Backcalculation of Dynamic Modulus

from Falling Weight Deflectometer Data

Y. Richard Kim, Ph.D., P.E., F. ASCE

Zhe Zeng
Kangjin Lee, Ph.D.
Dept. of Civil, Construction, & Environmental Engineering
North Carolina State University

NCDOT Project 2017-03

FHWA/NC/2017-03
January 2021
Backcalculation of Dynamic Modulus from Falling Weight
Deflectometer Data

FINAL REPORT
Research Project No. HWY-2017-03

Submitted to:

North Carolina Department of Transportation

Office of Research

Submitted by:

Y. Richard Kim, Ph.D., P.E., F.ASCE

Jimmy D. Clark Distinguished University Professor
Campus Box 7908
Department of Civil, Construction, & Environmental Engineering
North Carolina State University
Raleigh, NC 27695-7908
Tel: 919-515-7758, Fax: 919-515-7908
E-mail: [email protected]

Zhe Zeng
E-mail: [email protected]

Kangjin Lee, Ph.D.

E-mail: [email protected]

Department of Civil, Construction, & Environmental Engineering

North Carolina State University
Raleigh, NC

January 2021
 North Carolina Department of Transportation
Research and Development

Technical Report Documentation Page

1. Report No. 2. Government Accession No. 3. Recipient’s Catalog No.
FHWA/NC/2017-03
4. Title and Subtitle 5. Report Date
Backcalculation of Dynamic Modulus from Falling Weight Deflectometer Data January 12, 2021
6. Performing Organization Code

7. Author(s) 8. Performing Organization Report No.

Y. Richard Kim, Zhe Zeng, and Kangjin Lee
9. Performing Organization Name and Address 10. Work Unit No. (TRAIS)
Campus Box 7908, Dept. of Civil, Construction, & Environmental Engrg.
NCSU, Raleigh, NC 27695-7908
11. Contract or Grant No.

12. Sponsoring Agency Name and Address 13. Type of Report and Period Covered
NC Department of Transportation Final Report
Research and Analysis Group August 2016 – July 2020
1 South Wilmington Street
14. Sponsoring Agency Code
Raleigh, NC 27601
RP2017-03
15. Supplementary Notes

16. Abstract
The Mechanistic-Empirical Pavement Design Guide (Pavement ME Guide) and its related software, AASHTOWare
Pavement ME Design, have been widely used in the asphalt pavement community. The Guide proposes three
analysis levels for highway agencies to use for rehabilitation design. However, as these methods were developed in
the early 2000s, certain aspects were not considered sufficiently in terms of their accuracy and efficiency. Over time,
researchers have offered various suggestions to modify the theories and guidelines in the Pavement ME Guide, and
from a practical point of view, some parts of the Guide still need to be improved, such as how best to perform
backcalculations and characterize multilayered existing asphalt concrete (AC) pavement. To this end, this study
focused specifically on improving methods to characterize damaged existing AC layers.
Based on North Carolina (NC) 96 highway data, North Carolina State University researchers evaluated and
compared the accuracy of the three levels in the Pavement ME Guide. Through tests and analysis of field cores, this
report provides specific guidelines for the North Carolina Department of Transportation to select representative
layers in multilayered AC pavements. The research team compared predicted damaged dynamic modulus
mastercurves to laboratory-measured values. The results show that Level 1 should always be the first choice for
highway agencies. Nonetheless, if Level 2 or 3 needs to be applied, then the transfer function that relates the damage
factor to the percentage of bottom-up cracking needs to be calibrated first to ensure that the results are consistent
with those of Level 1. Although sometimes the in situ modulus of the granular layer is measured during the
backcalculation process, the modulus value is not recommended to be fixed in the program in order to give the
algorithm enough freedom to satisfy the root mean square error criterion. Regarding the selection of the
representative layer, the total core, not the thickest layer, should be used for laboratory evaluation. The NCDOT can
thus save time and effort without sacrificing the accuracy of rehabilitated pavement performance predictions.

17. Key Words 18. Distribution Statement

Asphalt concrete, rehabilitation design, dynamic
modulus, Witczak’s predictive equation, falling
weight deflectometer (FWD), backcalculation
19. Security Classif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price
66
Form DOT F 1700.7 (8-72) Reproduction of completed page authorized

i
 North Carolina Department of Transportation
Research and Development

DISCLAIMER
The contents of this report reflect the views of the authors and are not necessarily the views of
North Carolina State University. The authors are responsible for the facts and the accuracy of the
data presented herein. The contents do not necessarily reflect the official views or policies of the
North Carolina Department of Transportation at the time of publication. This report does not
constitute a standard, specification, or regulation.

ii
 North Carolina Department of Transportation
Research and Development

ACKNOWLEDGEMENTS
This research was sponsored by the North Carolina Department of Transportation. The Steering
and Implementation Committee was comprised of Clark Morrison, Ph.D., P.E. (Chair), Josh
Holland, P.E., Lamar Sylvester, P.E., Scott Capps, P.E., Christopher A. Peoples, P.E., James B.
Phillips, P.E., and Mustan Kadibhai, P.E. (PM). These advisors have given invaluable direction
and support to the research team throughout the project. In addition, Prof. Murthy Guddati led
the theoretical development of the forward modeling and back-calculation algorithms in this
project, while Amin Sahafipourfard enhanced the formulations and implemented the algorithms.

iii
 North Carolina Department of Transportation
Research and Development

EXECUTIVE SUMMARY

Research Objectives

The overall objectives of this research are to:

1. Evaluate the reliability of the three analysis levels that are recommended in the
Mechanistic-Empirical Pavement Design Guide (Pavement ME Guide) to estimate the
dynamic modulus values of asphalt layers in existing pavements.

2. Recommend an effective way to determine the in situ dynamic modulus mastercurve by

comparing the predicted dynamic modulus values used in Witczak’s predictive equation
against the dynamic modulus values measured from laboratory tests of field cores.

3. Investigate the effect of the inputs used in the backcalculation process on the
backcalculated modulus values.

4. Develop recommendations for pertinent falling weight deflectometer (FWD) test

protocols and reliable methods to determine the dynamic modulus values of existing
asphalt layers for the North Carolina Department of Transportation (NCDOT)’s routine
use of the Pavement ME Guide’s rehabilitation design.

Research Methodology

To achieve the research goals, the North Carolina State University (NCSU) research team
applied various means to investigate efficient ways to characterize the dynamic modulus of
existing asphalt concrete (AC) pavements. North Carolina (NC) 96 was selected for this study
because it is a full-depth pavement that consists of multiple AC layers that have experienced
damage since construction, thus making this pavement a good candidate for the research
purposes. Seven pavement sections, each 1000 feet long, were chosen for the field tests, i.e.,
falling weight deflectometer (FWD) and dynamic cone penetration (DCP) tests. The NCSU team
recorded the air and pavement surface temperatures during testing and conducted a condition
survey to obtain the required information for Pavement ME Design.

The NCSU team extracted field cores from the same locations the field tests were conducted and
used various laboratory tests to determine the dynamic modulus mastercurves for these field
cores. The research team conducted two types of laboratory tests, i.e., tests required by
Witczak’s predictive equation and dynamic modulus tests. The former provides an indirect way
to determine the dynamic modulus value and the latter provides direct measurements. Also,
because field cores do not necessarily have a standard geometry or provide sufficient materials
for testing, the NCSU research team devised different solutions to ensure that the tests of the

iv
 North Carolina Department of Transportation
Research and Development

field cores were as close as possible to typical laboratory tests without sacrificing measurement
accuracy.

In addition to testing, the research team evaluated the three levels of analysis that are
recommended in the Pavement ME Guide using AASHTOWare Pavement ME Design software.
Level 1 analysis, which is the most accurate approach, not only involves the backcalculation of
FWD-tested deflections, but also requires measured data to satisfy Witczak’s predictive
equation. The research team also investigated ways to backcalculate the elastic modulus using
the Pavement ME backcalculation tool by testing three sets of subgrade inputs based on
experimental measurements, empirical predictions, and direct backcalculation, respectively. The
research team then determined the best way to use the Pavement ME backcalculation tool by
comparing the root mean square errors (RMSEs) of the inputs. In addition, due to the limited
amount of material that each field core could provide, the team carefully considered the order
that the tests were conducted to obtain maximum operational efficiency and data accuracy.

Instead of the FWD and backcalculation needed for Level 1 analysis, Level 2 analysis requires a
field condition survey of alligator cracking. Using the precalibrated transfer function given in the
Pavement ME Guide, the research team determined damage factors for the NC 96 selected test
sections. Then, the team developed the NC 96 in situ damaged dynamic modulus mastercurve in
combination with the dynamic modulus mastercurve predicted by Witczak’s equation. Note that,
in this report, ‘undamaged’ and ‘damaged’ mastercurves refer to the dynamic modulus
mastercurves of undamaged and damaged AC pavements, respectively. For Level 3 analysis,
instead of using a specific number to quantify the NC 96 field conditions, the research team used
ratings as inputs into AASHTOWare, coupled with typical Witczak’s predictive equation inputs
for North Carolina. In this way, the research team obtained the field dynamic modulus
mastercurve. Based on comparisons of the three analysis levels in terms of their accuracy and
practicability, the NCSU research team is able to provide final recommendations to the NCDOT
to characterize the dynamic modulus of an existing AC pavement.

Another important problem to be solved in this project was determining the best way to select the
representative layer of an existing AC pavement, especially when the pavement consists of
multiple layers. The NC 96 pavement has five layers (or even more in reality) and its thickest
layer is the top layer (3 in.). The research team categorized the field core testing into three cases:
top (top layer only), bottom (remaining layers after excluding the top layer), and the total core.
Considering each of these cases separately as the representative layer, the research team tested
the characterization method given in the Pavement ME Guide and predicted the rehabilitated
pavement performance using AASHTOWare. By comparing the differences in predicted
performance and considering the complexity of the test procedures, the research team is able to
provide suggestions to the NCDOT for selecting the representative layer of an existing AC
pavement.

v
 North Carolina Department of Transportation
Research and Development

Finally, the NCSU research team tried one more step to backcalculate the in situ dynamic
modulus mastercurve directly in an attempt to remove the burden of conducting the tests required
by the Pavement ME Guide. To achieve this goal, the dynamic modulus mastercurve had to be
obtained solely from FWD-measured deflections. Therefore, the team evaluated the conditions
that are required to determine the mastercurve and compared the mastercurve against the
information obtained by the FWD.

Conclusions

The main conclusions drawn from this research are:

• The undamaged mastercurve that is predicted using Witczak’s predictive equation has a
different shape and magnitude than the mastercurve that is constructed based on a field
core. Because the shape of the curve is linked to viscoelasticity and the damaged
mastercurve is merely the vertical shift of the undamaged mastercurve, the current
protocol in Pavement ME based on Witczak’s predictive equation could result in
erroneous damaged mastercurves.

• The results from Level 1, 2, and 3 analyses differ significantly in terms of damage factor
estimations and cracking performance predictions. Level 1 is recommended as the first
choice for agencies to use in practice. However, if Level 2 or 3 needs to be applied, then
the transfer function that relates the damage factor to the percentage of bottom-up
cracking needs to be calibrated first to ensure that the results are consistent with those of
Level 1.

• When the tests specified by Levels 1 and 2 need to be conducted for a multilayered AC
pavement, then the total core should be used for laboratory evaluation, not the thickest
layer. Dividing the existing pavement into multiple layers is possible only for Level 1,
but even in that case is not recommended because this approach requires multilayer
backcalculations and considerable time and resources to characterize the individual layer
materials.

