Local Calibration of the MEPDG Using Test Track Data 
 
by 
 
Xiaolong Guo 
 
 
 
 
A thesis submitted to the Graduate Faculty of 
Auburn University 
in partial fulfillment of the 
requirements for the Degree of 
Master of Science 
 
Auburn, Alabama 
December 14, 2013 
 
 
Keywords: MEPDG, Local Calibration, Transfer Function, Input Level 1 
 
Copyright 2013 by Xiaolong Guo 
 
 
Approved by 
 
David H. Timm, Chair, Professor of Civil Engineering 
Rod E. Turochy, Associate Professor of Civil Engineering 
Randy West, Director of National Center of Asphalt Technology 
 
ii 
 
 
 
 
 
 
 
ABSTRACT 
 
 
 The Mechanistic-empirical Pavement Design Guide (MEPDG), now known as 
AASHTOWare Pavement ME Design, is a state-of-art mechanistic-empirical design method. It 
has been adopted by the American Association of State Highway and Transportation Officials 
(AASHTO) as the new standard for pavement design in the U.S. However, the MEPDG was 
calibrated to various types of pavements located across the country representing diverse design 
conditions. Since the default transfer functions were nationally-calibrated, the performance 
prediction for local projects sometimes may not correlate well with the local performance 
measurements. The National Center for Asphalt Technology (NCAT), equipped with a full-scale 
accelerated Test Track, supported the evaluation and local calibration of the nationally-calibrated 
models. The study results showed the nationally-calibrated fatigue cracking model and the IRI 
model matched the measured values in the 2003 research cycle, and no better coefficients were 
found to improve the model predictions. However, applying the same fatigue coefficients to the 
2006 research cycle resulted in poor correlations to measured performance and no significant 
improvement to fatigue cracking prediction was found. The nationally-calibrated rutting model 
over-predicted rut depths, but the accuracy of the rutting model was significantly improved by 
local calibration. The research methods and findings may be useful for the transportation 
agencies to perform local calibration studies.   
iii 
 
 
 
 
 
 
 
ACKNOWLEDGMENTS 
 
 
 I would like to sincerely acknowledge my academic advisor Dr. David Timm for his 
professional guidance on my graduate study. His patience, inspiration, and rigorousness 
motivated me to reach deeper and be more involved in the research of civil engineering. I not 
only learned a wealth of knowledge and skills, but also shaped research perspectives based on his 
expertise of pavement design and asphalt technology. I truly appreciate his understanding of my 
program progress. It is his critical advice that strengthens my confidence to insist on the study 
until a completion of this program. 
I am truly grateful to Dr. Rod Turochy for his endeavor to support my study, and also 
thank his family for bringing me in community activities. I also would like to thank Dr. Randy 
West for taking time from his busy schedule to review my work and participate as one of my 
committee members. 
I am thankful to Dr. Raymond Powell, Dr. Mary Robbins, and Dr. Fabricio Leiva-
Villacorta for their assistance in conducting experiments. I would like to thank Derong Mai, Xiao 
Zhao, Jingwen Li, and Boyao Zhao, as well as other students of the civil engineering program 
here for their kind help with my work and my life. I also would like to acknowledge Miller 
Writing Center at Auburn University library for supporting the writing development. I appreciate 
the professional suggestions to my thesis writing from their tutors, especially Courtney Hewitt.   
Finally, I would like to thank my parents, my relatives and friends in China for their 
selfless care and encouragement during my study.  
iv 
 
 
 
 
 
 
TABLE OF CONTENTS 
 
 
LIST OF FIGURES ..................................................................................................................... vi 
LIST OF TABLES ....................................................................................................................... ix 
CHAPTER 1 INTRODUCTION .................................................................................................. 1 
 BACKGROUND .............................................................................................................. 1 
 OBJECTIVE ................................................................................................................... 12 
            SCOPE ............................................................................................................................ 12 
            ORGANIZATION OF THESIS ..................................................................................... 13 
CHAPTER 2 LITERATURE REVIEW ..................................................................................... 14 
            INTRODUCTION .......................................................................................................... 14 
 PAVEMENT DESIGN OVERVIEW ............................................................................. 14 
            TRADITIONAL DESIGN METHODS ......................................................................... 28 
            MEPDG........................................................................................................................... 36 
            LOCAL CALIBRATION CASE STUDIES .................................................................. 47 
CHAPTER 3 MEPDG INPUT CHARACTERIZATION AND PERFORMANCE DATA                                                                          
COLLECTION ........................................................................................................................... 69 
            TEST FACILITY INTRODUCTION............................................................................. 69 
 MEPDG INPUTS CHARACTERIZATION .................................................................. 70 
            PERFORMANCE MONITORING ................................................................................ 96 
CHAPTER 4 THE METHODOLOGY OF LOCAL CALIBRATION AND VALIDATION 100 
            INTRODUCTION ........................................................................................................ 100 
v 
 
 LOCAL CALIBRATION ............................................................................................. 102 
            VALIDATION .............................................................................................................. 107 
CHAPTER 5 A METHOD OF RUNNING THE MEPDG AUTOMATICALLY .................. 109 
            BACKGROUND .......................................................................................................... 109 
 METHOD DESCRIPTION .......................................................................................... 110 
            SUGGESTION ............................................................................................................. 114 
CHAPTER 6 RESULTS AND DISCUSSION ......................................................................... 115 
            INTRODUCTION ........................................................................................................ 115 
 RESULTS OF CALIBRATION ................................................................................... 115 
            RESULTS OF VALIDATION ..................................................................................... 120 
            SUMMARY .................................................................................................................. 125 
CHAPTER 7 CONCLUSION AND RECOMMEDATION .................................................... 127 
REFERENCES ......................................................................................................................... 130 
APPENDIX A ........................................................................................................................... 137 
APPENDIX B ........................................................................................................................... 152 
APPENDIX C ........................................................................................................................... 154 
 
  
vi 
 
 
 
