint
and crack). An example of joint-identification results is shown in Section 010A
[6], where joint locations were indicated with circles in the bottom chart and
joint spacings are plotted in the top chart. Once the joint locations are identi-
fied, the fault computation can be performed.
Fault Computation
As described in the AASHTO R3604 Standard Practice for Estimating Fault-
ing of Concrete Pavements [7], joint faulting can either be measured using a
physical device or be estimated using automated surveys. Two points on either
side of a joint are located approximately 300 mm apart along a wheel path
(avoiding spalled locations). The absolute value of the difference in height (ele-
vation) is defined as the faulting. Faulting values that are greater than 5 mm are
recorded and rounded to the nearest 1 mm.
The AASHTO Subcommittee of Materials (SOM) led the effort revising
the AASHTO R3604 to AASHTO R3612. The manual method in the latter
is modified from the AASHTO R3604 specification, including changes in

requirements for the spacing between Leg 2 and the joint (i.e., C), and the spac-
ing between Leg 3 and joint (i.e., D) adjustableas it may depend on site con-
ditions. Recommended values for C and D are between 76 mm and 226 mm
(3" and 8.9") (see Fig. 2). Currently, no such manual devices have been built to
meet the requirements in AASHTO R3612. Several designs of manual-fault-
measurement devices (including MDOT and FDOT) have been proposed to
validate high-speed profilers for implementing the automated method.
The following details the development of the ProVAL AFM as the basis
for the automated method in the draft AASHTO R3612.
The following joint-identification algorithms are reviewed:
MDOT methodby the Mississippi DOT,
FDOT methodby the Florida DOT, and
FHWA Curl-Warp project methodby the FHWA Curl-Warp research
team.
MDOT Method
The MDOT fault-computation methods include [3]:
Simulated Georgia Fault Meter (adjust for grade), and
AASHTO R3604 method (adjust for grade).
MDOT discovered that the simulated Georgia fault meter has several
issues: (1) the fault measurements need to be adjusted for grade, and (2) the
fault measurements may be erroneous when spalling presents under any foot of
the (simulated) device.
Also, MDOT discovered that the fault definition of the existing AASHTO
R3604 require correction if one were to emulate the process with pavement
profile elevation, unless it is adjusted for grade.
FIG. 2Manual faulting measurement with faultmeter.

FDOT Method
The FDOT fault-computation method is based on the combined algorithms of
AASHTO R3604 method and FDOT PCS specifications [4,5].
The FDOT algorithm merges the joint identification and fault computation.
The program recalculates all the joint locations and magnitudes using the best
SF, which yields the largest number of joints and saves this information into
the joint array.
FHWA Curl-Warp Method

The FHWA Curl-Warp method for fault computation is based on a simple
least-squares-fit method [6]. The procedures are as follows:
Load a profile and joint location information,
Crop segments of the profile on either side of each joint within the
section,
De-trend and de-mean the cropped profile,
Mask the profile within the joint window,
Perform curve fitting for the remaining profile for the approach slab
edge and leave slab edge,

Estimate linear slopes for the fitted shapes for both the approach and
leave sides, and

Estimate the faulting based on the AASHTO PP3604 specification.
The FHWA Curl-Warp method was proved to be effective to handle vari-

ous joint geometry conditions.
Joint-Fault-Computation Methodology in ProVAL AFM

After reviewing the above methods, the ProVAL team decided to implement
the ProVAL AFM with a method that is similar to the FHWA Curl-Warp
research. The FDOT method is not used in the ProVAL AFM implementation,
as it is considered compatible with the FHWA Curl-Warp method.
Based on the joint locations identified (and edited by users if desired) from
the first stage of the ProVAL, joint faults will be computed as follows:
Crop a profile segment that centers a joint with a length of 2.438 m
(8 ft).
Separate the profile slices for the approach slab and departure slab (i.e.,
1219 mm or 48 in. in length for each slice).

Allow users to select a range of offset values from the joint in down-
stream direction between 76 mm and 226 mm (3 in. and 8.9 in.).

For the profile slice from the approach slab, mask the area close to the
joint based on the joint-window input and perform least-squares fitting.

The fitting would extend to the leave side of the joint for 226 mm
FIG. 3Curve fitting of cropped profile segments and computation of faulting from multiple sampling points.
FIG. 4ProVAL AFM input screen.
FIG. 5Report of faults for individual joint or crack in charts in ProVAL

AFM.
(8.9 in). Obtain the elevation at the downstream end of the fitted line for
later fault computation.
For the profile slice from the departure slab, mask the area close to the
joint based on the joint-window input and perform least-squares fitting.

The fitting would be performed from the downstream end of the slice to-
ward the joint location. Obtain the elevations at every fitted data point
offset from the joint within 226 mm (8.9") (i.e., matching the exact hori-
zontal location of the elevation readout value from the above step).
Compute the average of the fault values obtained from elevation differ-
ences between the fitted lines from the above two steps between 76 mm
and 226 mm (3 in. and 8.9 in.) offset from the joint in downstream
direction.
An example from the above process is presented in Fig. 3. Currently, a
first-order polynomial is used for the least-squares fitting. Though the second-
order polynomial tends to fit the slab shapes better, the first-order one appear to
FIG. 6Report of positive and negative faults for individual joint or crack in
ProVAL AFM.
produce better correlation with the field measurements with physical devices.
The joint fault results from the ProVAL AFM are validated by MDOT with
their own implementation through years of research, development, and field
validation. Further revision of this method would allow a range of offset values
(e.g., 76.2 mm to 226 mm or 3 in. to 8.9 in.) toward downstream and automated
method would take the average of the fault values estimated within this range
as the representative faulting value.
In ProVAL AFM, users would be required to input: nominal joint spacing,
a reporting interval, a joint window (i.e., uncertainty of the exact joint locations
or possible joint defect area, default: 50 mm or 2.0 in.), selection of a joint-
detection method, whether skewed joints exist, and whether to include cracks
in the analysis (Fig. 4). Similar validation effort has also been conducted by
FDOT.
Examples of the ProVAL AFM outputs are shown in various different
forms: (1) Report of faults for individual joint or crack in bubble charts, where
the sizes of bubbles are relative to the fault severities (Fig. 5). (2) Report of
FIG. 7Report of maximum absolute values of faults for pavement segments

in ProVAL AFM.
FIG. 8Report of accumulated, absolute values of faults for pavement seg-

ments in ProVAL AFM.
positive and negative faults for individual joint or crack in bar charts (Fig. 6).
(3) Report of maximum absolute values of faults for pavement segments (Fig.
7). (4) Report of accumulated, absolute values of faults for pavement segments
(Fig. 8).
The above ProVAL AFM analysis results can be easily exported to PDF,
Excel, or text files to facilitate the data submission, say, to pavement-
management systems.
Conclusions
To address the need for an automated fault-measurement method based on
pavement-profile data to reduce the risk associated with manual faulting meas-
urements, the authors developed an automated-faulting-measurements (AFM)
module in the Profile Viewing and Analysis (ProVAL) software to allow prac-
tical application of this method. This paper describes the development of the
recommended profile-measurement methodology, AFM algorithm for joint

identification and joint fault computation, validation efforts of the AFM tool by
MDOT and FDOT, and usage/limitation of the AFM tool on the ProVAL plat-
form. Three methods were implemented in the ProVAL AFM to identify joint
locations, including the downward-spike-detection method derived from the
FHWA Curl-Warp study, the step-detection method based on the MDOTs
implementation, and the curled-edge-detection method developed under the
FHWA AFM project. The fault-computation method was based on a least-
squares fit of the profile slices on both sides of a joint and the fault measure-
ment was taken at an offset from a joint location that is compatible with the
proposed manual device in AASHTO R36. The AFM is shown to be an effec-
tive tool for implementation at all project levels. The AFM is also served as the
basis and a recommended tool in the AASHTO R3611 Standard Practice for
Estimating Faulting of Concrete Pavements and further revision in R3612.
Acknowledgments
The writers acknowledge the support from the FHWA and numerous fruitful
meetings with MDOT and FDOT. Specifically, the writers knowledge the field-
data-collection efforts by Alan Hatch and Cindy Smith of MDOT, AASHTO
R36-12 specification reviews and comments by Grady Aulman and of MDOT
and FDOT staff, and AFM beta testing and comments by MDOT and FDOT
staff.
References
[1] AASHTO R36-12, 2012, Standard Practice for Estimating Faulting of

Concrete Pavements, American Association of State Highway & Trans-
portation Officials, Washington, D.C.
[2] AASHTO R56-10, 2010, Standard Practice for Certification of Inertial
Profiling Systems, American Association of State Highway & Transpor-
tation Officials, Washington, D.C.
[3] Mississippi DOT; BATCHCALCFAULT Software User Guide 3.00
(March 2010).
[4] Nazef, A., Mraz, A., Iyer, S., and Choubane, B., Semi-Automated Fault-
ing Measurement Approach for Rigid Pavements Using High-Speed Iner-
tial Profiler Data, Transportation Research Board 88th Annual Meeting,
[5] Nazef, A., Mraz, A., and Choubane, B., Alternative Validation Practice
of an Automated Faulting Measurement Method, March 2010
(unpublished).
[6] Chang, G. K., Karamihas, S. M., Rasmussen, R. O., Merritt, D. K., and
Swanlund, M., Quantifying the Impact of Jointed Concrete Pavement
Curling and Warping on Pavement Uneveness, Proceedings of the Sixth

