THE NATIONAL ACADEMIES PRESS

This PDF is available at http://nap.edu/22766 SHARE

   

Assessment of Continuous Pavement Deflection Measuring

Technologies

DETAILS

0 pages | 8.5 x 11 | PAPERBACK

ISBN 978-0-309-43504-8 | DOI 10.17226/22766

CONTRIBUTORS

GET THIS BOOK Gerardo Flintsch, Brian Ferne, Brian Diefenderfer, Samer Katicha, James Bryce,
Simon Nell, and Trenton Clark; Strategic Highway Research Program; Strategic
Highway Research Program Renewal Focus Area; Transportation Research Board;
FIND RELATED TITLES National Academies of Sciences, Engineering, and Medicine


Visit the National Academies Press at NAP.edu and login or register to get:

– Access to free PDF downloads of thousands of scientific reports

– 10% off the price of print titles
– Email or social media notifications of new titles related to your interests
– Special offers and discounts


Distribution, posting, or copying of this PDF is strictly prohibited without written permission of the National Academies Press.
(Request Permission) Unless otherwise indicated, all materials in this PDF are copyrighted by the National Academy of Sciences.

Copyright © National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

The Second
S T R A T E G I C H I G H W A Y R E S E A R C H P R O G R A M

Report S2-R06F-RW-1

Assessment of Continuous Pavement

Deflection Measuring Technologies
Gerardo Flintsch, Samer Katicha, and James Bryce
Virginia Tech Transportation Institute

Brian Ferne and Simon Nell

Transport Research Laboratory

Brian Diefenderfer
Virginia Center for Transportation Innovation and Research

T R A N S P O R TAT I O N R E S E A R C H B O A R D
WASHINGTON, D.C.
2013
www.TRB.org

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Subscriber Categories
Construction
Highways
Maintenance and Preservation
Pavements

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

The Second Strategic Highway SHRP 2 Report S2-R06F-RW-1

Research Program
ISBN: 978-0-309-12960-2
America’s highway system is critical to meeting the mobility and
economic needs of local communities, regions, and the nation. © 2013 National Academy of Sciences. All rights reserved.
Developments in research and technology—such as advanced
materials, communications technology, new data collection tech-
nologies, and human factors science—offer a new opportunity Copyright Information
to improve the safety and reliability of this important national
Authors herein are responsible for the authenticity of their materials and for
resource. Breakthrough resolution of significant transportation obtaining written permissions from publishers or persons who own the copy-
problems, however, requires concentrated resources over a short right to any previously published or copyrighted material used herein.
time frame. Reflecting this need, the second Strategic Highway The second Strategic Highway Research Program grants permission to repro-
duce material in this publication for classroom and not-for-profit purposes.
Research Program (SHRP 2) has an intense, large-scale focus, Permission is given with the understanding that none of the material will be
integrates multiple fields of research and technology, and is used to imply TRB, AASHTO, or FHWA endorsement of a particular product,
fundamentally different from the broad, mission-oriented, method, or practice. It is expected that those reproducing material in this
document for educational and not-for-profit purposes will give appropriate
discipline-based research programs that have been the mainstay acknowledgment of the source of any reprinted or reproduced material. For
of the highway research industry for half a century. other uses of the material, request permission from SHRP 2.
Note: SHRP 2 report numbers convey the program, focus area, project number,
and publication format. Report numbers ending in “w” are published as web
The need for SHRP 2 was identified in TRB Special Report 260: documents only.
Strategic Highway Research: Saving Lives, Reducing Congestion,
Improving Quality of Life, published in 2001 and based on a
study sponsored by Congress through the Transportation Equity Notice
Act for the 21st Century (TEA-21). SHRP 2, modeled after the The project that is the subject of this report was a part of the second Strategic
Highway Research Program, conducted by the Transportation Research Board
first Strategic Highway Research Program, is a focused, time- with the approval of the Governing Board of the National Research Council.
constrained, management-driven program designed to comple- The members of the technical committee selected to monitor this project and
ment existing highway research programs. SHRP 2 focuses on to review this report were chosen for their special competencies and with regard
applied research in four areas: Safety, to prevent or reduce the for appropriate balance. The report was reviewed by the technical committee
and accepted for publication according to procedures established and overseen
severity of highway crashes by understanding driver behavior; by the Transportation Research Board and approved by the Governing Board of
Renewal, to address the aging infrastructure through rapid design the National Research Council.
and construction methods that cause minimal disruptions and The opinions and conclusions expressed or implied in this report are those of
the researchers who performed the research and are not necessarily those of the
produce lasting facilities; Reliability, to reduce congestion through Transportation Research Board, the National Research Council, or the program
incident reduction, management, response, and mitigation; and sponsors.
Capacity, to integrate mobility, economic, environmental, and The Transportation Research Board of the National Academies, the National
community needs in the planning and designing of new trans- Research Council, and the sponsors of the second Strategic Highway Research
Program do not endorse products or manufacturers. Trade or manufacturers’
portation capacity. names appear herein solely because they are considered essential to the object
of the report.
SHRP 2 was authorized in August 2005 as part of the Safe,
Accountable, Flexible, Efficient Transportation Equity Act: A
Legacy for Users (SAFETEA-LU). The program is managed by the
Transportation Research Board (TRB) on behalf of the National
Research Council (NRC). SHRP 2 is conducted under a memo-
randum of understanding among the American Association of
State Highway and Transportation Officials (AASHTO), the
Federal Highway Administration (FHWA), and the National
Academy of Sciences, parent organization of TRB and NRC.
The program provides for competitive, merit-based selection of SHRP 2 Reports

research contractors; independent research project oversight; and Available by subscription and through the TRB online bookstore:
dissemination of research results. www.TRB.org/bookstore

Contact the TRB Business Office:

202-334-3213

More information about SHRP 2:

www.TRB.org/SHRP2

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

The National Academy of Sciences is a private, nonprofit, self-perpetuating society of distinguished scholars
engaged in scientific and engineering research, dedicated to the furtherance of science and technology and
to their use for the general welfare. On the authority of the charter granted to it by Congress in 1863, the
Academy has a mandate that requires it to advise the federal government on scientific and technical matters.
Dr. Ralph J. Cicerone is president of the National Academy of Sciences.
The National Academy of Engineering was established in 1964, under the charter of the National Academy
of Sciences, as a parallel organization of outstanding engineers. It is autonomous in its administration and
in the selection of its members, sharing with the National Academy of Sciences the responsibility for
advising the federal government. The National Academy of Engineering also sponsors engineering programs
aimed at meeting national needs, encourages education and research, and recognizes the superior achieve-
ments of engineers. Dr. Charles M. Vest is president of the National Academy of Engineering.
The Institute of Medicine was established in 1970 by the National Academy of Sciences to secure the
services of eminent members of appropriate professions in the examination of policy matters pertaining
to the health of the public. The Institute acts under the responsibility given to the National Academy of
Sciences by its congressional charter to be an adviser to the federal government and, on its own initiative,
to identify issues of medical care, research, and education. Dr. Harvey V. Fineberg is president of the
Institute of Medicine.
The National Research Council was organized by the National Academy of Sciences in 1916 to associate
the broad community of science and technology with the Academy’s purposes of furthering knowledge and
advising the federal government. Functioning in accordance with general policies determined by the
Academy, the Council has become the principal operating agency of both the National Academy of
Sciences and the National Academy of Engineering in providing services to the government, the public, and
the scientific and engineering communities. The Council is administered jointly by both Academies and
the Institute of Medicine. Dr. Ralph J. Cicerone and Dr. Charles M. Vest are chair and vice chair,
respectively, of the National Research Council.
The Transportation Research Board is one of six major divisions of the National Research Council. The
mission of the Transportation Research Board is to provide leadership in transportation innovation and
progress through research and information exchange, conducted within a setting that is objective, interdisci-
plinary, and multimodal. The Board’s varied activities annually engage about 7,000 engineers, scientists, and
other transportation researchers and practitioners from the public and private sectors and academia, all of
whom contribute their expertise in the public interest. The program is supported by state transportation
departments, federal agencies including the component administrations of the U.S. Department of Transporta-
tion, and other organizations and individuals interested in the development of transportation. www.TRB.org

www.national-academies.org

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

SHRP 2 STAFF

Ann M. Brach, Director

Stephen J. Andrle, Deputy Director
Neil J. Pedersen, Deputy Director, Implementation and Communications
James Bryant, Senior Program Officer, Renewal
Kenneth Campbell, Chief Program Officer, Safety
JoAnn Coleman, Senior Program Assistant, Capacity and Reliability
Eduardo Cusicanqui, Financial Officer
Walter Diewald, Senior Program Officer, Safety
Jerry DiMaggio, Implementation Coordinator
Shantia Douglas, Senior Financial Assistant
Charles Fay, Senior Program Officer, Safety
Carol Ford, Senior Program Assistant, Renewal and Safety
Elizabeth Forney, Assistant Editor
Jo Allen Gause, Senior Program Officer, Capacity
Rosalind Gomes, Accounting/Financial Assistant
Abdelmename Hedhli, Visiting Professional
James Hedlund, Special Consultant, Safety Coordination
Alyssa Hernandez, Reports Coordinator
Ralph Hessian, Special Consultant, Capacity and Reliability
Andy Horosko, Special Consultant, Safety Field Data Collection
William Hyman, Senior Program Officer, Reliability
Michael Marazzi, Senior Editorial Assistant
Linda Mason, Communications Officer
Reena Mathews, Senior Program Officer, Capacity and Reliability
Matthew Miller, Program Officer, Capacity and Reliability
Michael Miller, Senior Program Assistant, Capacity and Reliability
David Plazak, Senior Program Officer, Capacity
Onno Tool, Visiting Professional
Dean Trackman, Managing Editor
Connie Woldu, Administrative Coordinator
Patrick Zelinski, Communications/Media Associate

ACKNOWLEDGMENTS
This work was sponsored by the Federal Highway Administration in cooperation with the American Asso-
ciation of State Highway and Transportation Officials. It was conducted in the second Strategic Highway
Research Program, which is administered by the Transportation Research Board of the National Acad-
emies. The project was managed by James Bryant, Senior Program Officer for SHRP 2 Renewal.
The authors would like to thank Chuck Taylor of the second Strategic Highway Research Program
(SHRP 2), the Project R06F technical expert task group members (Lynne Irwin, Erland Lukanen, Mark
McDaniel, Nadarajah Sivaneswaran, Thomas Van, and Tom Warne) for their guidance and informative
feedback, the U.K. Highways Agency for access to the HA traffic speed deflectometer measurements, and
the U.S. FHWA and Applied Research Associates for the rolling wheel deflectometer data.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

FOREWORD
James W. Bryant, Jr., Ph.D., P.E., SHRP 2 Senior Program Officer

The measurement of the response of a pavement to an applied load is a critical input for
(1) structural analysis of in-service pavements, (2) identification of sections with structural
capacity deficiencies at the network level, and (3) design of pavement renewal or rehabilita-
tion treatments at the project level. The most widely used method for measuring pavement
response to an applied load is the falling weight deflectometer (FWD), which is a time-
consuming technology and may not be practical for a network-level structure monitoring.
The use of continuous deflection measuring devices, which operate at speeds of 30 to 45 mph
in some cases, allows for better spatial coverage with less impact on traffic. This project
evaluated current technologies implemented in continuous deflection measuring devices.

The objective of this project was to critically assess the potential of existing continuous
deflection devices to be practical and cost-effective tools for use in the development of
optimum pavement rehabilitation strategies on rapid renewal projects. This assessment
included (1) the potential value of and demand for continuous deflection data by transporta­
tion agencies; (2) the technical capabilities (including accuracy and repeatability of test
results and ability to provide meaningful data), limitations (field applications, equipment
configuration, and operating and safety characteristics), and other impediments to imple-
mentation of existing devices; and (3) suggestions for improvements to currently available
technologies.
The main products of the project include a catalogue of existing continuous deflection
measuring technologies, detailed assessment of the capabilities of the most-promising devices,
case studies illustrating the application of the technology for supporting various pavement
management decision-making processes, a fact sheet describing the main technologies
identified for continuously measuring pavement deflections and their potential uses, training
materials for a workshop on the topic, research needs statements for the most-pressing research
identified (provided as an appendix), and a dissemination and implementation plan for this
technology.
Technologies for continuous deflection measurement are still evolving. This report provides
practical examples of how data from these devices can be used for network-level pavement
management applications. As budgetary pressures continue to place a high demand of the
effective allocation of resources, the ability to isolate areas for more-detailed and “-costly”
pavement assessments will become desirable. The data collected by the devices investigated
in this study can help a transportation agency ascertain the areas of pavement that need a
detailed condition assessment. Demand for this type of technology will continue to increase.
Information provided in this report helps to expand the knowledge base of what this
technology can do and provides confidence and examples of how the technology can be used.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Contents

1 Executive Summary
5 Chapter 1 Introduction
5 Objective
5 Critical Research Questions
6 Methodology and Report Overview
8 Chapter 2 Research Approach
8 Identification and Assessment of Available Technologies
8 Determination of User Needs
9 Selection of Candidate Devices
9 Data Collection
11 Data Analysis
11 Example Applications
11 Current and Future Developments
12 Chapter 3 Analysis and Findings
12 Catalogue of Deflection Measuring Devices
18 Survey of State DOT Practices and Needs
21 Selection of Candidate Devices
25 Detailed Description of the Selected Equipment
29 Phase I Assessment
38 Phase II Assessment
66 Operational Characteristics
74 Chapter 4 Example Applications and Recent Developments
74 Example Applications
86 Current and Future Developments
90 Chapter 5 Conclusions and Suggested Research
90 Findings
92 Conclusions
92 Recommendations for Implementation
95 References
97 Appendix A. Research Problem Statements

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Executive Summary

The measurement of pavement structural capacity is a critical input to perform structural analy-
sis of in-service pavements, identify sections with structural capacity deficiencies at the network
level, and design pavement renewal or rehabilitation treatments at the project level. Continuous
deflection measuring devices are increasingly being used to support these and other pavement
management business processes. These nondestructive pavement evaluation devices can measure
pavement deflections caused by a moving load and, in some cases, with little or no traffic control.
The ability to measure without disrupting traffic makes them more advantageous to use than
stationary devices such as the falling weight deflectometer (FWD). SHRP 2 Renewal Research
Project R06F, Assessment of Continuous Pavement Deflection Measuring Technologies, evalu-
ated current technologies implemented in different types of continuous deflection measuring
devices, identified the most promising devices for effectively supporting pavement management
decisions, evaluated the capabilities of these devices, and identified and illustrated applications
that could be useful for supporting pavement management. The main products of the project
include a catalogue of existing continuous deflection measuring technologies; detailed assessment
of the capabilities of the most promising devices; case studies illustrating the application of the
technology for supporting various pavement management decision-making processes; a fact sheet
describing the main technologies identified for continuously measuring pavement deflections
and their potential uses; training materials for a workshop on the topic; research needs state-
ments for the most pressing research identified, provided as an appendix; and a dissemination
and implementation plan for the technology.
The critical performance parameters for the available equipment were assessed by the follow-
ing research question: Is the technology capable of providing the quality information needed
to support the main pavement management business functions identified by potential users?
To answer this guiding question, the research team evaluated whether the devices can be used
for screening pavements at the network level and thus identify weak (i.e., structurally deficient)
sections, or whether the structural response information collected by the devices can be used to
differentiate sections that may be good candidates for preservation from those that would likely
require more substantial treatment.
For this report, a continuous deflection device is defined as a “constantly moving deflection
measuring device that can collect data at intervals of approximately 300 mm (1 ft) or smaller using
load levels typical of truck loading (i.e., 40 to 50 kN [9 to 11 kips] per wheel or load assembly).” The
project started with a review of literature, case studies, and ongoing research to identify deflection
measuring devices that had the potential to meet project requirements and to assess their potential
to meet user requirements. Devices that met the definition of a continuous deflection device
were the Portancemetre, the moving FWD, the Measuring Ball, the traffic speed deflectometer
(TSD), the rolling dynamic deflectometer (RDD), the rolling wheel deflectometer (RWD), the

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

airfield rolling weight deflectometer (ARWD), the road deflection tester (RDT), and the image
deflection measurement device (IDM). This list of continuous deflection devices included laser-
based devices that measure the deflection below a moving truck load, devices that apply a vibra-
tory load, and one system still in early stages of development that uses image analysis methods to
determine pavement deflections under loading.
The demand and potential value of continuous deflection devices for use in developing opti-
mum pavement rehabilitation strategies for rapid renewal projects were evaluated through a sur-
vey of state and provincial departments of transportation (DOTs). The survey not only included
questions to assess technical needs of the DOTs, but also aimed at determining the value assigned
by the agencies to the collected data. The survey showed that by February 2010 the majority of
agencies performed at least some deflection testing using the FWD. Most testing was performed
to support project-level decisions, and only five agencies had incorporated deflection data into
their pavement management system (PMS). Potential users in general agreed that the main
advantage of a continuous measuring device (as compared with a stationary device) is to support
network-level decisions. The assessment also identified the following parameters as important
in the evaluation of equipment: survey speed (safety), repeatability, accuracy (and feasibility
of establishing correlations with existing technologies, such as FWD), equipment cost, ease of
operation, customer service (availability of service and maintenance), ease of use of the data col-
lected, availability of software for interpretation of results, reliability, size of the vehicle, relevance
of the information (e.g., use in the Mechanistic–Empirical Pavement Design Guide (MEPDG) and
its software package known as DARWin-ME), and past experience. A survey speed close to the
speed of traffic was an important desired parameter identified in the survey, even if achieving
such a speed resulted in some loss in equipment accuracy. For this purpose, a minimum device
survey speed of 55 km/h (35 mph) was set as a critical selection criterion.
While responses to the initial survey suggested that users would like to be able to collect contin-
uous pavement response data to support project-level decisions, follow-up interviews showed that
respondents understood the current limitation of the technologies and agreed that network-level
applications are the more likely application in the near future. Furthermore, respondents agreed
on the need for pavement structural data to support network-level PMS decisions. At the network
level, the primary application of the continuous deflection device would be (1) to help identify
weak (i.e., structurally deficient) areas that can then be further investigated at the project level,
(2) to provide network-level data to calculate a structural health index that can be incorporated
into a PMS, and (3) to differentiate sections that may be good candidates for preservation (those
that have good structural capacity) from those that would likely require a more substantial treat-
ment (those that show structural deterioration or deficiencies). For the user, the ideal overriding
requirement is that network-level data be collected at highway speeds.
A more detailed assessment of the capabilities of candidate devices, based on criteria provided
by a survey of potential users, allowed further reduction of the list to two devices, the RWD and
the TSD. The rest of the devices were eliminated because they did not apply loads similar to that
of a heavy vehicle, they did not meet the survey speed requirement (a key requirement expressed
by potential users), or the existing prototypes had been decommissioned or alternatively reas-
signed to other uses. Both the RWD and TSD devices conduct measurements (deflection or
deflection velocity of a loaded pavement, respectively) under a truck axle and at close to highway
speeds. The RWD is mounted within a custom-designed semitrailer and measures the response
from one-half of an 80-kN (18-kip) single-axle load traveling at normal traffic speeds. It uses a
spatially coincident methodology for measuring pavement deflection by comparing undeflected
and deflected pavement laser scan profiles. The device can test approximately 320 to 480 lane-
km (200 to 300 lane-mi) per day. The TSD is mounted on an articulated truck but uses a rear-
axle load of 100 kN (22 kips). The TSD model evaluated uses four Doppler lasers mounted on
a height-adjustable rigid beam to record the vertical deflection velocity of a loaded pavement
under one of the dual-wheel assemblies. One laser is mounted away from the loaded wheels to
measure the unwanted movement of the rigid beam. This movement is then subtracted from

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

3  

the other laser measurements to provide the vertical pavement deflection velocity at each sensor.
The deflection velocity is divided by the instantaneous vehicle speed to give a measurement of
deflection slope, which is generally expressed in mm/m.
The capability of continuous deflection technology to support network-level pavement
management decisions was evaluated through a combination of literature review and data col-
lected in combination with other efforts. The detailed field evaluation included measurements
on pavement sections in the United Kingdom on different types of pavements, using various
operational conditions, and reference FWD deflection testing equipment where possible. The
sites evaluated included flexible, composite, and rigid pavement sections, and all attempts
were made to include subsections with good, fair, and poor functional conditions within each
of these pavement types. Some sites were measured several times. A similar evaluation was
planned for the RWD in the United States, but field verification was not possible because the
device had not been available during the second phase of the project.
The analysis of collected data showed that the repeatability of both systems depends on the
aggregation length and appears to be appropriate for network-level applications. Except for a few
sections with significant surface deterioration, the TSD repeatability was independent of the type
of pavement and the value measured, ranging from 0.065 to 0.201 mm/m (mean = 0.089 mm/m)
for 10-m averaging, and from 0.022 to 0.114 mm/m (mean = 0.028 mm/m) for 100-m averaging.
On the basis of limited data collected in Virginia, repeatability of the RWD was evaluated to be
51.4 µm (2 mils) for 160-m (0.1-mile) averaging. However, it must be noted that the data used
for assessing the RWD were collected using a previous version of the equipment. It is likely that
recent enhancements made to the device have improved the quality of the measurements.
The direct analysis of the relationship between the TSD and FWD measurements showed that
there are two distinct relationships between TSD deflection slope and FWD deflection depend-
ing on pavement type (one for flexible and composite pavements and another for rigid pave-
ments). This could be expected, as two different quantities (deflection slope and deflection)
are measured by each device. These quantities are affected differently by the pavement type.
The comparability of the TSD with FWD was assessed by using two surface indices, the surface
curvature index (SCI) and base distress index (BDI), quantities that can be obtained from both
devices. In this case, the relationship between the quantities measured with the FWD and those
measured with the TSD (measurements averaged over 10 m in length) was the same for all pave-
ment types and reasonably close to the equality line. However, there is a significant variation and
bias in this relationship. For example, for an average SCI or BDI value of 300 µm, the bias was
30 µm (FWD values lower than TSD values) and the comparability was 380 µm. The analysis of
RWD comparability with FWD (using measurements averaged over a 160-m [0.1-mi] interval)
showed, based on limited data from an earlier version of the RWD, that the coefficient of varia-
tion (cov) was found to be relatively unchanged as a function of deflection. The relationship
showed a bias of 11.6% (FWD deflection lower than RWD deflection) and a repeatability of
64.6% (range of [-43.2%; 21.4%]). FWD results tended to be lower than the results of either
TSD or RWD.
The example applications demonstrated that continuous deflection measurement devices can
be used to estimate many parameters important to modern pavement management applications.
The analysis showed that, at least for the section investigated, the strains at the bottom of the
asphalt layer estimated with measurements using the FWD and TSD resulted in an approximately
one-to-one relationship. Similarly, the effective structural number (SN) estimated with measure-
ments obtained from TSD testing at two sites broadly matched the expected SN calculated from
the layer composition and surface condition.
Within this study it has not been possible to examine in detail the operational characteristics
of the equipment being assessed. Although external factors, such as temperature, road geom-
etry, road profile, texture profile, moisture, acceleration, deceleration, and so forth, may have
been recorded during the surveys, it was not possible to control these factors. Therefore, it was
not easy to assess their effects on the measurements. This report briefly discusses the potential

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

impact of speed, road geometry, texture profile, dynamic loading, acceleration, and deceleration
determined on the basis of a limited examination of the data collected. Both devices use laser-
based noncontact sensors, which fail to measure correctly when the road is damp or wet. Laser
reflection is degraded by water on the surface. Other potential limitations include difficulties in
measuring along sharp curves (especially for the RWD), impact of surface texture on measure-
ments, and variations on the load applied because of the vehicle dynamic (response to the road
profile, acceleration, and deceleration).
Updated versions of the devices have become available during the final phases of the project.
Recent modifications to the RWD have placed the lasers in a temperature-controlled enclo-
sure and have added a sensor to the RWD at a second position further away from the rear axle
to provide some information about the deflection bowl shape. The second generation of TSD
equipment includes a custom trailer, more sensors (up to seven, enabling the derivation of the
full deflection bowl), and a more robust system to measure vehicle speed, which is a vital part
of the measurement process. The sensor mounting beam has also been installed on longitudinal
rails so that its position can be varied, thus enabling more reliable calibration.
More research is recommended for conducting additional field tests with the latest versions
of both the RWD and TSD at the same locations and with different pavement designs, for
developing tools for using technology to support network-level pavement management busi-
ness functions, for assessing the potential for using output from the selected equipment to
provide advice on pavement rehabilitation alternatives (e.g., preservation versus renewal), and
for verifying the accuracy of testing equipment by conducting measurements on instrumented
sections to compare measured parameters with the response of in-situ transducers measuring
absolute deflection and strain.
In conclusion, the study performed in this project has demonstrated that at least one con-
tinuous deflection measurement device, the TSD, can (1) provide adequate repeatability
for network-level data collection, (2) collect deflection measurements and indices that are
broadly comparable to those collected by traditional measurement devices such as the FWD,
and (3) provide measurements that can be used for supporting some of the most critical
network-level applications identified by the potential users. Although information collected
in the first phase of the project suggests that the RWD may be able to provide the same type
of capabilities, this has not been confirmed because of the unavailability of the equipment
for detailed evaluation in the second phase of the project.
However, the technology is only just maturing. Future research would further assess the mea-
surement capabilities of these devices and the usefulness of the collected data. For example, at
least one assessed device demonstrated some ability to identify localized structural deteriora-
tion, which may indicate a potential for project-level use. Potential enhancements to the devices
that may help improve the quality of information obtained from measurements and widen the
range of possible applications include (1) providing a more complete deflection bowl shape;
(2) enhancing the quality of measurement signals so that local structural deterioration can be
reliably identified, that is, down to 1-m (3-ft) features; (3) providing pavement layer thickness
measurement capability, for example, by adding ground-penetrating radar equipment; and
(4) measuring the dynamic load on the loading-wheel assembly.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Chapter 1

Introduction

A growing area of interest in pavement research is the devel- 2012) and it has recently completed a survey of 18,000 km
opment of technologies that are well suited for nondestructive in Australia (Baltzer et al., 2010).
assessment of the pavement structure without causing delays
to the traveling public. These technologies are needed as the Objective
nation’s highway infrastructure systems continue to age and
the determination of appropriate rehabilitation strategies The objective of the SHRP 2 R06F project was to carry out
becomes ever more important to network mobility preser- a critical assessment of (1) the potential of existing continu-
vation. The measurement of the response of a pavement to ous deflection devices as practical and cost-effective tools for
an applied load is a critical input for (1) structural analysis use in the development of optimum pavement rehabilitation
of in-service pavements, (2) identification of sections with strategies on rapid renewal projects, and (2) the capability of
structural capacity deficiencies at the network level, and these devices for screening structural deficient sections and
(3) design of pavement renewal or rehabilitation treat- scoping their needs at the network level.
ments at the project level. A recent survey showed that most To accomplish these two objectives, the research team
state departments of transportation (DOTs) routinely use examined the following: (1) the potential demand by and
deflection measurements, obtained mainly with falling value to public agencies; (2) the technical capabilities (includ-
weight deflectometer (FWD) testing, at the project level ing repeatability, ability to be compared with FWD test
and some are also starting to use them at the network level results, ability to provide meaningful data), limitations (field
(Figure 1.1). applications, equipment configuration, operating and safety
Although the FWD is a very useful tool for assessing the characteristics, costs), and possible impediments to imple-
pavement structural or bearing capacity, and for determin- mentation of existing devices; and (3) further development
ing the moduli of the component layers, this technology of the technology, including both hardware and software,
has the limitation of allowing only stationary measure- needed to make these tools practical for use. Potential practi-
ments at discrete points along the pavement sections. Since cal uses of the technology are showcased through example
the test requires the equipment to remain stationary on the applications.
road for a short period of time (typically 1 to 4 minutes
depending on the protocol), it disturbs traffic and requires
Critical Research Questions
traffic control. This limits productivity and the number of
data points at which measurements can be obtained. The The critical performance parameters for the available equip-
use of continuous deflection measuring devices, which in ment were assessed using a combination of available exist-
some cases operate at traffic speed, allows for better spatial ing information and data from field trials. The trials were
coverage with less negative impact on mobility. The cur- organized in cooperation with existing activities to control
rently available continuous devices are becoming increas- costs and avoid duplication of efforts. The evaluation of the
ingly popular as practical alternatives to stationary FWD equipment was guided by the following research question: Is
devices, especially for network-level structural monitoring. the technology capable of providing the quality information
For example, the traffic speed deflectometer (TSD) is cur- needed to support the main pavement management business
rently being used in four countries in Europe (Greenwood, functions identified by the potential users?

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Source: Flintsch and McGhee, 2009.

Figure 1.1. Summary of current practices for pavement data

collection in North America.

To answer this guiding question the research team evaluated FWD measurements and the devices’ repeatability and opera-
the following: tional limitations.

• Whether the devices can be used for screening pavements at

Methodology and
the network level and thus identify weak (i.e., structurally
Report Overview
deficient) sections, which could then be further investigated
at the project level (e.g., using FWD). The report summarizes the results of the two phases of
• Whether the structural response information collected the project; although the project could be considered as
by the devices can be used to differentiate sections that part of a three-phase process. The first phase identified two
might be good candidates for preservation (good struc- continuous deflection devices based on a literature review
tural capacity) from those that would likely require a and user needs obtained from a survey and from follow-
heavier treatment (showing structural deterioration or up interviews, and a plan for assessment of the devices
deficiencies). was prepared and approved by SHRP 2. The second phase
included limited site trials on selected sections of highway.
Evaluating the equipment required assessing how accurately The results of these trials highlighted the capabilities of
and consistently the most promising continuous deflection the technology to meet user needs and identified needed
devices identify the location of structurally weak sections of improvement and implementation routes for such equip-
the road network, whether the measurements can help distin- ment. The third phase of this process would require com-
guish between sections with good and poor structural capac- prehensive evaluation trials of one or more recommended
ity, and whether the measurements of the candidate devices devices, ideally as side-by-side comparisons. These trials
can provide a structural health index, or indices, which can were not conducted as part this project but are recom-
be incorporated into a pavement management system (PMS). mended for a subsequent phase.
The consistency of a device in locating weak sections can be The report is organized into four main chapters and an intro-
assessed by examining how regularly test sections fall within a duction, Chapter 1. Chapter 2 presents the research approach
threshold range in repeat surveys. The assessment of accuracy and Chapter 3 the main analysis and findings of the research,
of locating weak sections depends on examining the relation- including the results of the literature review, the survey of
ship between the continuous deflection measurements and state DOTs, the devices selected for further evaluation, and
reference structural condition information. Since no direct the analysis of the data collected. Chapter 4 presents example
structural condition information is available, information applications of the TSD, new devices that emerged during the
derived from FWD measurements was taken as a proxy. Other analysis phase, and recent developments of the devices evalu-
secondary objectives to support the primary focus included ated. Chapter 5 summarizes the main findings and conclusions
evaluating the agreement of the measurements with actual and gives recommendations.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

7  

Two important issues addressed in this report are device deviation of the difference between two measurements.
repeatability and reproducibility. Repeatability is a measure Comparability (used instead of reproducibility because we
that answers the following question: If two deflection mea- are comparing different technologies) was similarly defined
surements at the same location are obtained from two dif- as the 95% interval on the difference between two measure-
ferent runs, in general, how much different will these two ments obtained using two different devices. This definition of
measurements be? Repeatability is evaluated as the 95% repeatability (or comparability) has many advantages over a
confidence interval on the difference between two repeated definition based on calculating the correlation between two
measurements calculated as 1.96sd where sd is the standard measurement series.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Chapter 2

Research Approach

This chapter presents the approaches used for the following: as irrelevant to this project, and an abridged version was
produced grouping similar devices (e.g., variations on the
• Identification of the most promising technologies; deflectograph). Information presented in this table includes
• Determination of user needs; equipment type, model, characteristics, survey speed, and
• Selection of candidate devices; development status (whether the device is a current or former
• Data collection, data analysis, and interpretation; production model or a working prototype). The devices were
• Development of example applications; and further subcategorized into three groups: static measurement
• Identification of recent developments and possible devices, moving measurement vehicles with stationary mea-
improvements. surement apparatus, and moving measurement vehicles with
nonstationary measurement apparatus.
Identification and Assessment
of Available Technologies Definition of Continuous Deflection Device
The project started with a review of literature, case studies, and The purpose of this project was to identify current measure-
ongoing research in order to (1) identify the deflection measur- ment devices capable of continuously measuring pavement
ing devices that had the potential to meet the project require- bearing capacity without the need for the vehicle or measure-
ments and (2) assess their potential to meet user requirements. ment equipment (relative to the vehicle) to remain stationary
The review was completed in two stages. The first stage con- while surveying. For this report, a continuous deflection device
sisted of an Internet search for relevant papers, performed by has been defined as a deflection measuring device constantly
searching for related key words such as “continuous deflec- moving that can collect data at intervals of approximately
tion device”; the second stage consisted of using the same key 300 mm (1 ft) or smaller using load levels typical of truck load-
words to search the English language International Transport ing (i.e., 40 to 50 kN [9 to 11 kips] per wheel or load assembly);
Research Documentation (ITRD) and Compendex databases the ideal solution would be a device requiring no traffic con-
for any relevant papers stored therein. This was complemented trol. It should be noted that this definition covers the interval
with the extensive personal archives of the members of the at which the data are collected, not reported. The latter is often
research team who have had many years of experience in this much greater than the former because of the significant mea-
research field. Once these searches were completed, the papers surement noise level generated during data collection. Averag-
and documents were used to compile a list of available survey ing over these longer lengths reduces the noise considerably
vehicles capable of measuring pavement deflection, whether at while retaining the important pattern of the road’s deflection
traffic speed or otherwise. in response to strength variations.