 
 
 
 
LIST OF FIGURES 
 
 
Figure 1.1 Scheme of the M-E design .......................................................................................... 9 
Figure 1.2 NCAT Test Track ...................................................................................................... 12 
Figure 2.1 Fatigue cracking ........................................................................................................ 16 
Figure 2.2 Rutting ....................................................................................................................... 17 
Figure 2.3 Longitudinal cracking ................................................................................................ 17 
Figure 2.4 Transverse cracking ................................................................................................... 18 
Figure 2.5 Low pressure tire and high pressure tire.................................................................... 21 
Figure 2.6 Design chart for full depth asphalt concrete .............................................................. 30 
Figure 2.7 Nomograph ................................................................................................................ 32 
Figure 2.8 Chart for estimating effective subgrade resilient modulus ........................................ 34 
Figure 2.9 Pavement structure .................................................................................................... 35 
Figure 2.10 Measured and predicted alligator cracking in western Washington ........................ 53 
Figure 2.11 Measured and predicted alligator cracking in eastern Washington ......................... 53 
Figure 2.12 Measured and predicted longitudinal cracking in western Washington .................. 54 
Figure 2.13 Measured and predicted longitudinal cracking in eastern Washington ................... 54 
Figure 2.14 Measured and predicted transverse cracking in western Washington ..................... 55 
Figure 2.15 Measured and predicted transverse cracking in eastern Washington ...................... 55 
Figure 2.16 Measured and predicted rutting in western Washington ......................................... 56 
Figure 2.17 Measured and predicted rutting in eastern Washington .......................................... 56 
vii 
 
Figure 2.18 Measured and predicted IRI in western Washington .............................................. 57 
Figure 2.19 Measured and predicted IRI in eastern Washington ................................................ 57 
Figure 2.20 Measured versus predicted distresses ...................................................................... 61 
Figure 2.21 Measured versus predicted HMA total rutting before local calibration .................. 63 
Figure 2.22 Measured versus predicted HMA total rutting after local calibration ..................... 65 
Figure 2.23 Measured and predicted rutting for Section 2 before local calibration ................... 67 
Figure 2.24 Measured and predicted rutting for Section 2 after local calibration ...................... 67 
Figure 3.1 Triple trailer truck...................................................................................................... 72 
Figure 3.2 Box trailer truck ......................................................................................................... 72 
Figure 3.3 Number of axles per vehicle ...................................................................................... 73 
Figure 3.4 Hourly traffic distribution for the 2003 research cycle ............................................. 75 
Figure 3.5 Hourly traffic distribution for the 2006 research cycle ............................................. 76 
Figure 3.6 Test Track on-site weather station ............................................................................. 77 
Figure 3.7 Hourly climatic database file ..................................................................................... 78 
Figure 3.8 Cross section of structural sections in the 2003 research cycle ................................. 79 
Figure 3.9 Cross section of structural sections in the 2006 research cycle ................................. 80 
Figure 3.10 Measured vs. predicted E* data ............................................................................... 87 
Figure 3.11 The MEPDG E* master curve for N1-1 lift in the 2003 research cycle.................. 88 
Figure 3.12 Construction and lab record for N1-1 lift in the 2006 research cycle ..................... 89 
Figure 3.13 Cross sections comparison ...................................................................................... 92 
Figure 3.14 Vertical pressures in N1 subgrade in the 2006 research cycle ................................ 95 
Figure 3.15 Sample crack map from section N6 at the end of the 2006 research cycle ............. 98 
Figure 3.16 Dipstick profiler....................................................................................................... 99 
viii 
 
Figure 3.17 ARAN inertial profiler ............................................................................................ 99 
Figure 4.1 Reasonable goodness of fit for model predictions .................................................. 101 
Figure 4.2 Unreasonable goodness of fit for model predictions ............................................... 101 
Figure 4.3 Trial and error calibration steps ............................................................................... 103 
Figure 4.4 Curves in similar shape ........................................................................................... 104 
Figure 4.5 Curves not in a similar shape................................................................................... 104 
Figure 4.6 Project templates in the MEPDG ............................................................................ 106 
Figure 5.1 VB script flow chart ................................................................................................ 111 
Figure 5.2 SSE result in a message box .................................................................................... 112 
Figure 5.3 Dialog box pop-up ................................................................................................... 113 
Figure 5.4 Excel error warning ................................................................................................. 114 
Figure 6.1 Model predictions vs. measured values for fatigue cracking .................................. 116 
Figure 6.2 Fatigue cracking model prediction optimization ..................................................... 117 
Figure 6.3 Model predictions vs. measured values for rutting ................................................. 117 
Figure 6.4 Rutting model prediction optimization  ................................................................... 119 
Figure 6.5 Model predictions vs. measured values for IRI ....................................................... 120 
Figure 6.6 Model predictions vs. measured values for fatigue cracking .................................. 121 
Figure 6.7 Model predictions vs. measured values for rutting ................................................. 122 
Figure 6.8 Model predictions vs. measured values for rutting (excluding N1 and N2) ........... 123 
Figure 6.9 Model predictions vs. measured values for IRI ....................................................... 124 
  
ix 
 
 
 