Symposium on Pavement Surface Characteristics (SURF), Slovenia,
2008.
[7] AASHTO R36-04, 2004, Standard Practice for Estimating Faulting of
Concrete Pavements, American Association of State Highway & Trans-
portation Officials, Washington, D.C.
STP 1555, 2012
DOI:10.1520/STP104432
Stephen N. Goodman,1 Rhiannon Arsenault,2 and Tom Dziedziejko3
Proactive Pavement Smoothness from the

Base Upward
REFERENCE: Goodman, Stephen N., Arsenault, Rhiannon, and Dzied-

ziejko, Tom, Proactive Pavement Smoothness from the Base Upward,
Pavement Performance: Current Trends, Advances, and Challenges on De-
cember 5, 2011 in Tampa, FL; STP 1555, B. Choubane, Editor, pp. 238247,
2012.
ABSTRACT: The importance of high initial smoothness on newly constructed
or rehabilitated pavement facilities has been identified by many researchers
and recognized by many road agencies. Pavements with high initial smooth-
ness display retarded roughness progression and, therefore, longer service
lives. In Ontario, Canada, the Ministry of Transportation (MTO) implemented
a bonus/penalty system for smoothness in the mid-1990s. Initially based on
the profile index and scallops (i.e., bumps) measured with a California Profil-
ograph, the specification is now based on the International Roughness Index
analyzed locally (for bumps) and over regular intervals. The potential bonus
for high initial smoothness in Ontario is considerable, as is the potential pen-
alty for excessive roughness. Furthermore, the ability to correct areas of ex-
cessive roughness after placement is greatly limited by the MTO because the
public does not wish to observe a new pavement surface being damaged
by a milling machine. Given such, contractors are proactively requesting
smoothness testing on base pavement layers so that areas of excessive
roughness can be addressed prior to the placement of the surface layer.
However, once the smoothness of the base layer is determined, the contrac-
tor must decide whether areas of excessive roughness will be sufficiently
smoothed by the surface hot mix asphalt (HMA) layer or whether milling is
first required (at additional cost). In addition to a brief overview of the Ontario
smoothness specification, this paper presents the change in pavement
October 2012.
1
AME Materials Engineering, Ottawa, ON K2H 9C1, Canada (Corresponding author).
2
AME Materials Engineering, Ottawa, ON K2H 9C1, Canada.
3
AME Materials Engineering, Caledon, ON L7C 3M6, Canada.
Copyright V
19428-2959.
238
GOODMAN ET AL., doi:10.1520/STP104432 239
smoothness with increasing HMA layers measured at a select project loca-

tion in order to identify how much additional smoothing a contractor should
anticipate. Of particular interest is the change in smoothness from an in-
place recycled layer to a base HMA layer and, finally, a surface HMA layer.
The results will allow contractors to optimize pavement smoothness (and
their associated bonus) while minimizing costly corrective milling.
KEYWORDS: pavement smoothness, inertial profiler, International Rough-
ness Index, bonus and penalty system
Introduction
The Importance of Pavement Smoothness

The importance of high initial smoothness on newly constructed or rehabili-
tated pavement facilities has been identified by many researchers and recog-
nized by many road agencies. Pavements with high initial smoothness display
retarded roughness progression and, therefore, longer service lives. Pioneering
work in this regard was conducted through National Cooperative Highway
Research Program Project 1-31, the results of which indicated that an increase
in pavement smoothness of 25 % resulted in an overall increase in pavement
life of 9 % [1].
Increased pavement smoothness also contributes to reduced fuel consump-
tion. An excellent article on the issue by Marks [2] reports that reductions of
between 10 % and 20 % in fuel consumption with increased smoothness were
commonly noted in the literature. A 2000 study by the Federal Highway
Administration at Westrack reported fuel savings of 4.5 % on smooth (i.e.,
post-rehabilitation) versus rough (pre-rehabilitation) pavements [3].
Based on these and other findings, road agencies quickly adopted specifica-
tions to encourage contractors to place the smoothest pavements possible. A
survey of practices by Canadian provincial road agencies in 1999 by the Cana-
dian Strategic Highway Research Program indicated that all provinces incorpo-
rated some type of smoothness specification, with many offering a bonus for
enhanced smoothness and/or penalties for substandard work [4].
MTO Smoothness Requirements

In Ontario, Canada, the Ministry of Transportation (MTO) implemented a
bonus/penalty system for smoothness in the mid-1990s. Initially based on the
profile index and scallops (i.e., bumps) measured with a California Profilo-
graph, the specification is now based on the International Roughness Index
(IRI) analyzed locally (for bumps) and over regular intervals.
The MTO smoothness specification is evolving, and the most recent
version is provided in Non-Standard Special Provision No. 103F31M [5].
Under that specification, smoothness is measured using an inertial profiler
conforming to the minimum requirements of an ASTM Class 1 Profilometer
(ASTM E950-09 [6]). Furthermore, the MTO requires the use of a RoLine
3 kHz laser sensor, which effectively produces a 100 mm wide transverse laser
line to filter out the pavement macrotexture.
Three runs of the same road length are conducted. An average IRI value
(average of the left and right wheel path) is calculated using ProVal software
and is reported at 100 m sublot intervals. Table 1 provides the sublot pay fac-
tors for the IRI according to the MTO specification. Up to a 20 % pay bonus
can be achieved for sublots with an IRI less than 0.5 m/km, with a reduced bo-
nus achieved for sublots with an IRI between 0.5 and 0.65 m/km. Full pay (a
pay factor of 1.0) is applied for an IRI between 0.65 and 1.0 m/km, with penal-
ties calculated for IRIs between 1.0 and 1.25 m/km. Sublots with IRIs in excess
of 1.25 m/km are considered rejectable and require repair.
Repair options include diamond grinding, the placement of additional hot
mix asphalt (HMA) padding (if in a depressed area), or removal and replace-
ment with HMA. Repaired areas are assigned a pay factor of 1.0 regardless of
the improvement in smoothness attained.
As previously mentioned, additional penalties are also applied for
localized roughnessi.e., bumpsthat are identified in all three of the data
collection runs. Table 2 provides the penalties associated with localized rough-
ness incidents. As shown, a penalty of $1500 is applied for each incident of
localized roughness between 3.0 and 3.5 mm, and a penalty of $3000 is applied
to incidents between 3.55 and 4.7 mm. Incidents in excess of 4.7 mm are con-
sidered rejectable and must be repaired, unless the total allowable repair area is
exceeded, in which case they are assigned a penalty of $3000 also.
TABLE 1Ontario Ministry of Transportation smoothness pay factors.
Average IRI for a Sublot from a Set of

Three Measurements Taken by an Inertial
Profiler in Each 100 m Sublot, m/km Sublot Pay Factor
0.500 1.200a
>0.500 to 0.650 1.867 (1.333 IRI)a
>0.650 to 1.000 1.000
>1.000 to 1.250 1.40 (0.4 IRI)b
>1.250 REJECTEDb,c (requires repairs)
a
The payment factor will be equal to 1.000 for subsequent IRI measurements that are taken after repairs,
regardless of the reason for the repairs.
b
Sublots with IRIs between 1.000 and 1.350 m/km and located in areas of hot-in-place recycling being used
as a surface course (including hot-in-place with an integral overlay and hot in-place recycled premium mix)
will receive a payment factor of 1.00. Sublots with IRIs greater than 1.350 m/km and located in areas of hot-
in-place recycling being used as a surface course (including hot-in-place with an integral overlay and hot in-
place recycled premium mix) shall be repaired if within the total allowable repair area limit.
c
Repairs to a sublot will be permitted only up to the total allowable repair area limit. Any rejectable sublot
that is not allowed to be repaired because the total allowable repair area limit is exceeded will receive a pay-
ment factor of 0.500, in addition to any other penalties assessed for incidents of localized roughness within
that sublot that are also not allowed to be repaired.
TABLE 2Ontario Ministry of Transportation localized roughness penalties.
Incidents of Localized Roughness