Prescreening of Candidate Devices Determination of User Needs

The list of survey vehicles was then summarized in a compre- The demand and the potential value of continuous deflec-
hensive table detailing the important aspects of each device (e.g., tion devices for use in developing optimum pavement reha-
survey speed, measurement interval, personnel requirements, bilitation strategies for rapid renewal projects were evaluated
etc.). After reviewing this table, some devices were disregarded through a survey of state and provincial DOTs. The survey

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

9  

included questions to assess technical needs and also endeav- applicable for network-level data, and a subset of these sec-
ored to determine the value assigned by the agencies to the tions was subject to a detailed evaluation as discussed in the
collected data. following sections (illustrated by Figure 2.1). The final exper-
The survey of state and provincial DOTs was divided into imental design focused mainly on evaluating the capabilities
two stages. In the first stage, a web survey was sent to the differ- of the general technology of traffic-speed continuous deflec-
ent DOTs. The survey was divided into five sections based on tion measurements and the application of this technology for
the type of information requested. The first section collected supporting pavement management decisions.
contact information about the respondents. The second sec- From the network-level testing routes, a few carefully
tion focused on current practices and uses of deflection testing selected evaluation sections were also assessed using FWD
by DOTs. The third section inquired about pavement reha- measurements. The data collected in these locations were used
bilitation design procedures and how, if applicable, deflection to assess the ability of the systems for detecting weak spots.
testing results are used for this purpose. The fourth section
focused on whether and how deflection testing results are used
TSD Data Collection
in pavement management applications. The fifth and final
section gave respondents the opportunity to provide general To answer the main evaluation question posed earlier (that is,
comments. In the second stage of the survey, follow-up phone the capability of the devices to support network-level pave-
interviews were conducted with nine states where continu- ment management decisions), researchers assessed the TSD in
ous deflection devices have already been used or the states are the United Kingdom by identifying a number of evaluation sec-
facing substantial renewal challenges on high-traffic-volume tions. Each section measured approximately 2 to 4 km (1.25 to
roadways. These states were identified based on the web survey 2.5 mi) in length and generally incorporated weak subsections;
results. researchers conducted repeat TSD surveys of these sections, as
well as FWD surveys on most sections, at spacing of up to 20 m
(60 ft). The accuracy and consistency with which the TSD iden-
Selection of Candidate Devices
tifies any strong and weak sections was also evaluated. Sites were
Following the literature review, an assessment of the capa- chosen by examining the construction and structural condi-
bilities of available continuous deflection devices (moving tion of a significant sample of the English road network. These
measurement vehicles with nonstationary measurement appa- evaluation sections covered a variety of structural designs and
ratus), and considering the information obtained during the ages. In general, road sections were selected to cover a range of
user needs survey, two devices that offer the most promising structural conditions as shown by the deflection response.
technologies to address the needs of end users were chosen for Table 2.1 lists these sites and a summary of the key parame-
further study. These devices are the rolling wheel deflectome- ters of each site. The nominal deflection responses are equiva-
ter (RWD) and the traffic speed deflectometer (TSD). Sample lent peak central FWD deflections at a load of 50 kN (11 kips).
data from the two devices were collected and processed as part The table includes the following:
of the preliminary assessment of their capabilities.
• Flexible sites, where the main structural layers are of asphalt
or granular construction and can be broadly classified as
Data Collection
fully flexible (i.e., with asphalt upper layers and granular
To assess the capability of continuous deflection technology to lower layers);
support network-level pavement management decisions, rela- • Composite sites, where asphalt and cement-bound layers are
tively long sections with uniform and variable structural con- both structural layers (i.e., with asphalt upper layers, usually
ditions were selected for site testing. These network-level sites greater than 150 mm [6 in.] thick, hydraulically bound base
included flexible, composite, and rigid pavement sections. All layers without joints, usually termed “lean concrete” in the
attempts were made to include subsections with good, fair, United Kingdom); and
and poor functional conditions within each of these pavement • Rigid sites, with pavement quality concrete as the primary
types. This enabled the evaluation of the capabilities of the load-bearing layer, sometimes with asphalt upper surfacing
devices for network-level use. Some sections were measured layers, less than 75 mm (3 in.) thick.
several times in succession in a single day, as discussed in the
following sections. Some of these sites, at least two from each pavement type,
The evaluation included testing on different types of pave- were each surveyed repeatedly during the course of one day.
ments, under various operational conditions, and reference Comparisons with FWD measurements were made on three
FWD deflection testing equipment where possible. The plan of the flexible sites, one of the composite sites, and two of the
included sections that were evaluated following a protocol rigid sites.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

10

Network-level routes

Evaluation sections

Sub-sections of interest
(detailed project-level analysis)

Figure 2.1. Schematic of the section selection.

Table 2.1. Summary of Potential Network-Level TSD Sites

Nominal
Cement Bound Deflection
Length Asphalt Thickness Thickness Response Structural
Site mi (km) in. (mm) in. (mm) mils (mm) Variability Site Use

Flexible
  UK_F1 1.25 (2.0) 3 (75) NA 8–32 (0.2–0.8) Very high R and C

  UK_F3 2.0 (3.2) 10.5 (270) NA 4–8 (0.1–0.2) Low R and C

  UK_F5 2.0 (3.2) 12 (310) NA 4–10 (0.1–0.25) High R and C
  UK_F6 2.2 (3.5) 12.6 (320) NA 4–12 (0.1–0.3) Low R
Composite
  UK_C1 2.5 (4.0) 7 (175) 7 (175) 2–14 (0.05–0.35) High R
  UK_C2 2.1 (3.4) 6.25 (160) 8.25 (210) 2–8 (0.05–0.2) Low R
  UK_C3 1.3 (2.1) 6.25–11.5 (160–290) 6–14.2 (150–360) 2–14 (0.05–0.35) High R and C
Rigid
  UK_R2 0.9 (1.4) 4 (100) 8 (200) 8–80 (0.2–2.0) High R and C
  UK_R3 1.9 (3.0) 3 (70) 9 (225) NA Low R and C
Note: R = repeatability assessment; C = comparison with FWD measurements for comparability assessment; and NA = not available.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

11  

RWD Data Collection slope measurements is investigated. This smoothing splines

denoising methodology shows potential to improve the fre-
The original plan included the collection of RWD data fol-
quency at which useful information can be obtained (i.e., data
lowing a protocol similar to the one used for the TSD. The
averaging distance). In this report, repeatability (or compa-
data analyzed in Phase I of the project were collected with
rability) is defined as the 95% confidence interval of the dif-
an older version of the device; this allowed only a prelimi-
ference between repeated measurements (difference between
nary assessment of the device’s capabilities. The device has
measurements of TSD and FWD or RWD and FWD).
since been upgraded, by modifying the lasers ensemble
and truck suspension, and was expected to perform bet-
ter than it had in the various applications reported in the Example Applications
literature. However, the upgraded device had some prob-
lems and was not fully operational during the second phase Based on the data collected and the analysis, a number of exam-
of the project. Thus, the planned tests were not possible ple applications for the data produced by the devices have been
within the time frame available and only data collected in developed and are presented in Chapter 4. These applications
the first phase of the project were used for the preliminary illustrate the use of continuous deflection measurements for
assessment of the capabilities of the RWD. Data from only network-level pavement management. They are not meant to
two routes, one in Virginia and one in New Mexico, were be comprehensive, but rather show how deflections can be used
used for assessing the repeatability and comparability of to address a number of issues applicable to modern pavement
the RWD, respectively. The equipment operator, Applied management practices.
Research Associates (ARA), in cooperation with the Federal The applications presented include using measurements
Highway Administration (FHWA), is working to correct obtained from continuous deflection measurement devices
the problems. Once they have been corrected, FHWA plans to do the following:
to support additional field testing.
• Segment pavement sections into homogeneous sections;
• Estimate the strain at the bottom of the asphalt layer and
Data Analysis the effective structural number (SN) of the pavement;
and
Results from previous testing were analyzed in Phase I of the • Identify relatively weak pavement sections as well as weak
project. This preliminary analysis was performed using com- pavement sections defined by absolute thresholds.
mon methods such as correlation and regression analysis.
These methods provide a quick analysis; they do not allow The segmentation was performed using a statistically based
for accurate evaluation of equipment capabilities. For Phase binary segmentation algorithm. The strain at the bottom of
II of the project, the data analysis was expanded to allow for the asphalt layer and the effective structural number were
accurate evaluation of repeatability and comparability. The estimated using the difference between two deflection slope
collected data were analyzed to evaluate the repeatability of measurements from the TSD. The identification of relatively
the TSD and RWD and the comparability of both devices by weak pavement sections was used to demonstrate the abil-
comparison with the FWD. Results of different studies that ity of the device to locate anomalies or locally weak sections
evaluated the RWD (conducted in the United States) and the within a network, whereas the identification of weak pave-
TSD (conducted in Europe) were analyzed. It was noted in ment sections based on thresholds demonstrated the ability
this analysis that repeatability and reproducibility were not of the device to repeatedly identify weak sections.
uniformly defined across all those studies. Measures and
methodologies used to evaluate the devices included corre-
lation, regression analysis, standard deviation, and in some
Current and Future
Developments
cases subjective visual inspection of plots. The drawbacks
associated with the use of correlation and regression analysis Since Phase I of this study was completed, the research team
to evaluate repeatability and comparability are discussed in has discovered one additional device that might meet the
this chapter. Then, repeatability and comparability analysis original requirements of this study. However, as this informa-
based on the limits of agreement (LOA) method suggested by tion was discovered since the commencement of the study, it
Bland and Altman (1986) is recommended and used to evalu- has not been included in the data acquisition and comparison
ate the continuous deflection devices. A method of evaluat- Phase II of the project. Furthermore, both the RWD and the
ing repeatability from one run is presented and compared to TSD have been developed further since Phase I was completed.
the method based on the LOA. Finally, the use of smoothing A description of these developments is presented in Chapter 4,
splines as a tool to remove the noise from TSD deflection along with recommendations for further improvement.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Chapter 3

Analysis and Findings

This chapter summarizes the results of the literature review, Among the devices listed in Table 3.3, the TSD, RWD, RDT,
survey of practice, and selection of the most promising tech- ARWD, RDD, and IDM can apply loads similar to that of a
nologies. It also provides a comprehensive assessment of truck, but only the first three are capable of surveying with-
these technologies, including repeatability, comparability, out the need for traffic control (each capable of surveying at
and operational characteristics. 70 km/h [45 mph] or faster). The Portancemetre, the Measur-
ing Ball, and the RDD operate at walking pace and are based
Catalogue of Deflection on a vibrating wheel whose acceleration is doubly integrated
Measuring Devices to produce deflections. The TSD, RWD, RDT, and ARWD use
laser measurements to determine the pavement deflection or
Using this definition, several devices can be removed from deflection slope. Finally, the IDM, which uses image analysis
consideration. Static impulse loading devices (e.g., FWD methods to determine pavement deflections under loading, is
variations and Dynaplaque) are capable of sample intervals still in the early stages of development. Descriptions of each
of less than 300 mm (1 ft) but must be stationary to record device are presented below.
measurements. The FWD was thus removed from consider-
ation as a continuous measurement device, but it is used as
a reference device for field testing for the continuous deflec- Overview of the Most Promising Devices
tion devices. Plate loading devices, certain rolling wheel load Portancemetre
devices (Benkelman beam, Dehlen curvature meter), vibrat-
ing load devices (e.g., Dynaflect), and the Flexigraphe laser The Portancemetre continuously measures the bearing capacity
were also omitted from consideration. The devices consid- of a road. A 10-kN (2.2-kips) test wheel is mounted on a specific
ered are listed in Table 3.1. trailer using a retractable axle (Figure 3.1). A system comprising
There are also several devices that, although the vehicle is a hydraulically unbalanced mass makes the wheel vibrate at a
nonstationary while testing, keep the measurement equip- 35 Hz frequency providing an additional 6 kN (1.3 kips) load-
ment stationary while sampling. Because this does not co- ing. The instrumentation allows the measurement of the ver-
incide with the definition of continuous used for this report, tical acceleration components of the vibrating and suspended
these devices were also omitted from further investigation. masses. Double integration of the vertical acceleration signal
These include the traveling deflectometer (6 to 11 m [20 to determines the vertical load applied to the ground and the cor-
36 ft] spacing), Deflectograph variations (3 to 10 m [10 to responding deflection. This method allows the measurement of
33 ft] spacing), and the Curviametre (5 m [16 ft] spacing). the rigidity of the structure. Since the vibrating wheel is pulled
These devices are listed in Table 3.2. at a slow speed (3 to 4 km/h [2 to 2.5 mph]), measurements are
Devices that met the definition of a continuous deflection taken every 30 mm (1.2 in.), and peak deflection is normally
device were the Portancemetre, the moving FWD, the Mea- reported at 1-m (3.3-ft) intervals.
suring Ball, the traffic speed deflectometer (TSD), the rolling
dynamic deflectometer (RDD), the rolling wheel deflectome-
Measuring Ball and Moving FWD
ter (RWD), the airfield rolling weight deflectometer (ARWD),
the road deflection tester (RDT), and the image deflection The Measuring Ball is a vibrating steel wheel mounted in a
measurement (IDM) device. These are presented in Table 3.3. two-wheel, one-axle trailer towed by a car at about 5 km/h

12

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

13  

Table 3.1. Static Measurement Devices

Nominal Load
Equipment Speed While Status
Generic Name Type Model Equipment Characteristics Testing (km/h) as of 2010

Impulse loading Falling weight Falling weight deflectometer Automated impulse load 0 Production model
device deflectometer
Heavy weight deflectometer 0 Production model
(HWD)
Light weight deflectometer 0 Production model
(LWD)
Loadman LWD 0 Production model
Dynaplaque Impulse generator developed by 0 Production model
LCPC for foundation assessment
Rolling wheel na Benkelman beam 0 Production model
load
Dehlen curvature meter a.k.a. South African curvature 0 Production model
meter
Plate load test Plate load test NA Static load applied by circular 0 Production model
plate; primarily for foundation
assessment
The Thumper NA Plate test affixed to van 0 Production model

Vibrating load Dynaflect Dynaflect Oscillating load applied through 0 Production model
device steel wheels
Schwinger Swiss version of Dynaflect 0 Production model
Road Rater Similar to Dynaflect but higher 0 Production model
loading
Laser-based Flexigraphe NA Laser and photocell 0 Production model
device laser
Note: na = not applicable; NA = not available.

Table 3.2. Moving Measurement Vehicles with Stationary Measurement Apparatus

Nominal
Load Speed
Equipment While Testing Status
Generic Name Type Model Equipment Characteristics (km/h) as of 2010

Rolling wheel na Traveling California vehicle with 2 Benkelman beams 1–1.5 Former production
load deflectometer attached model
Deflectograph Deflectograph Automated Benkelman beams 2.4 Production model
(original model) developed by LCPC, France
Double-beam LaCroix deflectograph developed for use on 2.4 Production model
deflectograph rigid pavements with both beam arms on
same side
Pavement deflection U.K. version of LaCroix deflectograph 2.5 Production model
data logging
machine (PDDL)
Deflecto Variation on LaCroix deflectograph 3.5 Production model
PASE (pavement Australian version of LaCroix 4 Production model
strength evaluator) deflectograph
Deflectolab Australian version of LaCroix 5 Production model
deflectograph
Flash Updated, faster LaCroix deflectograph by 3–10 Production model
LCPC
na Curviametre Geophones mounted on chain stationary on 18 Production model
pavement measuring vertical deflection
Note: na = not applicable.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

14

Table 3.3. Moving Measurement Vehicles with Nonstationary Measurement Apparatus

Nominal
Load Speed
Equipment While Testing Status
Generic Name Type Model Equipment Characteristics (km/h) as of 2010

Vibrating mass na Portancemetre Vertical accelerations measured from steel 3.6 Production model
loading wheel vibrations developed by LCPC, France
Moving FWD Developed by KUAB, Sweden 30 Early prototype
Measuring Ball Similar to Portancemetre 5 Early prototype
Rolling dynamic Vibrating load applied through coated steel 5 Prototype
deflectometer wheels developed in Texas
Rolling wheel na Airfield rolling weight Loaded and unloaded longitudinal profiles 35 Decommissioned
load deflectometer measured by lasers; developed by prototype
QuestUSA for U.S. Air Force
na Road deflection Loaded and unloaded transverse profiles 70 Prototype
tester measured by lasers; developed by VTI,
Sweden
na Rolling wheel Loaded and unloaded longitudinal profiles; Up to 80 Prototype
deflectometer developed by ARA for FHWA
na Traffic speed Doppler laser sensors measuring vertical 60–80 Prototype
deflectometer pavement velocity; developed by Green-
wood, Denmark
Image na IDM device Developed by LCPC, France, using structured 4 Early prototype
deflection light pattern; tested in laboratory and
measurement statically on test track
Note: na = not applicable.

(3 mph). The vertical vibration of the wheel is measured by acceleration peak and the resulting sinusoidal acceleration
means of an accelerometer mounted at the wheel hub. The signal is calculated. The result is a measure of the relative
measurement principle is based on the idea that the stiff- stiffness of the ground and is expressed in terms of a scale
ness of the ground will cause an acceleration at the wheel. from 0 to 150. The peak load generated by the vibration is
The acceleration is processed in a computer housed in the not known but is likely to be significantly less than a typical
towing vehicle, and the relationship between the highest heavy vehicle wheel load.

Figure 3.1. The Portancemetre.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

15  

Through private correspondence it has been found that

a device was developed by KUAB of Sweden. This device
was referred to as a moving FWD and measured, without
stopping, one test point every 15 m (50 ft) at 30 km/h (19 mph).
The device did not measure with the same quality as an FWD,
but more like a deflectograph or Benkelman beam. It was not
developed further into a commercial model because it was
not as accurate as the FWD, and it no longer exists.

Rolling Dynamic Deflectometer Source: Arora et al., 2006.

The RDD is a heavy truck weighing about 200 kN (45 kips) Figure 3.2. Rolling dynamic deflectometer.
that surveys at 4.8 km/h (3 mph). It carries a servo-hydraulic
vibrator capable of producing dynamic loads up to 310 kN
(70 kips) in the frequency range of 5 to 100 Hz super- four Doppler lasers mounted on a servo-hydraulic beam to
imposed on a static load that can be selected within the range record the deflection velocity of a loaded pavement. Three
of 65 to 180 kN (15 to 40 kips). The load is transmitted to Doppler lasers are positioned such that they measure deflec-
the road using two sets of dual wheels mounted side by side tion velocity at a range of distances in front of the rear axle:
on separate axles with a spacing of 1,180 mm (4 ft) between 100, 200, and 300 mm (4, 8, and 12 in.); and 100, 300, and
them, meaning they are rolling inside of the road wheels of 750 mm (4, 12, and 30 in.) in the two present prototypes.
the truck (Figure 3.2). Deflections are measured by means of The fourth sensor, acting as a reference laser, is positioned
accelerometers mounted between further sets of dual wheels 3.6 m (12 ft) in front of the rear axle largely outside the
rolling between the loaded wheel sets and isolated from the deflection bowl. The beam on which the lasers are mounted
dynamic system. The deflections are obtained by the double moves up and down in opposition to the movement of the
integration of the acceleration signal. Using an accelerometer, trailer in order to keep the lasers at constant height from
however, means that only deflections caused by the dynamic the pavement surface. To prevent thermal distortion of the
load variations can be detected. steel measurement beam, a climate control system main-
tains the trailer temperature at a constant 20°C (68°F). Data
is recorded at a survey speed of 70 km/h (45 mph) at a rate of
Traffic Speed Deflectometer
1000 Hz, that is, a 20-mm (0.8-in.) spacing of the raw mea-
The TSD (Figure 3.3) is an articulated truck with a rear axle surements. These results are usually reported as averaged over
load of 100 kN (22 kips), which, in the model evaluated, uses 10 m (33 ft).

Figure 3.3. Two TSD devices at the Transport Research Laboratory (TRL) test track and a computer-generated
schematic.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

16

Measurement Methodology

Source: FHWA, 2009.

Figure 3.4. Rolling wheel deflectometer and its measurement principle.

Rolling Wheel Deflectometer The device uses four sensors to estimate the deflection
due to an applied wheel load. The ARWD places one sen-
The RWD (Figure 3.4) is based on the spatially coincident
sor near the load wheel and three sensors ahead of it in
method for measuring pavement deflections. Three lasers line with the first sensor and beyond the deflection bowl
located in front of the dual tires (away from the applied (Figure 3.5).
load and, therefore, deflection bowl) are used to measure Distances to the pavement surface are measured by the
the unloaded pavement surface, and a fourth laser (addi- first three sensors and then again by the second, third, and
tional lasers have been added in a newer version, as dis- fourth sensors. The measurements are timed so that they
cussed later in the report) located between the dual tires and are spatially coincident. The sensors are placed 2.74 m (9 ft)
just behind the rear axle measures the deflected pavement apart based on the idea that the deflection bowl in most
surface. Deflection is calculated by comparing “spatially pavements at highway speeds is generally less than 2.74 m
coincident” scans as the RWD moves forward. The RWD (9 ft) in radius. This implies that the beam in which the sen-
applies a 40-kN (9-kips) load through 2 wheels spaced sors are mounted must be greater than 8.22 m (27 ft) long.
330 mm (13 in.) apart and surveys at speeds up to 80 km/h However, the deflections of the beam tend to cause signifi-
(50 mph). The deflection profile is obtained by subtracting cant errors, which are magnified in computations. To over-
the profile of the deflected shape from that of the unde- come these limitations, the ARWD uses a laser beam that is
flected shape measured in the same location. The RWD
surveys with a 2 kHz sampling rate, that is, every 11 mm
(0.5 in.) at 80 km/h, and averages the deflection values over
longer sections, typically 160 m (0.1 mi), to produce a single
deflection measurement.

Airfield Rolling Weight Deflectometer

The ARWD was designed to measure runway deflections
under a wheel load of 40 kN (9 kips) moving at a speed of
35 km/h (20 mph). The ARWD was developed by Quest Inte-
grated, Inc. in the 1990s for the U.S. Air Force. The equipment
was updated in the mid 2000s with the help of Dynatest after
the equipment was transferred to the U.S. Army for repair.
The equipment was decommissioned in 2010 by the Air Force
after failed attempts to repair the system. The sensors were
returned to Quest. Figure 3.5. Airfield rolling weight deflectometer.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

17  

passed inside the physical beam as a reference to measure the of a structured pattern on the road surface. A camera cap-
deflection of the physical beam and makes corrections for tures the surface; and software analyzes the pattern defor-
this deflection in the computations. This process overcomes mation, thereby measuring the pavement deflection. The
the problem of thermal expansion and vibrational bending technique has been checked in the laboratory, in static
of the beam. tests, and with a load moving at 4 km/h (2.5 mph). The
latter tests were carried out on the LCPC circular acceler-
ated loading facility at Nantes (Figure 3.7). Development is
Road Deflection Tester
ongoing to turn this device into a robust operational mea-
The RDT (Figure 3.6) consists of a truck that has been retro- surement tool.
fitted with two arrays of laser range finders, each consisting
of 20 sensors arranged in a line transverse to the direction
Summary of Promising Devices
of travel. The first array is positioned 2.5 m (8 ft) behind
the front wheels, and the second array is placed 0.5 m Several continuous deflection devices exist that can measure
(1.6 ft) behind the rear wheels. Thus, the distance between when constantly moving and can collect data at intervals of
the two arrays is approximately 4 m (13 ft). The first array approximately 300 mm (1 ft) or smaller using load levels typi-
measures the transverse deflection profile largely outside cal of truck loading (i.e., 40 to 50 kN [9 to 11 kips] per wheel
of the deflection basin; the second measures the deflection or load assembly). These include the following three main
profile near the center of the deflection basin. The truck types of devices:
has two weights of 4 kN (1 kip), which can be moved back
and forth. During testing, these weights are moved to the • Laser-based devices that measure the deflection below a
rear of the truck. The weights are moved back to the front moving truck load—including the TSD, RWD, RDT, and
of the truck during transportation for better weight distri- ARWD;
bution. The engine of the truck is also placed in the rear, • Devices that apply a vibratory load—including the Por-
and together with the weights it can produce a force of 40 tancemetre, the Measuring Ball, and the RDD; and
to 70 kN (9 to 15.7 kips) on the rear axle. The sampling • The IDM device, which uses image analysis methods to
hardware operates at a sampling frequency of 1 kHz, and determine pavement deflections under loading; this repre-
at a speed of 70 km/h (45 mph). Samples are stored every sents a very promising technology, but it is still in the early
20 mm (0.8 in.), but are normally reported at 50-m (165-ft) stages of development.
intervals.
Only the devices in the first group are currently capable of
surveying without the need for traffic control. The vibra-
Image-Based Deflection Measurement Device
tory devices operate at walking pace, and the IDM was
This equipment has been recently developed by the LCPC still being tested in a stationary mode at the time of the
Nantes and LRPC Strasbourg in France using the projection evaluation.

Figure 3.7. IDM prototype being tested at LCPC’s

Figure 3.6. Swedish road tester. pavement accelerated load facility.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

18

Survey of State DOT Table 3.5. Summary of Responses

Practices and Needs to Question: If Your Agency
Routinely Uses Deflection Testing,
The demand and the potential value of continuous deflec- Please Indicate an Estimate of the
tion devices were evaluated through a survey of state and Typical Testing Spacing in Feet
provincial DOTs. The two-stage survey included questions to
assess technical needs and also aimed to determine the value Response Response
Answer Option Average Count
assigned by the agencies to the collected data.
Project (ft)   437 24
Network (ft) 1,297 14
Stage I: Web Survey
Note: This question was answered by 25 respondents
For the web survey, a list of potential participants provided by and skipped by 19.

National Cooperative Highway Research Program (NCHRP)

for a similar project was complemented with personal contact devices (FWD), which probably explains why more testing
from the project team. The final list comprised 63 potential is performed at the project level rather than at the network
participants. A commercial web-based survey application, level. These results are consistent with those reported by
SurveyMonkey, was used. The web survey link was sent to Flintsch and McGhee (2009). The average spacing between
56 of the potential participants because seven had the survey FWD tests is 133 m (437 ft) and 396 m (1,297 ft) for project
service blocked. Of the 56 recipients, 44 completed the sur- and network level, respectively (Table 3.5).
vey, resulting in a response rate of 79%. A number of respondents suggested it would be very valu-
able to them to have a continuous deflection device, most
Practices and Uses of Deflection Testing importantly for network-level data collection. The main
desired uses for the continuous device would be to do the
Thirty-five of the survey respondents (84%) replied their following:
agency uses pavement deflection testing. As presented in
Table 3.4, most deflection testing is performed exclusively • Determine subgrade modulus (65%);
in-house (77%), while only 9% of deflection testing is exclu- • Calculate overlay thickness (65%); and
sively outsourced. A number of agencies (five, or 14%) rely • Select the most appropriate type of rehabilitation (50%).
on both in-house collection and outsourcing. The main uses
of the deflection testing are as follows: The main concerns for the adoption of a continuous deflec-
tion device, voiced by the respondents, were safety, accu-
• To determine the subgrade modulus or bearing capacity
racy, and cost and savings. The average costs reported for
(97%) for flexible pavements; and project- and network-level deflection measurements were
• To evaluate the joint/crack transfer efficiency (59%) for
variable ($28 to $3,000 and $10 to $790, respectively).
rigid pavements. These numbers were primarily based on best estimates, with
some respondents reporting typical agency values. The
Deflection testing is performed mainly at the project level, average current cost for network-level data collection was
but several agencies indicated that they are also doing so at estimated to be around $100/km ($167/mi), based almost
the network level. All respondents use static deflection testing exclusively on rough estimates.

Table 3.4. Summary of Responses Pavement Rehabilitation Design Applications

to Question: How Does Your Agency
Currently Collect Pavement Deflection Whether rehabilitation design procedures are based on
Data (Please Check All that Apply)? empirical or mechanistic–empirical (ME) methods, deflec-
tion measurements data can provide valuable information.
Response Response
Answer Option Percentage Count
This is supported by the survey responses, which indi-
cate that 85% (29 out of 34 respondents; 10 respondents
In-house collection 77.1% 27 skipped the question) of agencies incorporate deflection
Outsourced 8.6% 3 testing into their pavement rehabilitation design procedure.
Both 14.3% 5 The main uses of deflection measurements to support their
Note: This question was answered by 35 respondents and
pavement rehabilitation design procedures include subgrade
skipped by nine. modulus determination and overlay thickness determination

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

19  

Table 3.6. Summary of Responses to Question: What Are the Key Engineering
Parameters that You Would Wish to Derive from Deflection Testing?

Breakdown by Pavement Type

Number of
Key Engineering Parameter Respondents Percentage Flexible Composite JCP CRCP

Subgrade bearing capacity 31 86% 31 16 14 6

Deflection values 26 72% 26 14 12 5
Layer moduli 25 69% 25 12 8 3
Joint/crack transfer efficiency 19 53% 2 8 19 4
In-service structural number 18 50% 18 10 6 2
Deflection basin area 18 50% 17 7 7 1
Pavement remaining service life 14 39% 14 10 8 4
Depth to bedrock 6 17% 6 3 2 1
Note: JCP = jointed concrete pavement; CRCP = continuously reinforced concrete pavement.