 
 
 
 
LIST OF TABLES 
 
 
Table 2.1 Sensitivity of the MEPDG calibration coefficients .................................................... 50 
Table 2.2 Final local calibration coefficients .............................................................................. 52 
Table 2.3 Statistical summary of the rutting model local calibration results.............................. 60 
Table 2.4 Statistical summary of validation results .................................................................... 61 
Table 2.5 Recommended calibration coefficients ....................................................................... 62 
Table 2.6 Hypothesis testing before local calibration of the rutting model ................................ 63 
Table 2.7 Hypothesis testing after local calibration of the rutting model ................................... 64 
Table 3.1 Construction and traffic months ................................................................................. 71 
Table 3.2 Initial IRI values ......................................................................................................... 71 
Table 3.3 Surveyed layer thickness of the 2003 sections ........................................................... 83 
Table 3.4 Surveyed layer thickness of the 2006 sections ........................................................... 84 
Table 3.5 E* regression coefficients for the 2003 research cycle ............................................... 86 
Table 3.6 E* regression coefficients for the 2006 research cycle ............................................... 86 
Table 3.7 Unbound materials gradations for the 2003 research cycle ........................................ 91 
Table 3.8 New fill depths for the 2003 research cycle ............................................................... 93 
Table 3.9 Unbound materials moduli for the 2003 research cycle ............................................. 93 
Table 3.10 Unbound materials gradations for the 2006 research cycle ...................................... 94 
Table 3.11 Regression coefficients for unbound materials ......................................................... 96 
Table 3.12 Unbound materials moduli for the 2006 research cycle ........................................... 96 
x 
 
Table 6.1 Statistical analysis for the 2003 fatigue cracking predictions................................... 116 
Table 6.2 Statistical analysis for the 2003 rutting predictions .................................................. 118 
Table 6.3 Statistical analysis for the 2003 IRI predictions ....................................................... 120 
Table 6.4 Statistical analysis for the 2006 fatigue cracking predictions................................... 121 
Table 6.5 Statistical analysis for the 2006 rutting predictions .................................................. 123 
Table 6.6 Statistical analysis for the 2006 rutting predictions (excluding N1 and N2)  ........... 124 
Table 6.7 Statistical analysis for the 2006 IRI predictions ....................................................... 125 
Table 6.8 Best-fit calibration coefficients for three models in the MEPDG ............................ 125
1 
 
CHAPTER 1 
INTRODUCTION 
BACKGROUND 
Overview of Pavement Design 
Pavement design is critical to successful road construction. It provides a workable plan to 
specify the pavement section before it is built. The effectiveness of pavement design usually 
determines whether an application can serve for a long time with good quality at an affordable 
cost for owners. A good quality pavement means durable support, skid resistance, and energy 
conservation for road users. However, engineers are inevitably faced with a variety of design 
challenges when considering traffic, the environment, or subgrade soils with local characteristics. 
For a highway agency, pavement design relies on material specifications, pavement design 
policies, and construction practice. Obviously, it is never easy to achieve a pavement design that 
fulfills the requirements above. In this case, pavement design procedures or guides are needed 
for providing a paradigm of generating appropriate designs. The common procedures account for 
these main design factors such as traffic, materials, the environment, and performance 
predictions. 
Main design factors 
Traffic 
Pavements are designed to sustain recurring loadings of vehicles. Vehicles impose weight 
at the interface between tires and pavement, as a result, the stress is distributed into the pavement 
structure, attenuating from the top to the bottom. The heavier the vehicle is, the greater the 
stresses. It is a rule of thumb that doubled axle weight will increase induced damage by 16 times 
(Equivalent Single Axle Load, 2007). With repetitive vehicle loadings, the pavement endures a 
tremendous amount of cycles of deforming and rebounding. If the allowable volume of traffic 
2 
 
loading is reached, the pavement will experience a loss of serviceability and become distressed. 
From the standpoint of pavement design, the repetitive vehicle loadings in a period of time 
should be related to how much damage they impose. Therefore, traffic is decomposed into a 
number of axle groups based on configurations and weights, and the induced damage of each 
axle group is estimated. In this way, the total damage of traffic during a period of time can be 
determined by an accumulation of damage of these axle groups. To design a new pavement, 
designers need to derive future traffic predictions based on the current trend. However, traffic 
may fluctuate due to unexpected situations, such as rapid economic activity or population growth, 
and this should be within the concern of pavement design. 
Materials 
Materials physically comprise a layered pavement structure, containing surface, base, and 
subgrade. The properties of the materials reflect the capacity of the pavement to withstand 
damage. For example, a subgrade with high strength can prevent compressive or shear stress 
failure under heavy loads. Open-graded asphalt courses allow rainwater to drain through the 
pavement surface, reducing the amount of splash and spray on highways. The adoption of good 
quality material helps pavements to last longer, but it may greatly increase project costs. 
Therefore, engineers sometimes choose economical materials with local availability.  
Environment 
It is common sense that civil engineering structures must function within the environment 
where it is built. The material properties of structures are significantly affected by environmental 
variations. This principle can be applied to pavements. The viscoelastic properties of asphalt are 
affected significantly by pavement temperature changes and loading frequency. Furthermore, the 
asphalt binder will be oxidized if exposed in air and the asphalt mixture will become stiffer and 
3 
 