(Determined from Three Sets of
Measurements Taken by an Iner-
tial Profiler, Run through ProVAL
Version 2.73.0033 and Rounded to
the Nearest 0.05 mm) Penalty
3.00 to 3.50 The contractor shall receive a penaltya,b of $1500 for each inci-
dent of localized roughness located in multi-lane freeways and
a penalty of $1250 for each incident of localized roughness
located in all other highway types. Repairs will be allowedb for
any incident of localized roughness in this amplitude range.
3.55 to 4.70 The contractor shall receive a penaltya,b of $3000 for each inci-
dent of localized roughness located in multi-lane freeways and
a penalty of $2500 for each incident of localized roughness
located in all other highway types. Repairs will be allowedb for
any incident of localized roughness in this amplitude range.
>4.70 All incidents of localized roughness shall be repaired in this
range.b For any additional incident of localized roughness that
is not allowed to be repaired because the total allowable repair
area limit has been exceeded, the contractor shall receive a
penalty of $3000 if it is located in a multi-lane freeway and
$2500 if it is located in any other highway type.
a
Incidents of localized roughness with a deviation between 3.00 and 4.70 mm that are located in areas where
hot-in-place recycling is used as a surface course (including hot-in-place recycling with an integral overlay
and hot in-place recycled premium mix) will receive no penalties.
b
As long as the total area for all surface smoothness-related repairs has not exceeded the total allowable
repair area limit, and subject to the restrictions on repairs stated in clause 313.08.01.05.04.01 entitled
Repairs, the contractor may repair an incident of localized roughness with a deviation greater than 3.00
mm. If the repair removes that incident of localized roughness, then the penalty for that incident of localized
roughness will be waived. Any additional incident of localized roughness that is not allowed to be repaired
because the total allowable repair area limit has been exceeded shall receive the above-specified penalty.
The potential bonus for high initial smoothness in Ontario is therefore con-
siderable, as is the potential penalty for excessive roughness. Furthermore, the
ability to correct areas of excessive roughness after placement is greatly limited
by the MTO, because the public does not wish to observe a new pavement sur-
face being damaged by a milling machine. The total allowable repair area is
currently limited to 5 % of the project area.
Contractor Perspective and Objective of Investigation

Paving contractors want to construct smooth pavements not only in order to
maximize the amount of any applicable bonus, but also as a point of pride of
workmanship. Many rehabilitation projects in Ontario do not simply entail
milling and the placement of a single layer of new HMA but involve in-place
recycling, multiple lifts of new HMA, or a combination of both. As a result,
there is an opportunity for contractors to measure smoothness on intermediate

layers in order to identify areas of excessive roughness for repair prior to the
placement of the surface HMA. By providing a smooth base, the contractor
helps to ensure that the final surface layer is as smooth as possible.
However, once the smoothness of a base layer is determined, the contractor
must decide whether identified areas of elevated roughness will be sufficiently
smoothed by the surface HMA layer or whether milling or grinding is first
required (at additional cost). The primary objective of this investigation was
therefore to determine the incremental improvement in smoothness that may
be expected with each additional layer of HMA placed on an in-place recycled
base.
Project Description
A large rehabilitation project involving in-place recycling of the existing HMA
followed by multiple layers of new HMA was considered the ideal candidate
for this investigation. The project selected involved the rehabilitation of High-
way 417 from a point 0.7 km west of Panmure Road. (Interchange 163) to a
point 1.0 km east of Highway 7 (Interchange 145) under MTO Project
20104000. This 18.6 km section of Highway 417 is a four-lane divided free-
way with a posted speed limit of 100 km/h. Highway 417 is a major eastwest
corridor connecting Canadas fourth and second largest cities, stretching from
the city of Ottawa in eastern Ontario (1.13 106 people) for 181 km eastward
to the border of the province of Quebec, with access to the city of Montreal
(3.64 106 people).
The rehabilitation strategy involved cold-in-place recycling of the existing
eastbound lanes to a depth of 110 mm, followed by the placement of 50 mm of
new Superpave 19 Traffic Category D HMA with PG 64-34 asphalt cement,
which was in turn followed by 40 mm of new Superpave 12.5 Friction Course
2 Traffic Category D HMA with PG 64-34 asphalt cement. In Ontario, Traffic
Category D represents a 20 year equivalent single axle load design life of
10 106 to 30 106. For this project, the in-place recycled mix was stabilized
with expanded asphalt, as opposed to emulsion, and is referred to as cold-in-
place recycled expanded asphalt mix (CIREAM).
An AMES high speed inertial profiler incorporating RoLine 3 kHz laser
sensors was used to collect pavement smoothness data on the surface of the
CIREAM, the surface of the Superpave 19, and the final surface of the Super-
pave 12.5 FC2. The average IRI value of the left and right wheelpaths was gen-
erated and reported at 100 m sublot intervals using ProVal software as required
by the MTO.
As previously indicated, the primary objective of this investigation was to
assess the incremental improvement in pavement smoothness obtained with
each HMA layer, so that contractors can better determine whether base repairs
are warranted prior to placement of the surface course. The results of the inves-
tigation are now presented.
Results and Discussion
Sublot International Roughness Index

Summary statistics for the 100 m sublot IRI values collected in the driving lane
and the passing lane are provided in Table 3.
Although the CIREAM material is placed with a paver, the recycling train
must follow the profile of the existing roadway to obtain sufficient material,
and it therefore has a limited ability to improve smoothness relative to the
existing condition (in this case, the existing smoothness was not measured).
The average CIREAM sublot IRI was 1.81 and 1.55 m/km for the driving and
passing lanes, respectively.
A considerable improvement in smoothness was observed after placement
of the Superpave 19 base HMA. As shown in Table 3, the average sublot IRI
was reduced to 0.91 and 0.83 m/km for the driving and passing lanes, respec-
tively, representing reductions of 45 % and 42 %. The improvement in smooth-
ness between the CIREAM and the Superpave 19 HMA base layer was
anticipated, as the HMA paver utilizes a 10 m sonic sensor to improve the lon-
gitudinal profile.
A further reduction in the sublot IRI was observed with the subsequent
placement of the Superpave 12.5 FC2 surface layer. As shown in Table 3, the
average sublot IRI was reduced to 0.64 and 0.60 m/km for the driving and pass-
ing lanes, respectively, representing reductions of 23 % and 25 %.
TABLE 3Sublot International Roughness Index (IRI) summary statistics for Highway 417.
CIREAMa Superpave 19 Superpave 12.5

Base, m/km Base, m/km Surface, m/km
Driving lane Average 1.81 45% 0.91 23% 0.64
Minimum 0.80 Average 0.65 Average 0.45
Maximum 4.22 Reduction 1.49 Reduction 1.23
Standard deviation 0.63 0.19 0.15
Passing lane Average 1.55 42% 0.83 25% 0.60

Minimum 0.75 Average 0.52 Average 0.37
Maximum 3.04 Reduction 1.56 Reduction 1.28
Standard deviation 0.42 0.22 0.17
a
Cold-in-place recycled expanded asphalt mix.
Localized Roughness
Summary statistics for the number of deviations (i.e., instances of localized
roughness), the total length of deviations, and the absolute value of the depth
(or height) of the deviations are presented in Table 4.
As shown, a reduction of almost 1 order of magnitude in the number of
localized roughness deviations was observed from the CIREAM layer (approx-
imately 200) to the Superpave 19 base HMA layer (approximately 20) to the
Superpave 12.5 surface HMA layer (approximately 5).
A similar relationship was observed for the total length of the measured
deviations (i.e., the distance over which deviations were measured). The CIR-
EAM layer displayed 207.2 and 138.5 m of deviations for the driving and pass-
ing lanes, respectively, whereas the length of deviations for the Superpave 19
was 23 and 35.5 m for the driving and passing lanes, respectively. The length
was further reduced with the placement of the Superpave 12.5 FC2 layer, in
which only 7.4 and 8.7 m of deviations were measured in the driving and pass-
ing lanes, respectively.
The magnitude of the deviations was likewise greatly reduced. The abso-
lute values of the depths (or heights) of deviations measured were summed and
are also presented in Table 4. As shown, approximately 1000 mm of total devi-
ations were measured on the CIREAM surfaces; this value dropped to approxi-
mately 100 mm for the Superpave 19 surfaces and approximately 20 mm for
the final Superpave 12.5 FC2 surfaces.
For the remaining deviations measured on the surface HMA, it was of in-
terest to know whether an associated deviation was present in the underlying
base course HMA and therefore transferred upward to the surface, or whether
the deviation was imposed during the paving of the surface layer. The data pre-
sented in Table 5 suggest that some of the remaining deviations might have
been transferred upward from the underlying base HMA, although most appear
to have been imposed solely by the surface paving operation.
TABLE 4Localized roughness summary statistics for Highway 417.
CIREAMa Superpave Superpave

Base 19 Base 12.5 Surface
Driving lane Number of deviations 213 20 5
Total length of deviations, m 207.2 23 7.4
Absolute value of depth of deviations, mm 1003.2 87 21.3
Passing lane Number of deviations 206 30 6
Total length of deviations, m 138.5 35.5 8.7
Absolute value of depth of deviations, mm 901.1 120.4 26.0
a
TABLE 5Localized roughness deviations on surface HMA for Highway 417.
Superpave 19 Base Superpave 12.5 Surface
Deviation Depth, Deviation Length, Deviation Depth, Deviation Length,

mm m mm m
Driving lane 6.38 2.1 5.29 1.5
No associated deviation in base HMA 3.4 0.9
4.23 1.1
4.37 2.6
3.97 1.3
Passing lane 3.82 0.8 6.17 2.1
4.441 0.5 5.17 1.9
3.468 0.5 3.35 0.8
No associated deviation in base HMA 4.05 1.8
3.67 1.3
3.63 0.8
Note: Negative deviation values suggest a bump, whereas positive deviation values suggest a dip in the
surface.
Pay Factors
Using the specification ranges provided in Tables 1 and 2, the pay factors asso-
ciated with the measured smoothness values were then determined. Pay factors
would not normally be calculated for the base layer smoothness, as only the
final surface is evaluated. However, for comparison purposes, pay factors for
all surfaces evaluated were calculated.
Table 6 provides the average sublot pay factor for each layer, as well as the
number of rejectable lots and additional penalties associated with localized
roughness. As shown, a considerable pay reduction would be associated with
the CIREAM surface if it represented the final product.
TABLE 6Pay factors for sublot IRI and penalties for localized roughness on Highway 417.
CIREAMa Superpave Superpave