(65% of responses each), followed by determination of type Stage II: Follow-Up Interviews
of pavement rehabilitation (50% of responses). In most cases
Using results of the survey, the research team identified
(91% of responses), agencies use deflection testing results to
a subset of states to interview. A more detailed question-
determine multiple parameters (two or more).
naire was prepared, and interviews were conducted over
Most of the respondent agencies still rely on an empiri-
the phone with the following nine state DOTs: Arizona,
cal pavement design methodology, mainly the AASHTO
Florida, Indiana, Kansas, Montana, New Hampshire, New
1993 methodology or some modification of it (26 of
Mexico, Oregon, and Virginia. This list includes states that
34 respondents or 76%). Although several agencies (16 out
use network-level deflection testing in their PMS (Arizona,
of 34 respondents or 47%) also use mechanistic–empirical
Florida, Indiana, Kansas, and Virginia), as well as states that
design methods, only four agencies (12%) were exclusively
have some experience with a continuous deflection device,
using an ME design procedure. mainly the RWD (Indiana, Kansas, New Hampshire, Oregon,
Table 3.6 summarizes the main engineering parameters and Virginia). Questions were divided into three categories:
that survey respondents would like to derive from the (1) desired capabilities and applications of a continuous mea-
deflection measurements. Flexible pavements’ subgrade suring device, (2) use of deflection data within the PMS, and
structural bearing capacity was the most frequently men- (3) experience with the RWD.
tioned parameter, followed by deflection values and layer
moduli (in bold).
Desired Uses and Capabilities

Pavement Management Applications The responses indicate that most respondents envision using
a continuous deflection device for network-level data col-
Perhaps the primary benefit from a continuous deflection lection. Within this framework, speed is perceived as the
measuring device is its ability to provide an overall assess- most critical characteristic even if it means sacrificing some
ment of the structural condition of the pavement network. accuracy, as long as results are comparable to static deflec-
Deflection test results can be incorporated into an agency’s tion measurements, such as with the FWD. However, a few
pavement management system (PMS) to support main- respondents indicated a desire to obtain a deflection basin,
tenance and rehabilitation strategy scoping and resource area parameter, or some other parameter that can be used
allocation decisions, among other asset management busi- to assess the structural capacity of the various pavement lay-
ness functions. Although most of the respondent agencies ers or detect hidden problems (e.g., stripping) in some of
(93%) have implemented a PMS, only five incorporate the undersurface layers. Ground-penetrating radar (GPR)
the results of deflection testing into their PMS. The dollar to determine layer thicknesses was reported as a desired fea-
amount that agencies are willing to pay to obtain continu- ture that could be easily added to the system. Other desir-
ous deflection is in the same range as the amount they cur- able characteristics, which would facilitate the adoption of
rently pay for FWD measurements, around $6 to $125/km the technology, include: (1) ease of operation, (2) availabil-
($10 to $200/mi). ity of fast data processing software and service support, and

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

20

(3) data format compatibility with the current agency data- Experience with Existing Continuous Deflection
base structures. Measuring Equipment
The primary application of the continuous deflection
A number of interviewed state DOTs (Indiana, Kansas, New
device at the network level would be to do the following:
Hampshire, Oregon, and Virginia) have had some experi-
ence with the RWD, mostly through FHWA-sponsored dem-
• Help identify “weak” (i.e., structurally deficient) areas that
onstration projects. In general, the representatives found
can then be investigated further at the project level;
RWD test results to be repeatable, successful in identifying
• Provide network-level data to calculate a “structural health
problem areas, and generally well correlated with FWD test
index” that can be incorporated into a PMS; and
results (except in the case of Virginia). The main data col-
• Differentiate sections that may be good candidates for
lected included maximum deflections every 0.1 mi (tempera-
preservation (good structural capacity) from those that
ture corrected) and location, along with speed in some cases.
would likely require a heavier treatment (showing struc-
Some reports provided by state DOT representatives also
tural deficiencies).
included repeated test and correlations with the FWD; these
are discussed in detail in the following sections.
A desired application at the project level would be to provide
In addition to the demonstration projects, the Kansas
input for rehabilitation pavement design (e.g., input to the
DOT independently contracted measurements on one seg-
Mechanistic–Empirical Pavement Design Guide (MEPDG)/ ment of a four-lane rural interstate highway, I-70. Testing
DARWin-ME or other overlay thickness design method). was conducted as a screening tool to detect potential hid-
Other desired applications mentioned include determination den problems along the highway corridor. Although no sur-
of long-term trend in structural capacity, overall evaluation of prises were found, the assessment was that the equipment
bounded layers (e.g., detecting stripping), and calculation of performed well.
remaining service life.
Important parameters that users indicated should be
considered in evaluation of the equipment include speed Summary of User Needs
(safety), repeatability, accuracy (and feasibility of estab- The majority of agencies perform at least some deflection
lishing correlations with existing technologies, such as the testing using the FWD. Most testing is performed to support
FWD), equipment cost, ease of operation, customer service project-level decisions, and only a small number of agen-
(availability of service and maintenance), ease of use of the cies (five) have incorporated deflection data into their PMS.
data collected, availability of software for interpretation Potential users in general agree that the main advantage of
of results, reliability, size of the vehicle, relevance of the a continuous measuring device would be for supporting
information (e.g., use in MEPDG/DARWin-ME), and past network-level decisions. The assessment of user needs sug-
experience. gests the following:

Current PMS Uses • Important parameters that users indicate should be

considered in the evaluation of the equipment include:
The parameters that have been used by DOTs that currently speed (safety), repeatability, accuracy (and feasibility of
use FWD data in a PMS include the effective structural num- establishing correlations with existing technologies, such
ber (SN) and layer moduli for flexible pavements, and AREA as FWD), equipment cost, ease of operation, customer
and k-value for rigid pavements (Virginia). Results of deflec- service (availability of service and maintenance), ease
tion testing have also been used to compute a structural index of use of the data collected, availability of software for
(Indiana) and as part of a decision tree for project scoping interpretation of the results, reliability, size of the vehi-
(Indiana and Virginia), as well as for pavement overlay design cle, relevance of the information (e.g., use in MEPDG),
and pavement deterioration monitoring (Virginia). The and past experience.
Kansas DOT uses the FWD’s center and last deflections to • While the responses to the initial survey suggested users
make remaining-life calculations. would like to be able to collect continuous pavement
Some reasons cited for not incorporating results of response data to support project-level decisions, the
deflection testing into the PMS include cost associated follow-up interviews showed that respondents under-
with data collection, technical issues such as software and stand the current limitation of the technologies and agree
programming, and organizational issues related the agen- that network-level applications are more likely in the near
cy’s structure (i.e., one division does the data collection and future. Furthermore, respondents agreed on the need of
another runs the PMS; planning division versus maintenance pavement structural data to support network-level PMS
division). decisions.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

21  

• At the network level, the primary application of the the adopted definition of a continuous deflection device. The
continuous deflection device would be to (1) help iden- device was defined as a deflection measuring device constantly
tify “weak” (or structurally deficient) areas that can be moving that can collect data at intervals of approximately
then investigated further at the project level; (2) provide 300 mm (1 ft) or smaller using load levels typical of truck load-
network-level data to calculate a “structural health index” ing (i.e., 40 to 50 kN [9 to 11 kips] per wheel or load assembly).
that could be incorporated into a PMS; and (3) differenti- Devices that met the definition of a continuous deflection
ate sections that may be good candidates for preservation device were evaluated in detail in Tables 3.7, 3.8, and 3.9. The
(good structural capacity) from those that would likely Portancemetre and the Measuring Ball were eliminated from
require a heavier treatment (showing structural deteriora- consideration because they do not apply loads similar to that
tion and deficiencies). of a heavy vehicle. The Portancemetre measures the response of
a pavement under an oscillating load with an average value of
10 kN (2.2 kips) and an amplitude of 6 kN (1.3 kips) at a speed
Selection of Candidate Devices
of 3 to 4 km/h (2 to 2.5 mph). Similarly, the Measuring Ball
A more detailed summary highlighting the current knowl- is towed by a car at about 5 km/h (3 mph) and applies a load
edge of the capabilities of the continuous deflection devices significantly lower than a typical heavy vehicle wheel load. Since
is presented in Tables 3.7, 3.8, and 3.9. These tables provide a the magnitude of the load is small and the speed slow, these
somewhat subjective evaluation of the various technologies devices are primarily used for quality checks on unpaved sur-
in the following broad categories: faces. The RDT and ARWD were also eliminated from fur-
ther consideration because the existing prototypes have been
• Measurement capability; decommissioned or reassigned to other uses.
• Types of pavements suitable for measuring; The IDM system appears to be very promising; however,
• Sampling rate; an operational prototype is not yet available and, at pres-
• Accuracy; ent, does not meet the survey speed requirement. The most
• Operating conditions; recent published information on the IDM device described
• Development status; trials in which the measuring device was stationary and
• Available interpretation methods for different types of only the loaded wheel was moving, at just 4 km/h. The pre-
applications; and liminary trial concentrated on measuring the bowl shape
• Extent of usage for different applications. in the area of maximum change, partly for convenience
and partly because associated modeling and analysis has
It should be noted that these tables have not been updated suggested that such information is a useful supplement to
since they were produced for Phase I of this study in early maximum deflection when assessing pavement condition.
2010. The FWD is also included because it was adopted as In an ongoing project, developers hope, by mounting a ver-
the reference device by which to evaluate the comparability sion of the system on a heavy truck and with the aid of a
of the continuous deflection measurement devices. Although fast camera, to measure deflection in a continuous fash-
the authors understand that this technology has limitations, ion. Benefits of this system are the potential for continuous
it represents the most common mobile deflection measur- measurements and the potential to measure across joints
ing device available worldwide with some degree of stan- if the geometry of measurement close to the loaded wheel
dardization. Thus, the FWD can be used as a reasonable can be resolved.
reference in both U.S. and European assessments of the The RDD was also originally identified as a good candi-
equipment. date, especially for measurements on concrete pavements;
Based on information collected in the literature review however, it was not selected for further evaluation because
and summarized in Tables 3.7, 3.8, and 3.9, the research team the user needs survey indicated that the majority of the users
identified two devices as the most promising to deliver the would prefer a device that can measure at traffic speed, that is,
information needed by users under operating conditions com- at least at 55 km/h (35 mph). The original machine operated
patible with SHRP 2 objectives. These devices are the rolling at around 1.5 km/h (1 mph); it has recently been updated to
wheel deflectometer (RWD) and the traffic speed deflec- operate at 5 km/h (3 mph). This is still far short of operating
tometer (TSD). at a speed that does not require traffic control on busy roads.
As indicated in the Catalogue of Selection Measuring Nevertheless, the combination of a suitable frequency (gener-
Devices, static loading devices in which the vehicle is non- ally between 5 and 100 Hz) and relatively low survey speed
stationary while testing but keeps the actual measurement enables the assessment of the deflection response at the joints
equipment stationary while sampling (e.g., the deflecto- in concrete pavement that provides valuable guidance as to
graph) were not considered because they did not conform to required rehabilitation measures.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

22

Table 3.7. Detailed Device Evaluation Table

Measurement Device

IDM Portancemetre Measuring

Short Description FWD (Denmark) (France) (France) Ball (Sweden)

Measurement Measures nominal peak deflection Yes Unknown Yes Unknown

capability or equivalent
Measures nominal deflection bowl Yes Unknown No Unknown
shape or equivalent
Transverse measurement position 1 (mid-vehicle) 1 1 (mid-vehicle) 1
Type of load Impulse on Unknown Oscillating at 35 Hz on Unknown
300-mm 1-m (3-ft) diameter
diameter plate wheel
Peak load level 50 kN (11 kip) Unknown 10 ± 6 kN Unknown
(2.2 ± 1.3 kip)
Pavement type Fully flexible Yes Unknown No Unknown
suitability
Rigid Yes Unknown No Unknown
Granular Yes Unknown Yes Unknown
Sampling rate Normal sampling rate na Unknown 30 mm (1.2 in.) at 3 km/h Unknown
Normal reporting rate na Unknown 1 m (3 ft) Unknown
Accuracy Repeatability Very good Unknown Moderatea Unknown
Comparability Good (many Unknown Moderatea (several Unknown
devices) devicesa)
Relation to other devices na Unknown Unknown Unknown
Operating conditions Typical survey speed 0 4 km/h 3 km/h (2 mph) 5 km/h
(2.5 mph) (3 mph)
Surface conditions Any Drya Anya Unknown
Traffic management required Yes Yes Yes Yes
Status Production/development status Production model Laboratory Production model Unknown
prototype
% commercially developed 100 10 100 Unknown
Available interpretation Flexible pavements Yes No No Unknown
methods for
Rigid pavements Yes No No Unknown
pavement types
Granular pavements Yes No Yes Unknown
Use Distance surveyed Unknown 0 Unknown Unknown
Surveying Flexible/rigid/composite Unknown 0 Unknown Unknown
pavements
Screening structurally deficient Yes Unknown No Unknown
sections
Defining rehabilitation strategies Yes Unknown No Unknown
Designing rehabilitation treatments Yes Unknown No Unknown
Note: na = not applicable.
a Author’s estimate, which could not be verified by available information at this stage.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

23  

Table 3.8. Detailed Device Evaluation Table

Measurement Device

Moving FWD ARWD (United

Short Description RDD (United States) (Sweden) States)

Measurement capability Measures nominal peak deflection or Yes Yesa Yes

equivalent

Measures nominal deflection bowl shape or Yes (4 points) Unknown No

equivalent

Transverse measurement position 1 (mid-vehicle) 1 1 (mid-vehicle)

Type of load Oscillating at 30 Hza Impulsea load every Fixed dual wheel
on 300-mma (12-in.) 15 m at 30 km/h assembly
diameter wheel

Peak load level 55 ± 20 kN (12 ± Unknown 40 kN (9 kip)

4.4 kip) on rigid
pavement

Pavement type suitability Fully flexible Yes Yes Yes

Rigid Yes No No

Granular Yes Yes No

Sampling rate Normal sampling rate 13 mm at 5 km/h Unknown 3 m at 35 km/h

(0.5 in.) (9 ft)

Normal reporting rate 0.6–0.9 m (2–3 ft) 15 m at 30 km/h 25 m (90 ft)

(50 ft)

Accuracy Repeatability Gooda Unknown Unknown

Comparability na (Only one device) Unknown na (Only one device)

Relation to other devices Strong with FWD Unknown Unknown

Operating conditions Typical survey speed 1.5–5 km/h (1–3 mph) 30 km/h (19 mph) 35 km/h (22 mph)

Surface conditions Any a

Unknown Dry

Traffic management required Yes Yes Probably

Status Production/development status Working prototype Decommissioned Decommissioned

prototype prototype

% commercially developed 80 Unknown 60

Interpretation methods Flexible pavements Yes a

Unknown No
available for . . .
Rigid pavements Yes a
Unknown No

Granular pavements No Unknown No

Use Distance surveyed Unknown Unknown Unknown

Flexible/rigid/composite pavements surveyed Unknown Unknown Unknown

Screening structurally deficient sections Yesa Unknown Unknown

Defining rehabilitation strategies Ya Unknown Unknown

Designing rehabilitation treatments Y Unknown Unknown

Note: na = not applicable.
a Author’s estimate, which could not be verified by available information at this stage.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

24

Table 3.9. Detailed Device Evaluation Table

Measurement Device

Short Description RWD (United States) RDT (Sweden) TSD (Denmark)

Measurement capability Measures nominal peak Yes Yes Yesa

deflection or equivalent
Measures nominal deflec- No Yes Yesa (3 points)
tion bowl shape or
equivalent
Transverse measurement 1 wheelpath 2 wheelpaths 1 wheelpath
position
Type of load Fixed dual wheel assembly Fixed dual wheel assembly Fixed dual wheel assembly
Peak load level 40 kN (9 kip) 40–70 kN (9–16 kip) 50 kN (11 kip)
Pavement type suitability Fully flexible Yes Yes Yes
Rigid No No Maybe
Granular No No No
Sampling rate Normal sampling rate 11 mm at 80 km/h (0.4 in.) 20 mm at 70 km/h (0.8 in.) 20 mm at 70 km/h (0.8 in.)
Normal reporting rate 30 m (100 ft) 50 m (165 ft) 10 m (30 ft)
Accuracy Repeatability Gooda Moderateb Goodb
Comparability na (only one device) na (only one device) Goodb (two devices)
Relation to other devices Strong with FWD Poor with FWD and Strong with FWD and
deflectograph deflectograph
Operating conditions Typical survey speed 80 km/h (50 mph) 70 km/h (45 mph) 70 km/h (45 mph)
Surface conditions Dry Dry Dry
Traffic management No No No
required
Status Production/development Working prototype Prototype reassigned to Two working prototypes
status other uses
% commercially developed 80a 60 95
Interpretation methods Flexible pavements Yesb No Yesb
available for pavement
Rigid pavements No No No
types
Granular pavements No No No
Use Distance surveyed 12,000+ km (7,500+ mi) Unknown 30,000+ km (19,000+ mi)
Flexible/rigid/composite Mostly Flexible/composite Unknown Mostly Flexible/composite
pavements surveyed
Screening structurally Yes No Yesb
deficient sections
Defining rehabilitation Yesb No Yesb
strategies
Designing rehabilitation No No No
treatments
Note: na = not applicable.
a Measures vertical deflection velocity, which is converted to deflection slope by dividing by the horizontal vehicle velocity. Three deflection slopes enable maximum

deflection and part of the bowl shape to be estimated.

b Author’s estimate, which could not be verified by available information at this stage.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

25  

Detailed Description of the mounted within a custom-designed 17-m (53-ft) semitrailer.

Selected Equipment The measured deflection is the response from one-half of an
80-kN (18-kip) single-axle load traveling at normal traffic
This section expands on the characteristics of the selected speeds. In a previously tested version, an aluminum reference
devices and provides a preliminary assessment of their tech- bar, suspended beneath the trailer, contained four laser sensors
nical capabilities based on the information collected from to measure the distance to the pavement surface (Figure 3.8).
trials and demonstration projects from the United States and The RWD uses a spatially coincident methodology for measur-
Europe. Both selected devices, the RWD and TSD, conduct ing pavement deflection. This method was originally developed
measurements (deflection or deflection velocity of a loaded by the Transport and Road Research Laboratory (TRRL) and
pavement, respectively) under a truck axle and at speeds close implemented by Dr. Milton Harr at Purdue University.
to that of traveling traffic. In the evaluated model, three lasers are used to measure the
By contrast, the technology currently in use, the FWD, applies distance to the unloaded pavement surface (i.e., forward of
an impulse load (using known weights that are dropped from and outside the deflection basin), and a fourth laser, located
specific heights onto a load plate) and measures the response between the dual tires and just behind the rear axle, measures
(deflections) to those loads at seven to nine surface locations, the distance to the deflected pavement surface. The deflec-
starting at the center of the loading plate and extending radially tion is calculated by comparing the laser scans profile as the
up to 1.8 m (72 in.) from the load plate center. The deflection RWD moves forward. The beam uses a curved extension to
basin at each test location is indicative of the stiffness of the pass under and between the dual tires, placing the rearmost
underlying pavement structure and given the various sensors, laser approximately 150 mm (6 in.) rear of the axle centerline
allows a quantification of the structural capacity of various lay- and 178 mm (7 in.) above the roadway surface. The wheels
ers or layer types. A production rate of approximately 4 lane-km are spaced a safe distance from the laser and beam using cus-
(2.5 lane-mi) per day is typical, assuming testing at an interval tom lugs spacers (Steele and Vavrik, 2006). The equipment
of 23 m (75 ft). This testing scheme is best suited for project- has recently been retrofitted to enclose the laser sensors and
level testing. Agencies employing FWD for network-level FWD an additional sensor has been added, as discussed in New
test protocol may adjust this test spacing to achieve a produc- Improvements to the Evaluated Equipment in Chapter 4.
tion rate of approximately 32 to 40 lane-km (20 to 25 lane-mi) At 89 km/h (55 mph), the RWD’s 2-kHz lasers take readings
per day, assuming an interval of 320 m (0.2 mi) (Diefenderfer, approximately every 11 mm (0.5 in.), resulting in extremely
2010). large data sets. The average deflection every 160 m (0.1 mi)
is typically reported; this averaging helps reduce scatter and
file size. Figure 3.9 shows an example of the data collected
Rolling Wheel Deflectometer (RWD)
during the demonstration in Virginia. Each square represent
Detailed Description
the average deflection for each 160 m (0.1 mi); the continu-
The RWD has been developed by Applied Research Associates, ous line represents a 1.6-km (1-mi) moving average. The
Inc. (ARA), with support from the FHWA. The RWD system is data is generally filtered to eliminate outliers due to bridges,

Source: Diefenderfer, 2010; and FHWA, 2009.

Figure 3.8. RWD during testing in Virginia and close-up of laser sensor placed between dual tires.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

26

RWD Deflection (0.1 mile average) Moving Average

Sectioning (based on last resurfacing)
14

12

RWD Deflection (mils)

10

0
105 110 115 120 125 130 135 140
Mile Marker

Figure 3.9. Example of deflection data collected on eastbound I-64 in

Virginia in 2006.

sudden changes in speed, and so forth, before the analy- According to documented test reports, the total mile-
sis is performed. Additional details of the RWD deflection age tested exceeded 11,300 km (7,000 lane-mi). Each state
measurement process are presented elsewhere (ARA, 2005a; agency self-developed its test plan. Several of these tests
Steele and Hall, 2005). Recent upgrades to the RWD include included FWD measurements on the same sections; how-
improved laser sensors that are located within a temperature- ever, not all were conducted at the same time as the RWD
controlled housing. The result of these improvements on measurements were taken. A few of the demonstrations
comparison testing with FWD is ongoing and unknown at included multiple runs to assess the repeatability of the
this time. device.
The RWD technology can test approximately 320 to 480 km One of the earliest RWD test reports was authored by Arora
(200 to 300 lane-mi) per day. A potential benefit of the RWD is et al. (2006) and described testing in Texas in 2004. The RWD
that the load, loading mechanism, and loading rate of the RWD was used to test approximately 425 km (264 lane-mi) (a vari-
are thought to match more closely the actual dynamic effects ety of state routes with five repeat runs per roadway) with
on pavements caused by vehicle loading. In addition, the RWD some companion FWD testing. FWD deflection values ranged
testing is conducted at or near highway speeds with limited from approximately 100 to 1,300 microns (4 to 50 mils). The
or no traffic control requirements and minimal interruption authors stated that the RWD testing was repeatable, based on
to the highway users. However, the RWD does not currently visual observation of the plotted deflection results. In dis-
allow for some of the structural capacity analysis offered by the cussing a relationship of RWD to FWD deflection results, the
FWD. In its current state of development, it is anticipated that authors stated that some relationship exists although the data
the RWD could be used to prescreen the pavement network to “shows some scatter especially at smaller deflection values.”
identify areas that might require additional and more detailed The authors suggested that lower deflection values might be
study at the project level using traditional techniques such as measurable only at lower speeds.
the FWD, or to identify segments that could be good candi- Gedafa et al. (2008) reported on RWD testing of 333 km
dates for pavement preservation. (207 mi) of non-interstate highway in Kansas in 2006. The
The FHWA has sponsored RWD demonstration projects results of the RWD testing were compared to FWD testing
throughout the United States. Testing has been conducted that was conducted from 1998 to 2006. RWD testing was per-
in coordination with at least 16 U.S. state highway agencies, formed at 89 km/h (55 mph) with deflection readings aver-
including those in California, Colorado, Connecticut, Indiana, aged every 160 m (0.1 mi). The FWD data were collected at
Iowa, Kansas, Kentucky, Minnesota, New Hampshire, New 5 to 10 points per mile. The average FWD center deflec-
Jersey, New Mexico, Ohio, Oregon, Texas, Virginia, and West tion value ranged from approximately 0.13 to 0.45 mm (5 to
Virginia; on a federal road under the jurisdiction of the fed- 18 mils) (40 kN [9 kips] load). The results showed that the
eral lands (Natchez Trace); and on several test tracks including RWD deflection reading and the FWD center deflection value
the National Center for Asphalt Technology (NCAT), Virginia were statistically similar based on a significant difference test
Smart Road, and MnRoad. statistic. A linear regression analysis was also performed that

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

27  

showed a strong correlation between FWD and RWD deflec- axle. The fourth sensor is positioned 3.6 m (12 ft) in front of
tion readings. the rear axle, largely outside the deflection bowl, and acts as
Virginia reported RWD testing on portions of two inter- a reference laser. The beam on which the lasers are mounted
state routes and a loop consisting of primary rural highways moves up and down in opposition to the movement of the
in 2005 (Diefenderfer, 2010). All RWD testing was done at trailer in order to keep the lasers at a constant height from
or near the prevailing traffic speed. Companion FWD test- the pavement surface. To prevent thermal distortion of the
ing was conducted in 2006 on the two interstate test sections. steel measurement beam, a climate control system maintains
The FWD deflection values ranged from 0.08 to 0.38 mm the trailer temperature at a constant 20°C (68°F). Two proto-
(3 to 15 mils), with a majority less than 0.2 mm (8 mils) types had been developed at the time of the evaluation by
(40 kN [9 kips] load). The two interstate test sections com- the manufacturer, Greenwood Engineering A/S of Denmark.
prised hot-mix asphalt (HMA) (200–300 mm [8–12 in.]) over One is owned and operated by the Danish Road Institute
compacted aggregate and HMA (100–150 mm [4–6 in.]) over (DRI); the other is owned by the U.K. Highways Agency (HA)
CRCP (200 mm [8 in.]). Statistical testing of RWD repeat- and operated on their behalf by the U.K. Transport Research
ability was performed by use of a non-paired t-test assuming Laboratory (TRL). Newer production devices have incorpo-
equal variances. The results showed that for 8 of 15 trials on rated more Doppler laser sensors.
interstate highways and all non-interstate test sections, the The lasers are mounted at a small angle to measure the hori-
RWD data were repeatable. A poor linear correlation was zontal vehicle velocity, the vertical and horizontal vehicle sus-
found between the RWD and FWD measurements (adjusted pension velocity, and the vertical pavement deflection velocity.
R2 values less than 0.2). However, the FWD measurements Due to its location midway between the loaded trailer axle and
were taken several months after the RWD measurements and the rear axle of the tractor unit, the reference laser is expected
only on interstate sections with relatively low and uniform to measure very little vertical pavement deflection velocity, and
deflections. The results suggested that the deflection value its response can therefore be used to remove the unwanted sig-
may be influenced by surface texture as the standard devia- nals from the three measurement lasers. When accurately cali-
tion varied approximately at locations where the HMA surface brated, the TSD produces measurements of deflection velocity
mixture also varied. that depend on driving speed. To remove this dependence, the
deflection velocity is divided by the instantaneous survey speed
to give a measurement of deflection slope, as illustrated in Fig-
Traffic Speed Deflectometer (TSD)
ure 3.10. Deflection velocity is measured in mm/s while survey
Detailed Description
speed is measured in m/s; therefore, deflection slope measure-
The TSD (Figure 3.10) is mounted on an articulated truck ments are given in units of mm/m (Ferne et al., 2009b).
with a rear axle load of 100 kN (22 kips), which, in the model
evaluated, uses four Doppler lasers mounted on a servo-
The DRI Machine
hydraulic beam to record the vertical deflection velocity of
a pavement as it is loaded by one of the dual wheel axles. The DRI and Greenwood jointly developed the TSD, initially
Three Doppler lasers are positioned such that they measure called the high speed deflectograph (HSD), and have pub-
deflection velocity at a range of distances in front of the rear lished a number of papers on this work. An early independent

Figure 3.10. Computer render of the TSD, with the TSD in

operation (inset).

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

28

operating the device on their network, having covered their

main network from 2005 to 2007. Based on this experience,
Baltzer (2009) reported the DRI daily survey coverage for the
device of around 170 to 225 km (105 to 140 mi).

The Highways Agency TSD

The second prototype TSD currently in operation is owned
by the U.K. Highways Agency (HA) and operated by the
U.K. Transport Research Laboratory (TRL). The TRL has
reported its development and performance in recent confer-
ence papers (Ferne et al., 2009a and 2009b). An example of
typical survey results expressed over 10-m (33-ft) and 100-m
(1/16-mi) lengths are provided in Figure 3.12.

Figure 3.11. DRI device measuring side by side with Equipment Status
the U.K. TSD at TRL in 2008.
At the time of completion of the Phase I evaluation (Febru-
ary 2010), the RWD was a working prototype, with only one
evaluation by the Laboratoire Central des Ponts et Chaussées such prototype in existence. This prototype has been recently
(LCPC) in 2003 (Simonin et al., 2005) showed that even upgraded by adding an additional laser sensor and provid-
though the early prototype had limitations, it demonstrated ing temperature control to the beam that supports the lasers.
good repeatability in the short term and a good degree of There were also two working TSDs with new models (with
correlation with the maximum deflection recorded by other more sensors) under construction. At the time of the cur-
devices such as the FWD and the deflectograph. Other DRI rent report, two additional TSDs have been constructed, for
publications confirm some aspects of this work when assessing agencies in Italy and Poland. A fifth device is currently under
a developed version of this device that they currently own and construction for the South African Highway Administration.
operate. The DRI (Figure 3.11) also have practical experience The latter three devices incorporate improvements to the

1.000
10m averages
0.900
100m averages

0.800

0.700
Deflection Slope [mm/m]

0.600

0.500

0.400

0.300

0.200

0.100

0.000
0 500 1000 1500 2000 2500 3000 3500 4000
Chainage [m]

Figure 3.12. TSD deflection slope at 10 m (33 ft) and 100 m (330 ft) means against distance.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

29  

earlier two prototypes, such as additional velocity sensors interpretation method for either device although recent
and improved calibration facilities. The website of the manu- research has suggested that the TSD equipment may have a
facturer, Greenwood Engineering A/S, currently designates role in the preliminary evaluation of the joint condition of
the TSD as a production model and gives details about some rigid pavements.
features added to the equipment.
Use
Available Data Interpretation Methods
Table 3.10 and Table 3.11 summarize the status of survey cov-
Fully developed methods of data interpretation are not avail- erage for each type of device as of February 2010. For the
able specifically for either device. In principle, the RWD RWD, most of this testing was conducted on flexible pave-
deflection should be usable as an input for any procedure that ments with a total survey length of more than 12,100 km
requires only a maximum deflection response as its pavement (7,500 mi). For the TSD, close to 100% is of flexible construc-
response input. It has been proposed that the velocity mea- tion with a total surveyed length of more than 21,000 km
surements from the Danish device configuration can be used (13,000 mi) in the United Kingdom.
to produce surface curvature index values that are akin to The Danish device has covered well over 10,000 km
those measured by an FWD. Therefore, they should be viable (6,500 mi) in Denmark. In 2010 the device was commissioned
as input for procedures that require only surface curvature to cover 20,000 km (12,500 mi) of the road networks in two
index (SCI) as a pavement response input. Australian states, as described by Baltzer et al. (2010). To date,
Greenwood Engineering has developed a method to inter- little data have been explicitly used for specific pavement
pret the deflection velocity from the TSD by using a beam on management activities, so it is not possible to determine the
elastic foundation approach. The model builds a full deflection appropriate use of data such as screening structurally defi-
basin using a two-parameter function and information from cient sections, defining rehabilitation strategies, or designing
three deflection slope measurements. The model proposed for rehabilitation treatments.
the deflection basin is given below (Krarup et al., 2006):

−A Phase I Assessment
d(x) = (Cos ( Bx ) + Sin( Bx )) e − Bx (3.1)
2B The devices were further evaluated to determine their capa-
bilities based on existing data found in the literature review
Where: and obtained from interviews with DOT officials. Both can-
d(x) = deflection at any point within the basin, didate devices have been used in pilot projects over multiple
x = distance of deflection from center of load, and locations, and evaluation of accuracy and repeatability has
A and B = constants to be optimized. been conducted and reported. This section presents the past
The implications of the model are that the deflection slope research conducted on the devices.
directly under the load is zero. This can be seen directly by
differentiating the model with respect to the variable x, and Accuracy
evaluating it at zero.
In the United Kingdom, the main method for interpreting Equipment accuracy has many interpretations, whether
pavement deflection response uses the maximum deflection considering individual measurement accuracy or the over-
measured by a slow-moving deflectograph to estimate resid- all accuracy of the device. Therefore, accuracy is considered
ual lives and strengthening requirements. Research in the under a number of factors: choice of averaging length, short-
United Kingdom, reported earlier in this section, has shown term repeatability, long-term repeatability, effect of external
that equivalent deflectograph values can be estimated from variables, comparability, and comparison with other deflec-
TSD measurements, thus providing an approximate interpre- tion measures. The term “short-term repeatability” indicates
tation methodology for the English strategic road network. that the surveys have been repeated as quickly as possible in
Routine network surveys of this network started in Novem- order to minimize the effect of external environmental con-
ber 2009, and the measurements are being converted to one ditions such as temperature changes on the results. When
of four structural condition categories before being stored in assessing long-term repeatability, the surveys were carried
the Highways Agency Pavement Management System for use out over a period of several days or even weeks, so the results
by the agents responsible for the various parts of the network could potentially include the external effects. For each factor,
to assist them with their management of the network. the capability of the two devices is considered on the basis of
The above information refers to just flexible pavements. available information collected in Phase I of this study and
For rigid and unpaved roads, there is as yet no explicit is detailed in this section. This section presents preliminary

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

30

Table 3.10. Summary of Tests Conducted in the U.S. as of February 2010

FWD
FWD Data Sampling Road Functional
Location Date Lane-mi Availability Frequency Repeat Runs Class

Louisiana 2009 NA Good NA No NA

Kansas 2008 466 Good 0.1 No U.S. and state
New Mexico September 2008 443 Good 0.1 No U.S.
Colorado October 2008 230 Partial 0.1 No Int., U.S., and state
New Hampshire July 2007 712 NA NA No Int., U.S., and state
Connecticut September 2007 204 NA NA No Int., U.S., and state
Kansas July–August 2006 506 Good 0.1 Research sites U.S. and state
Iowa July 2006 278 Good 0.1 No Int., U.S., and state
Oregon June–July 2006 579 Partial 0.1 No Int., U.S., and state
California June–July 2006 685 NA NA Research sites Int., U.S., and state
Virginia October 2005 488 Partial 0.1 3 interstate, Int., U.S., and state
2 primary
New Jersey October 2005 803 Partial Varied No Int., U.S., and state
Minnesota September 2005 NA a Partial 0.1 MnRoad sites U.S., state, and
county
Kentucky–Ohio–West Virginia September 2005 437 Good in OH 0.1 No Int., U.S., and state
Indiana September 2004 688 NA NA Yes U.S. and state
Natchez Trace November 2004 800+ NA NA No U.S. park service
NCAT July 2005 NA b
NA NA Yes Test track
Texas July 2003 264 Good NA c
Yes 38 test sections;
U.S. and state
routes
Total 7,583+
Note: NA = not available; Int. = Interstate.
a Testing on county roads and MnRoad facility; mileage not recorded.

b Testing at varying speeds on 1.3-mi test track; mileage not recorded.

c FWD, MDD, and RDD testing on specific spots (see FHWA, 2009).

values based on a limited dataset collected for previous stud-

ies; a more rigorous calculation based on a larger dataset is
Table 3.11. Lane Lengths Surveyed by U.K. HA given later.
TSD by Survey Type (September 2005 to
February 2010)
Choice of Averaging Length
Approximate
Length The RWD demonstration at the Eastern Federal Lands
Surveyed Highway Division (EFLHD) included a comparison using
Survey Type km mi
160- and 32-m (0.1- and 0.02-mi) intervals for averaging the
results. Figure 3.13 illustrates the effect of a shorter averaging
European continent 350 220
interval on one of the tested sections. The figure suggests that
TRL track 1,300 800 decreasing the sample unit length does not significantly affect
Local roads 10,400 6,500 the overall trend (or the mean deflection for the overall sec-
Scottish road network 800 500 tion), but it does increase variability over the section length.
On the basis of these limited results, the manufacturer has
English trunk road and motorway network 8,200 5,100
cautioned against using a sampling interval that is too small
Total 21,050 13,100
to reduce random error sufficiently (ARA, 2005b).

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

31  

Source: ARA, 2005b.

Figure 3.13. Effect of sample unit length on RWD deflections.