brittle. Therefore, the material property changes due to environment should be considered in the 
pavement design process.   
Performance 
Pavements are affected by a combination of traffic loading and environmental effects. As 
a result, the performance degrades over time. Engineers need an efficient way to define the 
performance of pavement so that they know whether an in-service pavement is still in good 
condition. From the view of pavement design, a clear definition of performance can prepare 
engineers to set their design goal. In fact, the pavement performance is defined from the 
experience, but in different ways at project and network levels. At a project level, performance is 
defined by distress, loss of serviceability index, and skid resistance. On a network level, it is 
defined by overall condition of the network and even policy and economics (Lytton, 1987). 
Generally, serviceability, defined as the current ability of a pavement to serve the traffic, is taken 
as an overall indicator to represent the pavement performance.  
Approaches to Pavement Design 
Approaches to pavement design are required to provide a link between the inputs of a 
design scenario (i.e., traffic, environment, materials, and tolerable degradation of pavement) with 
an output design (i.e., material type and layer thickness).  The approaches currently available can 
be generally divided into two categories: empirical methods and mechanistic-empirical (M-E) 
methods.   
Empirical method 
Empirical methods correlate important variables based on experiments or experience. The 
important variables account for design factors in a measurable or quantitative way. Empirical 
methods were an effective method for early engineering practice; however, because it is 
4 
 
statistics-based or experience-based, the applicability of the model would be limited by the range 
of the experimental data or experience. For example, the 1993 AASHTO (American Association 
of State Highway and Transportation Officials) method is an empirical approach. The empirical 
relationship between important variables and a thickness-related ?structural number? is described 
by design equations aimed to determine pavement thickness. This method is built on the limited 
conditions and materials present in the 1950s, so it is not widely applicable to other conditions or 
current pavement materials. 
Advantages 
Through many trial-and-errors, the empirical approach can build upon what is already 
observed, and respond more appropriately to real world design scenarios. It does not need cogent 
theoretical derivation to prove the reasonableness of empirical relationships.  
Disadvantages 
The validity of the empirical approach is usually limited by the dataset on which the 
approach was based. The dataset inherently specifies the value range of the important variables 
or considers a limited set of contributing variables. Thus, the extreme condition, for instance, in 
which a variable falls out of the pre-defined value range or the neglected variable that may have 
a big impact, is not accounted for. Additionally, a fairly solid empirical model takes a long time 
to build and needs calibration before use. 
Mechanistic-empirical method 
In contrast with the empirical methodology, mechanistic-empirical methods integrate 
mechanistic analysis of pavement structures into a design procedure. Mechanistic analysis can 
determine the response of pavement layers (i.e., strain, stress, or deflections) and these responses 
5 
 
are highly related with the propagation of various pavement distresses. The relationship between 
them can be statistically defined by empirical regression equations called transfer functions. 
Advantages 
Mechanistic-empirical methods contain broader variable sets to characterize design 
inputs more realistically, so they has ability to account for new materials, advanced traffic 
technology, and changes of environmental conditions. Thus, designers are able to compose 
pavement designs for a wide variety of local areas and conditions. Additionally, the M-E method 
is able to predict various distresses by sophisticated algorithms and enables engineers to set a 
failure criterion for each of them.  
Disadvantages 
Along with advanced design concepts, a large number of inputs will require designers 
much time to collect the design input information. Since many agencies are still using old 
AASHTO methods, they will need to upgrade technical capabilities and to provide training for 
implementation of the mechanistic-empirical methods. It is also important that the M-E method 
requires local calibration to ensure its effectiveness. 
Development of Empirical Method 
The development of the AASHTO empirical method dates back to a first application of 
the Highway Research Board (HRB) soil classification in estimating the subbase and total 
pavement thickness. An method with a strength test was first used by the California Highway 
Department in 1929 (Porter, 1950). The thickness of pavement was related to the California 
Bearing Ratio (CBR), which defines the penetration resistance of a subgrade soil relative to a 
standard crushed rock. In the 1960s, AASHTO initiated to develop design equations and 
nomographs for both flexible and rigid pavements, primarily based on the regression of the data 
6 
 
from the AASHO (American Association of State Highway Officials) Road Test. These 
empirical methods had been adopted until the appearance of M-E method (Huang, 1993). 
AASHO road test and AASHTO methods 
In the late 1950s, AASHO conceived a grand plan of building real-world testing facilities 
to examine the performance of pavement structures of known thickness under moving loads of 
known magnitude and frequency. A long list of pooled-funding sponsors spent $27 million (1960 
dollars) to complete the construction of 6 loops of two-lane pavements, with half asphalt and half 
concrete. The road test provided the foundation for analytical evaluation of pavement response, 
load equivalencies, climate effects, and much more.  The precious performance information 
greatly helped to build the regression equations for AASHTO design methods. The initial version 
was completed in 1961 relating loss in serviceability, traffic, and pavement thickness and it was 
revised in 1972 by introducing the consideration of foundation, material, and environmental 
conditions. Then, the next version released in 1986 improved the characterization of the 
subgrade and unbound materials, considered drainage effects, and incorporated a reliability 
factor. The latest update was released in 1993 and had some additions for pavement 
rehabilitation designs. 
The AASHTO methods have gained popularity through relatively straightforward 
procedures; however, the design equations reflect a single traffic volume, single environmental 
condition, and limited design variables described in the late 1950s. Studies have shown that 
despite adjustments made over the years to the design equation in attempts to expand its 
suitability to different climate regions and materials, the design of flexible pavements still lacks 
accuracy in performance predictions and compatibility to include different materials and their 
complex behavior (Schwartz, 2007). The performance indicator PSI (Present Serviceability 
7 
 