Base 19 Base 12.5 Surface
Driving lane Average sublot IRI pay factor 0.731 0.994 1.107
Number of rejectable sublots 45 0 0
Penalties for localized roughness $496 500 $52 500 $15 000
Passing lane Average sublot IRI pay factor 0.846 1.006 1.128
Number of rejectable sublots 51 5 0
Penalties for localized roughness $486 000 $75 000 $16 500
a
Note: Penalties for localized roughness on CIREAM and Superpave 19 provided only for order of
magnitude and not actually applicable to project.
Of interest is the pay factor for the Superpave 19 base HMA, which hov-
ered between a slight penalty and a slight bonus for the driving and passing
lanes, respectively. Many rehabilitation projects involve only in-place recy-
cling and a single layer of new HMA, and it is questionable whether the results
obtained on the Superpave 19 in this investigation would be similar to those
obtained on a project in which a Superpave 12.5 mix was paved directly on top
of a CIREAM base.
The second and final layer of HMA (Superpave 12.5 FC2) produced a sur-
face that was well into the pay bonus category, eligible for bonuses ranging
from 10.7 % to 12.8 % for the driving and passing lanes, respectively.
Conclusions
Although only one project was evaluated during this investigation, a number of
tentative conclusions and points for further discussion may be drawn as
follows:
For this multi-layer pavement rehabilitation project, considerable
improvement in smoothness was observed with each additional layer

constructed. A reduction in the IRI of 42 % to 45 % was observed from
the in-place recycled (CIREAM) surface to the Superpave 19 base HMA
surface, and a reduction in the IRI of 23 % to 25 % was observed from
the Superpave 19 base HMA surface to the final Superpave 12.5 FC2
surface. Contractors may now use these percentage improvement values
to determine whether or not repairs should be made to the base layers
prior to the placement of the surface HMA.
With respect to localized roughness (i.e., bumps and dips), a reduction
by 1 order of magnitude in the number, length, and magnitude of the

deviations was observed with each additional layer constructed.
Localized roughness deviations measured in the final surface layer
mostly did not appear to be related to deviations in the underlying base

HMA and appeared to be the result of the surface paving operation
alone.
Although a considerable improvement in smoothness was observed on
the base HMA layer, the resulting pay factor was essentially 1.0 (i.e.,
full payment but no bonus), whereas a considerable bonus (10 % to 12 %)
was achieved with the additional surface HMA layer. Given that many
rehabilitation projects involve a single layer of new HMA over a recycled
surface, one must wonder whether it is possible to achieve a bonus in that
scenario.
Of course, continued data collection and analysis on additional
rehabilitation projects are recommended in order to confirm these initial
findings.
References
[1] Smith, K. L., Smith, K. D., Evans, L. D., Hoerner, T. E., and Darter,
M. I., Smoothness Specifications for Pavements, National Cooperative
Highway Research Program (NCHRP) Web Document 1, Transportation
Research Board, Washington, D.C., 1997.
[2] Marks, H., Smoothness Matters: The Influence of Pavement on Fuel
Consumption, HMAT: Hot Mix Asphalt Technology, Vol. 14, No. 6,
2009, pp. 1819.
[3] Sime, M., Ashmore, S. C., and Alavi, S., WesTrack Track Roughness,
Fuel Consumption, and Maintenance Costs, FHWA-RD-00-052, Federal
Highway Administration, Washington, D.C., 2000.
[4] Canadian Strategic Highway Research Program (C-SHRP), Summary of
Pavement Smoothness Specifications in Canada and Around the World,
Technical Brief #16, Transportation Association of Canada, Ottawa, ON,
Canada, 1999.
[5] Non Standard Special Provision No. 103F31M, Asphaltic Concrete Pay-
ment Adjustment for Surface Smoothness Based on Quality Assurance
Measurements Taken by an Inertial Profiler, Ontario Ministry of Trans-
portation, Downsview, Ontario, Canada, March 2012.
[6] ASTM E950/E950M09, 2009, Standard Test Method for Measuring the
Longitudinal Profile of Traveled Surfaces with an Accelerometer Estab-
lished Inertial Profiling Reference, Annual Book of ASTM Standards,
Vol. 04.03, ASTM International, West Conshohocken, PA.
STP 1555, 2012
DOI:10.1520/STP104420
Steven M. Karamihas1
Golden-Car Simulation Speed and Its

Implications to the Relevance of the IRI
REFERENCE: Karamihas, Steven M., Golden-Car Simulation Speed and Its

Implications to the Relevance of the IRI, Pavement Performance: Current
Trends, Advances, and Challenges on December 5, 2011 in Tampa, FL;
STP 1555, B. Choubane, Editor, pp. 248266, doi:10.1520/STP104420,
ABSTRACT: The road-monitoring community is considering alternatives to
the International Roughness Index (IRI) that are based on the Golden-Car
model, but perform the simulation at a speed other than standard value of
80 km/h. Many state and federal highway agencies apply pay adjustments to
new road and road resurfacing contracts based on the IRI, but expect signifi-
cantly higher prevailing travel speeds on very smooth limited-access roads.
Inertial profilers also continue to propagate as a tool for monitoring pavement
conditions on lower-speed roads, where the posted speed limit or the prevail-
ing traffic speed may be much lower than 80 km/h. The broad application of
the IRI to higher and lower-speed roads has created interest by practitioners
in the issue of Golden-Car model simulation speed. Changing the simulated
forward speed causes predictable changes in the frequency response of the
Golden-Car model: the temporal frequency response changes very little, but
the spatial frequency response shifts in proportion to speed. However, the
quantitative effect on the resulting index values and their relationship to other
relevant responses, such as tire load variations and passenger acceleration,
are less obvious. This paper explores the potential consequences of changes
to the Golden-Car model simulation speed, including changes to the output
roughness values and meaning of the roughness scale and roughness
threshold values. Most of the discussion relies on numerical and theoretical
analysis, with some sample measurements provided for elucidation.
KEYWORDS: road roughness, International Roughness Index, road profile,
vehicle simulation
1
Univ. of Michigan, Dept. of Mechanical Engineering, 2901 Baxter Rd., Ann Arbor, MI 48109-
2150, e-mail: stevemk@umich.edu
Copyright V
19428-2959.
248
KARAMIHAS, doi:10.1520/STP104420 249
Introduction
The International Roughness Index (IRI) developed over about a decade of
domestic and international research. Early work for the National Cooperative
Highway Research Program (NCHRP) produced a roughness index based on a
reference quarter-car simulation (RQCS), which would serve as a stable cor-
relation reference for response-type road roughness measuring systems
(RTRRMS) (e.g., Mays meters, etc.) [1]. The definition of the RQCS included
five settings that were optimized for correlation to the existing fleet of
RTRRMS (see Fig. 1). The RQCS predicted the average rectified velocity
(ARV) across the suspension of a vehicle with standard properties in response
to a measured longitudinal road profile.
At the time, researchers explained the trade-off between ARV and its spa-
tial equivalent, average rectified slope (ARS) [2]. ARV provided a closer repre-
sentation of the way a vehicle and its occupants experienced a road, because it
was a temporal quantity and a higher value meant a higher intensity of vehicle
vibration [3]. However, the ARV included the influence of both the road and
the forward speed in the judgment of road roughness. ARS, which was simply
ARV divided by forward speed, was more readily viewed solely as a geometric
property of the road surface. As such, it provided an avenue for casting
RTRRMS output to a standard scale for comparing roads.
The development of the RQCS into the IRI occurred under the auspices of
the World Bank [4]. The World Bank research finalized the IRI by fixing the
RQCS speed at 49.7 mph (80 km/h) and specifying average rectified slope as
the scale [5]. This gave the IRI a well-defined relationship to the longitudinal
profile of the road and cast it onto a scale with meaning that would not change
over time. The selection of 49.7 mph (80 km/h) over lower alternatives
FIG. 1The RQCS (after Sayers, 1995).

involved a trade-off between conflicting considerations [5]. Nevertheless, the

choice owed to:
Compatibility with other roughness indices that were in use at the time,
Practical considerations related to RTRRMS measurement efficiency,
Superior correlation to subjective ratings, and
The expectation that the IRI would be implemented widely on high-
speed, high-volume roads