• The dynamic modulus mastercurve cannot be backcalculated solely from FWD data due
to the limited frequency range that the FWD can capture. Therefore, to determine the
damaged mastercurve, some other means, such as FWD testing at multiple times of a day
or multiple seasons or laboratory measurements, need to be employed to obtain more
information outside the FWD frequency range. Because the Pavement ME Guide already
provides a way for agencies to determine a damaged mastercurve and its software is
convenient for the NCDOT to access, instead of developing a completely new protocol,
the NCSU research team suggests that the NCDOT use the Pavement ME method that
has been modified in this project based on NC 96 data.

vi
 North Carolina Department of Transportation
Research and Development

TABLE OF CONTENTS
1. INTRODUCTION .................................................................................................................. 1
2. PAVEMENT ME REHABILITATION DESIGN METHOD ............................................... 4
2.1. Input Level 1: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve
(First Method) ............................................................................................................................. 4
2.2. Input Level 2: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve
(First Method) ............................................................................................................................. 6
2.3. Input Level 3: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve
(First Method) ............................................................................................................................. 8
2.4. Input Level 1 – Approach to Determine Field-Damaged Dynamic Modulus Mastercurve
(Second Method) ......................................................................................................................... 8
2.5. Input Level 2 – Approach to Determine Field-Damaged Dynamic Modulus Mastercurve
(Second Method) ......................................................................................................................... 9
2.6. Input Level 3: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve
(Second Method) ....................................................................................................................... 10
2.7. Problems Associated with the Pavement ME Guide .......................................................... 10
3. MOTIVATIONS ................................................................................................................... 12
4. FIELD TESTS ...................................................................................................................... 13
5. LABORATORY TESTS ...................................................................................................... 23
5.1. Testing Required to Determine Input Properties for Witczak’s Predictive Equation ........ 24
5.1.1. Volumetric Property Tests and Gradation Analysis .................................................... 24
5.1.2. Extraction, Recovery, and Binder Viscosity Measurements ....................................... 25
5.2. Dynamic Modulus Testing ................................................................................................. 26
6. DYNAMIC MODULUS TEST RESULTS .......................................................................... 28
7. COMPARISON OF MEASURED DYNAMIC MODULUS VALUES VERSUS
BACKCALCULATED MODULUS VALUES ........................................................................... 35
8. EVALUATION OF THE REHABILITATION ANALYSIS LEVELS IN THE
PAVEMENT ME GUIDE ............................................................................................................ 38
8.1. Level 1 Analysis ................................................................................................................. 39
8.2. Level 2 Analysis ................................................................................................................. 39
8.3. Damaged Mastercurve Comparison ................................................................................... 40
9. EVALUATION OF REPRESENTATIVE LAYERS........................................................... 43
9.1. Damaged Mastercurve Comparison ................................................................................... 43
9.2. Pavement Performance Predictions .................................................................................... 49
10. DYNAMIC MODULUS MASTERCURVE BACKCALCULATION ............................... 51
11. CONCLUSIONS................................................................................................................... 52
12. REFERENCES ..................................................................................................................... 54

vii
 North Carolina Department of Transportation
Research and Development

LIST OF TABLES
Table 1. Flexible Pavement: Overall Estimate of Surface Cracking ............................................ 10
Table 2. Pavement Structure Information for Test Sections ......................................................... 14
Table 3. Field Test Information .................................................................................................... 15
Table 4. Height of Each Field Core (in.) ...................................................................................... 22
Table 5. Distress of Each Field Core ............................................................................................ 22
Table 6. Coefficients of Mastercurves .......................................................................................... 34
Table 7. Backcalculated Modulus Values ..................................................................................... 36
Table 8. Alligator Cracking Condition Survey Results ................................................................ 38
Table 9. Damage Factor Results ................................................................................................... 40
Table 10. Pavement ME Coefficients ........................................................................................... 41
Table 11. Damage Factors Resulting from Dividing Existing Layers into Two Sublayers ......... 47
Table 12. Four Ways to Determine Damaged Mastercurve ......................................................... 49
Table 13. Inputs and Outputs for Pavement Performance Simulation.......................................... 51

viii
 North Carolina Department of Transportation
Research and Development

LIST OF FIGURES
Figure 1. Locations of field cores in Section 1 - Section 4. .......................................................... 15
Figure 2. Locations of field cores in Section 5 - Section 7. .......................................................... 15
Figure 3. Falling weight deflectometer testing on NC 96. ............................................................ 16
Figure 4. Dynamic cone penetrometer testing on NC 96.............................................................. 16
Figure 5. Field cores obtained from two NC 96 test structures: (a) field core extracted from
pavement Structure 1 and (b) field core extracted from pavement Structure 2. ........... 17
Figure 6. Deflections versus pavement surface temperature. ....................................................... 18
Figure 7. Deflections versus pavement thickness. ........................................................................ 18
Figure 8. Different patterns of converted resilient modulus values using DCP data: (a) pattern 1,
(b) pattern 2, (c) pattern 3, (d) pattern 4, and (e) pattern 5. .......................................... 21
Figure 9. Field cores selected in Sections 1 through 7 for dynamic modulus testing................... 23
Figure 10. Flowchart of test plan for dynamic modulus testing. .................................................. 24
Figure 11. Dynamic modulus test configuration........................................................................... 27
Figure 12. Example of conditioning time determination (104°F)................................................. 27
Figure 13. Dynamic modulus results: (a) 1-2, (b) 1-10, (c) 2-1, and (d) 2-12. ............................. 28
Figure 14. Dynamic modulus results: (a) 3-1, (b) 3-13, (c) 4-1, and (d) 4-13. ............................. 29
Figure 15. Dynamic modulus results: (a) 5-3, (b) 5-12, (c) 6-1, and (d) 6-13. ............................. 30
Figure 16. Dynamic modulus results: (a) 7-2, (b) 7-11. ............................................................... 30
Figure 17. Phase angle results: (a) 1-2, (b) 1-10, (c) 2-1, and (d) 2-12. ....................................... 31
Figure 18. Phase angle results: (a) 3-1, (b) 3-13, (c) 4-1, and (d) 4-13. ....................................... 32
Figure 19. Phase angle results: (a) 5-3, (b) 5-12, (c) 6-1, and (d) 6-13. ....................................... 33
Figure 20. Phase angle results: (a) 7-2, (b) 7-11. .......................................................................... 33
Figure 21. Comparison between backcalculated modulus values versus lab-measured modulus
values. ........................................................................................................................... 37
Figure 22. Cracks on NC 96 pavement surface. ........................................................................... 38
Figure 23. Undamaged and damaged predicted mastercurves and measured mastercurve: (a) at
core Location A, (b) at core Location B, and (c) at core Location C. .......................... 43
Figure 24. Mastercurves for core Location A: (a) predicted by Witczak’s predictive equation, (b)
measured, (c) damaged mastercurves at Level 1, and (d) damaged mastercurves at
Level 2. ......................................................................................................................... 44
Figure 25. Mastercurves for core Location B: (a) predicted by Witczak’s predictive equation, (b)
measured, (c) damaged mastercurves at Level 1, and (d) damaged mastercurves at
Level 2. ......................................................................................................................... 45
Figure 26. Mastercurves for core Location C: (a) predicted by Witczak’s predictive equation, (b)
measured, (c) damaged mastercurves at Level 1, and (d) damaged mastercurves at
Level 2. ......................................................................................................................... 46
Figure 27. Mastercurves obtained by dividing existing layers into two sublayers: (a) at core
Location A, (b) at core Location B, and (c) at core Location C. .................................. 49

ix
 North Carolina Department of Transportation
Research and Development

1. INTRODUCTION
The dynamic modulus (|E*|) is an important material property that describes the time- and
temperature-dependent behavior of viscoelastic material, such as asphalt concrete (AC).
Dynamic modulus values measured at multiple loading frequencies and temperatures can be used
to develop a dynamic modulus mastercurve, which allows for the prediction of AC’s responses
to any loading history and temperature within the linear viscoelastic range. Therefore, the
dynamic modulus is widely accepted by the asphalt pavement industry as a fundamental stiffness
property of AC. For example, the Mechanistic-Empirical Pavement Design Guide (referred to as
Pavement ME Guide) and its related software, AASHTOWare Pavement ME Design, use the
dynamic modulus as the stiffness property of asphalt mixtures in the design of new pavements as
well as for rehabilitation design. In either new pavement design or rehabilitation design, the user
is required to input the dynamic modulus values of the existing asphalt layers using one of three
analysis approaches (Level 1 to Level 3). Of these three approaches, Level 1 is the most accurate
approach, but it requires significant work and testing. Levels 2 and 3 are simplifications of Level
1 and are easier to implement but lack the same degree of accuracy.

When an overlay needs to be designed, the first step in all three analysis levels is to establish the
‘undamaged’ modulus mastercurve using Witczak’s predictive equation. Levels 1 and 2 require
laboratory tests of field cores to obtain the equation’s inputs whereas Level 3 uses typical values.
If Level 1 analysis is chosen, a falling weight deflectometer (FWD) is employed to detect
deflections in the existing pavement. Then, the in situ elastic modulus value is backcalculated to
represent the field conditions. By assuming the FWD loading frequency (typically from 5 Hz to
30 Hz) (Ayyala et al. 2018), the damage factor (dAC) is calculated at the assumed frequency.
Next, this damage factor is used to shift the entire undamaged mastercurve downwards to
determine the so-called ‘damaged’ mastercurve, which reflects the material’s stiffness under
field damage. If Level 2 is chosen, instead of using a FWD, resilient modulus testing or a
cracking condition survey is used to calculate the damage factor. Level 3 is a simplification of
Level 1 and Level 2 where, instead of conducting the tests and analyses, the user merely rates the
pavement’s alligator cracking as excellent, good, fair, poor, or very poor. These ratings are
related to the extent of damage in the existing pavement, which eventually determines the
position of the damaged mastercurve. Note that, in this report, ‘undamaged’ and ‘damaged’
mastercurves refer to the dynamic modulus mastercurves of undamaged and damaged AC
pavements, respectively.

Since the development of the Pavement ME Guide in the early 2000s, technical limitations have
led researchers not to recommend taking direct dynamic modulus measurements of field cores in
the laboratory (ARA 2004). Instead, researchers use Witczak’s predictive equation to predict the
dynamic modulus mastercurve using a series of methods, such as gradation analysis and binder
viscosity measurements. For new pavement design, the determination of the dynamic modulus is
not difficult or time-consuming because the amount of available material is sufficient for

1
 North Carolina Department of Transportation
Research and Development

specimen fabrication and the mixture has not yet undergone field aging. However, these
advantages are not the case for rehabilitation design. Not only are tests of field cores (that
provide only limited materials) required, but also the in situ materials have been negatively
affected by aging and various types of distress while in service. These factors hinder the
determination of accurate field modulus values. Currently, well-established AASHTO
specifications (i.e., AASHTO TP 79-15/PP 61-13, T 342-11, and R 62) allow the dynamic
modulus values of asphalt mixtures to be determined accurately by testing laboratory-fabricated
specimens using a Superpave gyratory compactor. However, these test specifications require
100-mm diameter, 150-mm tall specimens, so they are not necessarily applicable for field cores.
In order to address this shortcoming, Kim et al. developed viscoelastic solutions that allow the
determination of dynamic modulus values from indirect tensile tests (Kim et al. 2004). Pape et al.
developed small geometry test specimens that also solved problems associated with dynamic
modulus tests of field cores (Pape et al. 2018). Other researchers also have measured field cores
directly using cylindrical specimens (Loulizi et al. 2007, Habbouche et al. 2018). These test
protocols have provided new technical means to determine field dynamic modulus values and,
compared to Witczak’s predictive equation, these protocols may be even easier for highway
agencies to implement in practice.

In addition, because pavement construction projects occur over different years, the existing
pavement always consists of multiple AC layers. Such multilayer pavement structures increase
the difficulty of using the Pavement ME method because Witczak’s predictive equation is based
on single-layer material. When this equation was developed, gradation parameters, binder
viscosity, and other properties were all based on a single mix design process. Hence, whether or
not the prediction accuracy would decrease when applying the equation to multilayered existing
asphalt pavement was unknown. In addition, the backcalculation of FWD-measured data is
sometimes tricky, especially when multiple AC layers are involved. In this scenario, whether or
not the existing AC layers should be regarded as a single layer becomes a question, because
multilayer backcalculation may present technical issues and compromise accuracy. Also, when
backcalculation is conducted, whether the base and/or subgrade modulus values should be
determined via backcalculation or via field tests needs to be investigated. The direct
measurement of the base/subgrade modulus may be more reliable, but setting this value in the
backcalculation algorithm can cause the program not to converge. Therefore, the best way to use
the backcalculation program correctly and efficiently should be evaluated as well.

Based on the background information, the overall objectives of this research are to:

1. Evaluate the reliability of the three analysis methods (Levels 1, 2, and 3) that are
recommended in the Pavement ME Guide to estimate the dynamic modulus values of
asphalt layers in existing pavements.

2
 North Carolina Department of Transportation
Research and Development

2. Recommend an effective way to determine the in situ dynamic modulus mastercurve by

comparing the dynamic modulus values predicted by Witczak’s predictive equation to the
dynamic modulus values measured from laboratory tests of field cores.