Index) cannot address today?s need in the situation that road designers and managers require 
more accurate evaluation of pavement performance regarding distresses and overall smoothness.  
Development of Mechanistic-Empirical Methods 
Brief introduction 
It was noted that Kerkhoven and Dormon in 1953 first suggested using vertical 
compressive strain on the surface of subgrade as a critical indicator of permanent deformation 
(Huang, 1993). In 1960, Saal and Pell recommended the use of horizontal tensile strain at the 
bottom of asphalt layer as a critical predictor of fatigue cracking (Huang, 1993). Afterwards, the 
use of these concepts was first presented for pavement design in the US by Dormon and Metcalf 
in 1965. These two criteria have been adopted by Shell Petroleum International (Claussen et al, 
1977) and the Asphalt Institute (Shook et al., 1982) in their proposed mechanistic-empirical 
methods. 
The concepts and principles of M-E design were extensively discussed in details in the 
National Cooperative Highway Research Program (NCHRP) 1-26 project ? ?Calibrated 
Mechanistic Structural Analysis Procedures for Pavements.? In this project, the available M-E 
design at that time was assessed, evaluated, and applied on a nationwide level, and researchers 
recognized that the M-E method represents one step forward from empirical methods. In the 
1990s, several M-E design approaches by computer software were developed, including 
MnPAVE adopted by Minnesota Department of Transportation, the 2002 AASHTO 
Mechanistic-Empirical Pavement Design Guide (MEPDG), EverSeries by Washington 
Department of Transportation, and Michigan Flexible Pavement Design System (MFPDS) by 
Michigan State. 
 
8 
 
M-E design framework 
The basic outline of M-E design includes calculating the pavement response under the 
impact of traffic by mechanistic models and applying transfer functions to predict number of 
loads to failure. Miner?s hypothesis (Miner, 1945) was integrated afterwards as an incremental 
approach to calculate the cumulative damage in the pavement life. The approach allows the 
summation of fatigue or rutting damage by axle loads of varying magnitudes under varying 
environmental conditions.  
As Figure 1.1 shows, the pavement structure specifies the material and thickness of layers, 
which constitutes the basic elements for mechanistic analysis. Environmental conditions impact 
the properties of materials by temperature, precipitation, wind speed, and so on. Traffic applies 
vehicle loading to induce pavement response depending on axle types and magnitudes. In order 
to predict long-term performance, the transfer functions are used to link dynamic pavement 
response with allowable repetitions of loads. It is also worthy to note that the magnitude of 
loading significantly affects mechanistic properties of stress-sensitive materials, such as resilient 
modulus of unbound materials. Further, for one of the strain levels, the damage is quantified as 
the ratio of actual repetitions of loads to allowable repetitions of loads. Thus, the total damage in 
pavement life is a summation of damage ratios for all levels of strain. If the total damage is 
greater than 1, the pavement is ?under-designed?, so it is necessary to modify the original 
pavement design, such as increasing the thickness of layers or adopting more durable materials. 
The modification of pavement design should be re-evaluated, and this iterative procedure is 
processed until the damage ratio is less than 1. Otherwise, if the total damage is much lesser than 
1, which indicates pavement is ?over-designed?, it is necessary to propose a less conservative 
design by reducing the thickness of layers or choosing poorer quality materials. 
9 
 
 
Figure 1.1 Scheme of the M-E design 
MEPDG 
Background 
At the time when the 1986 AASHTO method was adopted, the benefits of 
mechanistically-based pavement design were clearly recognized and it had become apparent for 
transportation agencies that there was a great need for an advanced design method that accounts 
for modern traffic, current materials, and diverse climates. Subsequently, the Washington State 
Department of Transportation (DOT), North Carolina DOT, and Minnesota DOT developed their 
10 
 
mechanistic-empirical procedures. The National Cooperative Highway Research Program 
(NCHRP) 1-26 project provided the basic framework for most of efforts attempted by state 
DOTs (Schwartz, 2007). To embrace the widely-accepted transition from empirical to 
mechanistic-empirical design methods, the AASHTO Joint Task Force on Pavements, in 
cooperation with NCHRP and FHWA (Federal Highway Administration), invited a group of top 
pavement engineers in the U.S. to a workshop on ?pavement design? in March, 1996 in Irvine, 
California. They were identifying means to develop the AASHTO mechanistic-empirical design 
procedure by 2002. Based on the conclusion and recommendation of the workshop, the NCHRP 
1-37A project, named ?Development of 2002 Guide for New and Rehabilitated Pavement: Phase 
II?, was awarded to the ERES Consultants division of Applied Research Associates Inc. in Feb, 
1998 for software generation. The version 0.7 of software was first released in 2004, named 
?Mechanistic-Empirical Pavement Design Guide (MEPDG)?.  
Adoption and implementation  
In order to facilitate MEPDG implementation among highway agencies, a list of NCHRP 
projects was carried out after the first release to deal with issues regarding traffic data, 
independent review, user manual, local calibration, technical assistance, models for predicting 
HMA overlay reflective cracking, models for predicting top-down cracking, sensitivity analysis, 
and so on (Qiang Li et al, 2011). Many of these have been completed and adopted as supportive 
strategies and implementation guides for application of the new software. The version 1.0 of 
MEPDG was issued in 2007 under the NCHRP 1-40D project, and then it was taken as the 
Interim Pavement Design Guide by AASHTO in 2008. By 2009, 80 percent of states in the U.S. 
had implementation plans for the MEPDG (?MEPDG Overview & National Perspective?, 
January 2009). The latest version of MEPDG is called ?AASHTOWare Pavement ME Design?. 
11 
 