The RQCS itself remained consistent since its introduction in the NCHRP
work, and was later given the nickname the Golden-Car model [2,6].
The road-monitoring community is considering alternatives to the IRI that
are based on the Golden-Car model, but perform the simulation at a speed other
than 49.7 mph (80 km/h). Many state and federal highway agencies apply pay
adjustments to new road and road resurfacing contracts based in the IRI, but
expect significantly higher prevailing travel speeds on very smooth limited-
access roads. These roads represent a contrast to the roads that were used in the
research that supported the development of the IRI, which were often very
rough, lower-speed roads. In addition, the IRI was finalized while a federally
imposed maximum speed limit of 55 mph (88.5 km/h) was in effect in the
U.S., and it gained acceptance during ongoing state-by-state speed limit
increases from 55 mph (88.5 km/h) [7]. Inertial profilers also continue to propa-
gate as a tool for monitoring pavement condition on lower-speed roads, where
the posted speed limit or the prevailing traffic speed may be much lower than
49.7 mph (80 km/h). The broad application of the IRI to higher and lower-
speed roads has created interest by practitioners in the issue of Golden-Car
model simulation speed.
Changing the simulated forward speed causes predictable changes in the
frequency response of the Golden-Car model: the temporal frequency response
changes very little, but the spatial frequency response shifts in proportion to
speed. However, the quantitative effect on the resulting index values and their
relationship to other relevant responses, such as tire load variations and passen-
ger acceleration, are less obvious. This paper explores the potential consequen-
ces of changes to the Golden-Car model simulation speed, including changes
to the output roughness values and meaning of the roughness scale and rough-
ness threshold values. Most of the discussion relies on numerical and theoreti-
cal analysis, with some sample measurements provided for elucidation.
The Golden-Car Model

An IRI value for a given longitudinal elevation profile is obtained in four steps
[6]:
Step 1. Convert the elevation profile to slope profile using the difference
between adjacent points.
Step 2. Apply a moving average with a base length 9.84 in. (250 mm).
Step 3. Pass the smoothened profile through the Golden-Car model using a
simulated forward speed of 49.7 mph (80 km/h). This predicts the rate of
suspension stroke in a vehicle with standard properties (see Fig. 1).
Step 4. Calculate the average rectified value from the Golden-Car model output
trace.
The influence of the moving average (Step 2) is a fixed function of spatial
frequency, and will interact with the profile the same way regardless of simu-
lated travel speed. In contrast, the response of the Golden-Car model depends
on temporal frequency, and the way it responds to the profile depends heavily
on travel speed.
Step 1: Conversion to Slope

Step 1 is often ignored when users of the IRI interpret its meaning. The Golden-
Car model at the core of the IRI algorithm includes four state variables: the verti-
cal position and rate of change of vertical position of each mass. The output of
the model is the difference between the vertical position of the sprung mass and
the vertical position of the unsprung mass, calculated at a constant longitudinal
distance interval. However, since the input trace provided to the model is a slope
profile, rather than an elevation profile, the output trace is the spatial rate of
change of suspension stoke, rather than the suspension stroke, versus distance.
While the IRI is often described as a measure of predicted suspension stroke
divided by distance traveled, it is actually the average rectified value of the out-
put trace after the filters in Steps 1 through 3 have been applied. In this regard,
the IRI algorithm is a bandpass filter with output in units of slope (e.g., in./mi,
m/km, etc.). The composite frequency response was tailored for broad relevance
to vehicle dynamics, including a legacy of correlation to several RTRRMS.
Step 2: Moving Average

The moving average simulates tire envelopment. More complicated models
of tire envelopment are possible, but the moving average was found to repro-
duce test results for wavelengths greater than the base length of the moving
average [1].
Figure 2 shows the sensitivity of the moving average as a function of wave
number. A notch appears at the reciprocal of the base length and its multiples.
For wave numbers below the first notch, the moving average may be
described as a low-pass filter with a cut-off wave number of 0.540 cycles/ft
(1.772 cycles/m), or a cut-off wavelength of 1.851 ft (0.564 m). The 9.84-in.
(250-mm) base length of the moving average is linked to the geometry of the
tire, and it does not change with forward speed. As a result, the sensitivity of
the moving average depends on spatial frequency, but its relationship to tempo-
ral frequency is affected by travel speed.
FIG. 2Gain for a 9.84-inch (250 mm) moving average.
Step 3: The Golden-Car Model

Figure 3 shows the gain for profile slope of the Golden-Car model as a function
of temporal frequency. The gain function includes peak responses at 1.41 Hz
and 9.69 Hz. The peaks correspond to resonance vibration of the sprung mass
and unsprung mass, respectively. The frequencies at these peaks provide some
insight into the behavior of the Golden Car. However, they should be
FIG. 3Golden-Car filter gain.

interpreted with caution. The settings used in the Golden-Car model do not
describe a specific passenger car [6].
The IRI was intended to provide broad relevance to as much of the prevail-
ing vehicle fleet as possible. To do this, the researchers that developed the
Golden-Car model increased the suspensions damping coefficient to dull the
resonance peaks and flatten out the frequency response over the range of inter-
est [1]. The Golden-Car model frequency response offered a contrast to previ-
ous RTRRMS simulations used in practice at the time, which simulated a
specific vehicle of interest or a RTRRMS host vehicle. For example, a well-
known quarter-car simulator that preceded the development of the RQSC used
parameters that represented a 1968 Chevrolet Impala [8]. Because of the more
representative, but softer damping coefficient used in this simulation, the res-
onance peaks in the frequency response are much more localized.
The flat, relatively uniform frequency response of the Golden-Car model
helps maintain the relevance of the index to as many vehicles as possible, and
prevent strong interactions between roughness that is isolated at a given wave-
length and resonance vibration of the Golden Car (i.e., tuning). With this in
mind, the term Golden-Car model provides a more instructive name for
Step 3.
The Golden-Car model is most sensitive to inputs from 0.77 Hz to 17.48
Hz. Temporal frequency (f), wave number () (i.e., spatial frequency), and
wavelength (k) are related by travel speed (V) as follows:
1 f
(1)
k V
At the standard travel speed of 49.7 mph (80 km/h) for the IRI, the Golden-Car
model is sensitive to wave numbers from 0.0105 cycles/ft (0.0346 cycles/m) to
0.240 cycles/ft (0.787 cycles/m) and wavelengths between 94.8 ft (28.9 m) and
4.17 ft (1.27 m) [9].
Travel speed determines the level of the interaction between the moving
average and the Golden-Car model [3]. At 49.7 mph (80 km/h), the peak
response of the Golden-Car model occurs at a wavelength of 7.52 ft (2.29 m).
At this wavelength, the moving average attenuates content in the profile by 2
% (i.e., the gain of the response function is 0.98). When the speed is reduced to
24.9 mph (40 km/h), the peak occurs at a wavelength of 3.76 ft (1.15 m). The
moving average attenuates content at this wavelength by 8 %, and attenuates
content at the border of the sensitive range described above by nearly 24 %.
Step 4: Average Rectified Output
The trace produced by Steps 1 through 3 is reduced to a single index value by
calculating the average rectified slope. The average rectified value was selected
instead of the root mean square (RMS) because:
It allows for more convenient aggregation of IRI values from adjacent

segments [6].
Characterization of localized roughness with a more convenient inter-
pretation is possible [1012].
RTRRMS accumulated contributions to their summary index values the
same way [1].
Theoretical Analysis
Basic Linear Theory

Theoretical examination of the index values produced by the Golden-Car
model requires simple linear analysis. If a stationary random signal passes
through a linear process the spectral density of the output profile is related to
the spectral density of the input signal by
Gin jH j2 Gout (2)
where:
Gin spectral density of the input signal, expressed as a function of wave
number,
H linear response gain, and
Gout spectral density of the output signal.
Figure 4 provides a corresponding graphical example similar to other tuto-
rial explanations of the quarter-car model and the IRI as an illustration of the
inputoutput relationship implied by Steps 1 through 3 [1,1316].
The plot at the left of Fig. 4 shows an idealized spectrum of road profile
slope with a common shape for flexible constructions [17]. Since the input is
provided as profile slope, it includes the effect of Step 1.
FIG. 4Linear analysis with the quarter-car model.