3. Investigate the effect of inputs on the backcalculated modulus values in the

backcalculation process.

4. Develop recommendations for pertinent FWD testing protocols and reliable methods to
determine the dynamic modulus values of existing asphalt layers for the North Carolina
Department of Transportation’s (NCDOT’s) routine use of Pavement ME rehabilitation
design.

3
 North Carolina Department of Transportation
Research and Development

2. PAVEMENT ME REHABILITATION DESIGN METHOD

The Pavement ME Design Guide indicates that the following information can be obtained from
FWD backcalculation:

• Time- and temperature-dependent properties of hot mix asphalt layers

• Resilient modulus values for unbound base/sub-base and subgrade materials

• Elastic modulus values of bedrock, if present

The determination of the asphalt layer dynamic modulus for rehabilitation design follows the
same general concepts as for new or reconstruction design, with the exceptions noted in the
following subsections that describe the procedural steps for each of three input levels
recommended in the Pavement ME Design Guide. Note that two types of rehabilitation design
methods are described in the National Cooperative Highway Research Program (NCHRP) 1-37A
report with regard to damage factor calculation and implementation of the different analysis
levels (ARA 2004). The first method is documented in Chapter 2 and the second method is
documented in Chapter 6. For the first method, the applications of the three levels are presented
in the following subsections.

2.1. Input Level 1: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve

(First Method)
1. Conduct nondestructive testing in the outer wheel path using the FWD for the project to
be rehabilitated and compute the mean backcalculated asphalt bound modulus value, Ei,
for the project. Be sure to include cracked as well as uncracked areas. The corresponding
asphalt pavement temperature at the time of testing also should be recorded. Perform
coring to establish the layer thickness throughout the project pavement. Layer thickness
also can be determined using ground penetrating radar (GPR). Backcalculate the asphalt
bound modulus value by combining layers with similar properties at each FWD test point
in the project (with known pavement temperature).

2. Perform field coring and establish the mix volumetric parameters (air void content,
asphalt volume, gradation, and asphalt viscosity parameters to define A-VTS values that
represent two coefficients of the linear function between viscosity and temperature) that
are required for computing the dynamic modulus value using Witczak’s predictive
equation, i.e., Equation (1). Equation (1) is based on the classical sigmoidal function form
where each parameter can be expressed as a function of the asphalt mixture properties.


log (| E* |) =  +  + ( log tr )
(1)
1+ e
where
4
 North Carolina Department of Transportation
Research and Development

|E*| = dynamic modulus,

tr = time of loading at the reference temperature,

α, δ = fitting parameters; for a given set of data, δ represents the minimum value of the
dynamic modulus and α + δ represents the maximum value of the dynamic
modulus, and

β, γ = parameters that describe the shape of the sigmoidal function.

Detailed descriptions of the parameters in Equation (1) are given below:

 Vbeff 
 = 3.750063 + 0.02932 200 − 0.001767 (  200 ) − 0.002841 4 − 0.058097Va − 0.802208 
2

Vbeff + Va 
 = 3.871977 − 0.00214 + 0.00395838 − 0.000017 382 + 0.00547034

 = −0.603313 − 0.393532 log (T r

)

( ( ))
log ( tr ) = log ( t ) − c log ( ) − log Tr

 = 0.313351

c = 1.255882

where:

ρ200 = percentage passing the No. 200 sieve,

ρ34 = cumulative percentage retained on the 3/4 sieve,

ρ38 = cumulative percentage retained on the 3/8 sieve,

ρ4 = cumulative percentage retained on the No. 4 sieve,

Va = air void content,

Vbeff = effective binder content in terms of volume,

t = time, s,

η = viscosity, 106 Poise, and

ηTr = viscosity at reference temperature, 106 Poise.

5
 North Carolina Department of Transportation
Research and Development

3. Develop an undamaged dynamic modulus mastercurve from the data collected in Step 2
using Equation (1).

4. Estimate AC damage, dAC, expressed as Equation (2).

Ei
dAC = 1 − (2)
| E*|
where

Ei = the backcalculated modulus value at a given reference temperature recorded in the

field, and

|E*| = the predicted modulus value at the same temperature as the temperature recorded
in the field.

5. Determine α’ as shown in Equation (3).

 ' = (1 − d j )  (3)
6. Determine the field-damaged mastercurve using α’ instead of α used in Equation (1).

The following tests and procedures are required to determine the dynamic modulus mastercurve
based on the asphalt layers of the existing pavement using the Level 1 approach:

• FWD testing

• Coring

• Binder extraction and recovery

• Viscosity tests of extracted binder

• Mixture bulk specific gravity (Gmb) and maximum specific gravity (Gmm) measurements
of field cores

• Sieve analysis

• Specific gravity tests of aggregate

2.2. Input Level 2: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve

(First Method)
1. Perform field coring and establish the mix volumetric parameters (air void content,
asphalt volume, gradation) and asphalt viscosity parameters to define the A-VTS values.
6
 North Carolina Department of Transportation
Research and Development

2. Develop an undamaged dynamic modulus mastercurve using Equation (1).

3. Conduct indirect resilient modulus (Mri) laboratory tests of field cores using the revised
protocol developed at the University of Maryland for NCHRP 1-28A (Witczak 2004).
Use two to three temperatures below 70℉.

4. Estimate damage, dAC, at similar temperatures and loading frequency conditions using
Equation (4).

M ri
dAC = 1 − (4)
| E* |
where

Mri = laboratory-estimated resilient modulus value at a given reference temperature, and

|E*| = dynamic modulus value predicted at the same temperature as the given reference
temperature.

5. Determine α’ using Equation (3).

6. Determine the field-damaged mastercurve using α’ instead of α that is used in Equation

(1).

The following tests and procedures are required to determine the dynamic modulus mastercurve
from the asphalt layers of an existing pavement using the Level 2 approach:

• Coring

• Indirect tensile resilient modulus testing

• Binder extraction and recovery

• Viscosity tests of extracted binder

• Gmb and Gmm measurements of field cores

• Sieve analysis

• Specific gravity tests of aggregate

7
 North Carolina Department of Transportation
Research and Development

2.3. Input Level 3: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve

(First Method)
1. Use typical estimates of the mix parameters (mix volumetrics, gradation, and binder type)
to develop an undamaged mastercurve with aging for the in situ pavement layers using
Equation (5).

log log = A + VTS log TR (5)

where

η = viscosity, cP,

TR = temperature, Rankine,

A = regression intercept, and

VTS = regression slope of viscosity temperature susceptibility.

2. Use the results of the distress/condition survey to obtain estimates of the pavement rating:
excellent, good, fair, poor, very poor.

3. Use the pavement rating to estimate the (asphalt bound) pavement layer damage factor,
dAC.

4. Determine α’ using Equation (3).

5. Develop the field-damaged mastercurve using α’ rather than α in Equation (1).

Only the distress survey data are needed to determine the dynamic modulus mastercurve from
the asphalt layer of the existing pavement using the Level 3 approach.

For the second method in the Pavement ME Guide, the applications of the three levels are
presented in the following subsections.

2.4. Input Level 1 – Approach to Determine Field-Damaged Dynamic Modulus

Mastercurve (Second Method)
This Level 1 approach is the same as the method presented in Section 2.1 except that the damage
parameter, dAC, is estimated by Equation (6) instead of Equation (2). Then, the determined dAC is
used in Equation (6) to obtain the entire damaged mastercurve from the undamaged mastercurve.

| E*| −10δ
| E*|dam = 10δ + (6)
1 + e−0.3+5log( dAC )

8
 North Carolina Department of Transportation
Research and Development

where

|E*|dam = damaged modulus.

2.5. Input Level 2 – Approach to Determine Field-Damaged Dynamic Modulus

Mastercurve (Second Method)
1. Perform field coring and establish the mix volumetric parameters (air void content,
asphalt volume, gradation) and asphalt viscosity parameters to define the A-VTS values.

2. Develop an undamaged dynamic modulus mastercurve using Equation (1).

3. Use the results of the distress/condition survey to obtain the amount of alligator cracking
that initiated at the bottom of the asphalt layers and was measured at the pavement
surface, as expressed in Equation (7).

 6000  1
FCbottom =  C1C1 +C2C2 log10 ( d AC 100) 
(7)
 1+ e  60
where

FCBottom = area of alligator cracking that initiates at the bottom of hot mix asphalt layers, %
of total lane area,

C1 = 1.31,

C2 = 5 for asphalt layer thickness (hac) less than 5 in.; 3.9666 for hac greater than 12 in.;
otherwise, equal to 0.867+0.2583×hac,

C1 ’ = −2C2’, and

C2 ’ = −2.40874−39.748 (1+hac) −2.856.

4. Obtain the entire damaged mastercurve from the undamaged mastercurve using Equation
(6).

The following tests and procedures are required to determine the dynamic modulus mastercurve
from the asphalt layers of existing pavement using the Level 2 approach:

• Coring

• Condition survey

• Binder extraction and recovery

9
 North Carolina Department of Transportation
Research and Development

• Viscosity tests of extracted binder

• Gmb and Gmm measurements of field cores

• Sieve analysis

• Specific gravity tests of aggregate

2.6. Input Level 3: Approach to Determine Field-Damaged Dynamic Modulus Mastercurve

(Second Method)
The Level 3 approach is the same as that described in Section 2.3; Table 1 presents the criteria
used to evaluate the pavement conditions (AASHTO 2015). After the condition rating is
determined, an approach similar to that for Level 2 is used to determine the damage factor and
damaged mastercurve.

Table 1. Flexible Pavement: Overall Estimate of Surface Cracking

Category Structural Condition
Excellent < 5% area cracked
Good 5%–15% area cracked
Fair 15%–35% area cracked
Poor 35%–50% area cracked
Very Poor > 50% area cracked

2.7. Problems Associated with the Pavement ME Guide

Note that the current AASHTOWare Pavement ME Design software adopts the second method
in its algorithm for asphalt pavement rehabilitation design. Therefore, in the latter part of this
report, only the second method is used and evaluated. Based on the procedures outlined in
Sections 2.1 through 2.6, the Pavement ME Guide method does not require direct dynamic
modulus measurements from field cores. Instead, its approach follows three steps:

1. Depending on the analysis level selected, either use typical values or collect field cores and
conduct laboratory tests to obtain the gradation, air void content, effective asphalt content, and
binder viscosity of the existing AC layers. Next, input the measured properties into Witczak’s
predictive equation to obtain the undamaged modulus mastercurve.

2. If Level 1 is chosen, conduct FWD tests in the field and backcalculate the field-damaged
modulus values at a single temperature and frequency using a linear elastic program (ARA
2004). By comparing the undamaged modulus and damaged modulus at the FWD test frequency,
the damage factor can be estimated. If Level 2 or Level 3 is chosen, then the damage factor is
evaluated using a transfer function as defined in Pavement ME guideline to match the amount of
alligator cracking that initiated at the bottom of the pavement.
10
 North Carolina Department of Transportation
Research and Development

3. Using this damage factor to shift the undamaged mastercurve, the new curve is regarded as
the damaged mastercurve that represents the field conditions.

Although this protocol is widely used by highway agencies and contractors, the possibility of
significant errors is present for all three approaches. First, the accuracy of the dynamic modulus
values obtained using Witczak’s predictive equation (i.e., Equation (1)) varies from case to case.
Applying a damage factor that is estimated from the predicted dynamic modulus value to the
entire dynamic modulus mastercurve could amplify the effects of a dynamic modulus prediction
error. Importantly, because Level 1 and Level 2 both require coring, dynamic modulus tests of
the cores at different temperatures and frequencies can be conducted to obtain the dynamic
modulus mastercurve. The amount of laboratory testing effort that this direct approach requires
may be less than that required for the various tests needed to adopt the Level 1 approach, which
include binder extraction and recovery testing, sieve analysis, specific gravity tests of the
aggregate, and viscosity tests of the extracted binder. Even though the Level 2 approach does not
require FWD testing, it requires a field condition survey in addition to the laboratory tests
required in the Level 1 approach. Hence, the Level 2 approach, as it is described in the Pavement
ME Guide, does not dramatically reduce the required effort compared to Level 1.