It is the next generation of AASHTO pavement design software with many improvements and is 
a production-ready tool to support day-to-day operations for engineers.  
To take an example of MEPDG implementation, the Indiana DOT began their program 
on January 1st, 2009. They initiated a steering committee, in cooperation with design engineers, 
FHWA, pavement associations, and contractor associations to discuss details and issues in 
practice. The Indiana DOT also provided support for engineers and consultants to facilitate the 
use of the MEPDG software. Most of designers and consultants gained familiarity with the new 
design procedure within six months. From January to December 2009, the Indiana DOT 
designed more than 100 pavement sections by using the MEPDG. The thickness of most 
concrete pavements on the Interstate and U.S highway systems was reduced by 2 inches 
compared with using the AASHTO 1993 method; a less prominent reduction applied to 
pavements on state routes. For 11 HMA projects of US highway and state route, the average 
thickness reduction was 2 inches and the savings were estimated to be $3,637,000 (TR News, 
Nov-Dec 2010).     
Need of MEPDG Local Calibration 
For the M-E design methods, the transfer functions link the bridge the mechanistic and 
empirical analyses, but the NCHRP 1-26 report concluded this to be the weak link in M-E design. 
Studies conducted in Washington State (Li et al, 2009), North Carolina (Muthadi and Kim, 2008), 
and Minnesota (Hoegh et al, 2010) suggested that transfer functions in MEPDG produce 
predictions that contained significant bias or variance. Indeed, the accuracy of the transfer 
functions needs to be improved so the field verification and local calibration are strongly 
recommended before it is used locally. To satisfy the need of implementing MEPDG for 
southeastern states, especially Alabama, local calibration is needed.  
12 
 
OBJECTIVE 
The main objective of this research was to evaluate the validity of the nationally-
calibrated transfer functions built in the MEPDG version 1.1 which were developed based on the 
Long Term Pavement Performance (LTPP) program database. The second objective was to 
calibrate these transfer functions to the local conditions at the NCAT Test Track.  
SCOPE 
NCAT Test Track 
The NCAT Test Track (Figure 1.2), a 1.7-mile oval accelerated pavement testing track 
located in Opelika, Alabama, was used in this research. The outside lane of track was divided 
into forty-six test sections, each 200 feet long, and sponsored on three-year cycles. During each 
cycle, NCAT operates a truck fleet to apply 10 million Equivalent Standard Axle Loads (ESALs) 
and collects field performance data on a continuous basis. In this study, eight sections from the 
2003 cycle were simulated in the MEPDG to evaluate nationally-calibrated models and 
subsequently to adjust the transfer functions for the Test Track. Eleven sections from the 2006 
cycle were utilized to validate results of the local calibration.  
 
Figure 1.2 NCAT Test Track 
13 
 
Evaluation and Calibration Process 
The process of evaluation was to first characterize traffic, material properties and 
environmental conditions at the NCAT Test Track and generate inputs for the MEPDG. The next 
step was to run the software and to compare the predicted performance from the MEPDG with 
observed performance on the Test Track. If the differences were small, it means the MEPDG 
performed very well for the Test Track conditions. Otherwise, poor predictions were obtained 
which means the MEPDG needs to be locally calibrated. The calibration was an iterative process 
that required varying the calibration coefficients in the transfer functions to calibrate the 
MEPDG predictions to the observed values. 
ORGANIZATION OF THESIS 
Chapter 2 will further provide a background literature review about traditional design 
methods, the MEPDG, and local calibration efforts conducted by some states. Chapter 3 covers 
an explanation of the datasets involved in the study such as the NCAT facilities, traffic operation, 
climate file generation, and data collection methods. Next, the methodology of validation and 
local calibration is described in Chapter 4. A method of programming for running the MEPDG 
automatically is introduced in Chapter 5. This method greatly reduces labor intensity involved in 
running the software. Chapter 6 presents the results of calibration and validation. Finally, the 
conclusions and recommendations are presented in Chapter 7.      
 
 
 
 
 
  
14 
 
CHAPTER 2  
LITERATURE REVIEW 
INTRODUCTION 
This chapter provides a general overview of pavement concepts, including pavement 
types, distresses, and design factors. Traditional design methods and the MEPDG are 
investigated to describe the evolution of pavement design and understand the purpose and 
significance of this study. Furthermore, MEPDG local calibration efforts by Washington State, 
North Carolina, Ohio and Minnesota were researched to guide the conduction of this study. 
PAVEMENT DESIGN OVERVIEW 
Pavement Types 
Pavement is usually classified into two categories based on surface materials, flexible and 
rigid. The flexible pavement is selected as the first choice in the U.S., which is determined by 
state highway agencies based on policy, economics, or both. In regard to service life, flexible 
pavement requires maintenance or rehabilitation every 10 to 15 years, while rigid pavement can 
last for 20 to 40 years. However, owing to high cost of rehabilitation, rigid pavement is narrowly 
adopted in urban and high traffic areas.  
Flexible pavement 
Flexible pavement is surfaced with the mixture of mineral aggregate and asphalt.  
Mineral aggregate, as the major component, accounts for 92 to 96 percent by volume and it 
provides pavement surfaces with qualities related to hardness, texture, and resistance to 
stripping. As a binding medium, asphalt becomes a viscous liquid at higher temperatures but 
turns to a solid at ambient air temperatures. Due to the effect of asphalt, the pavement will bend 
or deflect when traffic loads are applied without breaking; thus it is ?flexible?. A flexible 
15 
 