The plot in the center of Fig. 4 provides the combined influence of Steps 2
and 3. Sayers et al. [3] and Gillespie et al. [1] provided expressions for the
transfer function of the moving average, Hma(), and the Golden Car, HGC(),
respectively
sinpB
Hma (3)
pB
K1 x2
HGC x 2pV
lx4 K1 K2 1 lx2 K1 K2 jCxK1 x2 1 l
(4)
where the function H in Eq 2 is the product of the two transfer functions. Fig-
ure 4 corresponds to the IRI calculation, so it uses the standard values of V, B,
K1, K2, C, and l (see Fig. 1).
The plot at the right side of Fig. 4 is the spectrum of the output signal after
application of Steps 1 through 3. The figure shows that the spectral properties
of the output trace possess properties of the input trace and the transfer func-
tion. For example, the skew of the output toward low-frequency content is
caused by the skew in the profile slope spectral density toward low-frequency
content. Further, the two rounded peaks and the lack of content at the extremes
of the frequency range under display reflect the shape of the transfer function.
Unlike the input, the output is band limited. Steps 1 through 3 apply high
weighting to those features of the longitudinal profile that are considered most
relevant and filter out those features that are not. This eliminates contributions
from the very long and very short wavelength range.
Like the input, the output has units of slope, and the output spectrum has
units of ft/cycle. The mean square of the output signal is equal to the integral
under the output spectrum. Step 4 specifies that the IRI is the average rectified
value of the output signal. If the output signal is Gaussian with zero mean, the
average rectified value is proportional to the variance, which is in turn the
square root of the integral under the output spectrum
r
2
GCARS r (5)
p
where GCARS is the Golden-Car average rectified slope. When the speed (V) is
set to 49.7 mph (80 km/h) (with proper units), GCARS is equal to the IRI. For
other choices of forward speed, an appropriate subscript is added (e.g., GCARSV).
Idealized Input
Sayers [18] proposed a spectral model for road profile that combines white
noise elevation, white noise slope, and white noise (spatial) acceleration. For
spectral density of slope, the model is
Ge Ga
Gin 2
Gs (6)
2p 2p 2
where Ge, Gs, and Ga are constants that are typically fitted to a measured spec-
tral density plot.
For the analysis that follows, four coefficient sets are applied:
1. A wavy road with a share of white noise acceleration versus white
noise slope at the extreme of what has been observed on smooth new
pavement with marginal grade control,
2. A standard coefficient set equivalent to the suggested spectral model
for flexible constructions by La Barre et al. [17] used in Fig. 4,
3. Pure white noise slope, and
4. A choppy road with a share of white noise elevation versus white
noise slope at the extreme of what has been observed on old pavement
with significant distress.
Table 1 lists the pertinent numerical values for each sample road.
Golden-Car Average Rectified Slope

Figure 5 shows the change in GCARS with speed for three of the idealized
sample roads listed in Table 1. The roughness of each idealized profile was
scaled to produce an IRI value of 100 in./mi (1.58 m/km). These calculations
were performed numerically in the frequency domain using Eqs 26 and veri-
fied using time domain simulation over synthesized profiles with matching
spectral properties.
GCARS decreases with speed for the choppy road, increases with speed
for the wavy road, and comes to a minimum at about 47 mph (76 km/h) for the
standard road. Figure 5 demonstrates that GCARS is subject to changes with
speed, the magnitude of the change depends on the properties of the profile,
and that the rank order of a set of pavements with diverse properties is likely to
change with speed.
The effect of forward speed on the value of GCARS for a given profile
depends on three competing effects. First, the transfer function of the Golden-
Car filter, expressed as a function of wave number, shifts in inverse proportion
to speed. Figure 6 shows the change in the Golden-Car filter wave-number
TABLE 1Model coefficients for four idealized sample roads.
Description Ga=Gs (ft2) Ge=Gs (ft2)

Wavy 0.279 0
Standard 0.098 0
White noise slope 0 0
Choppy 0 0.86
FIG. 5GCARS versus speed.
response when the speed is changed from 50 mph (80 km/h) to 25 mph
(40 km/h). Reducing the speed by half increased the area under the Golden-Car
filter gain function by a factor of two. The estimated GCARS is proportional to
the square of the area under the output spectrum (see Eq 5). For a flat input
spectrum (i.e., white noise slope), the GCARS varies inversely with the square
root of speed (with some modification at low speeds caused by the moving
average). The physical explanation for this is that for the same (temporal) in-
tensity of output, more time is available for contributions to the GCARS to
build up over the same distance. Mathematically, this effect is a consequence
of scaling from velocity to slope.
Second, as speed decreases, the moving average attenuates a progressively
greater portion of the content in the waveband that the Golden-Car filter passes.
For example, on the choppy road the moving average reduced the GCARS by
2.1 % at 75 mph (120 km/h) and by 16.9 % at 25 mph (40 km/h).
Third, as speed changes, the spectral content of the profile determines
whether the content added to the output is greater than the content that is
dropped as the Golden-Car filter gain function shifts along the wave number
scale (see Fig. 6). This explains the difference in the trends among the three
examples in Fig. 5. In the case of the wavy road, the increased emphasis on
long wavelength content within the profile as speed increases trumps the
scaling effect described above, and the GCARS increases with speed. The
opposite is true of the choppy road, which includes spectral content that
increases in significance at short wavelengths. On the choppy road, the scaling
FIG. 6Golden-Car filter gain at two speeds.
effect and the profile content combine to cause an aggressive increase in

GCARS as speed decreases.
A less orderly relationship between GCARS and forward speed than Fig. 5
illustrates is possible if the spectral content of the profile is not as smooth as
the content implied by the Sayers model. Perera demonstrated common trends
for practical examples using Portland cement concrete (PCC) and asphalt con-
crete (AC) profiles [19]. In particular, when profile content is concentrated in a
narrow waveband, large changes in speed may cause the isolated content in the
profile to interact with one of the peaks in the Golden-Car filter response.
Golden-Car Average Rectified Velocity

As forward speed changes, GCARS does not necessarily rise or fall in propor-
tion to the change in vibration intensity. Figure 7 shows the change in Golden-
Car average rectified velocity (GCARV) with speed for three of the idealized
sample roads. Each point in the plot is simply the GCARS, shown in Fig. 5,
scaled by speed. On all three sample profiles, the GCARV increases with
speed. This shows that the relationship of the GCARS scale to the vibration
level present in a vehicle changes with speed. With that in mind, the GCARS
scale requires a unique set of threshold roughness targets for classifying the
ride quality of roads.
The exchange in trends between a spatial index (GCARS) and a temporal
index (GCARV) is consistent with measured responses. Figure 8 shows the
GCARS for 18 passes over 16 seconds of travel on a new PCC surface. The
passes include three repeat measurements at each of six speeds with a common
starting point. For each pass, the GCARS was calculated using the average
travel speed. For this pavement, GCARS20mph was more than 40 % greater
than GCARS70mph.
FIG. 7GCARV versus speed.
The sport utility vehicle that hosted the profile measurement system also
included sensors for measurement of accelerations at driver/vehicle interfaces.
Figure 9 shows the point vibration total for the location of the floor/foot inter-
face. The point vibration total combines the vertical, longitudinal, and lateral
using root mean square after the application of frequency weighting functions
and linear scaling coefficients specified by ISO 2631 for estimating comfort.
Unlike the GCARS, the measured temporal vibration at the floor/foot interface
FIG. 8GCARS versus speed for 18 passes over the same road.
FIG. 9Sprung mass acceleration versus speed for 18 passes over the same
road.
increases with speed. The point vibration total is more than 70 % higher at 70
mph (112 km/h) than it is at 20 mph (32 km/h). Vibration at the seat/buttock
interface and seat/back interface also increased with speed.
The decrease in GCARS with speed, while the intensity of vibration expe-
rienced in the vehicle increases shows that a threshold IRI value does not trans-
late to other forward travel speeds.
Other Golden-Car Responses

Suspension design involves balancing performance requirements for ride com-
fort, road holding, and suspension travel space [20]. Optimizing performance
in one category typically requires a designer to reduce performance in at least
one of the others. The quarter-car model is a common tool for examining these
trade-offs early in the vehicle development process [16], because it predicts
suspension stroke (for travel space), sprung mass acceleration (for ride com-
fort), and tire deflection (for road holding) [21,22].
The researchers that developed the IRI recognized that using the rate of
suspension stroke as an output for the Golden-Car model helped maintain rele-
vance of the index to ride comfort and road holding. This is because, at a given
speed, GCARS demonstrated a better relationship to both sprung mass acceler-
ation and tire deflection than they demonstrated to each other. Sayers [18] pro-
vided transfer functions for all three Golden-Car responses, and showed that
they are all sensitive to the same overall range of frequencies. Sprung mass
acceleration is somewhat more sensitive to long wavelength content (lower fre-
quencies) and tire deflection is somewhat more sensitive to short wavelength
FIG. 10GC RMS sprung mass acceleration versus speed.
content (higher frequencies). In this regard, the rate of suspension stroke