The reason the direct dynamic modulus testing approach is not recommended in the Pavement
ME Guide is simple. At the time the Pavement ME Guide was being developed, no laboratory
test method was available that allowed the dynamic modulus mastercurve to be determined from
thin field cores. However, this situation has since changed. Kim et al. (2004) developed the
viscoelastic solutions for the indirect tensile test and thus allowed the dynamic modulus testing
of thin disks that can be obtained from field cores. Park and Kim (2004) and Park et al. (2014)
used the 38 mm diameter cores obtained from field cores by horizontal coring to perform the
dynamic modulus testing. A more refined method of fabricating the 38 mm diameter specimens
was suggested by Pape et al. (2018) and verified against the 100 mm diameter specimens by Lee
et al. (2017).

Some researchers found that the measured modulus mastercurve is very close to the undamaged
mastercurve predicted by Witczak’s equation (Loulizi et al. 2007). Other researchers showed that
Witczak’s equation results in huge errors compared with laboratory-measured modulus values
(Habbouche et al. 2018). Because the reported accuracy varies, implementing the Pavement ME
Guide approach for dynamic modulus testing relates closely to the field conditions and in situ
material properties, which should be evaluated based on local resources.

With regard to implementing the different analysis levels, several studies have pointed out that
the determined damaged mastercurves do not overlap when using the three levels in the
Pavement ME Guide (Loulizi et al.2007, Harsini et al. 2013, Habbouche et al. 2018), indicating
that analysis level selection may also result in significant variation. In an attempt to resolve this
reported problem, Loulizi et al. suggested that Level 1 should be applied to all projects if the
inputs are available (Loulizi et al. 2007). Harsini et al. recommended relating the pavement
11
 North Carolina Department of Transportation
Research and Development

condition ratings to the modulus characterized in Level 1 in order to make the performance
predictions consistent (Harsini et al. 2013). Habbouche et al. concluded that the use of Witczak’s
predictive equation (Level 1) leads to damage overestimation and proposed a hybrid method to
improve accuracy; their method includes both Level 1 and Level 2 as well as laboratory dynamic
modulus tests of field cores (Habbouche et al. 2018). Also, Ayyala et al. investigated three
modified approaches to determine the damage factor using Long-Term Pavement Performance
(LTPP) program data: (1) calibrate Equation (7) so that it results in the same damage factor
calculated in Equation (6); (2) calibrate Equation (6) to match the damage factor in Equation (7);
and (3) change the equation expression so that the damaged mastercurve is obtained through
horizontal shifting instead of vertical shifting. Even though all these concepts make sense
logically, the researchers did not find a definitive relationship between the modulus-based
damage factor and the in situ cracking condition survey results (Ayyala et al. 2017 and 2018).

Furthermore, actual pavements always consist of multiple AC layers due to the construction of
either different lifts or mixes, which increases the uncertainty and complexity surrounding the
implementation of Witczak’s predictive equation because this equation is based on single mix
properties. As of yet, no clear solution is available for multilayer pavement systems. The only
guidance from Pavement ME is to conduct tests of a “representative lift” or combine different
layers to determine the inputs of Witczak’s equation (AASHTO Pavement ME Design Task
Force 2013), which may cause a large variation in the predicted modulus values. Some of the
aforementioned researchers encountered situations where field cores contained multiple AC
layers, but they did not provide details or guidance about representative lift selection and/or
testing. Therefore, this gap in the literature stills needs research to determine the possible effects
when different layer selections/combinations are employed.

3. MOTIVATIONS
Although various researchers have provided good information about the determination of
damaged mastercurves and have modified current methods in the right directions, research gaps
and problems remain. Specifically:

• The reported accuracy of the three analysis levels used in Pavement ME varies. Even
though a hybrid method (Habbouche et al. 2018) has been proposed to improve accuracy,
the effort required to complete all the Level 1 and Level 2 tests plus the suite of dynamic
modulus tests is difficult for agencies to rationalize.

• Not enough guidelines are provided for the dynamic modulus characterization of
multilayered AC pavements, especially in terms of representative lift selection that could
affect both experimental efficiency and performance predictions.

• No clear trend is evident between the damage factor obtained through FWD testing
(Level 1) and observed field cracking percentages (Levels 2 and 3) using a large database

12
 North Carolina Department of Transportation
Research and Development

such as the LTPP program database, suggesting that this relationship may relate
specifically to local materials and in situ conditions. Clear guidance based on the
agency’s own database would benefit local agencies.

Given these limitations and the need for localized guidance, the objective of this study is to
evaluate the three levels of Pavement ME for determining existing AC layer modulus values
based on North Carolina materials and field conditions. The NCSU research team will provide
guidance to the NCDOT regarding efficient and reliable ways to characterize existing AC layers.
The following goals are targeted in this research:

• Recommend a practical way to implement Pavement ME analysis methods with regard to

both accuracy and efficiency.

• Propose a method to determine damaged mastercurves for multilayer AC pavements that

includes the selection of a representative pavement layer.

4. FIELD TESTS
Two full-depth pavement structures in NC 96 were selected as the test sections for this study.
They are located in the section of NC 96 that crosses the county line between Wake County and
Franklin County in North Carolina (near 35.98°N, 78.45°W). Table 2 provides the pavement
structure information for these two pavement structures. A total of seven test sections are
identified from the two pavement structures. The first 6.5 test sections have the same pavement
structure (Structure 1 in Table 2) and the last 0.5 section has a different pavement structure
(Structure 2 in Table 2). Each section was 1000 ft long with 250 ft of coring sections at both
ends. The middle 500 ft is the condition monitoring section. In the condition monitoring sections,
five FWD tests were conducted in the middle of the traffic lane at every 100 ft. In each of the
coring sections, four FWD tests were conducted first and then field cores were extracted at the
same locations. Each FWD testing location had two FWD drops. The pavement surface
temperatures were measured during the FWD tests.

13
 North Carolina Department of Transportation
Research and Development

Table 2. Pavement Structure Information for Test Sections

Structure 1 (length: 1.57 mile) Structure 2 (length: 0.27 mile)
Construction Thickness Construction Thickness
Material Materialb
Year (in.) Year (in.)
a
2006 S 9.5B 3 2006 S 9.5B 3
1994 I-2b 1 2000 I-2 1
c e
1986 BCSC 1 2000 HDS 1.25
1970 BCSC 1 1992 I-2 1
1941 BSTd 0.5 1981 I-2 1
— Subgrade — 1941 BST 0.5
— — — — Subgrade —
Total — 6.5 Total — 7.75
a
S 9.5B = surface mix with 9.5 mm nominal maximum aggregate size (NMAS) designed for traffic level B; b I-2 =
light-duty Marshall surface mix; c BCSC = bituminous concrete surface course; d BST = bituminous surface
treatment; and e HDS = heavy-duty Marshall surface mix.

Table 3 provides information on FWD testing, coring, and DCP testing. Note that, in the coring
sections, DCP testing was done on every other coring location. Figure 1 and Figure 2 present
schematics of the field core locations for the first four test sections and the last three test
sections, respectively. After coring, dynamic cone penetrometer (DCP) tests were conducted to
measure the penetration index of the subgrade at some locations; these values were converted to
resilient modulus (Mr) values using the relationships expressed in Equations (8) and (9)
(Christopher et al. 2010, ARA 2004). Figure 3 and Figure 4 show photographs of the FWD
testing and DCP testing processes on NC 96, respectively. Figure 5 shows two field cores
extracted from each of the two test pavement structures in NC 96; these photographs show that
the core extracted from Structure 2 is longer than the core from Structure 1.

394
CBR = (8)
x 1.066
M r = 2555  CBR0.64 (9)
where

CBR = California bearing ratio, %,

x = penetration index, mm/blow, and

Mr = resilient modulus, psi.

14
 North Carolina Department of Transportation
Research and Development

Table 3. Field Test Information

Monitoring Sections
Coring Sections (250 ft at both ends)
(middle 500 ft)

FWD test FWD test Coring DCP test

Section 1 5 8 (4+4) 8 (4+4) 4 (2+2)
Section 2 5 8 (4+4) 8 (4+4) 4 (2+2)
Section 3 5 8 (4+4) 8 (4+4) 4 (2+2)
Section 4 5 8 (4+4) 8 (4+4) 4 (2+2)
Section 5 5 8 (4+4) 6 (3+3) 4 (2+2)
Section 6 5 8 (4+4) 6 (3+3) 4 (2+2)
Section 7 5 8 (4+4) 6 (3+3) 4 (2+2)

DCP DCP DCP DCP

250 ft 500 ft 250 ft

Figure 1. Locations of field cores in Section 1 - Section 4.

DCP DCP DCP DCP

250 ft 500 ft 250 ft

Figure 2. Locations of field cores in Section 5 - Section 7.

15
 North Carolina Department of Transportation
Research and Development

Figure 3. Falling weight deflectometer testing on NC 96.

Figure 4. Dynamic cone penetrometer testing on NC 96.

16
 North Carolina Department of Transportation
Research and Development

(a) (b)
Figure 5. Field cores obtained from two NC 96 test structures: (a) field core extracted from
pavement Structure 1 and (b) field core extracted from pavement Structure 2.

Because the FWD tests and DCP tests were conducted along the NC 96 pavement at multiple
locations, the variation in the test results along the project length could be investigated in this
research. The maximum FWD-measured deflections range from 4.4 mils to 19.6 mils, indicating
that the deflection basin varies significantly when the test location changes. All the measured
maximum deflections were plotted against the pavement surface temperature in Figure 6. No
clear trend between the maximum deflection and pavement surface temperature, except that
Section 6 shows the increase in the deflection as the temperature increases. This figure shows
that the variation in the deflections among the locations in the same pavement structure is larger
than the effect of temperature change in a day on deflections. Figure 7 shows the plot of
deflection versus pavement thickness. Only the locations that were cored are plotted because the
thickness was determined by measuring the core heights. In general, Structure 2 shows lower
deflections values than Structure 1, but no clear trend is evident regarding the deflections in
Structure 1 as a function of pavement thickness.

17
 North Carolina Department of Transportation
Research and Development

30
Sec 1 Sec 2
Sec 3 Sec 4
25 Sec 5 Sec 6
Deflection (mil)

Sec 7 - Structure 1 Sec 7 - Structure 2

20

15

10

0
20 30 40 50 60 70 80
Temperature ( F)
Figure 6. Deflections versus pavement surface temperature.

30
Sec 1 Sec 2
Sec 3 Sec 4
25 Sec 5 Sec 6
Deflection (mil)

Sec 7 - Structure 1 Sec 7 - Structure 2

20

15

10

0
5 7 9 11 13 15
Thickness (in.)
Figure 7. Deflections versus pavement thickness.

Figure 8 shows five different patterns in the DCP-measured resilient modulus results. Each
pattern presents a unique relationship between the measured resilient modulus and penetration
depth. Note that, even when the test locations are close, the test results show significantly
different patterns. This observation indicates that the field variability is large at the different
pavement locations.

18
 North Carolina Department of Transportation
Research and Development

60
(a)
Resilient Modulus (ksi) 50

40

30

20

10

0
0 5 10 15 20
Depth (in.)

60
(b)
Resilient Modulus (ksi)

50

40

30

20

10

0
0 5 10 15 20
Depth (in.)

19
 North Carolina Department of Transportation
Research and Development

60
(c)
Resilient Modulus (ksi) 50

40

30

20

10

0
0 5 10 15 20
Depth (in.)

60
(d)
Resilient Modulus (ksi)

50

40

30

20

10

0
0 5 10 15 20
Depth (in.)

20
 North Carolina Department of Transportation
Research and Development

25
(e)
Resilient Modulus (ksi) 20

15

10

0
0 5 10 15
Depth (in.)
Figure 8. Different patterns of converted resilient modulus values using DCP data: (a) pattern 1,
(b) pattern 2, (c) pattern 3, (d) pattern 4, and (e) pattern 5.

Table 4 and Table 5 present the height of each field core measured in the laboratory and the
distress type of each field core, respectively. Note that the thicknesses of the field cores are
always greater than the total AC thickness shown in Table 2, and the height difference between
the longest core and shortest core from the same structure is as much as 3.5 inches, indicating
significant construction variability in the field.