pavement structure is normally composed of flexible surface courses and several layers of base 
materials which can accommodate and distribute this ?flexing? over a larger area, downward to 
the subgrade.  
Rigid pavement 
Rigid pavement, also called portland cement concrete (PCC), is composed of aggregate, 
portland cement, and water. The aggregate accounts for 70 to 80 percent of portland cement 
concrete by volume and it can be either coarse aggregate (i.e., crushed stone and gravel) or fine 
aggregate (i.e., sand). The portland cement is the chief binding agent in rigid pavements. Once 
mixed with water, they harden into a solid mass. Because of high stiffness, rigid pavement tends 
to distribute the load over a relatively wide area of subgrade. The concrete slab itself provides 
most of a rigid pavement?s structural capacity.  
Pavement Distresses 
As discussed earlier, pavements are categorized into flexible or rigid, but only flexible 
pavement is discussed since it is the main interest of this study. The major distresses for flexible 
pavement include fatigue cracking, rutting, longitudinal/transverse cracking, raveling, stripping, 
and the loss of skid resistance.  
Fatigue cracking 
Fatigue cracking is prevalent among aged flexible pavements. It gets another name, 
?alligator cracking?, from its appearance in severely-damaged condition (Figure 2.1). The 
cracking initializes at the bottom of asphalt mixture layers and propagates upward to the surface, 
so it is also called ?bottom-up cracking?. The cracks allow moisture infiltration, roughness, and 
may further deteriorate to a pothole (Fatigue cracking, 2009). Thin HMA layers, stiff mixtures, 
and soft base are more likely to intensify the potential of bottom-up cracking (Schwartz and 
16 
 
Carvalho, 2007). The Highway Performance Monitoring System (HPMS) uses the percent of 
cracking area divided by the total area to measure the severity level of fatigue cracking. For 
example, the cracking of 500 ft length by 2 ft width in each wheelpath can be reported as 5% if 
the sample section is a single lane, 12 ft width by 1 mile length.  In addition to these causes, cold 
weather often quickens the propagation of cracking. In the laboratory, the flexural beam test is 
typically performed to investigate the fatigue of a HMA material, which can be related to its 
resistance to fatigue cracking. 
 
Figure 2.1 Fatigue cracking (Fatigue cracking, 2009) 
Rutting 
Rutting, also known as permanent deformation, is observed as surface depression in the 
wheel path, sometimes accompanied by pavement uplift along the sides of the rut (Figure 2.2). 
Ruts tend to steer a vehicle towards the rut path as its tires are riding on the rut, which can cause 
vehicle hydroplaning if ruts are filled with water (Rutting, 2008). It is a load-induced distress 
caused by consolidation or lateral movement of the materials due to cumulative traffic loading at 
moderate to high temperatures (Schwartz and Carvalho, 2007). The rut depth is usually measured 
by Straight Edge, Dipstick Profile, or Automated Laser Profile. In the laboratory, the tests for 
predicting the HMA rutting characteristics include the static creep test, the repeated load test, 
dynamic modulus test, empirical tests (i.e., the Hveem and Marshall mix design tests), and 
17 
 
wheel-tracking simulative tests. Each of them has limitations related to equipment complexity, 
expense, running time, and material parameters. 
 
Figure 2.2 Rutting (Rutting, 2008) 
Longitudinal/Transverse cracking 
Longitudinal cracking (Figure 2.3) often develops parallel to the pavement centerline 
when the paving lane joint is poorly constructed. Transverse cracking (Figure 2.4) usually 
propagates across the centerline. These types of cracks are not usually load-associated. They may 
be caused by the shrinkage of asphalt surface due to low temperatures, asphalt hardening, or 
result from reflective cracking caused by cracks beneath the asphalt surface (Huang, 1993).  
 
Figure 2.3 Longitudinal cracking (Longitudinal cracking, 2008) 
18 
 
 
Figure 2.4 Transverse cracking (Transverse cracking, 2006) 
Ride Quality 
Ride quality represents the overall performance of a pavement. The ride quality is 
evaluated by pavement roughness, which is defined as an expression of irregularities in the 
pavement surface. The roughness adversely affects the ride quality, delay costs, fuel 
consumption, and maintenance costs (Roughness 2007). The typical metrics of pavement 
roughness are Present Serviceability Rating (PSR) developed by AASHO Road Test, 
International Roughness Index (IRI) developed by World Bank, or other indexes, with IRI being 
the most prevalent (Roughness 2007). The measurement of IRI is made using a profilometer, 
which can measure small surface variations in vertical displacement as a function of longitudinal 
position. 
State highway agencies usually use the International Roughness Index (IRI) to rate 
pavement roughness, and schedule maintenance and rehabilitation plans. The lower the IRI is, 
the smoother the pavement surface.  
 