The TSD collects raw data at around 1000 Hz, but there of the true deflection profile are probably suppressed as the
is significant random noise in this raw signal. Even when averaging length increases from 1 m to 100 m (3.3 ft to 333 ft).
averaged over a 0.1-m (4-in.) length, this noise is noticeable, Therefore, in the United Kingdom, it has been decided to store
as illustrated by the black line in Figure 3.14. Also shown in results at 1-m (3.3-ft) intervals and generally report results
this figure are 1-m (40-in.), 10-m (33-ft), and 100-m (330-ft) as 10-m (33-ft) averages. From chainage (distance) 215 m
contiguous averages. This site is generally of a very variable to 250 m (705 ft to 820 ft) the construction changes to a rigid
and weak composite construction with corresponding very concrete construction, which has a relatively low and uniform
variable deflections. The figure illustrates how some features deflection response. This is demonstrated in Figure 3.15, which

Slope Vs. Chainage, TRL Large Loop (Long Straight) 23-Jun-09, P100
2.5

2
Deflection Slope [mm/m]

1.5

0.5

0
0 50 100 150 200 250 300
Chainage [m]

0.1m 1m 10m 100m

Figure 3.14. TSD deflection slope profile for Transport Research Laboratory track with
various averaging lengths.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

32

Slope Vs. Chainage, TRL Large Loop (Long Straight) 23-Jun-09, P100
2.5

Deflection Slope [mm/m] 2

1.5

0.5

0
200 205 210 215 220 225 230 235 240 245 250
Chainage [m]

0.1m 1m 10m 100m

Figure 3.15. TSD deflection slope profile for TRL track with various averaging
lengths: rigid section and transition.

shows a 50-m (165-ft) section of Figure 3.14 covering from the majority of cells, with standard deviations typically rang-
200 m to 250 m (655 ft to 705 ft). This exaggerated scale sug- ing from 50 to 75 microns (2 to 3 mils) (ARA, 2006).
gests that on weak composite pavement even a 10-m average In general, the various evaluations showed relatively good
length hides some true deflection variations. This is discussed repeatability that seemed to be appropriate for network-
further in the following repeatability section. level analysis. On the other hand, Diefenderfer (2010) con-
ducted statistical testing of RWD repeatability by use of a
non-paired t-test assuming equal variances and the results
Short-Term Repeatability
showed that the RWD data were repeatable for only 8 of
Several of the RWD demonstration projects included multiple 15 trials. Of the non-interstate test sections, 100% of the
runs. Figure 3.16 shows the results of conducting multiple runs trials were found to be repeatable. This raised some ques-
at the MnRoad test facility. RWD deflections are averaged over tions about the applicability of the system for detailed (e.g.,
15-m (50-ft) intervals. Figure 3.16a shows 10 repeat passes project-level) evaluations, especially in areas where low
on the inner lane of the low-volume road loop. This loop deflection ranges are expected. Figure 3.17 shows an exam-
included 11 asphalt concrete (AC) test sections with different ple of three repeated runs on a stretch of interstate highway
pavement structures. The sections included 4 cells (Nos. 27 in Virginia.
through 30) in very poor condition and one cell (No. 31) that The repeatability standard deviation for the average 0.1-mi
had been recently overlaid and was in excellent condition. segments is shown at the bottom of the chart; the average
These conditions were reflected in the deflection profile. The standard deviation was 20 microns (0.79 mils), or 17% of
repeatability standard deviations considering the individual the mean deflection. However, the repeatability standard
160-m (0.1-mi) segments ranged from about 25 microns deviation for the average values for the entire tested sections
(1 mil) for the section recently overlaid to approximately showed good repeatability (Table 3.12).
100 microns (4 mils) for the cells in poor condition. Fig- For the U.K. Highways Agency TSD, Ferne et al. (2009b)
ure 3.16b presents three repeated runs on the outer (driving) reported the results of testing conducted to investigate the
lane of the mainline experiment, which included AC test cells effect of testing speed. Measurements were taken on the TRL
of variable ages and AC layer thicknesses ranging from 100 to track over a range of speeds, and the results showed that as
380 mm (6 to 15 in.). Deflections were very uniform within the speed increased, a slightly lower value of deflection slope

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

33  

(a) Low-volume road (10 passes)

(b) Mainline experiment (3 passes)

Source: ARA, 2006.

Figure 3.16. RWD deflections at 50-ft intervals at MnRoad.

was recorded. This being the case, the testing speeds used Both the LCPC assessment of the first DRI prototype and
during further tests were strictly controlled to enable repeat- TRL’s assessment of the HA TSD suggest that the level of
able results to be obtained. Figure 3.18 shows a sample of repeatability is not particularly dependent on the mean level
the results of 6 runs on a 440-m (0.25-mi) length of the TRL of the slope. Therefore, in this section of the report they are
track, which had mainly a composite pavement but included given in absolute, not proportional, terms.
a 50-m (165-ft) length of jointed concrete at a nominal speed The consistency of the latest version of the HA TSD has
of 70 km/h (45 mph). The data showed reasonable short- been assessed on a small number of U.K. roads. Results of
term repeatability, with a relatively low standard deviation these tests in terms of the standard deviation of the mean
despite the relatively wide range of deflection slopes measured values of each of five runs of various lengths have been sum-
(i.e., changing by a factor of over seven). marized in Table 3.13 for the P100 and P300 TSD sensors.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

34

9
Correlation Matrix Pass 1 (P1)
8 P1 P2 P3 Pass 2 (P2)
P1 1.00 0.69 0.63
P2 0.69 1.00 0.65 Pass 3 (P3)
7 P3 0.63 0.65 1.00 Std Deviation (0.1 mi)

6
Deflection (mils)

0
0 2 4 6 8 10 12 14 16 18 20
Distance (miles)

Figure 3.17. Repeated runs of the RWD on I-64 in Virginia.

This good level of short-term repeatability of the TSD that changes in pavement temperature have only a small
that is achievable under controlled conditions can also be effect on the measured slope as surface temperature changed
observed graphically. Figure 3.19 shows a 20-m (66-ft) sam- from 4°C to 19°C (40°F to 66°F) during these surveys. This
ple length of the TRL track with the TSD P100 sensor results is not unexpected as the pavement is of flexible composite
calculated at 1-m (3-ft) intervals plotted against distance for construction. In the United Kingdom, deflection surveys on
all five repeat runs. The repeated identification of weak spots at strong flexible composite pavements are left uncorrected for
the same location (i.e., stations 187 to 188 m [613 to 617 ft] and pavement temperature.
197 to 198 m [646 to 650 ft]) is clearly seen. Figure 3.21 shows a 500-m selected section of the same
U.K. Site B as in Figure 3.20 but with 10-m (33-ft) averaging
used. Although the runs were performed over 5 months, all
Long-Term Repeatability
five surveys identify the weaker section in the same location,
Figure 3.20 shows a sample of five runs recorded over 5 months that is, from 2,350 m to 2,400 m (7,710 ft to 7,874 ft).
(September 2009 to February 2010) on 4 km (2.5 mi) of a U.K.
site, which is of flexible composite construction, with a nomi-
Comparability
nal testing speed of 70 km/h (45 mph). The data shown has
been averaged into 100-m (330-ft) lengths so that the change The comparability of the RWD cannot be assessed because
in deflection slope is more visible. only one such device has been produced. Even with two
Table 3.14 shows that the standard deviations of the mean devices, an assessment of true device reproducibility, such as
values of each of the five runs are very similar to those in with the TSD, is not possible. However, some limited com-
Table 3.13, meaning that repeatability apparently changed parisons have been made but not published. One such com-
little when assessed over longer periods of time. This suggests parison was made in September 2008 in the United Kingdom.

Table 3.12. Summary Statistics for Average Section RWD

Deflections of I-64 and I-81

Repeatability Repeatability
Average Std. Dev. Average Std. Dev.
Highway (mils) (mils) (microns) (microns)

Eastbound I-64 4.53 0.28 115.1 7.3

Westbound I-64 4.7 0.26 119.4 6.6
Northbound I-81 7.77 0.14 197.4 3.5
Southbound I-81 5.08 0.59 129.0 15.0

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

35  

1.75

Concrete
1.5

Std Deviation (10m)

1.25
Slope [mm/m]

0.75

0.5

0.25

0
0 50 100 150 200 250 300 350 400 450
Chainage [m]
Source: Ferne et al., 2009a.

Figure 3.18. Repeatability of deflection slope at 70 km/h (45 mph) on the TRL track.

Figure 3.22 illustrates the consistency between these two deflectograph that records peak deflections in both wheel-
devices when operating on the same 11-km (7-mi) length paths at the same time. Comparison of the two figures con-
of varying construction and deflection response, bearing in firms that any differences between the two devices are likely
mind that the two devices measure in different wheelpaths. explained by the different deflection responses of the two
Figure 3.23 illustrates differences between the wheel- wheelpaths.
paths as revealed by surveys conducted by a slow-speed
Comparison with Other Deflection Measures
Table 3.13. Repeatability Standard Deviation
of TSD for Five Runs in Terms of TSD Slope Investigations have been conducted comparing the RWD and
for Short-Term Repeatability TSD to other deflection measuring equipment, in particular
the FWD. However, since the FWD and rolling wheel devices
Repeatability Standard load the pavement in different ways, the relationship between
Deviation (mm/m)
them will not necessarily be one of equality.
Averaging Length Several of the RWD demonstrations included FWD mea-
surements on at least some sections; however, not all were
Overall 10 m 100 m 160.9 m
Site Length Sensor (33 ft) (330 ft) (1/10 mi) conducted at the same time that the RWD measurements
were obtained. Figure 3.24 presents examples of section-
TRL track 291 m/ P100 0.071 0.046 0.040
0.2 mi
level comparisons between RWD and FWD maximum
P300 0.053 0.038 0.034 deflections. In general, the RWD reports collected during
U.K. Site A 1,080 m/ P100 0.037 0.012 0.010 follow-up interviews suggest that the average results of the
0.7 mi RWD deflection measurements (normalized to a standard
P300 0.037 0.013 0.011
temperature) correlate relatively well with the average max-
U.K. Site B 3,871 m/ P100 0.054 0.025 0.023
2.4 mi imum FWD deflection when aggregated by homogeneous
P300 0.071 0.052 0.051
sections. The example from Texas (Figure 3.24d) suggests

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

36

Slope Vs. Chainage, TRL Large Loop (Long Straight) 23-Jun-09, P100 (1m data)
3.000

2.500

2.000
Deflection Slope [mm/m]

1.500

1.000

0.500

0.000
180 182 184 186 188 190 192 194 196 198 200
Chainage [m]

T120090623001 T120090623002 T120090623003 T120090623007 T120090623008

Figure 3.19. Selected 20-m (66-ft) length of TSD slope data as 1 m (3.3 ft) means on TRL track.

Slope Vs. Chainage, A33 SB, P300 100m Averages

0.700

0.600

0.500
Deflection Slope [mm/m]

0.400

0.300

0.200

0.100

0.000
0 500 1000 1500 2000 2500 3000 3500 4000
Chainage [m]

10-Feb-10 09-Nov-09 06-Nov-09 09-Oct-09 18-Sep-09

Figure 3.20. Long-term repeatability of deflection slope (P100) at 70 km/h (45 mph) on
U.K. Site B, 100-m (330-ft or 0.06-mi) averages.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

37  

Table 3.14. Repeatability Standard Deviation of first Danish Research Institute (DRI) prototype showed a
TSD for Five Runs in Terms of TSD Slope for strong correlation (R2 = 0.86) between the slope measured by
Long-Term Repeatability the DRI TSD and the peak central deflection measured by an
FWD over a range of sites in France.
Repeatability Standard
Deviation of TSD Slope
Comparisons in the United Kingdom between the HA
(mm/m) TSD and FWD measurements have been less common to date
because the main emphasis has been on comparison with
Averaging Length
the deflectograph, the prime deflection measuring device
Overall 10 m 100 m 160.9 m used in the United Kingdom. However, some comparisons
Site Length Sensor (33 ft) (330 ft) (0.1 mi) of deflection profiles on specific sites have been made. For
U.K. 3871 m P100 0.065 0.040 0.039 example, Figure 3.25 shows a comparison between an FWD
Site B (2.4 mi)
P300 0.063 0.038 0.038
central deflection profile at 2-m (6.6-ft) intervals compared
with a TSD deflection slope profile averaged over the same
intervals on a 400-m (1,300-ft) section of the TRL track.
The pavement structure includes both weak flexible compos-
that the correlation is better on sections with high deflec- ite materials and rigid concrete. Similarities in the shapes of
tions (having “weak” structural capacity). This is expected the two profiles are very encouraging despite the 4-year inter-
because a wider range of the dependent variable increases val between the surveys. It should be noted, however, that the
the correlation coefficient, as is discussed later in this report. vertical scales of the two parameters are relatively arbitrary
Additional analysis was performed on the New Mexico and have been adjusted to approximately align the two pro-
data that were provided for this project and is presented in files vertically.
the comparability section of the report. The data were col- In the United Kingdom, extensive comparisons have been
lected on U.S. Route 550 in New Mexico and were provided made between the TSD slope and peak deflection measured
by ARA for this project. by a U.K. deflectograph. Figure 3.26 illustrates the average
Many comparisons have been made between the TSD and relationship, together with 95% confidence limits, between
other deflection measuring devices. The early independent deflectograph (DFG) values and TSD slope values for the
evaluation by the LCPC in 2003 (Simonin et al., 2005) of the P300 sensor, which is located 300 mm (1 ft) from the center

Slope Vs. Chainage, A33 SB, P100 10m Averages

1.000

0.900

0.800
Deflection Slope [mm/m]

0.700

0.600

0.500

0.400

0.300

0.200

0.100

0.000
2200 2250 2300 2350 2400 2450 2500 2550 2600 2650 2700
Chainage [m]

10-Feb-10 09-Nov-09 06-Nov-09 09-Oct-09 18-Sep-09

Figure 3.21. TSD slope data, as 10 m (33 ft) means, for five repeat runs over a 5-month
period on U.K. Site B.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

38

Figure 3.22. Comparison of Highways Agency TSD and

Danish Research Institute TSD on a major U.K. road (70-km/h
[45-mph], 50-m [165-ft] averages).

of the load. The analysis covered almost 5,000 10-m (3.3-ft) standard deviation, and in some cases subjective visual inspec-
segments on a wide range of U.K. roads. tion of plots. Most studies also suggested that data averaging
length affected repeatability. For example, ARA recommends
that RWD results be averaged over 160 m (0.1 mi) but, in the
Phase II Assessment
United Kingdom, TSD test results are stored at 1-m (3.3-ft)
The data obtained in Phase I and collected in the field trials were averages and reported at 10-m (33-ft) averages. This section
analyzed to evaluate the repeatability of the TSD and RWD first discusses, evaluates, and highlights some of the draw-
and the comparability of both devices by comparison with backs associated with the use of correlation and regression
the FWD. In the Detailed Description of the Selected Equip­ analysis to evaluate repeatability and comparability. Then,
ment section, results of different studies that evaluated RWD repeatability and comparability analysis based on the limits
(conducted in the United States) and TSD (conducted in of agreement (LOA) method suggested by Bland and Altman
Europe) were analyzed. It was noted in this analysis that (1986) is recommended and used to evaluate the continuous
repeatability and comparability were not uniformly defined deflection devices. A method of evaluating repeatability from
across all those studies. Measures and methodologies used to one run is also presented and compared to the method based
evaluate the devices included correlation, regression analysis, on the LOA. Finally, the use of smoothing splines as a tool to

Figure 3.23. Deflectograph data for a major U.K. road at 50-m

(165-ft) averages (nearside indicates outside wheelpath,
offside indicates inside wheelpath).

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

39  

RWD Deflection (mils)

RWD Deflection (mils)

FWD Deflection (mils) FWD Deflection (mils)

(a) Kansas (b) Minnesota

RWD Deflection (mils)

RWD Deflection (mils)

FWD Deflection (mils) FWD Deflection (mils)

(c) Iowa (d) Texas
Source: FHWA, 2009.

Figure 3.24. Examples of RWD versus FWD comparisons.

remove the noise from TSD deflection slope measurements follows closely the one presented by Bland and Altman (2003).
is investigated. This smoothing splines denoising methodol- The reason it is included in this report is because the use of cor-
ogy shows potential to improve the frequency at which useful relation and regression is so pervasive in the pavement field that
information can be obtained (i.e., data averaging distance). their shortcomings (as will be illustrated) are often ignored.
In this report, repeatability (comparability) is defined as the The example supposes the true value of any measurement at
95% confidence interval of the difference between repeated 600 different locations (for example, pavement deflection) is
measurements (difference between measurements of TSD and known to be a sinusoidal wave varying between a minimum of
FWD or RWD and FWD). Correlation, cross-correlation, and 4 and a maximum of 6 units (Figure 3.27). Repeated measure-
regression are widely used to evaluate repeatability and com- ments, m1 and m2, are obtained using an instrument that is
parability in many pavement engineering applications such as known to produce measurements that are contaminated with
profile or friction measurements. Gaussian (from a normal distribution) noise with mean zero
and standard deviation of 0.5 units (Figure 3.28). Since the
relationship between m1 and m2 is known to be m1 = 1.0m2 + 0.0
Regression Analysis
(i.e., the line of equality), it is desirable that an appropriate sta-
For regression, the following example uses computer-generated tistical analysis can suggest with some confidence this one-to-
data that simulate repeated measurements. Because the correct one relationship.
answer is known, it illustrates how regression analysis can lead Figure 3.29 shows m2 versus m1 with the true relationship
to wrong conclusions. This argument about regression analysis (line of equality) and the regression line. The slope of the

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

40

TRL Track Long Straight.

FWD 2m spacings (17-Oct-2005), TSD P100 2m averaging (23-Jun-2009)
1000 2.5

900

800 2.0

700

TSD Deflection Slope [mm/m]

FWD Deflection [μm]

600 1.5

500

400 1.0

300

200 0.5

100

0 0.0
0 50 100 150 200 250 300 350 400
Chainage [m]
FWD D1 TSD P100

Figure 3.25. Comparison of HA TSD slope and FWD central deflection profiles on flexible composite pavement.

800

700

600

500
DFG [µm]

400

300

200

100

0
0 0.2 0.4 0.6 0.8 1 1.2
P300 [mm/m]

Figure 3.26. Deflectograph versus P300 with 95% confidence intervals.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

41  

5.5

True Value
5

4.5

4
0 10 20 30 40 50 60
Location

Figure 3.27. Sinusoidal function example for regression analysis.

8
m1
7 m2

6
Measurement

2
0 10 20 30 40 50 60
Location

Figure 3.28. Sinusoidal signal with added Gaussian noise.

8
Measurements
7 Orthogonal Regression
Line of Equality
Regression m2 vs m1
6
Regression m1 vs m2

5
m2

2
2 3 4 5 6 7 8
m1

Figure 3.29. Limitation of linear regression when errors are present in

the regressor. The relationship between m1 and m2 should be the line of
equality.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

42

regression line is 0.69, which is different from 1.0, and the Correlation measures have been extensively used to eval-
intercept is 1.55, which is different from zero. The 95% uate repeatability or “accuracy”—with respect to FWD—
(a = 0.05) confidence interval on the slope is [0.63; 0.75], while of measures of continuous deflection data. This use of
the 95% confidence interval on the intercept is [1.08; 2.03]. correlation is also prevalent in the analysis of pavement
Both slope and intercept are statistically different from 1.0 profile and friction data. However, the use of correlation
and 0, respectively, suggesting that the relationship between can be very misleading, as discussed by Bland and Altman
the two measurements does not follow the line of equality (1986, 2003). Correlation does not give agreement between
(which we know it does). Therefore, ordinary regression repeated measures. For example, two measures that vary
analysis is leading to the wrong conclusion. The cause for the exactly by any factor give a correlation of 1 (or -1, if the
failure of regression analysis in this case is the presence of factor is negative). A measuring device that gives repeated
error in the regressors (errors in m1), which violates the con- measurements that can vary by some factor is not one
ditions of the Gauss-Markov theorem. This violation leads to that is described as repeatable. Another drawback of
the least-squares regression coefficients to be biased (Myers, correlation is that it depends on the range of the true mea-
1990). The irony of this is that the more measurements that surement; the wider the range, the greater the correlation.
are obtained, the narrower the confidence interval on the In the extreme case, a pavement that is perfectly homo-
biased slope, which strengthens the wrong conclusion that geneous (i.e., strength is constant) will practically result
the relationship between the two measurements does not fol- in a zero correlation no matter how repeatable the device
low the equality line. is. This is because the calculated correlation in this case
The alternative to linear regression when errors are pres- is that of the error terms, which are randomly uncorre-
ent in both variables is the total least-squares regression (Van lated. Correlation should therefore be used with caution
Huffel and Wandewalle, 1991). When errors in both vari- when evaluating repeatability. This is not to say that cor-
ables have the same variance, total least-squares regression relation should never be used. For example, the proposed
is equivalent to orthogonal regression. The difference is that method of taking differences is not applicable when com-
while ordinary least squares minimizes the squared distance paring devices that measure two different physical quanti-
from the dependent variable to the fitted function, orthogo- ties (such as TSD and FWD). In this case, unless the two
nal regression minimizes the square of the perpendicular dis- measurements can be converted to the same quantity, cor-
tance to the fitted function. The orthogonal regression for m1 relation (or for that matter, linear regression) might be a
and m2 is presented in Figure 3.29. The slope of the orthogo- better choice.
nal regression line is 0.96, which is very close to 1.0 (the 95% The average correlations between the different repeated
confidence interval is [0.92; 1.00]). More information on this TSD measurements obtained in this study for each section
procedure can be found in Leng et al. (2007). are presented in Figure 3.30. The correlations are not the
Another way to look at this example is using the relation- same for the different sections. Interpreting the correla-
ship m2 = 0.69m1 + 1.6, calculate m1 = 1.45m2 - 2.32. Since tion as a measure of repeatability would give significantly
there is no specific reason to do the regression with m1 as the different repeatability results depending on the tested sec-
x-variable, it could be done with m2 as the x-variable. In this tion. As seen in Figure 3.30, for an averaging distance of
case, the relationship m1 = 0.66m2 + 1.7 is obtained, which is 1 m, the correlation varies from under 0.10 to almost 0.90.
different from m1 = 1.45m2 - 2.32. The two regressions, using Which correlation value in this range gives the repeatability
m1 or m2 as the x-variable, are presented in Figure 3.29. The of the device? The tested sections had a significant effect
relationship between m1 and m2 is not the same in each case. on the correlation. As expected, sections with low correla-
There is no reason to favor the use of m1 as the x-variable tions are those that had low variation in the measured slope,
to the alternative of using m2 as the x-variable. This clearly and sections with high correlations are those that had high
illustrates the inadequacies of linear regression to evaluate variation in the measured slope. For example, section F1
the repeatability of a given device. resulted in a significantly higher correlation than did all the
other sections; especially for sensor 100 and 1-m averaging
length. It can be concluded that correlation is a good indi-
Data Analysis Using Correlation
cator of the variability in the pavement section rather than
The drawbacks of using correlation are similar to the draw- in the repeatability of the device. This sentiment was some-
backs of regression analysis (although they are not com- what echoed, in more technical terms, by Bland and Alt-
pletely the same). Here, instead of using artificial data, the man (2003): “[T]he correlation coefficient is a measure of
actual repeated TSD slope measurements obtained on differ- the information content of the measurement.” This clearly
ent pavement sections are used to illustrate the drawbacks of shows how correlation can lead to false conclusions when
correlation. evaluating a device.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

43  

Figure 3.30. Correlation coefficient versus averaging length for three TSD
sensors.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

44

6 (a) 6 (c)

Slope (mm/m)

Slope (mm/m)
4 4

2 2

0 0

0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000
Distance (m) Distance (m)

6 (b) 6 (d)
Slope (mm/m)

Slope (mm/m)
4 4

2 2

0 0

0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000
Distance (m) Distance (m)

Figure 3.31. Measured deflection slope on two different flexible pavement sections: (a) first
run on Section UK_F5, (b) second run on Section UK_F5, (c) first run on Section UK_F1, and
(d) second run on Section UK_F1.

Figure 3.31 shows two different flexible pavement sections, and is therefore much less affected by the tested pavement
one with high variation in the deflection slope (UK_F1) and section.
another with low and uniform deflection slope (UK_F5). The
correlation between repeated runs is significantly different
for each section; UK_F1 had a high correlation (close to 0.9), Repeatability
while UK_F5 had a much lower correlation (less than 0.5). The definition of repeatability given by the British Standard
However, the measurements’ noise levels are comparable, as Institution (1979) was adopted in this report. It is defined as
can be observed from visual inspection of the plots. “the value below which the difference between two single test
Another observation is that the correlation varies with the results . . . may be expected to lie with a specified probability.”
distance between the sensor and the applied load. This results The specified probability was set at 95%, and repeatability was
from the fact that sensors closer to the loaded area measure calculated using the procedure suggested in a series of papers
higher slopes, which increases the correlation. As a sum- by Bland and Altman (Altman and Bland, 1983; Bland and
mary, correlation depends on the tested pavement (or range Altman, 1986, 2003, 2007). The main idea is to estimate the
of measurements), the instrument location (again, partly standard deviation of the difference between repeated mea-
caused by different range of measurements), and averaging surements from the same device (repeatability) or difference
length. In many cases, these factors have a much more signifi- between measurements from two different devices (compa-
cant effect on the correlation than does the effect of errors in rability) and construct the 95% confidence interval using
the measurements. 1.96sd, where sd is the standard deviation of the difference.
However, a device repeatability measure should be, as In the case of comparability, this 95% confidence interval is
much as possible, independent of the tested pavement. Not referred to as the limits of agreement (LOA) between the two
having this independence can lead to significantly different devices. In their procedure, Bland and Altman also specified
opinions about the suitability of the device. For example, calculating the bias between two different devices, while for
somebody evaluating the device on the F1 section would be the same device they incorporated this bias in the repeatabil-
very pleased with the performance based on the correlation ity measure. This report shows the results of incorporating or
and somebody evaluating the device on the C2 section would not incorporating the bias in the repeatability measure. For
be very disappointed in the device. The repeatability measure all practical purposes, the two methods resulted in the same
adopted in this study, in contrast, gives comparable results repeatability because the bias was negligible. For repeatability

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

45  

Scatter of Standard Deviation; Site F3 - 1m Averaging - Sensor 100

0.5
0.45
0.4

Standard Deviation
0.35
0.3
0.25
R² = 2E-05
0.2
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Mean Reading
Figure 3.32. Scatter of standard deviation versus mean for low average
readings.

measures between the FWD and TSD or the FWD and RWD, of two repeated measures, Altman and Bland (1983) suggested
the bias was not incorporated. plotting x1 - x2against (x1 + x2)/2. For the case where three
or more repeated measures were obtained, the plot shows
the standard deviation of the measurements at each loca-
TSD Repeatability
tion against the average of the measurements at each loca-
The analysis of TSD repeatability was performed for mea- tion (both plots give the same qualitative view). Figure 3.32
surements averaged over 1-m, 10-m, and 100-m distances. shows the standard deviation as a function of average slope
Five runs were obtained for each test section (except for Sec- for a relatively strong flexible pavement test section labeled
tion F1, which had four runs, and for Sections F3 and R2, F3 for slope measurements averaged over a 1-m interval. The
each which had three runs) resulting in five sets of slope mea- figure suggests the standard deviation (and therefore equip-
sures at each location. ment repeatability) is independent of the measured slope in
For repeatability (or comparability) analysis, it is impor- the range of measurements.
tant to check whether measurement repeatability depends on The observation was consistent for all other tested sections
the actual measurement level (in other words, the measure- except for the flexible section labeled F1 shown in Figure 3.33,
ment error depends on the actual measurement). In the case and the rigid section labeled R2. Both sections included

Scatter of Standard Deviation; Site F1 - 1m Averaging - Sensor 100

1.6
1.4
R² = 0.3572
1.2
Standard Deviation

1
0.8
0.6
0.4
0.2
0
-0.2 0.8 1.8 2.8 3.8 4.8 5.8 6.8 7.8
Mean
Figure 3.33. Scatter of standard deviation versus mean for high average
readings.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

46

relatively very weak spots, with significant deterioration. In important to note that this source of variability is not caused
Figure 3.33, the standard deviation is dependent on the asso- by the device.
ciated measured slope. A possible explanation for this depen- Repeatability is closely related to the estimation of error
dence could be that there are two main sources of variation in measurements obtained from a device. This error is com-
in the slope measurement. The first source is due to the error posed of variance and bias. Therefore, repeatability can be
from the sensors, vehicle dynamics, and any other factors that defined by either the variance or the error (variance and
can affect the TSD (temperature, moisture, etc.). This source bias). The advantage of defining repeatability in terms of
of variability is expected in most cases to be independent of variance and keeping bias as a separate measure is that causes
the associated measurement. The second source of variabil- of bias can often be identified and corrected. For example, in
ity is due to the spatial variability in the pavement strength. measurements on flexible pavement, bias can be the result of
Weaker pavements tend to have more distress factors, such as a temperature difference between repeated tests. More com-
cracking, which result in a greater spatial variability on the monly, bias is caused by different operational characteris-
strength and, therefore, the deflection measurements. It is tics or the equipment getting out of calibration. Table 3.15

Table 3.15. Consistency of TSD Measurements over Repeated Runs:

Incorporating Bias in Standard Deviation

Sensor 100 Sensor 300 Sensor 756

Data Average Average Average

Road Averaging Reading Std. Dev. Reading Std. Dev. Reading Std. Dev.
Section (mm/m) (mm/m) (mm/m) cov (mm/m) (mm/m) cov (mm/m) (mm/m) cov

F1 1 1.0468 0.2661 0.2542 0.6198 0.1918 0.3095 0.3169 0.144 0.4544

10 0.1054 0.1007 0.0717 0.1157 0.0383 0.1209
100 0.0625 0.0597 0.038 0.0613 0.0146 0.0461
F3 1 0.3226 0.1305 0.4045 0.2202 0.1376 0.6249 0.2279 0.1281 0.5621
10 0.0407 0.1262 0.0415 0.1885 0.0412 0.1808
100 0.0155 0.048 0.0139 0.0631 0.0137 0.0601
F5 1 0.4356 0.1166 0.2677 0.3691 0.1123 0.3043 0.3767 0.1096 0.2909
10 0.0401 0.0921 0.0383 0.1038 0.0367 0.0974
100 0.0147 0.0337 0.0128 0.0347 0.0111 0.0295
F6 1 0.4176 0.1254 0.3003 0.3252 0.121 0.3721 0.4316 0.1169 0.2709
10 0.0406 0.0972 0.0382 0.1175 0.0373 0.0864
100 0.0139 0.0333 0.0124 0.0381 0.0119 0.0276
C1 1 0.4786 0.1466 0.3063 0.3779 0.1398 0.3699 0.4386 0.1355 0.3089
10 0.0519 0.1084 0.0489 0.1294 0.0498 0.1135
100 0.0306 0.0639 0.0292 0.0773 0.0276 0.0629
C2 1 0.3469 0.1222 0.3523 0.2811 0.1226 0.4361 0.2949 0.1146 0.3886
10 0.0408 0.1176 0.0398 0.1416 0.039 0.1322
100 0.0142 0.0409 0.0146 0.0519 0.0143 0.0485
R2 1 0.5859 0.2072 0.3537 0.3972 0.1671 0.4207 0.3511 0.1217 0.3466
10 0.0643 0.1098 0.0558 0.1405 0.0512 0.1457
100 na na na na na na
R3 1 0.343 0.1483 0.4322 0.2356 0.1378 0.5846 0.3441 0.1419 0.4122
10 0.0432 0.1258 0.0404 0.1716 0.0433 0.1257
100 0.0153 0.0446 0.0136 0.0579 0.0148 0.0431
Note: cov = coefficient of variation. F = flexible; C = composite; R = rigid (road types); and na = not applicable.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

47  

presents results with the bias incorporated in the standard constant. Therefore, the relative effect of the bias becomes
deviation calculation. Table 3.16 presents results with the more significant.
bias taken out of the standard deviation. This effect was observed in the section labeled C1. In this
A comparison of the results presented in both tables sug- case, incorporating the bias into the standard deviation of
gests that the effect of bias in the obtained measurements is measurements averaged over a 100-m (330-ft) distance
negligible. This is confirmed by the bias results presented in resulted in a standard deviation that is twice as large as the
Table 3.16, which shows the bias to be small compared to the standard deviation calculated by not incorporating the bias
standard deviation. However, the effect of the bias becomes (compare Table 3.15 and Table 3.16 for C1). Figure 3.34
more significant for measurements averaged over longer shows that one of the runs has a significant systematic bias
distances (for example, when averaging over 100 m [330 ft] and is shifted up compared to the others. Note that the effect
compared to 1 m [3.3 ft]). This is because averaging over of the bias is much less pronounced for averaging distances of
longer distances reduces the variance while keeping the bias 1 m (3.3 ft) and 10 m (33 ft) for which the standard deviation

Table 3.16. Repeatability of TSD Measurements over Repeated Runs:

Excluding Bias from Standard Deviation

Sensor 100 Sensor 300 Sensor 756

Std. Std. Std.