(GCARV) is the response in the middle.
Figure 10 shows RMS sprung mass acceleration for the golden car versus
forward speed on the idealized wavy, standard, and choppy roads. As in Figs. 5
and 7, each profile is scaled to produce an IRI of 100 in./mi (1.58 m/km). At
49.7 mph (80 km/h), the wavy road produces a higher level of RMS sprung
mass acceleration than the choppy road for the same IRI value. However, at
lower speeds, where profile content at shorter wavelengths is more important,
the choppy road produces a higher level of RMS sprung mass acceleration than
the other roads. This shows that a different type of road may dominate the com-
munitys opinion of the highest GCARS value that can be permitted without
intervention.
Figure 10 demonstrates that setting threshold values for acceptable levels
of GCARS at alternative forward speeds requires consideration of a broad,
diverse set of roads. Yu et al. [23] examined this, and proposed a set of GCARS
threshold values for a range of speeds based on maintaining and equivalent
level of jolt (i.e., time derivative of acceleration) on a large set of sample
profiles. Extending the analysis for relevance to general vehicle response is sig-
nificantly more complex, because other important response quantities need to
be considered that will not change with speed the same way.
Figure 11 shows the percentage change in RMS tire deflection and sprung
mass acceleration on the choppy road when the Golden-Car simulation speed is
altered from the standard value of 49.7 mph (80 km/h). Consideration of ride
FIG. 11Trends in GC RMS response with speed.
quality alone may lead to a set of thresholds values for this type of road that
provide the public with poor performance in terms of road holding, and vice
versa. On the choppy road, setting threshold values using ride quality alone
could lead to poor road holding at high speeds, and setting threshold values on
road holding alone could lead to poor ride quality at low speeds.
In the case of white noise slope, the sprung mass acceleration increased
strongly with speed, but tire deflection remained consistent over much of the
range. On the other two sample roads, both responses increase with speed over
the range covered by Fig. 11, but not by the same percentage.
Discussion
An important strength of the IRI is that its scale is standard. The IRI is a mathe-
matical function of the longitudinal profile of the road, with a calculation pro-
cedure and settings that are a fixed part of the definition. This has helped the
IRI gain acceptance in North America, because the IRI provides engineers with
a way to quantify the roughness of the road and communicate the status of
roads and road systems to each other on a common, time-stable scale. Further,
the pervasiveness of the IRI has helped engineers build experience with the
scale. This supported the development of thresholds for repair of roads that
cause excessive vehicle response, and threshold values for pay adjustments on
new pavement surfaces based on value provided to the public in the form of ei-
ther superior or inferior ride quality.
The road-monitoring community must proceed with caution before using

alternative indices derived from the IRI. Adjusting the Golden-Car model simu-
lation speed to a value more like the expected traveling speed on a road may
improve the relevance of the output. However, the community should weigh that
improvement against the potential loss to the standard nature of the IRI and the
loss of all the engineering experience behind common IRI threshold values.
Certainly, the definition of the IRI should not change. Giving alternative
indices an abbreviation far different from IRI will help mitigate the confusion
caused by roughness values reported in the same units. This paper proposes
that when the IRI algorithm is applied with any speed other than 49.7 mph (80
km/h), the index is named GCARSV, where GCARS is a Golden-Car average
rectified slope, and the specific travel speed (V) appears as a subscript. This is
consistent with the research that led to the IRI, which coined the name
reference average rectified slope with a subscript (e.g., RARSV) [1].
Theoretical calculations demonstrated some of the potential consequences
of altering the simulated forward speed used in the Golden-Car model. An as-
pect of the IRI that mitigates the downside of the mismatch between prevailing
travel speed on the road and the simulated travel speed of the index is its flat
frequency response. This helps the IRI maintain its relevance for modest speed
changes.
For speeds that are much different from the 49.7-mph (80-km/h) standard,
the average rectified slope for a given pavement is likely to change, and the
direction and magnitude of the change depend on the pavement type. Further,
the rank order of roughness for a diverse set of pavements will change. For
example, pavement with poor grade control will experience a progressively
greater penalty than pavements with highly localized distress as GCARS
speed increases, and vice versa. Published spectral models show that rigid
pavement (i.e., PCC) is likely to fare better than flexible pavement (i.e.,
asphalt) for GCARS at very high speeds, and that the opposite is true at very
low speed [15,17].
The link between a given GCARS value and the intensity of vibration
experienced by a vehicle and its occupant depends directly on speed. Even
though GCARS decreases as speed increases on many pavements, temporal
vibration typically increases. In simpler terms, the same amount of inches per
mile feels much more intense at high speed because the motion is felt over a
shorter period of time. This means that a unique set of threshold values are
required at each alternative speed.
Other quantities of interest, such as tire loads and passenger acceleration,
will change with speed differently than GCARS as speed changes. A robust
attempt to establish GCARS thresholds for other travel speeds requires consid-
eration of overall ride quality, truck dynamic loading, tire load variations for
road holding, user costs, and progression of IRI values over time in light of
pavement structural condition.
In addition, the publics expectations for ride quality, the share of traffic
accounted for by heavy trucks, and the constraints placed on road builders
(intersections, design grade, depth of preparation under the final surface, etc.)
may be very different on low-speed roads than on limited-access high-speed
roads. Together, these factors complicate any attempt to set thresholds for
GCARS based on equivalent performance at alternative speeds.
Two avenues are available to proceed if a GCARS scale is needed at other
speeds. First, the community may gradually build experience with an alterna-
tive scale by retaining GCARS values for a proposed speed besides the IRI
until the scale becomes familiar, or enough objective data accumulate to lend
meaning to the scale. (This effort should include storage of the profile data, in
case still other speeds come under consideration.) However, if the profile data
are not available, statistical translation of IRI to GCARS based on a limited
data set may lead to large errors, because the change in GCARS with speed is
so sensitive to profile spectral content.
Second, specific studies of measured and simulated vehicle response on the
types of roads of interest could provide the engineering context needed to es-
tablish GCARS thresholds at other speeds. At most, the community is inter-
ested in a scale for very smooth roads traveled at very high speeds (e.g.,
GCARS75mph) and at speeds common on lower functional classes (e.g.,
GCARS35mph). Research for both ends of the scale can be conducted with the
same calculation tools and simulation software, but an appropriate set of sam-
ple roads are needed for each speed.
In summary, threshold IRI values for triggering intervention on in-service
roads or pay adjustment on new road surfaces have evolved over time based on
pavement engineers experience with the scale. At alternative simulated for-
ward speeds, no such experience exists. Further, translation to a GCARS scale
that uses other speeds requires consideration of a complex combination of sev-
eral vehicle response quantities and a diverse set of possible road profiles.
References
[1] Gillespie, T. D., Sayers, M. W., and Segel, L., Calibration of Response-
Type Road. Roughness Measuring Systems, National Cooperative High-
way Research Program Report 228, 1980, 81 pp.
[2] Sayers, M. W., Development, Implementation, and Application of the
Reference Quarter-Car Simulation, Measuring Road. Roughness and Its
Effects on User Cost and Comfort, ASTM STP 884, T. D. Gillespie and
M. Sayers, Eds., ASTM International, West Conshohocken, PA, 1985, pp.
2547.
[3] Sayers, M. W., Gillespie, T. D., and Queiroz, C. A. V., The International
Road. Roughness Experiment, World Bank Technical Paper Number 45,
1986, 453 pp.
[4] Sayers, M. W. and Karamihas, S. M., The Little Book of Profiling, Univer-
sity of Michigan Transportation Research Institute, 1998, 100 pp.
[5] Sayers, M. W., Gillespie, T. D., and Patterson, W. D. O., Guidelines for
Conducting and Calibrating Road. Roughness Measurements, World
Bank Technical Paper Number 46, 1986, 87 pp.
[6] Sayers, M. W., On the Calculation of International Roughness Index from
Longitudinal Road. Profile, Trans. Res. Rec., Vol. 1501, 1995, pp. 112.
[7] Baum, H. M., Wells, J. K., and Lund, A. K., Motor Vehicle Crash Fatal-
ities in the Second Year of 65 MPH Speed Limits, J. Safety Res., Vol.
21, 1990, pp. 18.
[8] Burchett, J. L., et al., Surface Dynamics Profilometer and Quarter-Car
Simulator: Description, Evaluation, and Adaptation, Kentucky Bureau of
Highways Research Report No. 465, 1977, 41 pp.
[9] Karamihas, S. M., et al., Guidelines for Longitudinal Pavement Profile
Measurement, National Cooperative Highway Research Program Report
434, 1999, 75 pp.
[10] Sayers, M. W., Profiles of Roughness, Trans. Res. Rec., Vol. 1260,
1990, pp. 106111.
[11] Swan, M. and Karamihas, S. M., Use of Ride Quality Index for Con-
struction Quality Control and Acceptance Specifications, Trans. Res.
Rec., Vol. 1861, 2003, pp. 1016.
[12] Karamihas, S. M. and Senn, K., Profile Analysis of LTPP SPS-6 Site in
Arizona, University of Michigan Transportation Research Institute
Report UMTRI-2010-17, 2010, 80 pp.
[13] Curtis, A. J. and Broykin, T. R., Jr., Response of Two-Degree-of-Freedom
Systems to White Noise Base Excitation, J. Acoust. Soc. Am., Vol. 33,
1961, pp. 655663.
[14] Crandal, S. H. and Mark, W. D., Random Vibration in Mechanical Sys-
tems, Academic Press, Orlando, FL, 1963.
[15] Sayers, M. W., Dynamic Terrain Inputs to Predict Structural Integrity of
Ground Vehicles, University of Michigan Transportation Research Insti-
tute Report UMTRI-88-16, 1988, 114 pp.
[16] Gillespie, T. D., Fundamentals of Vehicle Dynamics, Society of Automo-
tive Engineers, Warrendale, PA, 1992.
[17] La Barre, R. P., et al., The Measurement and Analysis of Road. Surface
Roughness, Motor Industry Research Association, Report 1970/5, 1970,
31 pp.
[18] Sayers, M. W., Characteristic Power Spectral Density Functions for Ver-
tical and Roll Components of Road. Roughness, Am. Soc. Mech. Eng.,
Vol. 80, 1986, pp. 113129.
[19] Perera, R. W. and Kohn, S. D., Effects of Variation in Quarter-Car Simu-
lation Speed on International Roughness Index Algorithm, Trans. Res.
Rec., Vol. 1889, 2004, pp. 144151.
[20] Dahlberg, T., An Optimized Speed-Controlled Suspension of a 2-DOF