To investigate the variation in AC stiffness throughout the different pavement sections, two field
cores were selected from each test section for dynamic modulus testing, as shown in Figure 9.
The selection of the cores was based mainly on the integrity of the samples. If the core did not
have debonding or severe cracking that would affect the dynamic modulus test operation, then it
could be selected as a test specimen. In addition, three different locations were selected for the
tests required by the Pavement ME method. Only three locations were chosen because the
Pavement ME test method requires large amounts of materials (details are provided later in this
report). These three locations are in Section 3, Section 6, and Section 7, respectively, and are
referred to hereafter as Locations A, B, and C. Locations A and B represent Structure 1 and
Location C represents Structure 2. In terms of the distress type at the selected sections, Location
A did not show severe debonding (based on visual observation) whereas Location B had serious
debonding issues. The debonded cores are marked as ‘D’ in Table 5 and had broken into parts
when extracted from the field. Similarly, if the core surface showed cracks or stripping, the core
is labeled as ‘C or ‘S’ in Table 5, respectively.

21
 North Carolina Department of Transportation
Research and Development

Table 4. Height of Each Field Core (in.)

Number of the field core
Section
1 2 3 4 5 6 7 8
1 9.4 9.1 8.1 9.3 7.9 7.3 8.7 7.5
2 9.1 9.1 8.3 7.9 9.4 9.8 10.6 9.4
3 9.4 9.1 9.4 9.1 7.9 8.5 8.7 8.5
4 7.9 8.9 8.7 8.3 7.9 8.3 8.3 9.1
5 8.5 8.7 8.7 8.3 9.1 8.7
6 8.7 8.3 7.9 10.2 9.1 8.7
7 7.9 7.1 7.9 12.0 14.6 14.2

Table 5. Distress of Each Field Core

Number of the field core
Section
1 2 3 4 5 6 7 8
1 D&C None None D None None None None
2 None D None None None C None C
3 None None D C C C C None
4 None C D&C None D&S C C D
5 None None None None None None None None
6 None D D None None D C D
7 S None S None None None None None
Note: C, D, and S stand for cracking, debonding, and stripping, respectively.

22
 North Carolina Department of Transportation
Research and Development

Figure 9. Field cores selected in Sections 1 through 7 for dynamic modulus testing.

5. LABORATORY TESTS
As described, 14 field cores (two from each of the seven test sections) were selected for dynamic
modulus testing. Analysis of the Pavement ME method was conducted based on the three
selected core locations (i.e., Locations A, B, and C). Because the NC 96 pavement consists of
multiple AC layers and most of them are less than 1.5 inches in thickness, the individual layers
could not be evaluated through laboratory testing. The top layer (S 9.5B) is relatively thick (3
23
 North Carolina Department of Transportation
Research and Development

in.) and could be evaluated easily based on visual observation of the core surface. To investigate
ways to select a representative layer, each core was divided into a top part (the S 9.5B layer) and
bottom part (the rest of the AC layers). Accordingly, tests of the top part, bottom part, and total
core were conducted separately to mimic scenarios in which each case (top, bottom, and total
core) was selected as the representative lift to determine the damaged mastercurve. To maximize
testing efficiency, the procedures and tests were conducted in a certain order, as shown in Figure
10. Details of each test are given in the following sections.

Figure 10. Flowchart of test plan for dynamic modulus testing.

5.1. Testing Required to Determine Input Properties for Witczak’s Predictive Equation

5.1.1. Volumetric Property Tests and Gradation Analysis

The bulk specific gravity (Gmb) of each core was measured before breaking down the specimen.
After this property was obtained, the core was heated to 230°F. Edge particles were removed
because these particles are affected by cutting and do not hold their original aggregate shape. The
maximum specific gravity (Gmm) was measured from the loose mix following AASHTO T 209.
24
 North Carolina Department of Transportation
Research and Development

Next, ignition oven tests were performed to obtain the binder content. The left-over aggregate
from the ignition oven tests was used for gradation analysis that provided the percentage of the
aggregate that was retained on each sieve. Finally, the specific gravity values of the coarse
aggregate and fine aggregate were measured to help determine the effective binder
content/volume.

Each test had two replicates. In order to mitigate field variation, every effort was made to ensure
that the materials used in each test came from the same location. Moreover, volumetric
measurements were taken only of the top part and bottom part of each core; then, the same data
were used to calculate the corresponding property for the total core. Despite using this procedure,
a single core was unable to provide enough material to satisfy the minimum mass requirement
for gradation analysis and specific gravity testing. Therefore, in such cases, nearby cores were
used to complete the tests.

5.1.2. Extraction, Recovery, and Binder Viscosity Measurements

For the selected specimens, vertical cores that contained all the targeted layers were used for
binder extraction and recovery. To make sure any solvent was removed completely following
binder recovery, a degassing oven and Fourier transform infrared spectroscopy were used to
check the amount of solvent that remained in the binder specimens. Once no solvent could be
detected, the extracted binder was put into a dynamic shear rheometer (DSR) to measure the
shear modulus values at 1.59 Hz and multiple temperatures (40°F, 55°F, 70°F, 85°F, 100°F,
115°F, and 130°F), as specified in the Pavement ME Guide (ARA 2004). Equation (10) was
employed to convert the shear modulus to viscosity.

4.8628
G*  1 
η=   1000 (10)
10  sin δ 
where

η = viscosity, cP,

|G*| = shear modulus, Pa, and

δ = phase angle, °.

Extraction and recovery were performed only for the top part and bottom part of each core. As
for the total core, instead of taking another core sample to run extraction and recovery, the
viscosity values of the top and bottom parts were used to estimate the viscosity of the total core
based on the assumption shown in Equation (11). This decision was made to maintain
consistency in the results because any variation due to sample selection or test operations could
influence the interpretation of the results dramatically. In addition, even though the total core
was extracted, homogenizing it for DSR testing would be difficult due to its small specimen
25
 North Carolina Department of Transportation
Research and Development

geometry (8 mm in diameter and 2 mm in height), which could cause errors in the sampling
process when the amount of extracted binder is large.
η w +η w
ηtotal = top top bottom bottom (11)
wtop + wbottom
where

ηtotal = viscosity of total core,

wtop = top part weight,

ηtop = viscosity of top part,

ηbottom = viscosity of bottom part, and

wbottom = bottom part weight.

5.2. Dynamic Modulus Testing

The selected cores were trimmed and smoothed before testing to ensure that they would make
good contact with the loading platens to ensure a uniform stress condition. Dynamic modulus
measurements of the total vertical core were taken first of specimens 150 mm (6 in.) in diameter.
The reason for not using 100-mm (4-in.) specimens is that the cores could debond easily at the
layer interface during the additional coring process. A Material Test System (MTS) was used to
conduct the dynamic modulus tests. The test temperatures used for the dynamic modulus tests
were 39.2°F, 68°F, and 104°F and the test frequencies were 0.1 Hz, 0.5 Hz, 1 Hz, 5 Hz, 10 Hz,
and 25 Hz. Figure 11 presents the test configuration. The reason to use this test configuration is
to include all the layers in the testing. Before the measurements were taken, a dummy specimen
of the same size as the test specimens was used to determine the conditioning time. Two thermo-
couples were installed on the dummy specimen by drilling and sealing: one on the surface and
the other at the center of the specimen. By placing the dummy specimen into the same
environment as that used for dynamic modulus testing, the temperatures could be recorded and
evaluated to determine the conditioning time. As an example, Figure 12 presents the 104°F data.
Based on this plot and data analysis, the conditioning time at this temperature was determined to
be five hours. Following the same procedure, the conditioning time of five hours also was
confirmed as a reasonable time at the other test temperatures.

26
 North Carolina Department of Transportation
Research and Development

Figure 11. Dynamic modulus test configuration.

Figure 12. Example of conditioning time determination (104°F).

For the cores selected for Pavement ME analysis, after measurements were taken of the whole
field core, the core was cut into the top and bottom parts. Two cores (4.3 in. × 1.5 in. or 110 mm
× 38 mm) were obtained from the top part by coring the field core horizontally, but the bottom
part was still tested vertically. The reason for using horizontal cores for the top part is based on

27
 North Carolina Department of Transportation
Research and Development

the vertical thickness limitation (Castorena et al. 2017, Pape et al. 2018). The test temperatures
and frequencies used to test these two specimen geometries were the same as the conditions used
for the total core.

6. DYNAMIC MODULUS TEST RESULTS

Figure 13, Figure 14, Figure 15 and Figure 16 present the dynamic modulus test results for the
selected 14 cores and Figure 17, Figure 18, Figure 19, and Figure 20 present the phase angle
results after shifting the measured modulus values and phase angles horizontally using the time-
temperature superposition principle. In the graphs, the specimen name is formatted as X-Y,
where X represents the section number and Y represents the location of the core. The measured
results clearly indicate that the modulus obeys the time-temperature superposition principle very
well. Most of the phase angle results also have smooth curves after horizontal shifting. The
measured stiffness values that the specimens display in the reduced frequency domain are
important because these values indicate the modulus variance throughout the NC 96 test sections.
In addition, the measured dynamic modulus values can be used in comparison with the
backcalculated modulus values when the Pavement ME backcalculation tool is applied.

Figure 13. Dynamic modulus results: (a) 1-2, (b) 1-10, (c) 2-1, and (d) 2-12.

28
 North Carolina Department of Transportation
Research and Development

Figure 14. Dynamic modulus results: (a) 3-1, (b) 3-13, (c) 4-1, and (d) 4-13.

29
 North Carolina Department of Transportation
Research and Development

Figure 15. Dynamic modulus results: (a) 5-3, (b) 5-12, (c) 6-1, and (d) 6-13.

Figure 16. Dynamic modulus results: (a) 7-2, (b) 7-11.

30
 North Carolina Department of Transportation
Research and Development

Figure 17. Phase angle results: (a) 1-2, (b) 1-10, (c) 2-1, and (d) 2-12.

31
 North Carolina Department of Transportation
Research and Development

Figure 18. Phase angle results: (a) 3-1, (b) 3-13, (c) 4-1, and (d) 4-13.

32
 North Carolina Department of Transportation
Research and Development

Figure 19. Phase angle results: (a) 5-3, (b) 5-12, (c) 6-1, and (d) 6-13.

Figure 20. Phase angle results: (a) 7-2, (b) 7-11.

A sigmoidal function was used to fit the dynamic modulus data to achieve a single smooth
mastercurve, as shown in Equation (12). Equation (13) shows the relationship between reduced
frequency and frequency. A second polynomial function was used to obtain the shift factor, as
shown in Equation (14). By fitting the mastercurves presented in Figures 13 through 18 using
these three equations, the coefficients of the model, shown in Table 6, could be determined.
Table 6 and the mastercurves for each core show that the mastercurve magnitude varies
according to pavement location, which is indicated by the sigmoidal coefficients and the
33
 North Carolina Department of Transportation
Research and Development

modulus values in the figures. However, the shapes of the mastercurves are all similar. In the
next step, these coefficients are used to estimate the elastic modulus value of each field core for
comparison against the backcalculated values.
c2
log E ( f R ) = c1 + − c3 − c4 log f R
(12)
1+ e
f R = f  T (13)
log T (T ) = aT 2 + bT + c (14)
where

f = frequency, Hz,

αT = shift factor,

fR = reduced frequency, Hz,

T = temperature, °C, and

a, b, c = fitting coefficients.

Table 6. Coefficients of Mastercurves

Name a b c c1 c2 c3 c4
1-2 -0.0002 -0.1036 2.1487 1.7530 2.5314 1.0094 0.4782
1-5 0.0005 -0.1524 2.8427 1.7516 2.6181 1.3210 0.4132
2-1 0.0009 -0.1741 3.1205 0.9318 3.2882 1.5687 0.3998
2-12 0.0004 -0.1552 2.9504 0.2346 4.2917 1.4242 0.3137
3-1 0.0010 -0.1773 3.1569 0.4487 3.8958 1.6360 0.3570
3-13 0.0001 -0.1196 2.3498 1.3130 3.0264 1.1796 0.4553
4-1 0.0005 -0.1384 2.5769 1.5036 2.2355 1.6042 0.4941
4-13 0.0004 -0.1386 2.6266 0.5525 3.5895 1.6241 0.3883
5-3 0.0004 -0.1498 2.8314 2.0491 2.4165 0.7302 0.4570
5-12 0.0011 -0.1995 3.5553 1.5256 3.0600 1.2600 0.4319
6-1 0.0006 -0.1587 2.9220 0.4325 3.8777 1.5187 0.3435
6-13 0.0009 -0.1628 2.9026 0.9860 3.4043 1.3096 0.3980
7-2 -0.0004 -0.0741 1.6276 2.0631 1.7126 2.0348 0.8492
7-11 0.0008 -0.1598 2.8595 0.8505 3.7548 1.1791 0.4226
Note: The specimen name is formatted as X-Y, where X represents the section number and Y
represents the location of the core.