Road Averaging Dev. Bias Rep.a Dev. Bias Rep.a Dev. Bias Rep.a
Section Length (m) (mm/m) (mm/m) (mm/m) (mm/m) (mm/m) (mm/m) (mm/m) (mm/m) (mm/m)

F1 1 0.2649 0.0328 0.5192 0.1771 0.0169 0.3471 0.1333 0.0054 0.2613

10 0.1024 0.2007 0.0653 0.128 0.0353 0.0629
100 0.0583 0.1143 0.0335 0.0657 0.0257
F3 1 0.1304 0.0068 0.2556 0.1376 0.0051 0.2697 0.1280 0.0055 0.2509
10 0.0404 0.0792 0.0414 0.0811 0.0410 0.0804
100 0.0149 0.0292 0.0137 0.0269 0.0132 0.0259
F5 1 0.1165 0.0067 0.2283 0.1122 0.0056 0.2199 0.1096 0.0026 0.2148
10 0.0397 0.0778 0.0381 0.0647 0.0367 0.0719
100 0.0136 0.0267 0.0122 0.0239 0.0111 0.0218
F6 1 0.1254 0.0044 0.2283 0.1209 0.0035 0.2370 0.1169 0.0035 0.2291
10 0.0405 0.0794 0.0381 0.0747 0.0372 0.0729
100 0.0136 0.0267 0.0122 0.0239 0.0117 0.0229
R2 1 0.2069 0.0161 0.405 0.1666 0.0186 0.3265 0.1217 0.0033 0.2385
10 0.0634 0.1243 0.0541 0.1060 0.0513 0.1005
100 na na na na na na na na na
R3 1 0.1482 0.0080 0.2905 0.1377 0.0062 0.2699 0.1420 0.0064 0.2780
10 0.0427 0.0837 0.0403 0.0790 0.0431 0.0845
100 0.0142 0.0278 0.01333 0.0261 0.0145 0.0284
C1 1 0.1430 0.0334 0.2803 0.1375 0.0262 0.2695 0.1332 0.0267 0.2611
10 0.0446 0.0874 0.0421 0.0825 0.0432 0.0847
100 0.0153 0.0300 0.0155 0.0304 0.0123 0.0241
C2 1 0.1221 0.0070 0.2393 0.1224 0.0103 0.2399 0.1145 0.0061 0.2244
10 0.0404 0.0792 0.0389 0.0762 0.0386 0.0757
100 0.0129 0.0253 0.0121 0.0237 0.0132 0.0259
Note: F = flexible; C = composite; R = rigid (road types); and na = not applicable.
a Rep. = repeatability or 1.96 * std. dev.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

48

Slope Vs. Chainage, UK_C1 (REA-S), P100 100m Averages

Figure 3.34. Measured slope averaged over 100-m length for Site C1.

of C1 is similar to that of the other sections. In this analysis, between runs 1 and 2, the bias and variances can be calculated
of all tested sections, C1 was the only one that had a statisti- as follows:
cally significant bias.
The results presented in Table 3.15 and Table 3.16 were N
yi1 − yi 2
obtained using the method of analysis of variance (ANOVA). Bias = ∑ (3.3)
i =1 N
The procedure can be illustrated using two repeated mea-
surements. For the TSD, each slope measurement consists of
the actual slope and an error term. This can be expressed as
N
σ d2 = ∑
[ Bias − ( yi1 − yi 2 )]2 (3.4.a)
follows: i =1 N −1

yij = sij + eij (3.2) 1 N [ Bias − ( yi1 − yi 2 )]

2

σ2 = ∑
2 i =1 N −1
(3.4.b)
where
yij = TSD slope measurement at location i for run j,
sij = actual (unknown) slope at location i during run j, and where
eij = error in TSD slope measurement at location i for run j. N = total number of data points per run,
s2d = variance of the differences, and
The first two runs for F5 are shown in Figure 3.35a. Figure s2 = variance of the measurements.
3.35b shows the difference between the two runs. The mean
of this difference is an estimate of the bias between the two If both bias and variance are used to estimate the repeat-
runs, while the variance of the difference represents the sum ability (see Bland and Altman, 1986), the mean squared dif-
of the error variances for each run (Bland and Altman, 1986). ference (MSD) can be calculated as follows:
Assuming the variance for each run is the same, the mea-
1 N [( yi1 − yi 2 )]
2
surement variance can be estimated by dividing the variance MSD = ∑ (3.5)
of the difference by two. For example, taking the difference 2 i =1 N

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

49  

1.2
Run1
1 Run2
0.8
Slope (mm/m)
0.6

0.4

0.2
0
(a)
-0.2
0 500 1000 1500 2000 2500 3000 3500
0.6
Slope Difference (mm/m)

0.4

0.2

-0.2

-0.4
(b)
-0.6
0 500 1000 1500 2000 2500 3000 3500
Distance (m)

Figure 3.35. Comparison of P100 measurements of Runs 1 and 2 for Site F5 (a), and difference
between the two runs (b).

The factor ½ in the calculation of the variance (or MSD) permits the establishment of a baseline repeatability value.
reflects that the measurement variance is half the variance of The analysis of repeatability was performed using three RWD
the difference. The calculated variances (in the case of Equa- repeated measurements averaged over 160 m (0.1 mi) col-
tion 3.3), using the difference between the first run and each lected over a distance of 32 km (20 mi) on I-64 for both east-
of the remaining 4 runs for F5 for the 1-m (3.3-ft) averag- bound and westbound directions. Figure 3.36 shows the test
ing distance, were 0.0141, 0.0132, 0.0130, and 0.0135 mm/m. results for the eastbound direction. The difference between
Levene’s test of equal variance with a = 0.05 showed that the the first and second runs is shown in Figure 3.37.
variances are equal. Because the calculated variances using the A test of normality showed that the differences between
difference between different pairs of runs are equal, the TSD the two measurements are normally distributed. The cal-
error variance can be estimated as the average of the calculated culated standard deviations of the difference (sd) between
variances. The TSD error standard deviation can then be calcu- Runs 1 and 2, Runs 1 and 3, and Runs 2 and 3 are 24.9, 27.3,
lated as the square root of the variance. The resulting standard and 27.2 µm (0.98, 1.07, and 1.07 mils), respectively, for the
deviation calculated for F5 and 1 m (3.3 ft) TSD measurement eastbound direction, and 26.1, 24.3, and 24.8 µm (1.03, 0.96,
averaging was 0.1165 mm/m. This procedure is essentially and 0.98 mils), respectively, for the westbound direction. Dif-
implemented in MATLAB in the function called “anova2.” ferences between the standard deviations were found not to
Calculation of the repeatability coefficient was performed be statistically significant at the 0.05 significance level using
using the 1.96sd estimate of confidence interval. This esti- Levene’s test of equal variances. Therefore, the error defined
mate was used after the error distribution was found to fol- as the standard deviation of RWD measurements (s) was cal-
low a normal distribution (using the Anderson–Darling test culated as 18.2 µm (0.72 mils). This standard deviation was
for normality). The repeatability calculated for all sections is computed by dividing sd by the square root of 2.
presented in Table 3.16. The bias between the runs is defined as the mean of the
difference. For the eastbound direction, the biases between
Runs 1 and 2, Runs 1 and 3, and Runs 2 and 3 are 3.93, 18.4,
RWD Repeatability Analysis
and 14.5 µm (0.15, 0.72, and 0.57 mils), respectively. For the
A similar analysis (which separates the bias and variance) was westbound direction, the biases between Runs 1 and 2, Runs 1
conducted for the RWD using the data obtained in Phase I for and 3, and Runs 2 and 3 are 14.2, 6.5, and -7.7 µm (0.56, 0.26,
Virginia. Although the device has been improved, this analysis and -0.30 mils), respectively. All these biases were found to

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

50

220
First Run
200 Second Run
Third Run
180

Deflection (µm) 160

140

120

100

80

60

40
0 5 10 15 20 25 30 35
Distance (km)

Figure 3.36. RWD deflections on eastbound I-64 in Virginia.

be statistically significant at the 0.05 significance level. Since of averaging length, the LOA roughly decreased (lower LOA
the distances covered are relatively long, the biases could be means higher repeatability) by a factor of L2 L1 going
due to differences in the testing conditions (e.g., pavement from the averaging length L1 to the averaging length L2. For
temperature). An average bias of 10.9 µm (0.43 mils) was cal- example, going from L1 = 1 m (3.3 ft) to L2 = 100 m (33 ft),
culated as the average of the absolute values of the biases. the repeatability roughly decreases by a factor of 100 1 = 10 .
Finally, the repeatability of RWD was calculated as 1.96sd, This occurs when measurement errors are uncorrelated and
which is equal to 50.4 µm (1.98 mils). The results of RWD therefore suggests that measurement errors are uncorrelated.
repeatability analysis are presented in Table 3.17. Furthermore, it was found that there is no significant bias
(systematic error) between repeated runs (except for one run
on Section C1). For Sites UK_F1 and UK_R2, the repeatability
Summary of Repeatability Evaluation
was found to depend on the actual deflection slope measure-
For the TSD, except for the two sites UK_F1 and UK_R2, ment. Higher measurements resulted in larger error standard
the evaluated LOA (repeatability) was the same for all tested deviation and therefore larger LOA (lower repeatability). A
sites and for the three sensors (see Table 3.16). As a function possible reason for this is that weaker sections, which result in

100

80

60
Deflection Difference (µm)

40

20

-20

-40

-60
0 5 10 15 20 25 30 35
Distance (km)

Figure 3.37. Difference of measurements versus average measurement.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

51  

Table 3.17. Repeatability Analysis of the RWD

Westbound Eastbound

1 and 2 1 and 3 2 and 3 1 and 2 1 and 3 2 and 3

Standard Deviation 24.9 27.3 27.2 26.1 24.3 24.8

Bias 14.2 6.5 –7.7 3.93 18.4 14.5
Repeatability 51.8 49.1
50.4
Note: Units in table in µm.

higher measurements, are much less homogeneous because (using the U.K. methodology) during FWD testing and TSD
of the presence of distresses (either on the surface or in the testing were included in the files. The method adopted was to
hidden layer). This results in larger LOA. correct the center deflection of the FWD first using the proce-
For the RWD, the analysis was limited to three runs per- dure described by the FHWA (1998). The correction assumes
formed on a single section. The effect of averaging length could that deflections 900 mm (36 in.) away from the center of the
not be evaluated as measurements were already averaged over load plate are not significantly affected by temperature. Thus,
160-m (0.1-mi) sections. Furthermore, there was a significant using the deflections at this point, along with a measured
bias between the three repeated runs. The repeatability of temperature, a value for asphalt stiffness, and the pavement
RWD for the section evaluated was 50.4 µm (1.98 mils). thickness, the center deflection of the FWD can be corrected
to a reference temperature. The latitude of the test site was
used to account for climatic differences.
Comparability
The corrected center deflection can be found with the fol-
Comparability is defined similarly to repeatability and is the lowing equations:
level of agreement between the two devices (TSD and FWD or
RWD and FWD). The difference is that comparability compares D0Corrected = D0Measured  TAF (3.6)
measurements from two different devices, whereas repeatabil-
ity compares measurements from the same device. The FWD where TAF = temperature adjustment factor calculated
is used as a reference point because it has become the de facto according to Equation 3.7.
standard for structural evaluation of pavement. Many engineer-
( D36 + Delta 36 @Tref )
ing parameters and properties are obtained from FWD testing. TAF = (3.7)
Most state DOTs have acquired enough experience to be able ( D36 + Delta 36 @Tmeas )
to effectively interpret and use FWD test results. Therefore, the
FWD can be used as the reference for evaluating any deflection where
measuring device (or structural capacity measuring device). D36 = deflection at 900 mm (36 in.) from the center
Because the TSD does not directly measure pavement deflec- of the load plate in µm,
tions, the analysis first investigates the correlation between Delta36@Tref = basin factor calculated at the reference
TSD slope and FWD deflection. To evaluate comparability, temperature, and
measurements need to be converted to the same physical Delta36@Tmeas = basin shape factor calculated at the measured
quantity. The physical quantities that can be obtained from temperature.
both devices are the surface curvature index (SCI) and the The basin shape factor is found by the following equation:
base damage index (BDI). The first step in the comparison
is to temperature correct FWD deflections to the TSD test log ( Delta 36 ) = 3.05 − 1.13 log ( ac ) + 0.502 log ( Theta ) log ( D36 )
temperature. The reason for correcting FWD deflections
and not the TSD or both devices is because the temperature- − 0.00487 (T ) log ( Theta ) log ( D36 )
correction procedure for the TSD is still under development. + 0.00677 (T ) log ( Theta ) log ( ac ) (3.8)

where
Temperature Correcting the FWD for TSD Sites
ac = total thickness of the HMA in mm,
FWD deflection values were first corrected to the temperature Theta = latitude of the pavement section, and
at the mid-depth of the asphalt layer during TSD testing. The T = temperature at middepth of the HMA in degrees
estimated temperatures near the mid-depth of the pavement Celsius.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

52

Distance from Center of Load Plate (mm)

0 200 400 600 800 1000
0

2 C
A

Deflection Slope (mm/m)

4
D
6

8
B
10

12

14
Figure 3.38. Temperature-correction technique for deflections
between D0 and D36.

After the center deflections were corrected, the deflections measurements from the two devices needs to be obtained. The
between 0 mm and 900 mm (0 in. and 36 in.) were also adjusted simplest relationship is the linear one. In this case, correla-
for temperature. The adjustment factor was assumed to vary tion is used to evaluate the strength of the linear relationship.
linearly along the distance from the applied from a maximum Table 3.18 shows the correlation between TSD measurements
calculated using Equation 3.5 at 0 to 0 at 900 mm away from the averaged over 10 m and selected FWD measurements (D0 and
applied load. Figure 3.38 illustrates how the correction factor D300) at 10-m (33-ft) intervals, obtained at six different sites
can be obtained. The calculation can be performed according (UK_R2, UK_R3, UK_F1, UK_F5, UK_C3, and UK_F3). The
to the following equation: correlation ranges from very good (~0.95) for Site UK_R2 to
relatively poor (~0.27) for Site UK_F3. One main drawback
( D0Corrected − D36) of correlation is that it is significantly affected by the range of
D ( x )Corrected = ( D ( x ) − D36 )  + D36 (3.9)
( D0 − D36) measurements: the wider the range of measurements, the
better the correlation. This is illustrated for the six sites in
where Figure 3.39 and Figure 3.40. Measurements collected on Site
D(x)Corrected = temperature-corrected deflection at location x, UK_F3 were gathered over a much smaller range than those
D(x) = measured deflection at location x, collected on the other sites, which explains why the correlation
D0Corrected = temperature-corrected center deflection, and for Site UK_F3 is lower than that of the other sites.
D0 = measured center deflection. While correlation values suggest a relatively good linear
relationship between FWD deflections and TSD slope, the
figures show that this relationship is pavement-type specific,
Comparison of TSD Slope and FWD Deflection
with flexible and composite pavements (F1, F3, F5, and C3)
The TSD and FWD measure two different quantities; therefore, exhibiting the same relationship, and rigid pavements (R2
measurements obtained from the devices cannot be directly and R3) exhibiting a different relationship (Figure 3.39 and
compared. To make a comparison, a relationship between Figure 3.40); that is, no single linear (or any function) model

Table 3.18. Correlation Between TSD Slope and

FWD Deflection

UK_R2 UK_R3 UK_F1 UK_F5 UK_C3 UK_F3

TSD100 and D0 0.9492 0.2798 0.8289 0.7252 0.8007 0.3716

TSD300 and D0 0.9514 0.3721 0.8260 0.7164 0.7942 0.2189
TSD100 and D1 0.9420 0.2839 0.7984 0.6941 0.7941 0.3338
TSD300 and D1 0.9448 0.4002 0.8223 0.6865 0.7948 0.1906

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

53  

5000
UK-R2
1000
4500 UK-R3
UK-F1
800
UK-F5
4000
600 UK-C3
UK-F3
3500 400 Linear Fit R Sites
Linear Fit F and C Sites
TSD P100 (µm/m)

3000 200

0
2500 0 100 200 300 400 500

2000

1500

1000

500

0
0 500 1000 1500 2000 2500
FWD D0 (µm)

Figure 3.39. TSD P100 slope versus FWD D0 deflection.

can adequately represent the relationship between FWD The calculated SCI is SCI300, which is defined as D0 - D300,
deflections and TSD slope for the two pavement categories. and the BDI is defined as D300 - D600. For the FWD, D0,
D300, and D600 are directly measured and calculation of the
SCI is straightforward. In the following section, we present
Comparability Between TSD and
the methodology used to calculate both the SCI and the BDI
FWD Test Results
from TSD slope measurements.
To evaluate the comparability between the TSD and the
FWD, the two measured quantities need to be converted to
Converting TSD Slope to SCI or BDI
the same physical quantity. The surface curvature index (SCI)
and the base damage index (BDI) were chosen because they The TSD measures the slope of the deflection bowl that
can be calculated from both the TSD and FWD measurements. results from the truck traveling over the pavement. Therefore,

3000 800

600

2500
400

200
2000
0
TSD P300 ( m/m)

1500 -200
0 50 100 150 200 250 300 350 400

1000

UK-R2
500 UK-R3
UK-F1
UK-F5
UK-C3
0
UK-F3
Linear Fit R sites
Linear Fit F and C Sites
-500
0 200 400 600 800 1000 1200 1400 1600
FWD D300 (μm)
0
Figure 3.40. TSD P300 slope versus FWD D300 deflection.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

54

deflection can, in principle, be obtained from the TSD slope 11.8 inches) from the applied wheel load, were investigated
by integration if a sufficiently detailed representation of the to fit measured TSD slope. The results showed that the
full deflection slope bowl is available. Integration is specified equations are not suitable to use in the sensor configura-
only up to a constant value; therefore, the deflection can- tion tested (sensors at 100, 300, and 756 mm [3.94, 11.81,
not be recovered without a reference deflection measure- and 29.8 in.] from the load). Therefore, numerical inte-
ment. The difference between two deflection readings can be gration was adopted (trapezoidal rule) to calculate the SCI
obtained; the constant cancels out, which gives the SCI or and BDI.
BDI. The relationship between slope, deflections, and SCI is To calculate the SCI and BDI from TSD measurements,
presented in Equation 3.10. the TSD slope was integrated numerically. Slope measure-
ments were obtained at 100, 300, and 756 mm (3.94, 11.81,
and 29.8 in.) from the applied load. The integration was per-
∫a s ( x ) Dx = D (b ) − D (a ) = SCI
b
(3.10)
formed using the trapezoidal rule. To calculate SCI300, the
integration interval is [0 mm; 300 mm] ([0 in.; 11.81 in.]).
where In this interval, TSD slope measurements were obtained at
s(x) = slope at location x, and 100 and 300 mm (3.94 and 11.81 in.) from the applied wheel
D(x) = deflection at location x. load and an assumption was needed on how the slope varies
Recent results by Thyagarajan et al. (2011) and Krarup et al. between 0 and 100 m (0 and 3.94 in.). Two assumptions illus-
(2006) suggest that SCI values, especially SCI300, correlate very trated in Figure 3.41 were investigated. The first assumption
well with tensile strain at the bottom of the asphalt layer. sets the slope to vary linearly throughout the interval from 0
Krarup et al. (2006) used a functional representation of the to 300 m (0 and 11.81 in.). This assumption is represented
deflection and slope presented in Equation 3.11 and Equa- by the red line between 0 and 100 mm (0 and 3.94 in.) and
tion 3.12 to perform the integration: the blue line between 100 and 300 mm (3.94 and 11.81 in.).
The integration in this case results in the area comprising the
A red and blue areas. The second, more realistic assumption
D( x ) = − [cos ( Bx ) + sin ( Bx )]exp ( − Bx ) (3.11) is to set the slope at 0 mm from the load equal to zero. This
2B
assumption is valid if the load is uniformly (or approximately
s ( x ) = D ′ ( x ) = A sin ( Bx ) exp ( − Bx ) (3.12) uniformly) applied over a specific area rather than being a
point load and if viscoelastic effects are neglected. The results
where A and B = experimentally determined constants. of both calculations are presented in Figure 3.42, which sug-
Equations 3.11 and 3.12, which were developed for a sen- gests the second assumption (slope at 0 equals 0) works bet-
sor configuration of 100, 200, and 300 mm (3.94, 7.87, and ter because the calculated SCI values (from FWD and TSD)

600

500

400
Slope ( m/m)

300

200

100

0
0 50 100 150 200 250 300
Sensor Location (mm)
Note: The SCI300 is the shaded area under the piecewise linear curve.

Figure 3.41. Effect of assuming the slope at 0 mm to be equal to zero on the

calculated SCI300.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

55  

1400

1200
Slope (0) 0
Slope (0) = 0
1000 Linear Regression
Orthogonal Regression
Equality Line
800
TSD SCI300 ( m)

600

400

200

-200
-200 0 200 400 600 800 1000
FWD SCI300 ( m)

Figure 3.42. Comparison of SCI300 calculated by using various assumptions on the slope.

better follow the line of equality. The second assumption was regression with FWD measurements as the independent
thus adopted in the analysis. variable, which had a slope of 0.79.
Figure 3.42 also shows the results of the ordinary and
orthogonal regressions between the SCI calculations based
SCI Comparisons
on the FWD and TSD measurements. Since both variables
contain errors, orthogonal (or total least-squares) regression Figure 3.43 shows the calculated SCI300 with the assumption
is more appropriate, as shown in the figure. The orthogonal of slope at 0 equal to 0 while Figure 3.44 shows the calcula-
regression line has a slope of 0.90, which is closer to the line tion of BDI (also known as SCI450, which is defined here as
of equality than the one obtained using the least-squares D300 - D600). The main advantage of comparing the indices

1000
200

900
150

800
100

700
50
TSD SCI300 ( m)

600
0
-20 0 20 40 60 80 100
500
UK-R2
400 UK-R3
UK-F1
300 Equality Line
UK-F5
200 UK-C3
UK-F3

100

0
-200 0 200 400 600 800 1000
FWD SCI300 ( m)

Figure 3.43. Comparison of SCI300 obtained from the TSD and FWD testing.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

56

900
200

UK-R2
800 UK-R3
150
UK-F1
700 Equality Line
100 UK-F5
UK-C3
600 UK-F3
50
TSD BDI ( m)

500
0
0 20 40 60 80 100

400

300

200

100

0
0 100 200 300 400 500 600 700 800 900
FWD BDI ( m)

Figure 3.44. Comparison of BDI obtained from the TSD and FWD testing.

is that both quantities have the same dimensions. The first line with the y variable taken as the difference in the SCI
important observation is that there is no obvious indication and the x variable taken as the average SCI. This regression
that the relationship is not the same for all tested pavement line is an estimate of the bias (as a function of average SCI)
types (Figure 3.43 and Figure 3.44). The plots also show that between the two methods. The variance can be estimated by
the results of the calculated SCI300 and BDI are comparable using the squared residuals obtained from the regression
for both the FWD and TSD; this is reflected by the close prox- analysis. This is done by (again) performing regression on the
imity of the observed measurements to the equality line. The squared residuals. The result of this regression is presented in
limits of agreement (LOA) method was then used to assess Figure 3.45. These results correspond to the square root of
the comparability between the two devices, as is presented in the squared residuals, and as such, the fitted lines represent
the next section. the standard deviation variation as a function of average
SCI. The reason for not plotting the squared residual is
because taking the square root results in a clearer visual repre-
Comparability and LOA of SCI
sentation. The LOA presented in Figure 3.46 can be calculated
and BDI Measurements
by using the derived standard deviation and constructing the
The plots presented in Figure 3.43 and Figure 3.44 suggest a 95% confidence interval around the bias.
relatively good agreement between both devices; however, to As recommended by Bland and Altman (1983), the plot of
compare measurements obtained from the two devices, one the difference of the SCI (FWD - TSD) versus the average SCI
would need to evaluate how well they agree with each other in ([FWD + TSD]/2) is the first step to evaluate the LOA. This plot
a quantifiable way. This is usually a judgment call, but there is presented in Figure 3.46 for both SCI300 and BDI (referred
are some parameters that provide guidance for this problem. to both as SCI). The figure suggests that in both cases, the
Generally, a discrepancy in measurement between the two difference depends on the SCI measurements. Thus, data
devices will be accepted if it is less than a certain limit, usually transformations were investigated to possibly remove this
referred to as the confidence limit. Confidence limits should dependence. If transformation is not successful, the repeat-
be narrow enough to expect that the result will not affect a ability can be defined as a function of the size of measurement
decision that is based on it. (Bland and Altman, 1983). A logarithmic transformation, as
To evaluate this dependence, the standard deviation of the well as a coefficient of variation transformation (difference
difference as a function of the SCI measurement was calcu- divided by average), on the data did not result in the removal
lated. This was done with the method presented by Davidian of this association. The repeatability was therefore defined as
and Carrol (1987). The procedure consists of fitting a regression a function of the size of measurements.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

57  

450

400
Residuals
350 STDEV (SCI<500)
STDEV
300

SCI Residuals (μm) 250

200

150

100

50

0
0 100 200 300 400 500 600 700 800 900 1000
Average SCI ( m)

Figure 3.45. Plot of absolute value on residual and the regression on the
residual squares.

Figure 3.45 and Figure 3.46 present two sets of results each. To interpret the results presented in Figure 3.46, the
The first trend line (red dashed line, STDEV) and confidence example average SCI of 300 µm (11.8 mils) can be exam-
interval used all the measurements to estimate the bias and ined. In this case, the SCI calculated from the FWD is
standard deviation. However, only seven measurements were expected to be (on average) 30 µm (1.18 mils) lower than
obtained for SCI values above 500 µm (19.7 mils), and the the SCI calculated from the TSD, with values ranging from
results of the analysis performed by using all measurements as much as 205 µm (8.07 mils) lower to 175 µm (6.89 mils)
are highly influenced by those seven measurements. The higher (for a range of 380 µm [15.0 mils]) expected to
same analysis was performed with only the set of points that occur 95% of the time. For an average SCI of 100 µm,
resulted in an average SCI below 500 µm (19.7 mils). As can the values computed using the FWD measurements
be seen in Figure 3.45 and Figure 3.46, the results of both are expected to be (on average) 50 µm (1.97 mils) lower
analyses are essentially the same for average SCI values below than the SCI calculated from the TSD, with values rang-
250 µm (9.84 mils), and differences increase with increasing ing from as much as 115 µm (4.53 mils) lower to 15 µm
average SCI values. (0.59 mils) higher (for a range of 130 µm [5.12 mils]),

1000

Difference
800 Bias
( m)

95% CI
600 95% CI
TSD

Bias (SCI<500)
95% CI (SCI<500)
- SCI

400 95% CI (SCI<500)

FWD
Difference SCI

200

-200

-400
0 100 200 300 400 500 600 700 800 900 1000
Average SCI ( m)

Figure 3.46. Plot of difference of SCI versus average SCI with bias and limits
of agreement.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

58

700
RWD
FWD

600

500
Deflections ( m)

400

300

200

100

0
0 20 40 60 80 100 120 140 160 180 200
Distance (km)

Figure 3.47. FWD and RWD measurements obtained from New Mexico.

expected to occur 95% of the time. To put these two numbers presented in Figure 3.47. In this case, RWD and FWD mea-
into perspective, the repeatability of TSD SCI300 measure- surements were obtained at 160-m (0.1-mi) intervals. The
ments for Sites UK_F1 and UK_F5 was calculated as 51 µm plot of the difference, FWD – RWD, versus the average is pre-
(2.01 mils) (range of 102 µm [4.02 mils]) for UK_F1 and 25 µm sented in Figure 3.48. The figure suggests the measurement
(0.98 mils) (range of 50 µm [1.97 mils]) for UK_F5. error depends on the associated measurement. Both loga-
rithmic and power transformations of the data were investi-
gated but did not result in removal of the error dependence
Comparability Between RWD and FWD
with the associated measurement. In the end, the normalized
Comparison of RWD and FWD testing performed in New difference calculated as (x1 - x2)/(0.5x1 + 0.5x2) was found
Mexico, which was provided by ARA for this project, is to be the error measurement parameter that was the least

400

300

200
Difference FWD - RWD ( m)

100

-100

-200

-300

-400

-500
0 100 200 300 400 500 600 700
Average RWD and FWD Measurements ( m)

Figure 3.48. Difference versus average measurement.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

59  

1.5
Difference
Linear Fit
1

Normalized Difference
0.5

-0.5

-1

-1.5
0 100 200 300 400 500 600 700
Average Measurement of FWD and RWD ( m)

Figure 3.49. Normalized difference versus average measurement.

dependent on the associated measurement. Figure 3.49 shows wider calculated bootstrap interval reflects the fact that the
the normalized difference as a function of the measurement. distribution in Figure 3.50 has a heavy tail.
The dependence on the associated measurement is not com-
pletely removed; however, it is appreciably lower than the
Direct Comparison of RWD Deflections
association in Figure 3.48.
and FWD Deflections
The average normalized difference of the difference
(FWD - RWD) is -0.2328 and the standard deviation is A comparison was made directly between the RWD data
0.3265. Figure 3.50 shows a distribution of the normalized and the FWD data collected. Figure 3.51 presents a more
difference. The Anderson–Darling test for normality showed detailed analysis of the data collected on U.S. Route 550 in
that this distribution does not follow the normal distribution. New Mexico.
The 95% bootstrap confidence interval of the normalized dif- The figure shows the entire length measured with the
ference was calculated as [-0.8642; 0.4286]. In comparison, RWD (Figure 3.51a) and FWD (Figure 3.51b) divided into
the 95% confidence interval using normal assumptions is homogeneous sections. Each plot displays the average and
[-0.8538; 0.3882]. The two intervals are close; however, the characteristic deflections (upper 95% confidence limit) for

90

80

70

60

50
Count

40

30

20

10

0
-1.5 -1 -0.5 0 0.5 1 1.5
Normalized Difference

Figure 3.50. Distribution of normalized difference.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

60

30 Deflection Data
std = 4.81
Uniform Sections
25 95% Upper Confidence Limit

std = 3.15

Deflection (mils)
20 std = 1.62
std = 2.86 std = 3.10 std = 1.71

15 std = 2.46

10

0
0 20 40 60 80 100 120
Distance (miles)
(a) RWD measurements

30
Deflection Data
std = 5.11 Uniform Sections
25 95% Upper Confidence Limit
Deflection (mils)

20
std = 3.07 std = 2.25
std = 1.58
15 std = 2.81
std = 1.47

10 std = 1.03

0
0 20 40 60 80 100 120
Distance (miles)
(b) FWD measurements

30
FWD Data
RWD Data
25 Correlation = 0.5993 Uniform FWD
Uniform RWD
Deflection (mils)

20

15

10

0
0 20 40 60 80 100 120
Distance (miles)
(c) Comparison of 0.1-mi averaged measurements and section averages

30 20
RWD Deflection (mils)

Correlation = y = 1.1*x + 1.7

RWD Deflections (mils)

0.5993
20 15 Correlation =
0.9132

Measurements
10 10
Line of Equality Uniform Section Measurements
Linear Trend
Line of Equality

0 5
0 10 20 30 5 10 15 20
FWD Deflections (mils) FWD Deflection (mils)
(d) Individual 0.1-mi segment comparisons (e) Homogeneous section comparisons

Figure 3.51. Analysis of New Mexico data.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

61  

each section. The sections were segmented visually, and the deflection measurements, and in 95% of the cases, FWD
average values are compared in Figure 3.51c. deflection measurements will be between -86.42% lower
Figure 3.51d and e compare the individual 160-m (0.1-mi) to 42.86% higher than RWD deflection measurements. This
averages and the homogeneous section averages, respectively. difference seems to be high however, and as in the case of the
The correlation between the individual measurements is rela- TSD comparison with FWD, the repeatability of the FWD
tively weak (R = 0.60, R2 = 0.36), but it improves significantly was not evaluated. Therefore, it was not possible to quantify
when the data are aggregated in homogeneous sections. The how much of this large range resulted from repeatability of
RWD deflections are consistently higher than the FWD ones, the FWD.
but the correlation is good (R = 0.91, R2 = 0.83).
TSD Repeatability from Single
Summary of Comparability Measurement Run
To evaluate the comparability between the TSD and FWD, The calculation of repeatability requires at least two runs of
the TSD slope measurements and the FWD deflection the TSD repeated on the same pavement section. Ideally, these
measurements were converted to the SCI and BDI. This was runs should be performed under the same conditions. For
performed because the FWD and TSD measure different example, for flexible pavement sections, the test temperature
physical quantities—deflection and deflection slope, respec- should, as much as possible, be the same. This section presents
tively. Both these quantities can be used to calculate the SCI a method to evaluate TSD repeatability from measurements
and BDI, which makes comparability evaluation between the obtained from a single run. Such a method can be very use-
two devices feasible. The LOA between the TSD and FWD ful in cases where repeated runs performed under the same
was found to depend on the average measurement of the two conditions are not practically or economically feasible. This
devices (see Figure 3.45 and Figure 3.46). Furthermore, there is probably includes the majority of applications of continuous
a bias between the TSD and FWD, SCI, and BDI measurements deflection devices, specifically the TSD.
that depends on the average measurement (see Figure 3.46).
To interpret the results presented in Figure 3.46, the example Difference Sequence Method for Standard
of average SCI of 300 µm (11.8 mils) can be examined. In Deviation Estimation
this case, the SCI calculated from the FWD is expected to be
(on average) 30 µm (1.18 mils) lower than the SCI calculated Difference sequence methods (DSMs) arose from the need to
from the TSD, with values ranging from as much as 205 µm estimate the error standard deviation in nonparametric regres-
(8.07 mils) lower to 175 µm (6.89 mils) higher (for a range of sion models. The calculated standard deviation can be used,
380 µm [15.0 mils]) expected to occur 95% of the time. For an among other things, for the computation of confidence bands
average SCI of 100 µm (3.94 mils), the value computed using the or the optimal choice of smoothing parameter (Munk et al.,
FWD measurements are expected to be (on average) 50 µm 2005; Brown and Levine, 2007). The general procedure is sum-
lower than the SCI calculated from the TSD, with values marized in Hall et al. (1990), who introduced difference-based
ranging from as much as 115 µm (4.53 mils) lower to 15 µm estimators of arbitrary order r using a difference sequence di of
(0.59 mils) higher (for a range of 130 µm [5.12 mils]) expected real numbers as follows:
to occur 95% of the time. It is important to point out that
r
the large range in the LOA between the two devices does not
imply that the TSD fails to give accurate measurements. The ∑ di = 0 (3.13)
i =0
LOA between the two devices depends on the repeatability of
the TSD, the repeatability of the FWD, and the comparability r
between the two devices. The repeatability of the FWD was ∑ di2 = 1 (3.14)
i =0
not evaluated.
The repeatability between the FWD and RWD was also 2
found to depend on the average deflection measurement. 1 n− r  r 
However, the cov was found to be relatively uniform across
σˆ = ∑  ∑
(n − r ) i=1  j =0
d j y j +1 

(3.15)
all measurement values and was therefore used as a measure
of repeatability. The average cov of the difference (FWD - It can easily be verified that, except for the case of r = 1, differ-
RWD) was calculated as -0.2328 (or -23.28%). The 95% ent combinations of the values di can be used and still satisfy
interval of the cov was calculated as [-0.8642; 0.4286] or, Equations 3.13 and 3.14. Here, the results of the best estimator
in percent, [-86.42%; 42.86%]. This means that on average, for the investigated data (Katicha et al., 2012), which was the
FWD deflection measurements are 23.28% lower than RWD second order estimator (r = 2), known as the Gasser estimator

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

62

Table 3.19. Standard Deviation Calculated with Second Order

Difference Sequences for Sensor P100

Slope Standard Deviation (mm/m)

Run
Difference
Site 1 2 3 4 5 Average Method

F1 0.2706 0.2623 0.2647 0.2750 na 0.2682 0.2661

F3 0.1260 0.1360 0.1226 na na 0.1283 0.1305
F5 0.1118 0.1173 0.1096 0.1152 0.1136 0.1135 0.1166
F6 0.1181 0.1253 0.1233 0.1226 0.1213 0.1221 0.1254
C1 0.1503 0.1429 0.1510 0.1357 0.1489 0.1459 0.1466
C2 0.1214 0.1184 0.1186 0.1172 0.1241 0.1200 0.1222
R2 0.2151 0.2087 0.1921 na na 0.2055 0.2072
R3 0.1566 0.1739 0.1501 0.1571 0.1601 0.1598 0.1483
Note: na = not applicable.