Vehicle Travelling on a Randomly Profiled Road., J. Sound Vibration,
Vol. 62, No. 4, 1979, pp. 541546.
[21] Chalasani, R. M., Ride Performance Potential of Active Suspension Sys-
temsPart I: Simplified Analysis Based on a Quarter-Car Model, Am.
Soc. Mech. Eng., Vol. 80, 1986, pp. 187204.
[22] Hrovat, D., Influence of Unsprung Weight on Vehicle Ride Quality,
J. Sound Vibration, Vol. 124, No. 3, 1988, pp. 497516.
[23] Yu, J., Chou, E. Y. J., and Yau, J.-T., Development of Speed-Related
Ride Quality Thresholds Using International Roughness Index, Trans.
Res. Rec., 2006, pp. 4753.
267
Author Index
A I
Akkari, A., 110-134 Izevbekhai, B., 110-134

Arsenault, R., 238-247
J
B
Jiang, Y., 59-76
Bennert, T., 135-149
Brown, H. J., 1-19
K
Bruinsma, J. E., 205-218
Karamihas, S. M., 248-266
C
L
Carvalho, R., 170-191
Chang, G. K., 219-237 Li, S., 59-76
Crouch, L. K., 1-19 Lu, Q., 150-159
D M
de Leon Izeppi, E., 20-39 McGhee, K., 20-39

Dziedziejko, T., 238-247 Miller, R., 192-204
F N
Flintsch, G., 20-39 Noureldin, S., 59-76
G O
Gaspard, K., 40-58 Orthmeyer, R., 219-237

Gedafa, D. S., 192-204
Goodman, S. N., 238-247 P
Pierce, L. M., 205-218

H
Haas, E. H., 135-149 R

Hencken, J., 135-149
Hossain, M., 192-204 Rasmussen, R. O., 96-109, 160-169
268
S W
Sandberg, U., 77-95 Watkins, J., 219-237

Selezneva, O., 170-191 Wiser, L., 170-191
Sivaneswaran, N., 205-218
Smith, K. D., 205-218 Z
Sohaney, R. C., 96-109, 160-169
Steele, D. A., 192-204 Zhang, Z., 40-58
Stubstad, R., 170-191 Zhu, K., 59-76
269
Subject Index
A J
aggregate, 1-19 jointed, 219-237

antilock brakes, 40-58
asphalt mixture, 150-159 L
automation, 219-237
laser sampling frequency, 59-76
B laser sampling rate, 59-76
lock-wheel friction tester, 40-58
baseline length, 59-76 loss-on-ignition, 1-19
bituminous, 1-19
bonus and penalty system, 238-247 M
C macrotexture, 77-95
material properties, 205-218
composite pavements, 110-134 mean profile depth, 59-76, 77-95
concrete, 160-169, 219-237 measurement method, 77-95
continuous friction, 20-39 mechanistic-empirical pavement
design, 205-218
E megatexture, 77-95
microtexture, 59-76
exposed aggregate surfaces, MPD, 77-95
110-134
N
F
network-level analysis, 170-191
falling weight deflectometer (FWD), noise, 160-169
170-191, 192-204, 205-218
faulting, 219-237 O
fixed-slip friction tester, 40-58
friction, 96-109 OBSI (on-board sound intensity),
friction measurements, 20-39 135-149, 150-159, 160-169
friction testing, 40-58
fuel consumption, 77-95 P
I pavement, 77-95, 96-109, 150-159,

160-169, 219-237
inertial profiler, 238-247 pavement characteristics, 77-95
International Roughness Index, pavement deflection, 170-191
238-247, 248-266 pavement friction, 59-76
270
pavement management system, slope variance, 59-76

170-191 smoothness, 96-109
pavement performance, 170-191 software, 219-237
pavement smoothness, 238-247 sound absorption, 110-134
pavement temperature, 192-204 sound intensity, 135-149
pavements, 20-39 spectral signature, 135-149
perpetual pavement, 192-204 speed, 59-76
polish resistance, 1-19 splash and spray, 96-109
profile, 96-109
profiling, 219-237
T
R texture, 77-95, 96-109,

160-169
rado model, 40-58
tire pavement interaction noise,
reference tires, 77-95
110-134, 135-149
rehabilitation, 205-218
tire-pavement noise, 96-109, 150-159,
ride quality, 110-134
160-169
road profile, 248-266
trailer, 77-95
road roughness, 248-266
rolling resistance, 77-95, 96-109
rolling-wheel deflectometer V
(RWD), 192-204
root-mean-square, 59-76 variable-slip tester, 40-58
vehicle simulation, 248-266
S
W
side-friction tester, 40-58
silica, 1-19 wavelength, 59-76

Stp1555-Eb 28948

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Stp1555-Eb 28948

Încărcat de

Drepturi de autor:

Formate disponibile

Copyright by ASTM Int'l (all rights reserved); Wed Aug 16 21:57:48 EDT 2017

Printed in the U.S.A.

ASTM Stock #: STP1555

Printed in Bay Shore, NY

Evaluating Polish Resistance of Tennessee Bituminous Surface Aggregates

L. K. Crouch1 and Heather J. Brown2

Evaluating Polish Resistance of Tennessee

REFERENCE: Crouch, L. K. and Brown, Heather J., Evaluating Polish Re-

4. Perform a preliminary evaluation of the new method.

Background and Literature Review

distress, macrotexture, and microtexture. The pavement distress modes of rutting

ineffective calibration procedures. Therefore, a new method for evaluating

Summary of Tennessee Terminal Textural Condition Method (T3CM)

FIG. 1Schematic T3CM uncompacted voids content apparatus modified

TABLE 1Laboratory test results for 32 project aggregates.

Correlation with Pavement Performance

Completion ADT Number of SN 40 Terminal

TABLE 3Summary of phase 4 aggregate source evaluations.

second best correlation coefficient, is simply an improved percent silica test,

TABLE 4Random production sample aggregate test results.

Aggregate Date sampled TTC ASTM C25 silica (%) BPN 9

FIG. 2Terminal skid number versus TTC.

FIG. 3Terminal skid number versus LOI.

FIG. 4Terminal skid number versus x-ray percent silica.

FIG. 5Terminal skid number versus ASTM C25 percent silica.

FIG. 6Terminal skid number versus BPN 9.

FIG. 7X-ray percent silica versus ASTM C25 percent silica.

Test method Correlation coefficient with available terminal skid numbers

FIG. 8ASTM C25 percent silica versus LOI.

FIG. 9X-ray percent silica versus LOI.

Number of X-ray ASTM C25

TABLE 7Range of laboratory test results of project aggregates (3 projects).

Number of X-ray ASTM C25

TABLE 8Coefficient of variation of laboratory test results of project aggregates (3 projects).

Number of X-ray ASTM C25

X-ray ASTM C25 Cumulative

TABLE 10Random production sample statistics.

[9] Tucker, C., 1997, Aggregate Terminal Textural Condition, Master of

Edgar de Leon Izeppi,1 Gerardo Flintsch,2 and Kevin McGhee3

Effect of Water, Speed, and Grade on

REFERENCE: de Leon Izeppi, Edgar, Flintsch, Gerardo, and McGhee,

To obtain the limits of agreement, sc is then multiplied by a factor of 1.96 to

Devices Used in this Study

TABLE 1Virginia Smart Road section pavement surfaces.

FIG. 2CFME units available from the friction loan program.

arrangements are made by VTTI. The agency is responsible for personnel to

FIG. 3CFME units driving uphill on the eastbound lane.

Complete Speed Tests Results

Statistic 404-40d 405-40d DYN-40d 404-40u 405-40u DYN-40u

TABLE 3Effect of grade with varying speed.

Grip tester 404 Grip tester 405 Dynatest HFT

40 km/h (25 mph) 65 km/h (40 mph) 90 km/h (55 mph)

Grip tester 404 Grip tester 405 Dynatest HFT

Complete Water-Film Tests Results

Conclusions and Findings

0.25 mm 0.50 mm 1.00 mm

[1] Henry, J. J., Evaluation of Pavement Friction Characteristics: A Synthe-

Kevin Gaspard1 and Zhongjie Zhang2

Ascertaining Parametric Correlations

REFERENCE: Gaspard, Kevin and Zhang, Zhongjie, Ascertaining Paramet-

authors opinion, this is a significant issue that must be addressed by the

FIG. 1Typical FN curve and Rado model.

Microtexture is primarily controlled by the surface properties of the aggregate