34
 North Carolina Department of Transportation
Research and Development

7. COMPARISON OF MEASURED DYNAMIC MODULUS VALUES

VERSUS BACKCALCULATED MODULUS VALUES
To compare the measured modulus values with the backcalculated values, the first step is to
convert the dynamic modulus to the elastic modulus. In previous studies, the FWD frequency
was normally assumed to be 5~30 Hz and 30 Hz was frequently used in literature (AASHTO
2017, Lee et al. 2017, Ayyala et al. 2017, 2018). For this study, the dynamic modulus at 30 Hz is
assumed to be approximately equal to the elastic modulus, as shown in Equation (15).

| E*| f =30Hz EElastic (15)

where

|E*|f=30Hz= dynamic modulus value when the frequency is 30 Hz, and

EElastic = elastic modulus.

Initially, only one AC layer and one subgrade layer were considered for backcalculation.
Nevertheless, the backcalculated results were found to be dependent on the way the material
properties and structural information were input. Specifically, using only a one-layer subgrade in
the backcalculation sometimes resulted in a large root mean square error (RMSE). Based on this
observation, scenarios with two subgrade layers (20 in. + infinity) also were tried. Meanwhile,
the subgrade modulus values were determined using three methods: DCP measurements,
Equation (16) predictions (AASHTO 1993), and direct backcalculation.

Esg =
(1 − μ ) P
2

(16)
πd r r
where

dr = deflection at distance r from the center of the load,

Esg = subgrade resilient modulus,

μ = Poisson’s ratio,

P = applied load, and

r = distance from center of load.

Instead of showing the results for each test section, the preselected locations A, B, and C are
chosen here for evaluation and comparison purposes. In total, seven scenarios were tried for
backcalculating the dynamic modulus values, as shown in Table 7. Because the RMSE must be
less than 5 to be considered for reasonable backcalculation results (AASHTO Pavement ME
35
 North Carolina Department of Transportation
Research and Development

Design Task Force 2017), the only scenario that satisfies this requirement is the last row in Table
7, which will be used for elastic modulus determination. Note that, although the results for the
other locations are not shown, their circumstances are similar to those explained above in terms
of modulus backcalculation.

Table 7. Backcalculated Modulus Values

Number of Determined Modulus Valuesf (ksi)
Subgrade Modulus
Subgrade
Determination Method A B C
Layers
1 DCP Measured 206 + 39 (83.1) 108 + 35 (34.6) 866 + 26 (6.6)

1 Predicted 118 + 97 (44.7) 166 + 24 (10.8) 966 + 25 (9.4)

1 Backcalculated 151 + 72 (39.2) 178 + 22 (9.0) 728+ 28 (3.2)

DCP +
2 250 + 39 + 97 (37.0) 124 + 35 + 24 (16.5) 943 + 26 + 25 (9.2)
Predicted
DCP +
2 245 + 39 + 73 (30.1) 128 + 35 + 22 (15.4) 746 + 26 + 28 (3.0)
Backcalculated
Backcalculated +
2 796 + 10 + 97 (14.6) 289 + 13 + 24 (1.6) 634 + 48 + 25 (6.9)
Predicted
Backcalculated +
2 498 + 12 + 74 (4.8) 292 + 13 + 24 (1.7) 848 + 16 + 30 (1.8)
Backcalculated
f
A and B represent two separate locations in Structure 1, C represents the location in Structure 2, and the
backcalculated modulus values are formatted as X + Y (Q) or X + Y + Z (Q) where X = modulus of AC, Y and Z =
moduli of subgrade, and Q = RMSE, %.

Another input that is required for the viscoelastic and elastic conversion is the mid-depth
temperature of the pavement. This temperature can be predicted using the LTPP program’s
BELLS3 equation, shown here as Equation (17) (Lukanen et al. 2000).

Td = 0.95 + 0.892 IR
+ ( log ( d ) − 1.25 ) ( −0.448IR + 0.621(1 − day ) + 1.83sin ( hr18 − 15.5 ) ) (17)
+0.042 IR sin ( hr18 − 13.5 )
where

Td = temperature at pavement depth, °C,

IR = infrared pavement surface temperature, °C,

d = depth at which the temperature needs to be predicted, mm,

1-day = average air temperature the day before testing, °C,

36
 North Carolina Department of Transportation
Research and Development

sin = sine function in an 18-hr clock system, and

hr18 = time of day calculated using an 18-hr AC temperature rise and fall time cycle.

Using the aforementioned method, the AC elastic modulus values at the 14 locations where the
dynamic modulus mastercurves were determined were backcalculated. Figure 21 presents a
comparison of the backcalculated modulus values versus the lab-measured modulus values. The
results show that the lab-measured modulus values are much higher than the backcalculated
modulus values for all locations. This outcome may be related to the fact that NC 96 is an old
pavement and has aged for a long time in the field. When conducting the FWD tests, the
technicians noticed that the pavement surface already had many cracks, as shown in Figure 22.
Therefore, when the FWD loading plate was dropped onto the pavement surface, the load
transfer throughout the pavement may have been affected by existing cracks. If debonding of the
layers was present beneath the road surface, then the modulus values measured via the FWD
would be affected accordingly. In comparison, the dynamic modulus specimens tested in the
laboratory did not have debonding; thus, the modulus values obtained from those measurements
should be higher than the field modulus values.

Figure 21. Comparison between backcalculated modulus values versus lab-measured modulus
values.

37
 North Carolina Department of Transportation
Research and Development

Figure 22. Cracks on NC 96 pavement surface.

To evaluate the amount of cracking on NC 96, the cracking condition survey data obtained from
the NCDOT were checked. Table 8 shows that the percentages of alligator cracking are high,
revealing that the FWD test results could be affected by these cracks when the stress wave
propagates throughout the pavement.

Table 8. Alligator Cracking Condition Survey Results

Name Structure 1 Structure 2
Amount of Alligator Cracking (%) 33 28
Ratings of Transverse Cracking Low None

8. EVALUATION OF THE REHABILITATION ANALYSIS LEVELS IN

THE PAVEMENT ME GUIDE
This part of the evaluation is based on the field cores that were extracted from the three
preselected locations (i.e., locations A, B, and C). The NCSU research team determined dynamic
modulus mastercurves for the top, bottom, and total core and conducted the laboratory tests
required by Witczak’s predictive equation following the procedures listed in Section 5.1.
Following the Pavement ME Guide, the research team also performed the different levels of
rehabilitation design analysis in this study. Sections 8.1, 8.2, and 8.3 provide details of the
Levels 1, 2, and 3 analyses, respectively.

38
 North Carolina Department of Transportation
Research and Development

8.1. Level 1 Analysis

The test results for the total cores were used to determine the inputs for Witczak’s predictive
equation to satisfy the equation for Level 1 analysis. In addition, FWD tests and backcalculations
were conducted to obtain the field modulus values following the protocol described in Section 7.
To determine the damage factor (dAC), the backcalculated AC modulus value was put into
Equation (6) for comparison with the undamaged modulus value predicted by Witczak’s
equation at the FWD testing temperature and frequency. The equivalent frequency of the FWD
test was assumed as 30 Hz. Table 9 lists the corresponding damage factor results.

8.2. Level 2 Analysis

The Pavement ME Guide requires a condition survey of alligator cracking for Level 2 analysis.
The NCDOT uses four ratings to evaluate the amount of alligator cracking in the recorded
automated distress data: none, ‘alligator low’, ‘alligator moderate’, and ‘alligator high’. In earlier
NCDOT research work, a conversion equation, shown here as Equation (18), was developed to
convert the NCDOT condition survey data to the LTPP condition survey data (Corley-Lay et al.
2010). Based on this equation, the alligator cracking for the three locations is shown in Table 9.

Total Alligator Cracking = X 1 (m 2 )( Low) + X 2 (m 2 )( Mod ) + X 3 (m 2 )( High) (18)

where

X1 = 5.9,

X2 = 1.7,

X3 = 13.9,

Low = amount of cracking that corresponds to the rating ‘alligator low’, m2,

Mod = the amount of cracking that corresponds to the rating ‘alligator moderate’, m2, and

High = the amount of cracking that corresponds to the rating ‘alligator high’, m2.

The condition survey data for alligator cracking were used in Equation (7) to estimate the
damage factor for Level 2 analysis. Table 9 also lists these calculated damage factors (dAC). The
dAC values at Level 1 and Level 2 differ significantly for all three locations, revealing that these
two levels can lead to very different results. Note that the Level 3 results are not shown in the
table because Level 3 requires only condition survey ratings and the AASHTOWare Pavement
ME Design software calculates the dAC internally.

39
 North Carolina Department of Transportation
Research and Development

Table 9. Damage Factor Results

Name Core Location A B C
AC Thickness (in.) 8.5 8.7 12.0
Mid-depth Temperature of Pavement (°F） 51.0 50.8 54.7
Level 1 Undamaged |E*| at FWD Test Temperature
2,736,286 2,703,510 2,525,988
and Frequency (psi)
Backcalculated Damaged |E*| (psi) 498,100 291,600 848,400
dAC 2.15 3.04 1.57
FCbottom (%) 48.4 3.7 5.4
Level 2
dAC 0.07 0.03 0.02
Note: |E*| is dynamic modulus, dAC is damage factor, and FCbottom is amount of alligator cracking that initiates at the
bottom of hot mix asphalt layers, %.

8.3. Damaged Mastercurve Comparison

Figure 23 (a), (b), and (c) present the resulting damaged mastercurves for the three core locations
(A, B, and C), respectively. Figure 23 shows that the damaged mastercurves of Level 1 are
shifted downwards from the original undamaged mastercurves. The amount of downward shift
represents the severity of the distress in the existing pavement. Figure 23 also shows that the
downward shift at Location B is the greatest among the three locations and is the least at
Location C. This observation matches the level of severity of the debonding distress observed
from the cores. Most cores from Location B (except the cores that are described in Chapter 7)
already had debonded when they were extracted, whereas cores from Location C showed no
visual signs of distress.

Another important observation from Figure 23 is that the damaged mastercurves generated by
Level 2 analysis almost overlap with the undamaged mastercurves, indicating that the field
modulus values are similar to the undamaged modulus values despite the fact that 48.4% of the
alligator cracking that initiated at the bottom of the hot mix asphalt layers, FCbottom, was detected
at Location A. The reason for this overlap may be related to the current coefficients used in
Equation (7) in AASHTOWare that were updated from the old coefficients. Originally, C1 and C2
were both 1.0 according to the Pavement ME Guide (AASHTO 2015). Hence, the research team
tried the old coefficients in Equation (7) in order to determine the damaged mastercurve; the
corresponding plots are shown in Figure 23 as well, labeled as ‘Predicted Damaged (Level 2 –
Old Coeff)’. As shown in Figure 23 (a), when the amount of alligator cracking is 48.4%, the
predicted damaged mastercurve at Level 2 is lower than the mastercurve generated by Level 1.

Another possible way to explain the unreasonable overlap is to calibrate Equation (7) to match
the damage factor from Equation (6). However, this solution works only for Location A because
the amount of alligator cracking (FCbottom) at Locations B and C is extremely low. No matter
whether calibrated coefficients are used or not, Equation (7) will not give the same damage
40
 North Carolina Department of Transportation
Research and Development

factor (dAC) values as Equation (6). In spite of the contradictory results obtained for Levels 1 and
2, Location A nonetheless can be used to develop a preliminary calibrated Equation (7) for North
Carolina, where C2 equals 1.11 and the other coefficients keep their original values. Different
sets of Pavement ME coefficients are shown in Table 10, and NC 96 calibrated coefficients are
recommended for NCDOT. In addition, the measured mastercurves for the cores lie lower than
the undamaged mastercurve predicted from Witczak’s predictive equation. This observation is
reasonable for the field cores because Witczak’s equation predicts the asphalt mixture’s modulus
value at the mixture’s virgin state immediately after mixture fabrication. In contrast, the
measured modulus values were obtained from field cores that had experienced field damage and
had lost some integrity.