(Gasser et al., 1986), are presented. This estimator for equally deviations obtained by using repeated runs (presented in the
spaced observations is given by the following equation: TSD Repeatability section). In general, the two estimates
of the standard deviation agree very well, which suggests that
2 n− 2  1
2 the DSM can be used to obtain the standard deviation from
1 
σˆ = ∑  yi − yi+1 + 2 yi+2 
3(n − 2 ) i =1  2
(3.16) a single run. Note that in the DSM, the bias is not accounted
for (because there is only one run investigated at a time, the
bias cannot be evaluated).
Calculation of Standard Deviation
Effect of Sampling Frequency on Accuracy of
Table 3.19 and Table 3.20 present the results of standard
Standard Deviation Estimation
deviation estimation for the measurements averaged over a
1-m (3.3-ft) length by using the Gasser estimator for P100 Although the results presented in the previous sections show
and P300, respectively. These are compared with the standard that the standard deviation estimated from a single run agrees

Table 3.20. Standard Deviation Calculated with Second Order

Difference Sequences for Sensor P300

Slope Standard Deviation (mm/m)

Run
Difference
Site 1 2 3 4 5 Average Method

F1 0.2053 0.2057 0.2027 0.2096 na 0.2058 0.1918

F3 0.1267 0.1363 0.1245 na na 0.1293 0.1376
F5 0.1137 0.1153 0.1071 0.1114 0.1077 0.1111 0.1123
F6 0.1117 0.1156 0.1215 0.1164 0.1181 0.1167 0.1210
C1 0.1491 0.1325 0.1426 0.1357 0.1467 0.1415 0.1398
C2 0.1199 0.1207 0.1201 0.1212 0.1263 0.1217 0.1226
R2 0.1674 0.1617 0.1482 na na 0.1593 0.1671
R3 0.1418 0.1550 0.1358 0.1473 0.1344 0.1431 0.1378
Note: na = not applicable.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

63  

with the standard deviation estimated by using repeated and Section R2, which had the most variation, resulted in
measurements, there are some limitations to the application the worse estimate of the standard deviation (see Table 3.21
of the DSM. For the method to be successful (which depends and Figure 3.52).
on how accurate the estimation of standard deviation needs As expected, the estimate is also worse for longer aver-
to be), the variation in the true measurement (deflection aging distance; for example, the results for 100-m (330-ft)
slope, in this case) profile should happen relatively smoothly. averaging are very different except for Site C2, which is a
In practical terms, for the most part of the profile, the differ- relatively uniform site. This shows one limitation of differ-
ence of the true deflection between adjacent points should ence sequence methods. However, the standard deviation
be small compared to the noise level in the measurements. for different averaging distances can still be estimated from
In a pavement structure, the deflection measured at a given the standard deviation for 1-m (3.3-ft) averaging using the
location is influenced by the pavement properties within a following formula:
certain distance, say of radius R, of that location. The radius
R depends on the pavement structure (number of layers, σ1
σn = (3.17)
thicknesses, and so forth) and the mechanics that govern n
the deformation of the pavement (e.g., multilayer analysis).
Results presented in the previous section suggest that, for all where
practical purposes, to obtain a good estimate of the standard n = section length in meters,
deviation from a single run, a distance of 1 m (3.3 ft) is within sn = the standard deviation for the sections of length n,
the radius of influence R for the tested pavements. and
To illustrate the effect of measurement distance, the stan- s1 = the standard deviation for 1-m averaging.
dard deviation for the case of P100 measurements averaged
over 10 m and 100 m (33 ft and 330 ft) were calculated by
Denoising and Data Aggregation
using the DSM. The results of this calculation, along with
the 1-m (3.3-ft) results, are presented in Table 3.21. The The purpose of continuous deflection devices is to measure
table shows that the calculated standard deviation for 10-m a physical characteristic of the pavement (e.g., deflection in
(33-ft) averaging derived with the DSM does not agree with the case of RWD and deflection slope in the case of TSD)
the calculated standard deviation derived with repeated at specific locations. Measurements are contaminated with
measurements. The DSM produces wrong estimates of noise; from these noisy measurements, the objective is to
the standard deviation for deflection averaged over a 10-m make inferences about the expected value of the true physical
(33-ft) section, probably because the average values can quantity being measured with a confidence interval on that
be significantly different from one point to another. The more expected value. For TSD measurements, this can be interpreted
variation there is in a tested section, the worse the estimate as finding out how the deflection slope varies as a function
of the DSM. This is confirmed by the fact that Section F1 of the measurement location (i.e., along the roadway). This

Table 3.21. Calculation of Standard Deviation for Measurements

Averaged over Distances (mm/m)

1 m (3.3 ft) 10 m (33 ft) 100 m (330 ft)

Difference Repeated Difference Repeated Difference Repeated

Site Sequence Runs Sequence Runs Sequence Runs

F1 0.2682 0.2611 0.2548 0.1054 0.3504 0.0625

F3 0.1282 0.1305 0.0436 0.0407 0.0256 0.0155
F5 0.1135 0.1166 0.0578 0.0401 0.0544 0.0147

F6 0.1223 0.1254 0.0448 0.0406 0.0227 0.0139

C1 0.1458 0.1466 0.0659 0.0446 0.0423 0.0153
C2 0.1200 0.1222 0.0409 0.0408 0.0168 0.0142
R2 0.2053 0.2072 0.1766 0.0643 na na
R3 0.1596 0.1483 0.0500 0.0432 0.0179 0.0153
Note: na = not applicable.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

64

6 0.6
4 0.4
0.2
2 0
-0.2
0
0 500 1000 1500 2000 0 500 1000 1500 2000 2500 3000
a b

1 0.8
0.6
0.5 0.4
0.2
0 0
0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 3000 3500
Slope (mm/m)

c d

1.5 0.8
1 0.6
0.4
0.5 0.2
0 0
0 1000 2000 3000 0 500 1000 1500 2000 2500 3000
e f

4 1
2 0.5
0
0
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200 1400
g h
Distance (m)

Figure 3.52. TSD slope measurements for Sensor P100 for U.K. sites: (a) F1;
(b) F3; (c) F5; (d) F6; (e) C1; (f) C2; (g) R2; and (h) R3.

can be done by using nonparametric regression. Common lated using the following model for the TSD deflection slope
regression analysis, which is extensively used by engineers, measurements:
is parametric regression where a (parametric) model (e.g.,
linear model) is postulated to represent the observed behav- yi = f ( x i ) + ε i (3.18)
ior and model parameters are obtained using a specified
criterion (e.g., least squares, maximum likelihood). In some where
cases, such as for continuous deflection measurements, the yi = TSD deflection slope measurements,
form of the regression curve is not known. In such cases i = 1, 2, . . . , n (the number of measurements),
nonparametric regression, which is not restricted to a given xi ∈ [0, 1],
form (such as linear and exponential), can be used. Essen- f(xi) = true deflection slope (which is not known and is to
tially, the collected data are used to infer on the regression be estimated), and
function. ei = i.i.d random variables with mean zero and known
The three most common methods of nonparametric regres- or unknown variance s2.
sion are kernel regression, smoothing spline regression, and
least-squares spline regression. The familiar moving average The smoothing spline method consists of finding the func-
falls under kernel regression. From a practical perspective, all tion g that minimizes the following formula:
three methods give very similar results, and selecting a particu-
lar one is often the result of individual preference and ease of 1 n
∑ ( g ( xi ) − yi )2 + λ ∫0 ( g (m) (u ))2 du
1
use. In this report, the method of smoothing splines is used (3.19)
n i =1
for its simplicity and ease of implementation. The main ques-
tion in smoothing spline regression is how much smooth-
where g(m) is the mth derivative of g.
ing should be performed. A number of objective methods
have been developed to answer this question. These methods In this report, m is taken as m = 2 (the typical value used for
consist of optimizing a parameter that controls the trade-off m). In this case, the solution to the minimization problem, g is
between smoothness and adherence to the measurements a cubic spline. The parameter l (the only parameter that needs
(i.e., controls variance and bias). The method can be formu- to be determined) is a smoothing parameter that controls the

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

65  
8 25
Measured Signal Measured Signal
True Signal True Signal
20
Estimated Signal Estimated Signal
6
15

10
4

2
0

-5
0
-10

-2 -15
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(a) (b)

Figure 3.53. Spline smoothing of sinusoidal function contaminated with (a) low and (b) high noise levels.

trade-off between the “roughness” of the solution and the of TSD slope variation in each site (see Figure 3.52). UK_F3
fidelity to the measured data. A popular method to estimate l, exhibited very low variation in the measured TSD deflection
which is used in this report, is the generalized cross validation slope, UK_F5 exhibited some variation in the measured TSD
(GCV) criterion first suggested by Craven and Wahba (1979). deflection slope, and UK_F1 exhibited the most variation in
the measured TSD deflection slope.
Illustrative Simple Examples with Figure 3.54 shows the measured first run on Site UK_F3,
a Sinusoidal Function with the smoothing spline regression function and the 95%
Bayesian confidence interval (Wahba, 1983; Nychka, 1988,
Two simple examples of spline smoothing using a sinusoidal 1990) superimposed to the measurements and the averages
function are first presented. The purpose of these examples is
over 100-m (330-ft) sections. The smoothing function
to give an intuitive feeling on the performance of the method.
gives results that are as “noise-free” as averaging results
The example compares synthetic measurements with relatively
over 100-m (330-ft) sections. One important difference
low noise levels (s = 1) and relatively high noise levels (s = 5)
between the smoothing function and the 100-m (330-ft)
of a sinusoidal function given by the following equation:
average sections is that the smoothing function is an esti-
mate of the deflection slope at 1-m (3.3-ft) intervals (com-
y = sin ( 2 π × 0.1x ) + 3 (3.20)
pared to an estimate of the average of a 100-m [330-ft]
Measurements are obtained for x between 0 and 60 and 0.01 section) and can therefore be used for project-level appli-
spacing (units are not important in the example; however, cations (compared to network-level applications). Further-
these can be meters). Figure 3.53a shows the results for the more, a confidence interval can be constructed around the
case of low noise levels (s = 1). From the measured signal, estimate. The operation of smoothing filters the noise from
the underlying signal is presumed to be fluctuating (such as a the data at the cost of introducing bias. The GCV criterion
sinusoidal). The smoothing spline estimate (dashed red line) finds the compromise between the bias and variance to be
is very close to the actual sinusoidal signal (continuous blue a trade-off. In practical terms, if more smoothing is per-
line). Figure 3.53b shows the measured signal for the case of formed, important features that are not spikes due to noise
high noise levels (s = 5). In this case, visual inspection of the will be smoothed out. This is better illustrated with Section
plot may not by itself suggest that the true signal is fluctuat- UK_F5.
ing. However, even for the level of noise in the signal, the Figure 3.55 shows the results of smoothing measurements
performance of the smoothing spline in estimating the true on Section UK_F5. In this case, an appreciable difference
function is still very reasonable. between the smoothed regression estimate and the results
averaged over 100-m (330-ft) sections can be observed.
The figure suggests that results for the 100-m (330-ft) aver-
Application to TSD Measurements
aged section are smoothing statistically significant features
In this section, the application of the smoothing spline is of the deflection slope profile. Therefore, these features are
demonstrated on three sets of TSD deflection slope measure- not likely due to random noise and perhaps (depending on
ments from the P100 sensor on Sites UK_F3, UK_F5, and whether they are important from an engineering perspective)
UK_F1. These were selected because of the different range should not be ignored (or smoothed out).

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

66

0.8

0.6

Deflection Slope (mm/m)

0.4

0.2

0
Measured
Smoothed
-0.2 Upper 95% CI
Lower 95% CI
100 m average
-0.4
0 500 1000 1500 2000 2500 3000
Distance (m)

Figure 3.54. Smoothing slope deflection measurements of

site UK_F3.

UK_F1 shows an even more extreme example (Figure 3.56). the capabilities of continuous deflection devices to be used
In this case, averaging over 100-m (330-ft) sections is likely to for project-level applications. Furthermore, the smoothing
result in smoothing out significant features (both statistically spline analysis can be used to identify features that vary with
and from an engineering perspective). Figure 3.56 also shows distances as small as 1 m within a certain level of confidence
that averaging over 10-m (33-ft) sections results in smooth- (using the confidence interval as shown in Figure 3.54).
ing out statistically significant features. This was expected, as Although the limited (in terms of number of sections and
UK_F1 shows significant variation in the measured deflec- section length) data used are only for the TSD, a similar type
tion slope. In this case, the noise in the signal is relatively small of analysis should be feasible for the RWD if more closely
compared to the signal itself. Therefore, smoothing has a much spaced data are provided. However, closely spaced (raw)
more significant effect on changing the signal than on reducing RWD measurements were not available for this project and
the level of noise and should be minimal. Again, how much this type of analysis could not be carried out for this device.
to smooth is controlled by the GCV criterion, which finds a
compromise between bias and variance.
Operational Characteristics

Summary Within this study it has not been possible to examine in detail
the operational characteristics of the equipment being assessed.
The results presented in this section suggest that the smooth- Although external factors such as temperature, road geometry,
ing (or averaging) deflection test results should not be set to road profile, texture profile, moisture, acceleration, deceleration,
a constant value, but rather controlled by the actual deflec- and so forth, may have been recorded during the surveys, it is
tion profile. Using an “optimal” smoothing may help improve not possible to control these factors and, therefore, not easy to

Measured
1
Smoothed
0.8 100 m average
Deflection Slope (mm/m)

0.6

0.4

0.2

-0.2
0 500 1000 1500 2000 2500 3000
Distance (m)

Figure 3.55. Smoothing slope deflection measurements of

Site UK_F5.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

67  

7 7
6 Measured
6 5
Smoothed

Deflection Slope (mm/m)

4
5 100 m average
3
2
10 m average
4 1
0
3 1200 1220 1240 1260 1280 1300

0
0 500 1000 1500 2000
Distance (m)

Figure 3.56. Smoothing slope deflection measurements of

Site UK_F1.

assess their effect on the measurements. Control of survey speed times, some differences for the asphalt pavement could be
is possible, but sometimes other uncontrollable effects confound due to variations in temperature.
the effort. The following sections discuss these issues together Repeat runs have been carried out with the HA TSD at
with the measuring capability of current and future devices on different speeds. A small change with increasing speed has
the typical range of pavement types encountered on a network. been shown, but data collected so far have been insufficient
to develop a correction procedure to a reference speed. Fig-
Operating Conditions ure 3.58 illustrates this by presenting the TSD P100 slope
profiles as 10-m (33-ft) means for a 1.5-km (1-mi) section with
Survey Speed flexible pavement at 60, 70, and 80 km/h (40, 45, and 50 mph).
The RWD is normally operated as close as possible to An analysis of network TSD measurements collected in the
80 km/h (50 mph). As yet, no method has been developed for United Kingdom during 2010 and the early part of 2011 over a
converting surveys at other speeds to the standard operating wide range of speeds, some outside the current recommended
speed. At present the TSD operates as close as possible to limits, is shown in Figure 3.59 in the form of a distribution plot.
70 km/h (45 mph) on divided highways, but a range of 60 to This suggests a reduction in response with increasing speed.
80 km/h (40 to 50 mph) is considered acceptable. For two- However, there are many other confounding effects present in
lane roadways in the United Kingdom, because of the speed this data, so it can be taken only as an indication of the likely
limit, the standard operating speed is 60 km/h (40 mph). effect of survey speed. The figure shows a two-dimensional ver-
The effect of survey speed has yet to be investigated fully in sion of a three-dimensional plot, in which the lines are con-
either device. The deflection response of a pavement will be tour lines containing given proportions of the total dataset. For
influenced by a number of factors, which include the speed example, 20% of the data were collected at speeds of around
of the loading wheel and the composition of the pavement. 61 km/h (38 mph) with slopes from 0.18 to 0.37 mm/m. The
In order to ascertain the extent of the effect, any experiment other concentration of data is just below 70 km/h (44 mph).
will need to be very carefully controlled in terms of operating These were the two target speeds in the surveys, but other
conditions; and the results are likely to vary with the proper- speeds were covered for various practical reasons.
ties, particularly the viscoelastic properties, of the pavement
layers. This kind of control was not possible within the scope Road Geometry and Profile
of the project. However, a limited examination based on
available data has been carried out for both devices. The specific effect of road geometry has yet to be investigated
The California RWD demonstration included repeated for either device. In particular, measurements on curves are a
runs at various speeds on flexible and rigid pavements. Fig- potential issue, especially with the RWD because of the way it
ure 3.57 shows 160-m (0.1-mi) RWD results on two of the compares the same texture profile from the deflected sensors
tested sections, with flexible and rigid pavements. The figures and the undeflected or reference sensor. The two sensors may
suggest that RWD deflections are relatively insensitive to not follow the same trajectory while measuring on curves.
truck speed for the speed ranges investigated (50 to 110 km/h A superficial examination of the effect of road curvature has
[30 to 70 mph]). Since the tests were conducted at different been carried out in the United Kingdom by examining the TSD

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

68

(a) AC test section on eastbound SR-16, Mileposts 14 to 15.5

(b) JPCP test section on southbound I-505, Exits 24 to 28

Source: ARA, 2007c.

Figure 3.57. Example of repeated RWD runs at different speeds.

network data collected during 2010 and 2011. This examination Surface Characteristics
showed no obvious relationships between longitudinal profiles,
gradient, transversal slope, or curvature. The data, with respect Research with the U.K. TSD has shown that surface type can
to longitudinal profile, are shown in the form of a contour map influence the response of the velocity sensors. In particular, new
of the distribution of almost 10,000 lane-km (6,250 lane-mi) of binder-rich surfaces can cause faulty operation of the velocity
data of deflection slope versus 3-m (10-ft) longitudinal profile sensors on the TSD but normal measuring performance returns
variance in Figure 3.60. after a few months of trafficking as the surfacing becomes less

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

69  

1.000

0.900
80km/h
0.800
70km/h

0.700 60km/h
Deflection Slope [mm/m]

0.600

0.500

0.400

0.300

0.200

0.100

0.000
1500 1700 1900 2100 2300 2500 2700 2900
Chainage [m]

Figure 3.58. TSD slope profiles for 10-m (3.3-ft) means on a U.K. road section at various
survey speeds.

Note: The blue area highlights the overall trend of the reduced deflection slope with increased velocity.

Figure 3.59. Effect of survey speed on TSD deflection slope.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

70

Figure 3.60. Effect of longitudinal profile variance on

TSD deflection slope.

reflective. Otherwise, the equipment functions reliably on a TSDs have the capability to routinely measure dynamic load-
wide range of surfaces, including rigid concrete. ing on the measuring load. The Danish Road Institute has been
This reliability has been illustrated by surveying on a length exploring this possibility with BASt in Germany using strain
of road with various types of road surfacing and in various gauges mounted on the rear axle of the trailer. The latest TSD,
conditions. This work established that the optics perform currently being constructed for the South African government,
better on lighter surfaces, such as jointed concrete, and least is said to include this technology.
well on new bituminous surfaces. This conclusion can be seen Some have argued that because the equipment has a suspen-
in Figure 3.61, where the data rate of the four Doppler lasers sion typical of other trucks, any additional deflection response
is shown for one particular survey along a 13-km (8.1-mi) caused by additional dynamic loads represents what normally
length of a U.K. motorway. The figure shows changes in data occurs at that point on the road and so provides a representative
rate (the rate at which laser signals are successfully returned estimate of the structural condition at that location. However,
from the road surface) over lengths of the jointed concrete knowledge of the dynamic load would provide a more complete
and a length of freshly laid thin surfacing in comparison to understanding of pavement behavior.
the rate returned from the old bituminous surface.
This phenomenon is likely due to optical properties of the Acceleration and Deceleration
different surfaces or surface condition. Although it is not fully
understood, it is important to identify such features because Longitudinal vehicle acceleration and deceleration are likely
they help to define the capabilities of the technology, and to affect the accuracy of deflection measurements, so oper-
such information will probably be employed in the quality ating limits for these parameters have been developed for
assurance of TSD data in the future. Information on the effect the U.K. HA TSD. When operating at normal survey speed,
of such parameters on RWD measurements is not available neither device requires traffic management. However, some
but is likely to be similar. form of traffic management will be necessary for operation
at slower speeds, as well as when operating in nonstandard
locations, such as in a U.K. motorway outer lane where trucks
Moisture
are not normally permitted.
Both devices use laser-based noncontact sensors that fail to
measure correctly when the road is damp or wet. The laser Acquisition and Operation Costs
reflection is degraded by the water on the surface.
Only the TSD is available commercially, at a cost between
$2 million and $2.5 million depending on the number of
Dynamic Loading
sensors requested. Although detailed operation and main-
The specific effect of dynamic loading has yet to be investigated tenance costs were not obtained for this study, experience
for either device. At present, neither the RWD nor any existing in the United Kingdom suggests that the cost of operating

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

71  

Exposed Concrete New bituminous surface

1200

1000

800

Data Rate [p/s]

600

400

P100 P200 P300 Pref Events

200

0
0 2000 4000 6000 8000 10000 12000
Chainage [m]

Figure 3.61. Effect of road surfacing type on data rate from

each TSD laser (p/s 5 number of successfully returned laser
pulses per second).

the TSD, not including the capital or maintenance costs, strong relationship to the peak deflections measured by other
is approximately $25 to $40 per mile. Adding the process- devices. The manufacturer, Greenwood A/S, has developed
ing cost is estimated to bring this to approximately $75 to an approximate relationship that can be fitted to the three
$90 per mile. offset slope measurements and thus enable estimation of the
surface curvature index of the pavement surface under load.
At present it is not possible to estimate the full deflection bowl
Measurement Capability from the current sensor configurations, but a device with more
sensors was delivered in 2010.
Bowl Shape Detail
Both devices currently measure in just one wheelpath in
The current RWD is designed to measure just the vertical between the two loaded wheels, which are mounted at a slightly
deflection close to the wheels in the outside wheelpath. This wider spacing than are standard truck dual wheels to enable
should be closely related to the maximum deflection response room for the measurement sensors. At present, both measure
of the pavement, although the location of this maximum will in the nearside (outside) wheelpath closest to the pavement
vary depending on the composition of the pavement and the outer edge in the countries in which they operate. However,
survey speed, due to any viscoelastic properties of the pave- measurement in both wheelpaths should be feasible with suit-
ment materials. The RWD as evaluated in this report did not able modifications and at an additional cost. The latest TSD
have the capability to measure the full deflection bowl, or a being constructed for the South African government is reported
sampled representation of it, as provided by the multiple sen- to have the capability of measuring in both wheelpaths.
sors fitted to most FWDs. However, recent modifications have At present the RWD operates with a 40-kN (9-kip) dual
added an additional sensor to the RWD at a second position wheel assembly load and the TSD operates with a 50-kN
farther away from the rear axle to provide such information (11-kip) load. Other loads can be employed relatively easily.
(see New Development to the Evaluated Equipment section).
At present the two fully operational versions of the TSD
Sampling and Reporting Intervals
measure the vertical velocity of the pavement response to
the dual wheel assembly loading at three offsets in front of Both devices sample the raw measurements frequently, the
the rear axle, which is then converted to deflection slope. RWD at around 2,000 Hz, equivalent to around 11 mm at
In the case of the Danish device, these offsets are 100 mm 80 km/h (0.4 in. at 50 mph), and the TSD at around 1,000 Hz,
(4 in.), 200 mm (8 in.), and 300 mm (12 in.). In the case of equivalent to around 22 mm at 80 km/h (0.8 in. at 50 mph).
the U.K. device, they are 100 mm (4 in.), 300 mm (12 in.), and However, there is much noise in the raw signal, so results are
756 mm (30 in.). The deflection sensors respond to velocity normally reported over much longer lengths. Some examples
(not displacement), which are converted to deflection slope, of different sampling lengths on the results are presented in a
and therefore cannot directly provide either the full deflection later section. Published material on the RWD suggests that the
or the maximum value. However as discussed later, the mea- device is suitable for measuring only on flexible and compos-
sured slopes closest to the axle, P100 and P300, have shown a ite pavements, normally presenting results at 160-m (0.1-mi)

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

72

TSD deflection
response

Height laser response

(5m transverse joint slots)

Figure 3.62. Longitudinal profile and deflection slope results from TSD survey on
jointed rigid pavement.

FWD LT % Bend TSD10K

10

200 8

6
150

Slope 100 (mm/m)

4
FWD LT(%)

100
Good

2
Moderate
Poor Poor
50
0
Poor Poor

0 -2
1 3 5 7 9 11 13 15 17 19
Chainage (m)

Figure 3.63. Raw TSD slope measured at 10 km/h (6 mph) and FWD joint load transfer efficiencies against
location.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

73  

intervals. The position with the TSD is similar except that results over a 90-m (300-ft) length. The red lower profile shows the
are normally presented at 10-m (33-ft) intervals. However, longitudinal height profile with spikes indicating the 5-m
some surveys have suggested that shorter length structural vari- slab joints. The upper blue trace shows the response of one
ations can be distinguished if the variability is sufficient. This TSD sensor with significant spikes at chainages of 2,827 and
has been discussed further in mathematical terms and examples 2,859, suggesting poor joint condition. In Figure 3.63, the
given earlier in the Denoising and Data Aggregation section. raw deflection slope profile of another section is shown in
comparison to load transfer efficiencies assessed with an
FWD. This preliminary investigation suggests that the TSD
Joint Load Transfer Efficiency
may have potential to assess the transfer efficiency at joints,
Limited measurements have been made on an unreinforced but further work is needed to make this a practical routine
jointed rigid pavement on a research track at a survey speed proposition. No information is yet available on the applica-
of just 10 km/h (6 mph). Figure 3.62 shows some results bility of either device to unsurfaced granular pavements.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Chapter 4

Example Applications and Recent Developments

This chapter presents a series of examples that illustrate poten- classified into severity categories such as very poor, poor, fair,
tial application of continuous deflection measuring technology. good, very good, and excellent. In general, different severity
It also identifies some recently developed devices and improve- levels would trigger different treatments to the pavement.
ments to the devices evaluated. Severity levels can be assigned to individual measurements;
however, this can have the following disadvantages:
Example Applications • Assigning a severity level to each reading can lead to sig-
Based on the data collected and the analysis performed in pre- nificant variations in the assigned severity levels at adjacent
vious sections, a number of example applications for the data points. This is not practical for pavement management pur-
produced by the continuous deflection measuring devices have poses, nor is it practical to apply different treatments at indi-
been developed. These applications illustrate the use of con- vidual points.
tinuous deflection measurements for network-level pavement • Each deflection (or slope) measurement consists of the actual
management. These applications are not meant to be compre- deflection (or slope) and a random error term. Therefore,
hensive, but rather are intended to show how deflections can a high level of uncertainty is associated with the severity level
be used to address issues applicable to modern pavement man- of each individual measurement. Suppose, for example, a
agement practices. given section is in fair condition. Because of random error
The applications presented within this section include using in measurements, deflection testing of this section can result
measurements obtained from continuous deflection measure- in readings that would fall in the poor, fair, or good category.
ment devices to divide pavement sections into homogeneous
segments, estimate the strain at the bottom of the asphalt layer, To address these shortcomings, the deflection (or slope)
determine the effective structural number (SN) of the pavement, profile can be segmented into uniform sections identified on
and identify relatively weak pavement sections as well as weak the basis of a statistical algorithm. The algorithm chosen to
pavement sections defined by absolute thresholds. The segmen- perform the segmentation is the circular binary segmentation
tation was performed using a statistically based binary segmen- (CBS) algorithm that tests for change-points in a given array
tation algorithm. The strain at the bottom of the asphalt layer, of data by using a maximal t-statistic with a permutation
along with the effective SN, was estimated by using the differ- reference distribution to obtain the corresponding P-value
ence between two deflection slope measurements from the TSD. (Venkatraman and Olshen, 2007). The approach taken by the
Identification of relatively weak pavement sections was used CBS algorithm is to compare the likelihood ratio of the null
to demonstrate the ability of the device to locate anomalies or hypothesis that there is no change in the data at a specific
locally weak sections within a network, whereas identification of point to the alternative that there is exactly one change-point
weak pavement sections on the basis of thresholds demonstrated at that location. The likelihood ratio is compared against the
the ability of the device to repeatedly identify weak sections. threshold value of the upper a-quantile of the distribution
(Olshen et al., 2004). The CBS algorithm searches for change-
points at every point within the array, and then returns each
Using Circular Binary Segmentation
change-point that has likelihood greater than the specified a.
to Identify Uniform Sections
The CBS algorithm, with a equal to 0.01 and the number
For pavement management purposes, pavement deflections of permutations equal to 10,000, was used to analyze homo-
(or slope measurements, in the case of the TSD) are often geneous sections for multiple runs of the TSD over the

74

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

75  

same section. This was completed for each site for 1-m aver- identify the peak that occurs between 1,200 m and 1,300 m, or
aging lengths. The values of a and the permutation number the peak at the end of the section. For Site C2, the minimum
were chosen on the basis of recommended values in Olshen length algorithm hides the detail of the peak at approximately
et al. (2004). 300 m for two of the runs with a 50-m (150-ft) minimum, and
Two sites are used to illustrate the segmentation: F1, which for four of the runs with a 100-m (330-ft) minimum length.
has the largest variation, and C2, which is one of the sites with This is expected because Site C2 is a site with very little struc-
the least variation. Figure 4.1 and Figure 4.2 present the seg- tural variability.
mentation of Sites F1 and C2, respectively. The analysis for The results of the segmentation indicate that the minimum
F1 yielded as many as 65 change-points. length of a pavement segment should be based on the variability
The segmentation presented in the figures was performed of the deflection profile or, in this case, the deflection slope
assuming no minimum lengths; thus if the variance changed profile. Practically, it would seem that a structurally variable
over very small sections, the algorithm identified the area as pavement segment would require more characteristic segments
the location of a change-point. This can be seen at Site C2 in to describe it and, therefore, require smaller specified minimum
Figure 4.2; the peak that is identified near 200 m (656 ft) is only lengths. However, it is not practical from a pavement man-
1 m (3.3 ft) long. In practical applications, a 1-m (3.3-ft) section agement standpoint for an agency to segment its network into
is too short and would be aggregated with the surrounding pavement sections that are very short. Some minimum length
sections, thus omitting the large peak. To address this, a post- should be specified. In reality, the choice of minimum length
processor was added to the CBS algorithm that combined sec- should be related to the normal maintenance practice of the
tions shorter than a specified length with surrounding sections highway administration involved.
on the basis of section length and average value.
The minimum length postprocessor was applied to the
Computation of Structural Health Indices
homo­geneous sections for Site F1 and Site C2, 1-m (3.3-ft)
averaged data for both 50-m and 100-m (150-ft and 330-ft) Pavement deflection measurements can be inputs into many
minimum lengths. The results for Site F1 can be seen in Fig- pavement condition assessment tools, including structural
ure 4.3. The figures seem to indicate that setting the minimum capacity indicators and tools to calculate the remaining service
length to 100 m does not provide adequate detail to clearly life of pavements. Therefore, it is believed that measurements

Figure 4.1. Homogeneous sections for Site F1, 1-m averages, Sensor 100.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

76

Figure 4.2. Homogeneous sections for Site C2, 1-m averages, Sensor 100.

Figure 4.3. Homogeneous sections for Site F1, 1-m averages and specified minimum length.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

77  

taken with the continuous deflection measuring devices can where

be surrogates for the FWD deflection values that are typi- eAC = strain at the bottom of the asphalt layer,
cally input into structural models. ARA reported results from HAC = thickness (mm) of the asphalt layer,
analyzing data gathered from 1,060 km (660 mi) of testing in SCI = surface curvature index (mm), and
Colorado and New Mexico and directly implementing the data BDI = base distress index (mm).
into a network-level decision process (Hausman and Steele,
These equations were developed for use with FWD measure-
2011). Among the benefits cited in the research was at least a
ments. They may not be appropriate for deflections measured
5% cost benefit from incorporating deflection data from the
by using a rolling wheel load moving at traffic speed, as in the
RWD into the network-level decision process.
case with the TSD.
It may be important to account for the consequences of
The equation to calculate the SCI from the TSD measure-
continuous deflection measurement vehicle dynamics on the
ments was based on the assumption that the slope at the
pavement response. Yoo and Al-Qadi (2007) discussed the effect
location of the applied load is zero. Because of the viscoelastic
of transient dynamic loads on pavement response in terms
nature of pavement, this assumption is not exactly true; but it is
of dynamic amplifications of strains and concluded that mass
expected to be a good assumption. The following equation was
inertia and damping forces generated by the transient dynamic
used to calculate the SCI from the TSD (trapezoidal integration):
loads should be considered for an accurate analysis of pave-
ment responses. Vehicle dynamics could play a significant role
SCI ( mm ) = 0.5 × ( P100 1, 000 ) × 100 mm + 0.5
in interpreting pavement response, especially when pavement
testing is being conducted on very rough roads. However, for × (( P100 + P 300 ) 1, 000 ) × 200 mm (4.3)
network-level analysis, which requires less detail than does
project-level analysis, the errors incurred by neglecting the where
vehicle dynamics are expected to be a small part of the mea- P100 = TSD measured slope deflection (mm/m) at 100 mm
surement. This is especially true when many measurements (3.94 in.), and
are averaged over a pavement section. P300 = TSD measured slope deflection (mm/m) at 300 mm
Given that the TSD does not measure pavement deflections (11.8 in.).
directly, some corrections may need to be made to the structural
The BDI is given by the following equation:
models to allow for the deflection slope to be input. However,
given the good level of comparability shown between the FWD BDI = D300 − D600 (4.4)
SCI (or BDI) and TSD SCI (or BDI) earlier in this report, the
structural capacity models based on difference between FWD
where
deflections should be applicable to the TSD slope measure-
D300 = deflection measured by the FWD 300 mm (11.8 in.)
ments, with some corrections.
from the center of the load plate, and
D600 = deflection measured by the FWD 600 mm (23.6 in.)
Asphalt Layer Strains from the center of the load plate.

The SCI can be directly calculated by using the TSD deflection The equation to calculate the BDI from the TSD deflection
slope measurements, as previously shown in this report. slope can be found with an equation similar to the equa-
This is because the SCI is based on taking the differences in tion used to calculate the SCI. The BDI is defined as the
measurements. Furthermore, the SCI calculated from FWD area under the curve defined by the deflection slope values
deflection measurements has been shown to be an excellent between 300 mm and 600 mm. The equation to calculate this
indicator of the strain of the asphalt layer in a flexible pavement area is the following:
(Xu et al., 2002; Thyagarajan et al., 2011). The relationships
developed for determining the strain for full depth pavements BDI ( mm ) = 0.5 × (( P 300 + P 600 ) 1, 000 ) × 300 mm (4.5)
and aggregate base pavements are presented in Equations 4.1
 P 300 − P 756 
and 4.2, respectively: P 600 =  × 300 mm + ( P 300 ) (4.6)
 300 mm − 756 mm 
Log ( εAC ) = 0.9977Log ( BDI ) + 3.3057 R 2 = 0.987 (4.1)
where
Log ( εAC ) = 0.5492Log ( SCI ) + 0.3850Log ( BDI ) P300 = TSD measured slope deflection (mm/m) at 300 mm,
P600 = TSD estimated slope deflection (mm/m) at 600 mm,
+ 0.7812Log ( HAC ) − 0.0017 HAC + 1.7353
and
R 2 = 0.994 (4.2) P756 = TSD measured slope deflection (mm/m) at 756 mm.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

78

Site F1 Evaluation of the HDM-4 Manual (Morosiuk and Riley, 2001). For illus-
To demonstrate the applicability of the methods described trative purposes, the method developed by Rhode (1994) has
earlier, the strain at the bottom of the asphalt layer was esti- been selected for the following analysis. This method for
mated for Site F1. As mentioned, Site F1 is a structurally vari- estimating the effective SN of a pavement is based on the pave-
able site with an estimated asphalt layer thickness of 75 mm ment thickness and the difference between the FWD center
(3 in.) and a relatively weak underlying cement stabilized layer. deflection and the deflection at 1.5 times the pavement depth.
The comparison of the SCI calculated from data gathered Using the difference in deflections between two points is advan-
by the TSD and FWD are presented in Figure 4.4, and the tageous for applying this method to the TSD because it is
BDI calculated from data from each device are compared in possible to integrate the deflection slope to find the difference
Figure 4.5. The relationship shown in each case was determined in deflection between the two points. The predictive model
using orthogonal regression, since both estimates contain mea- is based on the assumption that it is possible to estimate the
surement errors. These curves are similar to the ones presented deflection originating solely in the pavement structure know-
in the comparability between TSD and FWD test results but ing that 95% of the deflections measured on the surface of a
include only one individual road segment. pavement originate below a line deviating 34 degrees from
By use of the SCI, BDI, and thickness of the pavement, the the horizontal (Irwin, 1983). The steps for determining the
strain at the bottom of the asphalt layer was estimated along effective SN from FWD measurements are as follows:
the site for each measuring device. The comparison of the
strain at the bottom of the asphalt layer from the two devices 1. The FWD measurements should be normalized to 9,000-lb
is presented in Figure 4.6. load deflections.
These are very encouraging results considering the various 2. Determine the deflections at an offset of 1.5 times the
approximations, discussed earlier, that have been used in their pavement depth by using the following interpolation
derivation. formula:

Effective Structural Number ( x − B )( x − C ) ( x − A )( x − C )

D1.5 Hp = DA + DB
( A − B )( A − C ) ( B − A )( B − C )
There are many different ways of estimating the effective SN
(SNeff ) of a pavement from the results of nondestructive tests. ( x − A )( x − B )
+ DC (4.7)
A comprehensive review of this issue is published in Volume 6 (C − A )(C − B )

Figure 4.4. Site F1 comparison: TSD-calculated SCI versus FWD-calculated SCI.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

79  

Figure 4.5. Site F1 comparison: TSD-calculated BDI versus FWD-calculated BDI.

Figure 4.6. Site F1 comparison: Strain at the bottom of the asphalt layer estimated
by TSD and FWD.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

80

where calculated. The area under this curve defines the difference in
x = 1.5Hp, deflections between the two points between which the area is
Hp = depth of the pavement (mm), calculated. The calculation of this area was performed in a way
A, B, and C = points closest to x where the deflec- similar to the calculation of the SCI300, by using the assump-
tion is known, and tion that the deflection slope varies linearly throughout the
DA, DB, and DC = deflections at points A, B, and C deflection basin, and the deflection slope at the location of the
res­pectively. applied load is 0. The result is the SIP defined by Rhode et al.
3. Determine the structural index SIP of the pavement as (1994). With use of the SIP, the effective structural number
follows; for the section was estimated from the TSD measurements.
The results are presented in Figure 4.7.
SIP = D0 − D1.5 Hp (4.8) The results seem to indicate three distinct structural sections
within the site. These three sections were visually separated, and
where the mean section SNeff and upper 95% confidence interval
D0 = peak deflection under the 9,000-lb load of the data were plotted in the figure. Furthermore, the range
(microns), and of values estimated by the TSD seems to fall largely below
D1.5Hp = deflection at 1.5 times the pavement depth the range of expected values for pavement constructed with
(microns). these layers, as seen in Figure 4.8, as would be expected for
4. Determine the existing pavement SNeff as deteriorated pavement. The expected values were found by
SNeff = k1SIPk2 H pk3 (4.9) estimating the layer properties for each material defined in
Table 4.1 from appropriate literature sources.
where, for asphalt pavement, k1 = 0.4728, k2 = -0.4810,
k3 = 0.7581 (Rohde, 1994). Site F3
To further demonstrate the ability of the device to estimate the
Site F5 effective structural number of a section of flexible pavement,
Site F5 is a flexible site that was tested using both the TSD and the methodology was applied to Site F3, which was in good
FWD. The structure of Site F5 is shown in Table 4.1. The upper- condition. The structure of Site F3 is shown in Table 4.2.
bound and lower-bound layer coefficients are representative of The macadam is treated as an asphalt base layer.
values typically found in the design of new pavement sections Similar to the procedure for Site F5, the SIP was calculated
and were used to estimate initial SN values. The high modulus from the TSD deflection slope. The deflection slope at 1.5 times
roadbase is treated as an asphalt base layer. the pavement depth, or at 630 mm for Site F3, was estimated
To calculate the SIP from the TSD deflection slope, a similar with the following equation:
procedure to that used in estimating the SCI300 and BDI was
employed. First, the deflection slope at 1.5 times the pavement P 300 − P 756
P 630 = × 330mm + P 300 (4.11)
depth, or at 690 mm for Site F5, was estimated with the follow- 300mm − 756mm
ing equation:
After the deflection slope at 630 mm was estimated, the area
P 300 − P 756 under the curve defined by the deflection slope was calculated.
P 690 = × 390mm + P 300 (4.10) The result is the SIP defined by Rhode (1994). With use of the
300mm − 756mm
SIP, the effective structural number for the section was esti-
After the deflection slope at 690 mm (27.2 in.) was estimated, mated using the TSD. The results are presented in Figure 4.9,
the area under the curve defined by the deflection slope was along with the average and upper 95% confidence interval.

Table 4.1. Structure of Site F5

Thickness Upper-Bound Lower-Bound

Layer Material Name Date Laid (in.) Layer Coeff. SN 1 Layer Coeff. SN 2

4 Surfacing 05/28/2007 1.38 0.44 0.61 0.4 0.55

3 High modulus roadbase 05/28/2007 2.36 0.4 0.71 0.3 0.47
2 High modulus roadbase 05/28/2007 8.46 0.4 2.54 0.3 1.69
1 Type 1 granular 05/28/2007 5.91 0.12 0.71 0.1 0.59
Note: SN 1 = 5.65, SN 2 = 4.39.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

81  

Figure 4.7. SNeff estimated from the TSD for Site F5.

Figure 4.8. Effective SN from TSD for Site F5.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

82

Table 4.2. Structure of Site F3

Thickness Upper-Bound Lower-Bound

Layer Material Name Date Laid (in.) Layer Coeff. SN 1 Layer Coeff. SN 2

4 Hot rolled asphalt 7/10/1992 1.57 0.44 0.69 0.4 0.63

3 Heavy duty macadam 7/10/1992 2.36 0.4 0.87 0.3 0.71
2 Heavy duty macadam 7/10/1992 6.69 0.4 2.48 0.3 2.01
1 Type 1 granular 7/10/1992 5.91 0.12 0.71 0.1 0.59
Note: SN 1 = 5.02, SN 2 = 3.94.

The results from Site F3 seem to indicate that, structurally, it Identification of Weak Sections
is a relatively constant section. This matches what was reported
previously in this research. Furthermore, the range of values A critical need in network-level pavement management is the
estimated by the TSD fall generally within the range of expected ability to identify weak pavement sections within the network.
values, as seen in Figure 4.10. In many cases, the managing engineer may desire to identify
In general the results from estimating the effective SN at the weakest portions of the pavement relative to other pave-
the two sites indicate that the TSD gives a reasonable estimate. ment sections in the network. In a paper describing research
It is important to note that the model that was developed on moving pavement deflection testing devices, Rada et al.
by Rhode (1994) used in this approach should be further (2011) described this process as the ability of the device to
evaluated to determine if the constants are optimal for use identify pavement changes or anomalies at the network level.
with TSD data. There may be more-appropriate alternative They determined that an important application of the moving
approaches, as mentioned earlier in this section. However, deflection testing devices was the ability to determine relative
the methodology seems to produce results that reasonably changes of pavement structural condition with a network.
estimate the expected structural number values. In the case of this research, the question becomes whether

Figure 4.9. SNeff estimated by using the TSD for Site F3.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

83  

Figure 4.10. SNeff from TSD for Site F3.

the device consistently identified the same sections for each measured during the TSD testing using the procedure pre-
repeated set of measurements, and whether the same sections sented in the LTPP Guide to Asphalt Temperature Prediction
can be reproduced with a traditional measuring device like and Correction (FHWA, 1998).
the FWD. The weak values corresponded to an equivalent FWD center
deflection of greater than 995 microns (39 mils), the moderate
values corresponded to an equivalent FWD center deflection
U.K. TSD Site F1
of between 995 microns (39 mils) and 575 microns (23 mils),
Site F1 was a structurally variable site that could clearly be the strong values corresponded to an equivalent FWD center
split into several categories (i.e., weak, moderate, and strong) deflection of between 575 microns (23 mils) and 205 microns
after the site was segmented into homogeneous sections using (8 mils), and the very strong values corresponded to values
the CBS algorithm. This is shown in Figure 4.11, where weak less than 205 microns (8 mils).
sections were identified by a deflection slope value of over
2 mm/m, the strong threshold was set at 1.2, and the very
Discrimination Between Sections with
strong sections were identified by deflection values less than
Good and Poor Structural Capacity
0.5 mm/m. Setting the values for the weak, moderate, strong,
and very strong sections resulted in approximately 3% of An agency that already uses the FWD center deflection for
the data falling in the weak sections, 43% of the data in the network-level pavement management may wish to identify the
moderate category, 42% of the data in the strong category, and TSD deflection slope value that would correspond to agency
12% of the data in the very strong category. The areas identified thresholds for weak sections. If this is the case, the segments
as weak may be good candidates for localized patching. that are identified would no longer be considered different
To compare these sections to the data from the FWD, a relative to other pavement segments in the section, but weak
scatter plot was created comparing the center deflection of or strong on the basis of absolute thresholds. A benefit of
the FWD and the TSD P100 averaged over only the sections absolute thresholds is that they would most likely be mecha-
where FWD data were gathered (Figure 4.12). The FWD center nistically based, leading to a more objective classification of
deflections were corrected back to the temperatures that were pavement conditions.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

84

7.0

6.5
TSD Data Run 1 mean (10m), P100 TSD Data Run 2 mean (10m), P100
6.0
TSD Data Run 3 mean (10m), P100 TSD Data Run 4 mean (10m), P100
5.5

5.0

4.5
TSD P100 Sensor (mm/m)

4.0

3.5

3.0
Weak Sections
2.5

2.0
Moderate Sections
1.5

1.0
Strong Sections
0.5
Very Strong Sections
0.0
0 500 1000 1500 2000 2500
Chainage (m)

Figure 4.11. Site F1 segmented data.

4.5

4
TSD Deflection Slope (mm/m)

3.5

3 y = 0.0019x + 0.1086
R² = 0.702
2.5

1.5

0.5

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
FWD Center Deflection (microns)
Figure 4.12. Site F1 TSD P100 versus FWD-corrected center deflection.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

85  

Identification of Structurally Deficient Sections Table 4.3. Identification of Weak Sections

for Repeated Runs
To evaluate whether the device can consistently distinguish
structurally deficient sections from structurally adequate Percentage of
sections, the segmented data was compared with a set of Weak Sections
Identified by
threshold values. Typical threshold values for the RWD were
Site Average Length of Deficient Sections All Runs
obtained from a number of RWD demonstration reports
available at the FHWA website (FHWA, 2009). These thresholds F1 4,600 ft (1,400 m), 64% 85%
are based on a subjective rating of structural capacity (ARA, F3 No structurally weak sections identified
2007a and b). The threshold values were compared against a set F5 No structurally weak sections identified
of representative deflection sections, where the representative
F6 No structurally weak sections identified
deflection value was taken as the 95th percentile of readings
for one-tailed independent and identically distributed deflec- R2 525 ft (160 m), 13% 100%
tion profiles. The following threshold values were obtained R3 No structurally weak sections identified
from reports summarizing RWD demonstrations in Oregon, C1 No structurally weak sections identified
California, and Minnesota (note these values were used for
C2 No structurally weak sections identified
interstate and primary roads and might be different for sec-
ondary and local roads):

•
of 490 µm (19.3 mils). Similarly, the measurements between
Very good: <250 microns (10 mils);
• Good: 250–500 microns (10–20 mils); 95 m and 235 m (312 and 771 ft) are 533 µm, 536 µm, 544 µm,
• Fair: 500–750 microns (20–30 mils); and and 439 µm (21.0 mils, 21.1 mils, 21.4 mils, and 17.3 mils).
• Poor: >750 microns (30 mils). Setting the weak pavement threshold at 508 µm (20 mils) causes
these sections to seem not repeatable when identifying weak
The threshold values are in terms of equivalent 40 kN (9 kip) sections, even though the readings are all relatively close. This
FWD center deflections. In order to convert the TSD readings is shown in Figure 4.13, which shows repeated runs.
to equivalent FWD center deflection, an average relationship To overcome the shortcoming of applying set threshold
employed in the United Kingdom was used. Furthermore, only values to analyze whether the device could repeatedly identify
the 10-m (33-ft) averaged and segmented measurements were the same sections, an analysis of the number of locations in
segmented on the basis of the thresholds. The 1-m (3.3-ft) which identified uniform sections were within a certain range
data were deemed too noisy to provide consistent section- of each other was performed. The ranges chosen were 125 µm
ing on this basis, whereas the 10-m (33-ft) data maintained (5 mils), 250 µm (10 mils), and 500 µm (20 mils). Table 4.4
adequate detail while sufficiently reducing the noise level in summarizes results from this analysis.
the measurements. With the exception of Site F1, at least 98% of the repre-
sentative deflections are within 125 microns (5 mils) of each
Using Thresholds for Repeated other and the percentage is 100% for four of the tested sites.
Identification of Weak Sections Recall that Site F1 is a structurally variable site with large
peaks in the deflections at approximately 1,300 m and 2,100 m
A structurally deficient threshold of 500 microns (20 mils) (0.81 and 1.30 mi). The readings that are not within a 508-µm
was used to define structurally weak sections. This threshold (20-mil) range of each other occur at the locations of the large
value is based on demonstration reports from Connecticut peaks and are probably a consequence of the change-points in
(ARA, 2007b) and Indiana (ARA, 2005a). Table 4.3 summarizes the segmentation data occurring at slightly different locations
the number of sections with low structural capacity at each test for each run.
site, as well as the percentage of repeated runs that identified
the same structurally weak sections. Summary of Example Applications
In summary, the potential uses for the RWD and TSD mirror
Identification of Similar Sections
the current applications for the FWD network-level measure-
Setting a structurally weak threshold as a standard number ment. The ability to obtain deflections directly from RWD
may not provide adequate information about the repeated testing has been shown to provide cost savings in network-level
identification of weak sections. For example, for Site F1, the testing schemes. The TSD deflection slopes can be translated
measurements between 584 m (1,916 ft) and 694 m (2,277 ft) into the difference in deflection points, which allows for
for the repeated runs are 484 µm, 594 µm, 419 µm, and 439 µm many useful indices (i.e., the SCI and BDI) to be determined
(19.9 mils, 23.4 mils, 16.5 mils, and 17.3 mils) with an average from TSD testing. The difference in deflection points is used

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

86

140

120

Equivalent FWD Center Deflection (mils)

100

80

60

40

20
Structurally Weak Threshold

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Chainage (m)

Figure 4.13. Structurally weak threshold applied to representative

deflection sections at Site F1 for three different runs.

in many relationships including determining the strains at described in the following section. However, as this information
the bottom of the asphalt layer and determining the effec- was discovered after commencement of the study, it was not
tive structural number of a section of pavement. In terms of included in Phase II of the project, data acquisition and com-
network-level testing, each device has exhibited the ability to parison. In addition, both the RWD and the TSD have been
identify relative differences in the pavement structural condi- developed further since Phase I was completed. Therefore,
tion, which is useful for pavement management applications. a description of these recent developments is presented.

New Available Equipment

Current and Future
Developments Eyztek Continuous Runway Load Deflection
Evaluation Methodology
Since Phase I of this study was completed, the research team
has discovered one additional device that might meet the In 2006, Zybron Inc. (now Eyztek Inc.) was awarded a Phase 1
original requirements of this study. This equipment is briefly Department of Defense Small Business Innovative Research

Table 4.4. Identification of Similar Sections

Readings Within Readings Within Readings Within Maximum Range Range of

125-micron 250-micron 500-micron for All Locations All Readings
Site (5-mil) Range (10-mil) Range (20-mil) Range microns (mils) microns (mils)

F1 77% 95% 95% 1,712 (67.4) 3,226 (127)

F3 100% 100% 100% 74 (2.9) 168 (6.6)
F5 99% 100% 100% 137 (5.4) 201 (7.9)
F6 100% 100% 100% 109 (4.3) 152 (6.0)
R2 100% 100% 100% 112 (1.2) 803 (31.6)
R3 100% 100% 100% 112 (4.4) 112 (4.4)
C1 98% 100% 100% 30.5 (7.1) 269 (10.6)
C2 99% 100% 100% 236 (9.3) 251 (9.9)

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

87  

(a) (b)
Source: Arora et al., 2006.

Figure 4.14. Eyztek Inc. lighter weight rolling weight deflectometer prototype: (a) frame view
and (b) laser mounting beam.

award to develop an air-droppable lighter weight rolling approximated the values obtained from the center point of
weight deflectometer for the U.S. Army that can rapidly collect an FWD.
continuous data by using noncontact sensor technologies. Various refinements have been made to the design since 2003;
This was followed by a second award in 2008 to develop the the most significant was in 2009 when the laser devices were
concept further into a more robust prototype incorporating moved from a beam below the vehicle into the truck body and
ground-penetrating radar and thermal scanners for addi- mounted in a temperature-controlled enclosure (Figure 4.15).
tional in-depth information capability, such as estimating At that time, provisions were made to increase the number of
layer thicknesses and voids. The manufacturers consider that laser devices to provide two simultaneous traces of deflections
simultaneous measurement of deflection and pavement thick- in the pavement due to the loading, one close to the center
ness is essential for correct interpretation of results. The lighter of the load and one at a distance of 450 mm (18 in.) from the
weight RWD (Figure 4.14) can survey at speeds up to 110 km/h center. It is thought that these two traces will provide measure-
(70 mph) within the normal traffic flow, and traffic control is ments that will approximate two similar simultaneous readings
not normally necessary. As yet, no publications are available using an FWD.
on the outcome of these projects. Some further information
is available on the company website (Eyztek, 2011). Traffic Speed Deflectometer
Since the completion of Phase I, two further TSDs have been
Developments to Evaluated Equipment built to a modified design, the second-generation version, and
Rolling Wheel Deflectometer delivered to their owners. At the time this report was written, a
fifth device was under construction for the South African gov-
As stated earlier in the report, the deflection measuring ernment and was due to be delivered in July 2012. The modi-
system of the RWD consists of a laser-based measuring device fied design, which is shown diagrammatically in Figure 4.16,
mounted on a modified tractor trailer for measuring pavement has the following improved features:
deflections under the moving load of the truck. The load is set
to 80 kN (18 kips) on a single rear axle when measured stati- • The trailer is now custom built specifically for this device,
cally. The device was developed from the late 1990s through rather than being based on a container.
2003, when the first practical testing was conducted. • Rather than the three velocity-measuring lasers and one
The design of the device uses a series of industrial laser reference laser, the latest devices can be fitted with up to
measuring devices (Selcom) that provide measurements seven lasers.
at 2 kHz, and then averages these results over a selected sec- • The measurement of survey speed, which is a vital part
tion to eliminate the effects of texture on the readings. The of the measurement process, can now be accomplished
original device provided a single trace of measurements that with a more robust system. This device uses a motorcycle

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

88

(a) (b)

Figure 4.15. 2009 updated FHWA RWD: (a) side picture and (b) enclosed mounting for six laser sensors.

method of calibration. At the other extreme, the lasers can

be positioned so that they are on either side of the loaded
wheel. This movable beam also enables multiple test runs
with different intermediate positions, rather than single
runs with more sensors.

This revised configuration was first fitted to a device that

Figure 4.16. Diagram of second-generation version was delivered to the Italian concession agency National Auton-
of the TSD. omous Roads Corporation (ANAS S.p.A.) in June 2010. This
TSD is fitted with seven laser velocity sensors, one for refer-
ence and six for measurements. This device and its movable
wheel rather than a bicycle wheel as the main measurement measurement beam are shown in Figure 4.17.
component. As of December 2011, this device had surveyed around
• The mounting beam for the laser sensors is now mounted 5,000 km (3,100 mi) for the purposes of research, maintenance
on longitudinal rails so that its position can be varied. At identification, and quality control of new construction at sur-
one extreme, this enables the lasers to be positioned mid- vey speeds between 40 and 80 km/h (25 and 50 mph). The
way between the loaded wheels, providing an improved operators report a good correlation with FWD measurements

(a) (b)

Figure 4.17. Italian second-generation TSD: (a) side view and (b) movable measurement beam.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

89  

As of December 2011, this device had covered only around

20 km (12.5 mi) in comparison tests with the FWD.

Possible Improvements
On the basis of information collected, and especially given
the recent enhancement to the devices, it is assumed that the
existing systems can record the accurate location of the mea-
surements, measure the survey speed accurately, and measure
the surface temperature of the pavement accurately. Thus, the
main improvements that may help improve the information
obtained from the measurements and widen the range of
possible applications, include the following:
Figure 4.18. Polish second-generation TSD.
• Provide a more complete deflection bowl shape (e.g., by
adding additional sensors),
and a more practical calibration procedure, but no reports • Provide sufficient quality of measurement signals that local
have yet been published on their work. structural deterioration can be reliably identified (i.e., down
The fourth TSD was constructed for the Pavement Diagnos- to 1-m features), thus enabling the potential for use at a
tic Division of the Road and Bridge Research Institute (IBDiM) project or scheme level,
in Warsaw, Poland. The device is fitted with five laser velocity • Provide pavement-layer thickness measurement capability,
sensors, one for reference and four for measurements. The such as by adding ground penetrating radar equipment, and
device (Figure 4.18) was delivered to Poland in April 2011. • Measure the dynamic load on the loading wheel assembly.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Chapter 5

Conclusions and Suggested Research

The SHRP 2 R06F project described in this report evalu- From this group and on the basis of information collected in
ated current technologies implemented in different types the literature review, the research team identified two devices
of continuous deflection measuring devices, identified the as the most promising to deliver the information needed by
most promising devices for effectively supporting pavement the users under operating conditions compatible with the
management decisions, evaluated the capabilities of these SHRP 2 objectives. These devices are the RWD and the TSD.
devices, and identified and illustrated applications that can The vibratory devices were eliminated from consideration
be useful for supporting pavement management. This section because either they do not apply loads similar to that of a
summarizes the main findings and conclusions from the heavy vehicle or they measure at very low speed. The RDT
project. and ARWD were also eliminated from further consideration
because the existing prototypes have been decommissioned
Findings or reassigned to other uses. The IDM system appears very
promising in the detailed evaluation; however, a fully opera-
Equipment Selection tional prototype is not yet available and at present does not
There are a series of continuous deflection devices that can meet the survey speed requirement.
collect data at intervals of approximately 300 mm (1 ft) or
smaller by using load levels typical of truck loading (i.e., 40 to User Needs
50 kN [9 to 11 kips] per wheel or load assembly). These include
three main types of devices: The majority of agencies perform at least some deflection
testing using the FWD. Most testing is performed to support
• Laser-based devices that measure the deflection below project-level decisions and only a small number of agencies
an actual moving truck load—including the traffic speed (five) have incorporated deflection data into their pavement
deflectometer (TSD), the rolling wheel deflectometer management system. Potential users in general agree that the
(RWD), the road deflection tester (RDT), and the airfield main advantage of a continuous measuring device would be
rolling weight deflectometer (ARWD); for supporting network-level decisions. The assessment of user
• Devices that apply a vibratory load—including the Portan­ needs suggests the following:
cemetre, the Measuring Ball, and the rolling dynamic deflec-
tometer (RDD); and • Important parameters that users indicated should be con-
• The image deflection measurement (IDM) device, which uses sidered in the evaluation of the equipment include: speed
image analysis methods to determine pavement deflections (safety), repeatability, accuracy (and feasibility of estab-
under loading, which represents a very promising technology lishing correlations with existing technologies, such as the
but is still in the early stages of development. FWD), equipment cost, ease of operation, customer ser-
vice (availability of service and maintenance), ease of use of
Only the devices in the first group are capable of surveying the data collected, availability of software for interpretation
without the need for traffic control. The vibratory devices oper- of the results, reliability, size of the vehicle, relevance of the
ate at walking pace, and the IDM device was still being tested information (e.g., use in MEPDG/DARWin-ME), and past
in a stationary mode at the time of the evaluation. experience.

90

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

91  

• While responses to the initial survey suggested users would Except for a single run for Site C1, TSD repeated runs
like to be able to collect continuous pavement response data did not have a statistically significant bias.
to support project-level decisions, follow-up interviews 44 A method of calculating repeatability using a single TSD
showed that the respondents understand the current limita- run was also evaluated and found to give results that are
tion of the technologies and agree that network-level appli- in agreement with the method using repeated runs for
cations are more likely in the near future. Furthermore, measurements averaged over 1-m (3.3-ft) sections.
respondents agreed on the need of pavement structural data 44 The correlation of repeated runs varies from 0.05 to 0.89
to support network-level PMS decisions. for 1-m (3.3-ft) averaging, 0.15 to 0.98 for 10-m (33-ft)
• At the network level, the primary application of the con- averaging, and 0.55 to 0.99 for 100-m (330-ft) averaging.
tinuous deflection device would be to do the following: The correlation was found to be significantly affected by
44 Help identify weak (i.e., structurally deficient) areas that the range of measurements and is therefore not recom-
can be then investigated further at the project level; mended as a measure of repeatability.
44 Provide network-level data to calculate a structural health 44 The repeatability of the RWD was evaluated with data
index that can be incorporated into a PMS; and collected in Virginia and was evaluated to be 51.4 µm
44 Differentiate sections that may be good candidates for (2.0 mils). The RWD repeated runs had a statistically
preservation (those with good structural capacity) from significant bias of 10.9 µm (0.43 mils) on average.
those that would likely require a heavier treatment (those • Comparability
44 The analysis of TSD comparability with the FWD
showing structural deterioration or deficiencies).
(using TSD measurements averaged over a 10-m interval)
showed that there are two distinct relationships between
Detailed Technology Evaluation TSD deflection slope and FWD deflection depend-
ing on pavement type (one for flexible and composite
To assess the capability of the continuous deflection tech-
pavements, another for rigid pavements). This could be
nology to support network-level pavement management
expected, as two different quantities (deflection slope and
decisions, the experimental plan evaluated relatively long
deflection) are measured by each device. These quantities
selected evaluation routes with uniform and variable struc-
seem to be affected differently by the pavement type.
tural conditions. These network-level sites included flexible,
44 To better evaluate the comparability of the TSD with
composite, and rigid pavement sections. All attempts were
FWD, measurements from each device were used to
made to include subsections with good, fair, and poor func-
calculate the surface curvature index (SCI) and the base
tional conditions within each of these pavement types. These
distress index (BDI), quantities that can be obtained from
routes included segments that were measured repeatedly, each device. In this case, the relationship between the
several times in succession, in a single day, and with the FWD. quantities measured using the FWD and those measured
The evaluation included pavement sections on different types of using the TSD was the same for all pavement types.
pavement, using various operational conditions, and included Furthermore, the relationship between the indices mea-
reference FWD deflection testing equipment where possible. sured by the TSD and those measured by the FWD was
The analysis of the collected data resulted in the following reasonably close to the equality line, which is encouraging
observations: (although this might be a subjective opinion). However,
there is a significant variation as well as bias in this
• Repeatability relationship. For example, for an average SCI or BDI value
44 Repeatability of TSD slope for different averaging lengths of 300 µm (11.8 mils), the bias was 30 µm (1.18 mils)
was a mean of 0.089 mm/m with a range from 0.065 to (FWD values lower than TSD values) and the compara-
0.201 mm/m for 10-m (33-ft) averaging, and a mean of bility was 380 µm (15.0 mils).
0.028 mm/m with a range from 0.022 to 0.114 mm/m for 44 The analysis of RWD comparability with FWD (using
100-m (330-ft) averaging. Except for Site F1, the repeat- RWD measurements averaged over a 160-m [0.1-mi]
ability of the TSD evaluated with measurements at the interval obtained from New Mexico) was evaluated in
different tested sites was found to be similar and rela- terms of the normalized differences. This number was
tively independent of deflection slope value. For Site F1, found to be relatively unchanged as a function of deflec-
the measurement error (and therefore repeatability) was tion. The bias was 23.2% (FWD deflections lower than
found to be dependent on the deflection slope value. This RWD deflection), and the repeatability was 129.2%
could be due to the fact that sites with high deflection (range of [-86.4%; 42.8%]). Note that for both the TSD
slope values are deteriorated pavement sections and have and the RWD comparison with the FWD, FWD results
more variability due to distresses (cracking and rutting). tended to be lower than TSD and RWD results.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