Table 10. Pavement ME Coefficients

Pavement ME C C2 C1 ’ C2 ’
Coefficients 1

Old
1 1
Coefficients
5 for AC less than 5”
Current 3.9666 for AC greater than 12”
1.31 −2.856
Coefficients Otherwise, -2C2’ −2.40874 − 39.748 × (1+h )
AC
C2 = 0.867+0.2583×hAC
NC 96
Calibrated 1.31 1.11
Coefficients

Moreover, the measured mastercurve is higher than the Level 1 predicted damaged mastercurve
at all three core locations, especially at high reduced frequencies that represent low-temperature
behavior. Also, the measured mastercurve is steeper than the Level 1 predicted damaged
mastercurve at all three core locations, indicating greater temperature and rate dependency.
Three potential reasons for these observations are as follows. First, the method used to shift the
undamaged mastercurve downward to construct the damaged mastercurve is questionable. That
is, the shape of the mastercurve, which represents the temperature and rate dependence of the
material, may be affected by damage. Second, although the properties of aged binder (extracted
and recovered from field cores) were used in Witczak’s equation to generate the undamaged
mastercurve, the applicability of Witczak’s equation, which was calibrated using dynamic
modulus data of short-term aged mixtures, to aged mixtures has not been fully verified. Third,
the downward shift used in Level 1 is based on FWD deflections that were affected by the
presence of debonding in the pavement, whereas the measured mastercurve was developed based
on cores without debonding or at least without enough debonding to separate the layers by
manual force. Therefore, it makes sense that the Level 1 predicted damaged mastercurves lie
below the measured mastercurves.
41
 North Carolina Department of Transportation
Research and Development

42
 North Carolina Department of Transportation
Research and Development

Figure 23. Undamaged and damaged predicted mastercurves and measured mastercurve: (a) at
core Location A, (b) at core Location B, and (c) at core Location C.

Note that the current AASHTOWare Pavement ME Design software does not provide an option
for inputting measured modulus values with respect to the properties of existing layers. Based on
this limitation and the observations described here, the NCSU research team recommends to the
NCDOT that, for now, the Level 1 Pavement ME method should be used to determine the
dynamic modulus mastercurves for existing asphalt layers.

9. EVALUATION OF REPRESENTATIVE LAYERS

9.1. Damaged Mastercurve Comparison
As noted in Chapter 8, the Level 1 Pavement ME method is recommended to be used by the
NCDOT. This section addresses the question as to which layer(s) should be selected in a
multilayered existing pavement to determine the input properties for Witczak’s equation. As
described earlier, the selected field cores were investigated as three separate cases in this study:
the top and bottom parts, and the total core. Each case was considered as the representative layer
for all the AC layers. That is, the measured properties of each case were assumed to represent the
properties of all the existing AC layers.

Figure 24, Figure 25, and Figure 26 respectively show the mastercurve results obtained for
Locations A, B, and C. For both the predicted and measured results, most data points for the total
core are located in the middle of the curves for the top and bottom parts. This method of
presentation not only shows that the total core’s behavior tends to be the combined results of the
top and bottom parts, but also implies that the subjective choice of representative layers has the
potential to result in different mastercurves. To determine the damaged mastercurves, the Level 1
and Level 2 procedures were applied to the undamaged mastercurves that were predicted from
43
 North Carolina Department of Transportation
Research and Development

Witczak’s equation and the properties that were determined from the top part, bottom part, and
total core, respectively. The results are shown in the last two plots, i.e., (c) and (d), of Figure 24,
Figure 25, and Figure 26. Although the magnitude and vertical locations of the curves vary
among the core locations, the difference in modulus values among the three sets of data seems
not to be significant in logarithmic space. However, visual judgement may not necessarily reflect
the true scenario. Whether or not this amount of difference would result in a large performance
variation is evaluated later in this report.

Figure 24. Mastercurves for core Location A: (a) predicted by Witczak’s predictive equation, (b)
measured, (c) damaged mastercurves at Level 1, and (d) damaged mastercurves at Level 2.

44
 North Carolina Department of Transportation
Research and Development

Figure 25. Mastercurves for core Location B: (a) predicted by Witczak’s predictive equation, (b)
measured, (c) damaged mastercurves at Level 1, and (d) damaged mastercurves at Level 2.

45
 North Carolina Department of Transportation
Research and Development

Figure 26. Mastercurves for core Location C: (a) predicted by Witczak’s predictive equation, (b)
measured, (c) damaged mastercurves at Level 1, and (d) damaged mastercurves at Level 2.

In all the cases presented in Figure 24, 23, and 24, the existing layers are regarded as one single
layer and, therefore, only one in situ modulus value needs to be backcalculated for the AC layer.
In fact, even though the accuracy may be compromised when the number of backcalculated
layers is increased, the Pavement ME Guide does not prohibit the use of multiple AC layer
backcalculations in the current pavement characterization protocol (AASHTO Pavement ME
Design Task Force 2013). Hence, the research team tried another scenario in the analysis where
the existing layers were divided into two sublayers (top and bottom) and backcalculations were
conducted for each sublayer at Level 1. Table 11 presents the undamaged modulus values,
damaged modulus values, and damage factor values at the FWD testing temperature and
frequency. As this approach resulted in two sets of damaged mastercurves, a direct comparison
with the previous scenarios was not feasible. However, upon close inspection, the backcalculated
damaged modulus values indicate that the averaged values of the top layer and bottom layer for
46
 North Carolina Department of Transportation
Research and Development

all three locations (A, B, and C) are 476,700 psi, 306,550 psi, and 885,200 psi, respectively,
which are reasonably close to the values presented in Table 9 when the existing pavement is
considered as a single layer.

Table 11. Damage Factors Resulting from Dividing Existing Layers into Two Sublayers
Core Location A B C
Undamaged |E*| - Top Layer (psi) 2,444,663 2,613,167 2,083,527
Undamaged |E*| - Bottom Layer (psi) 2,793,981 2,645,637 2,730,193
Backcalculated Damaged |E*| - Top Layer (psi) 304,100 81,300 355,900
Backcalculated Damaged |E*| - Bottom Layer (psi) 649,300 531,800 1,414,500
dAC - Top Layer 2.82 5.62 2.38
dAC - Bottom Layer 1.99 2.17 1.11

Figure 27 (a), (b), and (c) present the results for the scenario in which the damage factors that
correspond to the separate sublayers are used to develop the damaged mastercurves for Locations
A, B, and C, respectively. Compared to the cases shown in Figure 24, 23, and 24, the magnitudes
of the damaged mastercurves for the top and bottom layers change because the vertical location
of the damaged mastercurves is dominated by the FWD backcalculated modulus values. Also,
now the mastercurves for the laboratory measurements are not necessarily located between the
undamaged curves and damaged curves. For example, Figure 27 (b) shows that the measured
mastercurve for the bottom layer almost overlaps with the predicted undamaged curve at low
reduced frequencies, and Figure 27 (c) shows that the measured curve is lower than the predicted
damaged curve at low reduced frequencies. If Level 2 or 3 is used, then only one damage factor
value can be determined using the cracking condition survey and Equation (7), which is the same
value as that shown in Table 9. As such, using Level 2 or 3 will increase the uncertainties and
errors of damage characterization because both the top and bottom sublayers need to use the
same damage factor in order to calculate the damaged mastercurve. Regardless of other
disadvantages, this feature should be considered carefully before dividing an existing pavement
into sublayers.

47
 North Carolina Department of Transportation
Research and Development

48
 North Carolina Department of Transportation
Research and Development

Figure 27. Mastercurves obtained by dividing existing layers into two sublayers: (a) at core
Location A, (b) at core Location B, and (c) at core Location C.

9.2. Pavement Performance Predictions

This section presents four ways to determine the damaged mastercurve. Table 12 provides a
summary of these four methods in terms of the layer(s) (‘Case’) used to determine the material
properties for the undamaged mastercurve generation and the backcalculation schemes (single
layer vs. two separate layers).

Table 12. Four Ways to Determine Damaged Mastercurve

Case Layer Used for Undamaged Mastercurve Backcalculation Scheme
Top Layer Top Layer All AC Layers in a Single Layer
Bottom Layer Bottom Layer All AC Layers in a Single Layer
Total Core Total Core All AC Layers in a Single Layer
Top and Bottom Layers
Top + Bottom Each of Top and Bottom Layers
Separated

In order to see the differences from another perspective other than modulus magnitude or curve
shape and location, the NCSU research team decided to predict the pavement performance using
AASHTOWare Pavement ME Design software. For this simulation, the annual average daily
truck traffic (AADTT) was obtained from the NCDOT’s historical records and the nearest station
to NC 96 was selected as the climate station. Default values were used for other required
information. Table 13 presents details of the Pavement ME software inputs. Note that Table 13
presents the critical performance results obtained by using different representative layers at Level
1. The results for Level 2 and Level 3 are not shown in the table because the predicted amount of
distress is zero in Level 2 and mostly zero (except for low numbers) for Location A in Level 3.

49
 North Carolina Department of Transportation
Research and Development

The reason for the low distress prediction in Level 2 is that the damaged mastercurve is
essentially the same as the undamaged mastercurve (see Figure 23). For Level 3, the condition
ratings for Locations A, B, and C are poor, excellent, and good, respectively, according to the
AASHTO Guide (1993). The resulting amount of distress derived from the Pavement ME
simulations for the overlay design shown in Table 13 is nearly zero.

50
 North Carolina Department of Transportation
Research and Development

Table 13 presents three sets of Pavement ME outputs. Note that other performance outputs (e.g.,
permanent deformation, thermal cracking) are not shown here because they either indicated
‘good’ condition all the time or they are not the main focus of this study. The results from
Locations A and C indicate that dividing the existing pavement into two sublayers (denoted as
‘Top + Bottom’ in Table 13) could lead to predictions of much worse cracking. As for the other
three representative layer choices (top, bottom, and core), their cracking performance varies
considerably at Location A but is relatively stable at Locations B and C. At Location B, 100%
cracking is predicted for all four cases, which is probably due to the fact that the backcalculated
in situ damaged modulus value at Location B is much lower than at Locations A and C, as shown
in Table 9. Such a low modulus value makes the corresponding damaged mastercurve exhibit
relatively low magnitude, which facilitates crack propagation. Thus, no matter which layer is
selected as the representative layer, the reflective cracking resistance is predicted to be very poor.
In terms of pavement life, the variation among the different representative layer choices is
relatively small. In fact, reflective cracking dominates the pavement life in all cases. From a
practical point of view, because pavement life does not vary much with respect to the choice of
representative layer, using the total core to characterize the existing pavement is recommended.
Based on the experience of the NCSU research team, the tests required for Witczak’s predictive
equation consume a substantial amount of materials, particularly to determine the gradations and
aggregate specific gravity. Therefore, using the total core to obtain the Witczak’s predictive
equation coefficients would save time and effort.

51
 North Carolina Department of Transportation
Research and Development

Table 13. Inputs and Outputs for Pavement Performance Simulation

Core Location A B C
Design Life (years) 20
Two-Way AADTT 1,560
Number of Lanes 2
Trucks in Design Direction (%) 50
Trucks in Design Lane (%) 100
Operational Speed (mph) 40
Traffic Growth Rate (%) 3
Traffic Growth Function Compound
Input
Climate Station US.NC (139555)
Overlay Thickness (in.) 3
Overlay Dynamic Modulus
Default Values at Level 2
(psi)
Milled Thickness (in.) 2
Measured Existing Layer 8.46 = 2.08 + 8.66 = 2.11 + 12.01 = 3.07 +
Thicknessa (in.) 6.38 6.55 8.94
Measured Existing Dynamic Measured Inputs for Levels 1 and 2
Modulus (psi) Default Values for Level 3
DCP Measured Subgrade
38,983 35,280 26,164
Modulus (psi)
Top Layer 48.6 100 0.16
Bottom Layer 1.5 100 0.03
FCbottom
Total Core 3.1 100 0.05

Critical Top + Bottom 95.2 100 99.1

Performance Top Layer 100 100 61.74
FCbottom Bottom Layer 76.1 100 65.03
Output
+
at FCrefb Total Core 78.7 100 63.99
Level 1 Top + Bottom 100 100 100
(%) Top Layer 6.0 1.3 7.5
Pavement Bottom Layer 5.8 1.8 6.9
Life Total Core 5.9 1.4 7.0
Top + Bottom 5.6 1.2 6.0
a
The format of the thickness input is L = M + N. If the existing pavement is regarded as one layer, the thickness
input is L. If the existing pavement is divided into two sublayers, the inputs are M and N for each of them.
b
FCref is the amount of reflective cracking (% lane area).