92

The repeatability of at least one continuous deflection mea- • Collect deflection measurements or indices that are com-
suring device (i.e., the TSD) can be considered adequate for parable to those collected by traditional measurement
network-level pavement management applications. The analy- devices such as the FWD; and
sis also showed that the device can collect deflection measure- • Provide measurements that can be used for supporting some
ments and indices that are comparable to those collected by of the most critical network-level applications identified by
traditional measurement devices such as the FWD. Additional the potential users and possibly assist project-level opera-
tests are needed to fully assess the repeatability and compara- tions by identifying localized structural deterioration.
bility of the RWD.
However, the technology is only just maturing. Future research
should be conducted to further assess the measurement
Example Applications
capabilities of these devices, the usefulness of the collected
The example applications demonstrated the ability of con- data, and the best way to interpret measurements from such
tinuous deflection measurement devices to estimate many devices. Enhancements to the devices that may help improve
parameters important to modern pavement management the quality of information obtained from measurements and
applications: widen the range of possible applications include providing a
more complete deflection bowl shape, enhancing the quality
• The results of the binary segmentation indicated that the of the measurement signal to achieve higher spatial resolution,
most appropriate minimum length for a pavement section providing pavement-layer thickness measuring capability, and
depends on the variability of the deflection profile, or the measuring the dynamic load on the loading wheel assembly.
deflection slope profile. Structurally variable pavement
sections required more characteristic segments to describe Recommendations
them; thus, some parameters were required to maintain for Implementation
minimum homogeneous segment lengths.
• The strain at the bottom of the asphalt layer was estimated This section summarizes the main products of the project
and provides some recommendations on how to move the
by using measurements obtained from testing using the
technology forward. Proposed actions include dissemination
FWD and the TSD. The comparison of the results yielded
efforts, implementation activities, and research needs.
a relationship between the devices’ output that approached
equality. The effective structural number (SN) was esti-
mated using measurements obtained from TSD testing, Value Added
and broadly matched the expected SN calculated from the
Although deflection testing is widely used by state DOTs at
layer composition and surface condition.
the project level, it is much less often used at the network
• The identification of relatively weak pavement sections
level. FWD testing is the prevalent technology. It allows only
was used to demonstrate the ability of the device to locate stationary measurements, which disrupt traffic flow, limit the
anomalies or locally weak sections within a network. The number of measurements, and cause safety issues to both the
results showed that relatively weak pavement sections could operators and the public using the highway. Although most
be consistently identified using repeated sets of measure- state DOTs would like to incorporate deflection testing into a
ments. The identification of weak pavement sections on the pavement management program, only a small number (five,
basis of thresholds demonstrated the ability of the device to according to our survey) have done so. The main concerns for
repeatedly identify the same weak sections. the adoption of a continuous deflection device, voiced by the
respondents, are accuracy, cost, and safety. This project focused
Conclusions on evaluating the accuracy of continuous deflection devices
and their application for supporting network-level pavement
Continuous deflection devices have become a valuable tool in management. This technology allows collecting data at highway
pavement analysis and management. Particularly promising speed, greatly improving the safety of collection personnel.
devices include the TSD and the RWD, due to their ability to As part of the research process, the research team performed
measure at traffic speed. The study performed in this project the following:
has demonstrated that at least one of the continuous deflection
measurement devices can do the following: • Conducted a survey and follow-up interviews with state
DOT personnel to request feedback about the need, poten-
• Provide repeatability for network-level pavement manage- tial uses, and value of continuous deflection measurements
ment applications; to their agencies; and

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

93  

• Tested and evaluated continuous deflection devices to eval- Implementation Plan

uate their accuracy and ability to identify homogeneous
Audience and Expertise for this Product Application.
sections and structurally weak sections. A major aspect,
when possible, was to compare these devices with FWD The audience for the research products is made up of the
test results (the current industry standard). stakeholders who are involved in the design, evaluation, and
management of roadways, including the following:
Dissemination Efforts
• Public sector (project owners). The main audience includes
Throughout the course of the project, the research team has state DOTs, FHWA, AASHTO, and local governments.
developed several journal and conference papers, presenta- Within these agencies, groups of interest include the divi-
tions, and workshops, including the following: sions or sections responsible for pavement assessment,
management, maintenance, evaluation, and rehabilitation
• Use of Continuous Deflection Measurement for Network- design.
Level Pavement Structural Assessment, a poster presented • Private sector (consultants, road operators, and road
at the Developing a Research Agenda for Transportation constructors). Groups interested in the research products
Infrastructure Preservation and Renewal Conference held include pavement design and management consultants,
in Washington, D.C., in November 2009. road operators (toll road operators, concessionaires, etc.),
• SHRP 2 R06F Development of Continuous Deflection Device road construction companies, and manufacturers of asset
Project Update, a presentation at the annual meeting of management data collection equipment.
the TRB Committee AFD80 Strength and Deformation
Characteristics of Pavement Sections, January 11, 2010.
•
Recommendations and Future Research
Use of Continuous Deflection Measurements for Network-
Level Pavement Analysis, a presentation at the Pavement The SHRP 2 R06F project summarized in this report assessed
Evaluation 2010 conference, held October 24 through 27 the capability of the continuous deflection measuring tech-
in Roanoke, Virginia. nologies for supporting pavement management decisions.
• Mini-workshop on continuous deflection measurement at However, the conclusions reached were limited by the lack of
the Pavement Evaluation 2010 conference, held October 24 any side-by-side testing of the two selected technologies on a
through 27 in Roanoke, Virginia. The workshop helped representative range of pavements and soils. Moreover, adop-
capture feedback from prospective users. tion of these technologies would be further facilitated by the
• SHRP 2 R06F Development of Continuous Deflection following recommended research (see Appendix A for more
Device Project Update, a presentation at the annual meeting complete descriptions of these projects):
of the TRB Committee AFD80 Strength and Deformation
Characteristics of Pavement Sections, January 2011. • Side-by-side testing of the RWD and TSD. This study has
• Evaluation of Traffic Speed Continuous Deflection Devices, shown the limitation of comparing equipment on differ-
presented at the 91st Meeting of the Transportation Research ent sites. Therefore, side-by-side field tests with the latest
Board and recommended for publication in the Transporta- versions of the selected technologies at a variety of loca-
tion Research Record: Journal of the Transportation Research tions and with a range of pavement designs are needed to
Board. definitively assess the capabilities of the individual devices
• Analyzing Repeatability of Continuous Deflection Device for network- and project-level applications.
Measurements, presented at the 91st Meeting of the Trans- • Developing tools for using the technology to support
portation Research Board. network-level pavement management business functions.
• Estimation of Pavement TSD Slope Measurements Repeat- This project would further develop and demonstrate pro-
ability from a Single Measurement Series, presented at the cedures and tools to convert the equipment measurements
91st Meeting of the Transportation Research Board. to engineering parameters that can be used to support
• Two contributions to the Workshop on Continuous Deflec- network-level pavement management decisions. As illus-
tion Measurements for Highway Infrastructure Assess- trated in Chapter 4, these applications can include screening
ments, held during the 91st Meeting of the Transportation sections that may be candidates for rehabilitation (and thus
Research Board: require project-level investigation) or candidates for pave-
44 SHRP 2 Research Update on Continuous Deflection ment preservation. The effort would include using maxi-
Study; and mum deflection equivalents (already available) to compute
44 Deflectograph and Traffic Speed Deflectometer: United a structural indicator, as well as using extra information
Kingdom Experience. that should be available from the two lasers on the latest

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

94

RWD and the three or more lasers on the TSD to estimate the absolute deflection. This effort would require design-
parameters that pavement engineers require for pavement ing an experiment and instrumenting pavement sections
management use (i.e., modified SN or SCI). with accelerometers and strain gauges (or use existing
• Assessing the potential for using the output from the selected instrumentation, e.g., at a research facility, possibly taking
equipment to provide advice on structural assessment advantage of those where relevant existing instrumentation
requirements (e.g., rehabilitation or surface treatment). might be available) to measure the in situ absolute deflection
• Verifying the accuracy of the testing equipment by con- and strain response of the survey equipment to try and verify
ducting measurements on instrumented sections to measure its accuracy.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

References

Altman, D. G., and J. M. Bland. 1983. Measurement in Medicine: the Bland, J. M., and D. G. Altman. 2003. Applying the Right Statistics:
Analysis of Method Comparison Studies. The Statistician, Vol. 32, Analyses of Measurement Studies. Ultrasound Obstet Gynecol. Vol. 22,
No. 3, pp. 307–317. pp. 85–93.
Applied Research Associates (ARA). 2007a. Rolling Wheel Deflectometer Bland, J. M., and D. G. Altman. 1986. Statistical Methods of Assessing
(RWD) Demonstration for the Kansas Department of Transportation. Agreement Between Two Methods of Clinical Measurement. The
FHWA, U.S. Department of Transportation, Topeka, Kans. Lancet. Vol. 327, No. 8476, pp. 307–310.
ARA. 2007b. Rolling Wheel Deflectometer (RWD) Results for the Con- British Standards Institution. 1979. Precision of Test Methods, Part 1:
necticut Department of Transportation. FHWA, U.S. Department of Guide for the Determination of Repeatability and Reproducibility for
Transportation, Newington, Conn. a Standard Test Method. BS 5497-1. London.
ARA. 2007c. Rolling Wheel Deflectometer (RWD) Results for the California Brown, L. D., and M. Levine. 2007. Variance Estimation in Nonpara-
Department of Transportation (Caltrans). FHWA, U.S. Department metric Regression via the Difference Sequence Method. The Annals
of Transportation, Sacramento, Calif. of Statistics. Vol. 35, No. 5, pp. 2219–2232.
ARA. 2006. Rolling Wheel Deflectometer (RWD) Testing for the Minne- Craven, P., and G. Wahba. 1979. Smoothing Noisy Data with Spline
sota Department of Transportation. www.fhwa.dot.gov/pavement/ Functions: Estimating the Correct Degree of Smoothing by the
management/rwd/mn/#LinkTarget_1664. Accessed January 2010. Method of Generalized Cross-Validation. Numerische Mathematik,
ARA. 2005a. Rolling Wheel Deflectometer Pilot Demonstration in Indi- Vol. 31, No. 4, pp. 377–403.
ana. www.fhwa.dot.gov/pavement/management/rwd/in/index.cfm. Davidian, M., and R. J. Carrol. 1987. Variance Function Estimation.
Accessed January 2010. Journal of the American Statistical Association. Vol. 82, No. 400,
ARA. 2005b. Rolling Wheel Deflectometer (RWD) Testing on the Natchez pp. 1079–1091.
Trace Parkway. www.fhwa.dot.gov/pavement/management/rwd/nt. Diefenderfer, B. K. 2010. Report No. VTRC 10-R5: Investigation of the
Accessed January 2010. Rolling Wheel Deflectometer as a Network-Level Pavement Struc-
Arora, J., V. Tandon, and S. Nazarian. 2006. FHWA/TX Report No. 06/0- tural Evaluation Tool. Virginia Transportation Research Council,
4380-1: Continuous Deflection Testing of Highways at Traffic Speeds. Charlottesville.
Center for Transportation Infrastructure Systems, University of Texas Eubank, R. 1999. Nonparametric Regression and Spline Smoothing, 2nd ed.
at El Paso. Marcel Dekker, New York.
Baltzer, S. 2009. Three Years of High Speed Deflectograph Measurements Eyztek. 2011. Rolling Weight Deflectometer. http://eyztek.com/products/
of the Danish State Road Network. Presented at 8th International RollingWeightDeflectometer.htm. Accessed December 2011.
Conference on the Bearing Capacity of Roads, Railways, and Airfields, Ferne, B., P. Langdale, N. Round, and R. Fairclough. 2009a. Development
Champaign, Ill. of a Calibration Procedure for the U.K. Highways Agency Traffic Speed
Baltzer, S., D. Pratt, J. Weligamage, J. Adamsen, and G. Hildebrand. Deflectometer. Presented at 88th Annual Meeting of the Transporta-
2010. Continuous Bearing Capacity Profile of 18,000 Km Australian tion Research Board, Washington, D.C.
Road Network in Five Months. Presented at 24th ARRB Conference: Ferne, B., P. Langdale, N. Round, and R. Fairclough. 2009b. Development
Building on 50 Years of Road and Transport Research, Melbourne, of the U.K. Highways Agency Traffic Speed Deflectometer. Presented
Australia. at 8th International Conference on the Bearing Capacity of Roads,
Bay, J. A., K. H. Stokoe, N. T. McNerney, S. Suttisak, D. A. Van Vleet, Railways, and Airfields, Champaign, Ill.
and D. K. Rozycki. 2000. Evaluation of Runway Pavements at the FHWA. 2009. Rolling Wheel Deflectometer Network-Level Pavement
Seattle–Tacoma International Airport Using Continuous Deflec- Structural Evaluation, Measuring Deflection at Highway Speeds.
tion Profiles Measured with the Rolling Dynamic Deflectometer. FHWA, U.S. Department of Transportation. www.fhwa.dot.gov/
Presented at 79th Annual Meeting of the Transportation Research pavement/management/rwd.
Board, Washington, D.C. FHWA. 1998. FHWA-RD-98-085: LTPP Guide to Asphalt Temperature
Bland, J. M., and D. G. Altman. 2007. Agreement Between Methods of Prediction and Correction. FHWA, U.S. Department of Transportation.
Measurement with Multiple Observations per Individual. Journal of www.fhwa.dot.gov/publications/research/infrastructure/pavements/
Biopharmaceutical Statistics. Vol. 17, No. 4, pp. 571–582. ltpp/fwdcd/delta.cfm#delta36. Accessed August 23, 2011.

95

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

96

Flintsch, G. W., and K. K. McGhee. 2009. NCHRP Synthesis of Highway Olshen, A. B., E. Venkatraman, R. Lucito, and M. Wigler. 2004. Circular
Practice 401: Quality Management of Pavement Condition Data Col- Binary Segmentation for the Analysis of Array-Based DNA Copy
lection. Transportation Research Board of the National Academies, Number Data. Oxford Journal of Biostatistics, Vol. 5, No. 4, pp. 557–572.
Washington, D.C. Rada, G., S. Nazarian, J. Daleiden, and T. Yu. 2011. Moving Pavement
Gasser, T., L. Sroka, and C. Jennen-Steinmetz. 1986. Residual Variance Deflection Testing Devices: State of the Technology and Best Uses.
and Residual Pattern in Nonlinear Regression. Biometrika, Vol. 73, Proc., 8th International Conference on Managing Pavement Assets,
pp. 625–633. Santiago, Chile.
Gedafa, D., M. Hossain, R. Miller, and D. A. Steele. 2008. Network Rohde, G. 1994. Determining Pavement Structural Number from
Level Pavement Structural Evaluation Using Rolling Wheel Deflec- FWD Testing. In Transportation Research Record: Journal of the
tometer. Presented at 87th Annual Meeting of the Transportation Transportation Research Board, No. 1448, TRB, National Research
Research Board, Washington, D.C. Council, Washington, D.C., pp. 61–68.
Gedafa, D., M. Hossain, R. Miller, and T. Van. 2010. Estimation of Simonin, J.-M., D. Lievre, S. Rasmussen, and G. Hildebrand. 2005.
Remaining Service Life of Flexible Pavements from Surface Assessment of the Danish High Speed Deflectograph in France.
Deflections. Journal of Transportation Engineering, Vol. 136, No. 4, Proc., 7th International Conference on the Bearing Capacity of Road,
pp. 342–352. Railways, and Airfields. Norwegian University of Science and Tech-
Greenwood Engineering A/S. Traffic Speed Deflectometer. www.greenwood nology (NTNU), Trondheim, Norway.
.dk/tsd.php. Accessed January 2012. Simonin, J.-M., L.-M. Cottineau, V. Muzet, C. Heinkele, and Y. Guillard.
Hall, P., J. Kay, and D. Titterington. 1990. Asymptotically Optimal 2009. Deflection Measurement: The Need of a Continuous and Full
Difference-Based Estimation of Variance in Nonparametric Regres- View Approach. Proc., 8th International Conference on the Bear-
sion. Biometrika, Vol. 77, pp. 521–528. ing Capacity of Roads, Railways, and Airfields. University of Illinois,
Hausman, J., and D. Steele. 2011. Using the Rolling Wheel Deflectom- Champaign.
eter to Improve Pavement Management Decisions. Proc., 8th Inter- Steele, D. A., and J. W. Hall. 2005. ARA Project No. 15874 Final Report:
national Conference on Managing Pavement Assets, Santiago, Chile, Rolling Wheel Deflectometer (RWD) Demonstration and Comparison to
November 15–19. Other Devices in Texas. Applied Research Associates, Champaign, Ill.
Huang, Y. H. 2004. Pavement Analysis and Design. Prentice Hall, Upper Steele, D. A., and J. W. Hall. 2004. ARA Project No. 15874 Final Report:
Saddle River, N.J.
Rolling Wheel Deflectometer Demonstration and Comparison with
Irwin, L. 1983. Report No. 83-8: User’s Guide to MODCOMP2. Cornell
Other Devices in Texas. ERES Consultants, Champaign, Ill.
University Local Roads Program, Ithaca, N.Y.
Steele, D. A., and W. R. Vavrik. 2006. ARA Project No. 16860/4: Rolling
Katicha, S. W., G. W. Flintsch, and B. W. Ferne. 2012. Estimation of
Wheel Deflectometer (RWD) Results for the Virginia Department of
Pavement TSD Slope Measurements Repeatability from a Single
Transportation. Applied Research Associates, Champaign, Ill.
Measurement Series. Presented at 91st Annual Meeting of the Trans-
Thyagarajan, S., N. Sivaneswaran, K. Petros, and B. Muhunthan. 2011.
portation Research Board, Washington, D.C.
Development of a Simplified Method for Interpreting Surface
Krarup, J., S. Rasmussen, L. Aagaard, and P. G. Hjorth. 2006. Output from
Deflections for In-Service Flexible Pavement Evaluation. Proc.,
the Greenwood Traffic Speed Deflectometer. Presented at 22nd ARRB
8th International Conference on Managing Pavement Assets, Santiago,
Conference Research into Practice, Canberra, Australia.
Leng, L., T. Zhang, L. Kleinman, and W. Zhu. 2007. Ordinary Least Square Chile.
Regression, Orthogonal Regression, Geometric Mean Regression and Van Gurp, C., A. Bode, M. Wennink, and A. Molenaar. 1998. Full Struc-
Their Applications in Aerosol Science. Journal of Physics: Conference ture Diagnosis of Low Volume Roads. Proc., 5th International Con-
Series, Vol. 78, No. 1. doi:10.1088/1742-6596/78/1/012084. ference on the Bearing Capacity of Roads and Airfields, Trondheim,
Morosiuk, G., and M. Riley. 2001. Modelling Road Deterioration and Works Norway.
Effects in HDM-4. Highway Development and Management Series Van Huffel, S., and J. Wandewalle. 1991. The Total Least Squares
Collection, Vol. 6. World Road Association, Paris. Problems: Computational Aspects and Analysis. SIAM Publications,
Munk, A., N. Bissantz, T. Wagner, and G. Freitag. 2005. On Difference- Philadelphia, Pa.
Based Variance Estimation in Nonparametric Regression When Venkatraman, E., and A. B. Olshen. 2007. A Faster Circular Binary
the Covariate Is High Dimensional. Journal of the Royal Statistical Segmentation Algorithm for the Analysis of Array CGH Data.
Society B, Vol. 67, Part 1, pp. 19–41. Bioinformatics, Vol. 23, No. 6, pp. 657–663.
Myers, R. H. 1990. Classical and Modern Regression with Applications, Wahba, G. 1983. Bayesian Confidence Intervals for the Cross-Validated
2nd ed. Duxbury Press, Boston, Mass. Spline. Journal of the Royal Statistical Society Series B, Vol. 45, No. 1,
National Cooperative Highway Research Program. 2004. NCHRP pp. 133–150.
Project 1-37A Final Report: Guide for Mechanistic Empirical Design Xu, B., R. Ranjithan, and R. Kim. 2002. New Relationships Between
Guide of New and Rehabilitated Pavement Structures. HRB, National Falling Weight Deflectometer Deflections and Asphalt Pavement
Research Council, Washington, D.C. Layer Condition Indicators. In Transportation Research Record: Journal
Nychka, D. 1990. The Average Posterior Variance of a Smoothing Spline of the Transportation Research Board, No. 1806, TRB, National Research
and a Consistent Estimate of the Average Squared Error. The Annals Council, Washington, D.C., pp. 48–56.
of Statistics, Vol. 18, No. 1, pp. 415–428. Yoo, P. J., and I. L. Al-Qadi. 2007. Effect of Transient Dynamic Loading
Nychka, D. 1988. Bayesian Confidence Intervals for Smoothing Splines. on Flexible Pavements. In Transportation Research Record: Journal of
Journal of the American Statistical Association, Vol. 83, No. 404, the Transportation Research Board, No. 1990, TRB, National Research
pp. 1134–1143. Council, Washington, D.C., pp. 129–140.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

A pp e n d i x A

Research Problem Statements

This appendix contains more complete descriptions of the time period (as close as possible). It is essential for the success
four follow-on research projects recommended to facilitate of any future comparison that testing of the devices is carried
the implementation of this technology. Problem statements, out on the same sites, with exactly the same compositions and
objectives, suggested approach, and urgency and payoff poten- spatial variability, and under the same operational conditions.
tial are described for each. Tests with the latest versions of the selected technologies at a
variety of locations and with a range of pavement designs are
needed to definitively confirm the capabilities of the individual
Problem 1: Side-by-Side
devices and investigate their potential use for network- and
Comparison of Continuous
project-level applications.
Deflection Measuring Devices
Research Problem Statement Suggested Approach
This SHRP 2 R06F project has identified the traffic speed Step 1: Testing of Pavement Sections
deflectometer (TSD) and the rolling wheel deflectometer
(RWD) as continuous deflection measuring devices that have The structural condition of pavement sections representing
the potential to meet agency needs in terms of deflection a wide range of designs (flexible, rigid, and composite) and
testing to support network-level and possibly project-level conditions (weak, strong, new, and old) will first be fully char-
decisions. The two devices measure different parameters; the acterized in terms of structural condition (e.g., by detailed
RWD directly measures pavement deflection, and the TSD FWD measurements, ground-penetrating radar [GPR] thick-
measures pavement deflection slope. Both of these parameters ness surveys, and coring). These sections will then be tested
are affected by measurement conditions such as speed and with both the TSD and the RWD. Tests will be performed
temperature, and these effects are in turn influenced by pave- concurrently to minimize the effect of external variables
ment composition. It is therefore important that the perfor­ (e.g., temperature) on test results. It is suggested that the raw
mance of the TSD is evaluated in U.S. conditions, ideally in a data from each device be analyzed (i.e., before any averaging
side-by-side comparison with the RWD, if the device becomes is performed). At least three passes for each device are recom-
available, on the same sites and in similar operational condi­ mended; more are preferred.
tions (i.e., with the same pavement structures, spatial variabil-
ity, and pavement temperature). Quality of the information Step 2: Data Analysis
obtained from each device can then be evaluated in terms
of signal-to-noise ratio, capability to distinguish weak from Data analysis should involve at least the following: (1) evalu-
strong sections, and capability to identify weak joints in con- ation of measurement noise level; (2) estimation of signal-
crete pavement, among other things. to-noise ratio; and (3) evaluation of different noise reduction
methodologies.
Objective
Step 3: Detailed Assessment of Device Capabilities
This proposed research will perform a side-by-side compari-
son of the TSD and RWD in the United States. Tests will be In this last step, each device will be evaluated for its capabilities
conducted on the same sections of pavement within the same to identify weak sections as well as weak joints. Furthermore,

97

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

98

each device will be evaluated for its capability to distinguish the SHRP 2 R06F project. This involves estimating the true
between sections in poor condition (in need of maintenance) deflection from noisy measurements by using available signal
and those in good condition and to produce structural param- processing and statistical methods and segmenting deflection
eters similar to those obtained from FWD testing. measurements to reflect different pavement conditions.

Urgency and Payoff Potential Step 2: Application to PMS

The proposed research will provide a much clearer indication The second step will develop methods for identification of
than is currently available of the relative performance of the structurally weak sections, screening sections needing different
two devices and, thus, whether the technologies are ready for treatments (rehabilitation or preservation), and calculating
adoption by transportation agencies. remaining service life. These methods can then be used to
develop and recommend the following: (1) a structural pave-
ment condition index that can be used for network-level
Problem 2: Integrating
pavement management, (2) an algorithm to scope pavement
Continuous Deflection
M&R projects at the network level, and (3) a framework for
Measurements into
specifying structural capacity thresholds.
Pavement Management
Business Functions
Urgency and Payoff Potential
Research Problem Statement
The proposed research will help establish a link between
As of February 2010, only five agencies have incorporated network-level and project-level resource allocation. This
deflection data into their pavement management system will improve the overall efficiency of pavement management
(PMS). One of the main reasons for this low number is that business decisions.
FWD equipment, which is currently used to obtain deflection
data, is a stationary measuring device that disrupts traffic flow
and has a low output rate. This technology provides a low spa- Problem 3: Pavement
tial coverage as well as a low production rate (around 20 km/h Structural Assessment
[12 mph] at 320-m [0.2-mi] intervals). Continuous deflection Methods Using Continuous
devices have a dense spatial coverage and a high production rate Deflection Measurements
and are therefore well suited to support network-level pavement
management decisions. The devices can be efficiently used to Research Problem Statement
screen sections that may be candidates for rehabilitation (and Continuous deflection measurement devices have high spatial
thus require project-level investigation) or to decide if a section coverage and production rates and are therefore well suited
is a good candidate for pavement preservation. to support network-level pavement management decisions.
Measurements collected by these devices are contaminated
Objective with noise, however, as illustrated in this report. The effect
of noise can be reduced using statistical signal processing
This proposed research will develop tools for using contin-
methods. These methods are essential for the potential use of
uous deflection measurement devices to support pavement
the output of continuous deflection measurement devices in
management business functions. The effort would use maxi-
providing advice on structural assessment requirements. This
mum deflection equivalents (already available) to compute
will possibly allow output of continuous deflection devices to
a structural indicator, and the extra information that should
be used for structural assessment requirements such as reha-
be available from the two lasers on the latest RWD and the
bilitation or surface treatments.
three or more lasers on the TSD to estimate parameters
that pavement engineers require for pavement management
(i.e., modified SN or SCI). Objective
This proposed research will evaluate the possibility (or at
Suggested Approach least determine how much improvement is needed) of using
Step 1: Data Processing and Interpretation data from selected devices to compute structural condition
indicators relevant for supporting structural assessment and
of Continuous Deflection Measurements
rehabilitation. Current methods of averaging measurements
The effort will first determine the best methods to process over specific distances are far from optimal. Much better
and use the data for PMS applications by using data from methods based on statistical analysis that trades off between

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

99  

the variance and bias are available to better estimate the (RWD). The results show that their measurements are broadly
pavement deflection from noisy measurements. Furthermore, comparable to the deflection measuring device currently widely
these methods can provide estimates of confidence intervals used in the United States, the falling weight deflectometer
for denoised measurements. With these confidence intervals, (FWD). However, there are differences between the measure-
engineers can suggest structural treatments based on acceptable ments, because the type and rate of loading are very differ-
risk levels. ent between traffic-speed rolling wheel devices (the TSD and
the RWD) and a static impulse loading device (the FWD).
Suggested Approach Therefore, to confirm that the devices are measuring what is
claimed, it is necessary to compare the measurements of the
Step 1: Review of Potential Data devices with the response of a pavement under that load. To
Analysis Methodologies achieve this, a number of pavement sections with a range of
different structural designs should be comprehensively instru-
There are a number of statistical signal processing tools for
mented and the response under the loading wheels of each
denoising of data, such as spline and kernel smoothing,
device should be compared with the measurements of the
frequency domain filtering, and wavelet denoising through
built-in instruments.
nonlinear thresholding or shrinkage of wavelet coefficients.
Each method has advantages and disadvantages. For example,
spline smoothing is a linear operation and, therefore, is not Objective
well adapted to large spatial variability; however, it is very
This proposed research will evaluate the accuracy of continu-
robust to the distributional properties of measurement errors.
ous deflection devices by using measured absolute deflection,
On the other hand, wavelet denoising is adaptive to spatial
acceleration, and strains from instrumented pavement sec-
variability; however, much of the theory has been developed
tions. This effort may include testing on existing facilities or
for the case of normally distributed errors, and handling dif-
specifically instrumented pavement sections.
ferent types of distributional errors can be challenging.

Suggested Approach
Step 2: Structural Assessment
Step 1: Design of Experiment
This step will concentrate on developing (empirical, semi-
empirical, and theoretical) methodologies to calculate relevant In the first step, candidate pavement sections will be selected
structural parameters from continuous deflection measure- for instrumentation with state-of-the-art accelerometers and
ments. A certain level of confidence (based on the statisti- strain gauges. These sections will reflect a wide range of pave-
cal analysis in Step 1) can be associated with those structural ment types and conditions. If possible, preliminary testing
parameters, so that a risk level can be associated with the dif- of the continuous deflection device on existing instrumented
ferent treatment and rehabilitation options. facilities should be performed. Results from this testing
should be used to better design the instrumentation of the
selected pavement sections.
Urgency and Payoff Potential
The proposed research will facilitate the incorporation of pave- Step 2: Testing and Evaluation
ment structural condition considerations into the pavement
management process, which can help improve the efficiency Testing will be conducted on the instrumented test sections
of pavement management business decisions. under various conditions, in accordance with the designed
experiment. These tests will allow the accuracy of the mea-
surements and the impact of various operational parameters
Problem 4: Evaluation of to be determined. Two of the most important operational
the Absolute Accuracy parameters that will be investigated are testing speed and
of Continuous Deflection pavement temperature.
Measuring Devices on
Instrumented Road Sections
Urgency and Payoff Potential
Research Problem Statement
The proposed research will allow a detailed evaluation of the
The SHRP 2 R06F study has evaluated the performance of accuracy of continuous deflection devices, which is critical
two continuous deflection measuring devices, the traffic speed for assessing long-term prospects of the technology and for
deflectometer (TSD) and the rolling wheel deflectometer developing required standards.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

TRB Oversight Committee for the Strategic Highway Research Program 2*

Chair: Kirk T. Steudle, Director, Michigan Department of Transportation
Members
H. Norman Abramson, Executive Vice President (retired), Southwest Research Institute
Alan C. Clark, MPO Director, Houston–Galveston Area Council
Frank L. Danchetz, Vice President, ARCADIS-US, Inc.
Stanley Gee, Executive Deputy Commissioner, New York State Department of Transportation
Michael P. Lewis, Director, Rhode Island Department of Transportation
Susan Martinovich, Director, Nevada Department of Transportation
John R. Njord, Executive Director, Utah Department of Transportation
Charles F. Potts, Chief Executive Officer, Heritage Construction and Materials
Ananth K. Prasad, Secretary, Florida Department of Transportation
Gerald M. Ross, Chief Engineer, Georgia Department of Transportation
George E. Schoener, Executive Director, I-95 Corridor Coalition
Kumares C. Sinha, Olson Distinguished Professor of Civil Engineering, Purdue University
Paul Trombino III, Director, Iowa Department of Transportation
Ex Officio Members
John C. Horsley, Executive Director, American Association of State Highway and Transportation Officials
Victor M. Mendez, Administrator, Federal Highway Administration
David L. Strickland, Administrator, National Highway Transportation Safety Administration
Liaisons
Ken Jacoby, Communications and Outreach Team Director, Office of Corporate Research, Technology, and Innovation Management,
Federal Highway Administration
Tony Kane, Director, Engineering and Technical Services, American Association of State Highway and Transportation Officials
Jeffrey F. Paniati, Executive Director, Federal Highway Administration
John Pearson, Program Director, Council of Deputy Ministers Responsible for Transportation and Highway Safety, Canada
Michael F. Trentacoste, Associate Administrator, Research, Development, and Technology, Federal Highway Administration

*Membership as of June 2012.

Renewal Technical Coordinating Committee*

Chair: Cathy Nelson, Technical Services Manager/Chief Engineer, Oregon Department of Transportation
Vice-Chair: Daniel D’Angelo, Recovery Acting Manager, Director and Deputy Chief Engineer, Office of Design, New York State Department
of Transportation
Members
Rachel Arulraj, Director of Virtual Design & Construction, Parsons Brinckerhoff
Michael E. Ayers, Consultant, Technology Services, American Concrete Pavement Association
Thomas E. Baker, State Materials Engineer, Washington State Department of Transportation
John E. Breen, Al-Rashid Chair in Civil Engineering Emeritus, University of Texas at Austin
Steven D. DeWitt, Chief Engineer, North Carolina Turnpike Authority
Tom W. Donovan, Senior Right of Way Agent (retired), California Department of Transportation
Alan D. Fisher, Manager, Construction Structures Group, Cianbro Corporation
Michael Hemmingsen, Davison Transportation Service Center Manager (retired), Michigan Department of Transportation
Bruce Johnson, State Bridge Engineer, Oregon Department of Transportation, Bridge Engineering Section
Leonnie Kavanagh, PhD Candidate, Seasonal Lecturer, Civil Engineering Department, University of Manitoba
John J. Robinson, Jr., Assistant Chief Counsel, Pennsylvania Department of Transportation, Governor’s Office of General Counsel
Ted M. Scott II, Director, Engineering, American Trucking Associations, Inc.
Gary D. Taylor, Professional Engineer
Gary C. Whited, Program Manager, Construction and Materials Support Center, University of Wisconsin–Madison
AASHTO Liaison
James T. McDonnell, Program Director for Engineering, American Association of State Highway and Transportation Officials
FHWA Liaisons
Steve Gaj, Leader, System Management and Monitoring Team, Office of Asset Management, Federal Highway Administration
Cheryl Allen Richter, Assistant Director, Pavement Research and Development, Office of Infrastructure Research and Development,
Federal Highway Administration
J. B. “Butch” Wlaschin, Director, Office of Asset Management, Federal Highway Administration
Canada Liaison
Lance Vigfusson, Assistant Deputy Minister of Engineering & Operations, Manitoba Infrastructure and Transportation

*Membership as of April 2013.

Copyright National Academy of Sciences. All rights reserved.

Assessment of Continuous Pavement Deflection Measuring Technologies

Geotechnical Solutions for Soil Improvement, Rapid Embankment Construction,

and Stabilization of the Pavement Working Platform (R02)
Precast Concrete Pavement Technology (R05)
Nondestructive Testing to Identify Delaminations between HMA Layers (R06D)
Performance Specifications for Rapid Renewal (R07)
Using Existing Pavement in Place and Achieving Long Life (R23)

Copyright National Academy of Sciences. All rights reserved.