10. DYNAMIC MODULUS MASTERCURVE BACKCALCULATION

The NCSU research team tried one more option to backcalculate dynamic modulus mastercurves
directly from FWD test data. To achieve this goal, the team developed a highly efficient forward
52
 North Carolina Department of Transportation
Research and Development

model by combining various traditional and recent computational techniques. Specifically, to

obtain the surface response of the pavement under FWD loading, the research team used a modal
expansion technique for the axisymmetric system of discretized layers in the frequency domain
(unlike Hankel transform that is common in many existing models). Such modeling was
facilitated with the help of the so-called ‘complex-length finite element method’ that reduced the
number of elements required to discretize the mesh in the vertical direction. The computational
cost was reduced further with the help of Padé interpolation across the frequency spectrum of the
load. In the end, the accurate surface response of a pavement under typical FWD loading could
be computed within less than 0.2 second on quad-core processors that are common in standard
laptop and desktop computers.

Although the main efficiency of the inversion approach is inherited from the forward model, the
number of forward solutions nonetheless was reduced while ensuring the robustness and
convergence of the backcalculation. The approach involved combining gradient-based
optimization with Monte Carlo-type global optimization techniques. By carefully comparing the
various approaches with respect to cost and performance, the research team determined that a
finite difference computation of the gradient was the most desirable approach and would work
well in a Gauss-Newton optimization framework. To help with convergence towards the global
minimum, the gradient iteration was embedded into a global multi-start optimization framework
based on a Monte Carlo approach. This method, coupled with other techniques such as nonlinear
transformation of the parameter space, led to the reduction of the number of function evaluations
and eventually the overall computational cost. This inversion approach takes 1 to 25 minutes,
most of the time at the lower end of the spectrum.

However, FWD tests involve low-frequency excitation, where the load is significant below 50
Hz to 100 Hz typically. Similarly, the response is also significant below this frequency. Given
that the measurement time is limited to 0.06 seconds, the lower limit of the frequency typically is
5 Hz to 10 Hz. Therefore, for a given pavement structure and known excitation, the response can
be reliably predicted only within this narrow frequency range. A natural consequence is that,
even in the absence of non-uniqueness issues, the pavement modulus can be backcalculated
reliably only for this frequency range. Because this range constitutes only a narrow portion of the
entire mastercurve, the research team concluded that the mastercurve cannot be backcalculated
solely from FWD data.

11. CONCLUSIONS
Based on the field tests and laboratory analysis of field cores extracted from NC 96, the NCSU
research team evaluated the accuracy and efficiency of the three analysis levels used in the
current Pavement ME Guide with regard to the determination of damaged mastercurves of
existing AC layers for rehabilitation design. In addition to comparing the three analysis levels,
the research team also investigated various ways that highway agencies conduct tests that
involve multilayered pavement structure systems. The team investigated four cases in terms of
53
 North Carolina Department of Transportation
Research and Development

undamaged mastercurve characterization and damaged mastercurve determination, and the

corresponding pavement performance predictions.

With respect to the three levels of Pavement ME rehabilitation analysis, the following
conclusions can be drawn from this study.

• For Level 1 applications, FWD measurements and backcalculations are needed to

determine the in situ modulus values. However, the backcalculated modulus values are
related to how the user inputs the structural information into the backcalculation program.
Based on the evaluation results, fixing a certain layer modulus value during
backcalculation is not recommended. Instead, the user should adjust the pavement
structure (i.e., divide the subgrade into two sublayers) to give the program enough leeway
to determine the field modulus values. The RMSE of the output should be lower than 5
percent.

• Comparing the backcalculated elastic modulus values with the laboratory-measured

modulus values, and assuming that the elastic modulus equals the dynamic modulus at 30
Hz, the laboratory-determined modulus value is much higher than the backcalculated
modulus value. The research team believes that this trend is due to the fact that the FWD
measurements could have been influenced by debonding or cracking observed from the
field cores, which was not reflected by the laboratory-measured dynamic modulus values.

• The undamaged mastercurve that is predicted using Witczak’s predictive equation has a
different shape and magnitude than the mastercurve that is based on measured field cores.
Because the shape of the curve is linked to viscoelasticity and the damaged mastercurve
is merely the vertical shift of the undamaged mastercurve, the current protocol in
Pavement ME based on Witczak’s predictive equation could result in erroneous damaged
mastercurves.

• The results from Level 1, 2, and 3 analyses differ significantly in terms of the damage
factor estimations and cracking predictions. Level 1 is recommended as the first choice
for agencies to use in practice. However, if Level 2 or 3 needs to be applied, then a
transfer function that relates the damage factor to the percentage of bottom-up cracking
needs to be calibrated first to ensure that the results are consistent with those of Level 1.
Based on the data used in this project, the calibrated new C2 equals 1.11 but the other
coefficients keep their original values for the transfer function (Equation (7)). However,
more North Carolina field data are needed to confirm the reasonableness of the calibrated
new C2 value.

With respect to multilayered pavement characterization, the following conclusions can be drawn
from this study.

54
 North Carolina Department of Transportation
Research and Development

• When the tests that are specified by Levels 1 and 2 need to be conducted using
multilayered AC pavements, the total core should be used for laboratory evaluation, not
the thickest layer. In this project, different representative layers (the top layer of the core,
bottom layer(s) of the core, and total core) were selected for the Pavement ME method
analysis. The results show that most dynamic modulus data points from the total core are
located in the middle of the curves for the top and bottom parts of the core, indicating that
the total core’s behavior tends to be the combined results of the top and bottom parts.

• Dividing the existing pavement into multiple layers is possible only for Level 1 analysis.
Even so, this approach is not recommended because it requires multilayer
backcalculations and considerable time and resources to characterize the individual layer
materials. However, in terms of pavement life, the variation among the different
representative layer choices is relatively small. In fact, reflective cracking dominates the
pavement life in all cases. Hence, using the total core is the most effective way to
characterize existing AC layers.

With regard to dynamic modulus mastercurve backcalculation, the conclusion drawn by the
research team is that the mastercurve cannot be backcalculated solely from FWD data due to the
limited frequency range that the FWD can capture. Therefore, to determine the damaged
mastercurve, some other means, such as FWD testing at multiple times of a day or multiple
seasons or laboratory measurements, need to be employed to obtain more information outside the
FWD frequency range. Because the NCDOT and other state highway agencies have already
adopted the Pavement ME Guide, instead of developing a completely new protocol, the NCSU
research team recommends that the NCDOT use the modified Pavement ME method based on
NC 96 data.

12. REFERENCES
AASHTO. 1993. Guide for Design of Pavement Structures. American Association of State
Highway and Transportation Officials (AASHTO).

AASHTO Pavement ME Design Task Force. 2013. Material and Design Inputs for Pavement
Rehabilitation with Asphalt Overlays. Presented at AASHTO Pavement ME Design Webinar.

AASHTO. 2015. TP 79-15, Determining the Dynamic Modulus and Flow Number for Asphalt
Mixtures Using the Asphalt Mixture Performance Tester. American Association of State
Highway and Transportation Officials.

AASHTO. 2015. PP 61-13, Developing Dynamic Modulus Master Curves for Asphalt Mixtures
Using the Asphalt Mixture Performance Tester (AMPT). American Association of State Highway
and Transportation Officials.

55
 North Carolina Department of Transportation
Research and Development

AASHTO. 2015. T 342-11, Determining Dynamic Modulus of Hot Mix Asphalt. American
Association of State Highway and Transportation Officials.

AASHTO. 2015. R 62-13, Developing Dynamic Modulus Master Curves for Asphalt Mixtures.
American Association of State Highway and Transportation Officials.

AASHTO. 2015. Mechanistic-Empirical Pavement Design Guide: A Manual of Practice.

American Association of State Highway and Transportation Officials.

AASHTO. 2016. T 209-12, Theoretical Maximum Specific Gravity (Gmm) and Density of Hot
Mix Asphalt (HMA). American Association of State Highway and Transportation Officials.

AASHTO Pavement ME Design Task Force. 2017. FWD Deflection Data Analysis and
Backcalculation Tool: BcT. Presented at AASHTO Pavement ME Design Webinar.

AASHTO Pavement ME Design Task Force. 2017. User Manual for Pavement M-E Deflection
Data Analysis and Backcalculation Tool. American Association of State Highway and
Transportation Officials.

ARA, Inc., ERES Consultants Division. 2004. Guide for Mechanistic-Empirical Design of New
and Rehabilitated Pavement Structures. Part 2. Design Inputs. Chapter 2. Material
Characterization. National Cooperative Highway Research Program, 2004.

Ayyala, D., H. S. Lee, and H. L. Von Quintus. 2017. Damage Characterization in Existing
Asphalt Layer Concrete for ME Rehabilitation Design. Presented at 96th Annual Meeting of the
Transportation Research Board, Washington, D.C.

Ayyala, D., H. S. Lee, and H. L. Von Quintus. 2018. Characterizing Existing Asphalt Concrete
Layer Damage for Mechanistic Pavement Rehabilitation Design. Publication FHWA-HRT-17-
059. FHWA, U.S. Department of Transportation.

Castorena, C., Y. R. Kim, S. Pape, and K. Lee. 2017. Development of Small Specimen Geometry
for Asphalt Mixture Performance Testing. National Cooperative Highway Research Program
Innovations Deserving Exploratory Analysis Project 181.

Christopher, B.R., Schwartz, C.W. and Boudreau, R.L., 2010. Geotechnical aspects of
pavements: Reference manual. US Department of Transportation, Federal Highway
Administration.

Corley-Lay, J., F. M. Jadoun, J. N. Mastin, and Y. R. Kim. 2010. Comparison of flexible

pavement distresses monitored by North Carolina Department of Transportation and Long-Term
Pavement Performance program. Transportation Research Record: Journal of the
Transportation Research Board, 2153(1): 91-96.
56
 North Carolina Department of Transportation
Research and Development

Habbouche, J., E. Y. Hajj, P. E. Sebaaly, and N. E. Morian. 2018. Damage assessment for ME
rehabilitation design of modified asphalt pavements: Challenges and findings. Transportation
Research Record: Journal of the Transportation Research Board, 2672(40): 228-241.

Harsini, I., W. C. Brink, S. W. Haider, K. Chatti, N. Buch, G. Y. Baladi, and E. Kutay. 2013.
Sensitivity of input variables for flexible pavement rehabilitation strategies in the MEPDG.
Airfield and Highway Pavement 2013: Sustainable and Efficient Pavements: 539-550.

Kim, Y. R., Y. Seo, M. King, and M. Momen. 2004. Dynamic modulus testing of asphalt
concrete in indirect tension mode. Transportation Research Record: Journal of the
Transportation Research Board, 1891(1): 163-173.

Lee, H.S., Ayyala, D. and Von Quintus, H., 2017. Dynamic backcalculation of viscoelastic
asphalt properties and master curve construction. Transportation Research Record, 2641(1),
pp.29-38.

Lee, K., S. Pape, C. Castorena, and Y.R. Kim. 2017. Evaluation of Small Specimen Geometries
for Asphalt Mixture Performance Testing and Pavement Performance Prediction. Journal of the
Transportation Research Board, 2631: 74-82.

Loulizi, A., G. W. Flintsch, and K. McGhee. 2007. Determination of in-place hot-mix asphalt
layer modulus for rehabilitation projects by a mechanistic–empirical procedure. Transportation
Research Record: Journal of the Transportation Research Board, 2037(1): 53-62.

Lukanen, E. O., R. Stubstad, R. C. Briggs, and B. Intertec. 2000. Temperature Predictions and
Adjustment Factors for Asphalt Pavement. Publication FHWA-RD-98-085. FHWA, U.S.
Department of Transportation.

Pape, S., K. Lee, C. Castorena, and Y. R. Kim. 2018. Optimization of the laboratory fabrication
of small specimens for asphalt mixture performance testing. Transportation Research Record:
Journal of the Transportation Research Board, 2672(28): 438-450.
https://doi.org/10.1177/0361198118790845.

Park, H.J. and Y.R. Kim. 2004. Primary Causes of Cracking of Asphalt Pavement in North
Carolina: Field Study. International Journal of Pavement Engineering, 16(8): 684-698.

Park, H., M. Eslamania, and Y.R. Kim. 2014. Mechanistic Evaluation of Cracking in In-Service
Asphalt Pavements. Materials and Structures, 47(8): 1339-1358.

Witczak, M.W. 2004. Laboratory determination of resilient modulus for flexible pavement
design (No. 285). National Cooperative Highway Research Program Project No. 1-28A.

57