19 
 
Design Factors 
Design factors were briefly discussed in Chapter 1; however, they are an essential part in 
the design method and crucial to the determination of a final design (i.e., thickness design). Thus, 
design factors are further specified with details in regard to traffic, the environment, material, 
failure criteria, and reliability herein. 
Traffic  
Traffic is the major cause of pavement distresses. It poses external forces to induce 
fatigue cracking progress and rutting accumulation during repetitive loadings. The effect of 
traffic is considered in terms of axle loads and the number of repetitions, contact pressure, 
vehicle speed, and traffic volume distribution.  
Axle loads and number of repetitions 
The response of pavement to applied traffic is determined by the magnitude of loads, axle 
types, and the spacing between axles. Load magnitude is a primary factor in the determination of 
damage; the relationship between axle weight and inflicted pavement damage is not linear but 
exponential. For instance, a 44.4 kN (10,000 lbs) single axle needs to be applied to a pavement 
structure more than 12 times to inflict the same damage caused by one repetition of an 80 kN 
(18,000 lbs) single axle (Loads, WSDOT Pavement Guide). As for axle types, two axles, if not 
grouped closely as tandem, tridem, or quad axle, should be considered independently because 
they are too far apart to superpose their effects (i.e., stresses or strains); otherwise, the influenced 
areas of these axles begin to overlap, as a result, the concern is no longer the single load but 
rather the combined effect of all the interacting loads. Obviously, it is not appropriate to treat 
those axles as one group as a single axle, nor is appropriate to treat each axle independently 
(Huang, 1993). In the AASHTO method, the load of an axle (i.e., single, tandem, tridem, or 
20 
 
quad) with a specific magnitude is normalized to a number of ?standard? loads based on its 
damaging effect on the pavement. The ?standard? loads are named as Equivalent Single Axle 
Loads (ESALs). In the M-E method, ?load spectra? replaced the ESALs to characterize traffic 
directly by numbers of axles, axle configuration, and axle weight. It provides precise load data 
for each passing axle in the mechanistic analysis.  
The number of repetitions describes how many axle loads are imposed during a long 
term. In the AASHTO method, the number of ESALs describes the repetitive traffic loadings in a 
uniform manner, serving as a fundamental input in pavement design. In the M-E method, the 
ratio of the number of actual loads over the number of allowable loads is used to quantify a long 
term effect of traffic loadings under the Miner?s hypothesis (Miner, 1945). It is assumed in 
Miner?s hypothesis that each axle load inflicts a certain amount of incremental damage, and the 
total damage can be estimated by a known number of load repetitions. If the pavement is 
imposed with a variety of load magnitudes, the damage ratio for each load magnitude can be 
specified separately based on its number of repetitions and then they can be combined together 
for predicting the total pavement damage. 
Contact pressure 
The contact pressure is the force spread over the interface between the tire and pavement, 
which equals to the force distributed to the pavement. In fact, the contact pressure is the result of 
composition of tire pressure and pressure due to tire wall. The contact pressure is greater than the 
tire pressure for low-pressure tires (Figure 2.5) because the wall of tires is in compression and 
the sum of vertical forces due to wall and tire pressure must be equal to the force due to contact 
pressure; otherwise, the contact pressure is smaller than the tire pressure for high-pressure tires 
21 
 
(Figure 2.5) (Huang, 1993). In mechanistic-empirical methods, the contact pressure is usually 
assumed to be the tire inflation pressure with a circular tire/pavement interface. 
 
Figure 2.5 Low pressure tire (left) and high pressure tire (right) (Huang, 1993) 
Vehicle speed 
The vehicle speed determines the frequency of moving loads, which may affect the 
properties of pavement materials. For example, the stiffness of asphalt mixture is sensitive to the 
frequency of loading. Slower speeds typically result in lower modulus values in asphalt concrete 
which, in turn, increase strain levels. 
Traffic volume distribution 
On any given road, one direction typically carries more traffic than the other. 
Furthermore, within one direction, each lane carries a different portion. The outer lane often 
carries most of the trucks and therefore is usually subjected to heavier loading, especially for 
Interstate highways. This requires the outer lane to be more resilient to load-induced damage. 
To prepare traffic data for designing a new pavement, agencies may need to monitor the 
characteristics of former traffic loads (i.e., past traffic volume, weight data, speed) along or near 
the roadway segment to be designed, thus, they can best forecast the traffic inputs for design 
purpose. Four main sources of traffic data are typically used for the traffic characterization in the 
MEPDG: weigh-in-motion (WIM) data, automatic vehicle classification (AVC), vehicle counts, 
22 
 
and traffic forecasting and trip generation models (NCHRP 1-37A Report, Part 2, Chapter 4, 
2004).  
Environmental effects 
The environment provides an independent medium that affects material properties. The 
environment can vary much across different regions in a country with seasonal or monthly 
changes. The key environmental concerns in the pavement design are typically temperature and 
precipitation, which are usually recorded by local weather stations. 
Temperature 
Temperature greatly affects the behavior of asphalt materials. When the pavement gets 
cold, the HMA becomes stiffer and tends to have fatigue damage because stiff materials often 
have a shorter fatigue life. The extreme low temperature even induces low-temperature cracking, 
which could happen regardless of traffic loading (Ksaibati and Erickson, 1998). In practice, this 
problem can be avoided by adopting modified asphalt materials. Otherwise, when the pavement 
gets hot, the HMA becomes softer and tends to deform under traffic loading so rutting is more 
likely to occur at wheel paths. 
Precipitation 
Precipitation impacts the quantity of surface water, which can change the location of the 
groundwater table once it infiltrates into the subgrade. The groundwater table should be kept at 
least 3 feet below the pavement surface (Huang, 1993); otherwise, the seasonal frost will have a 
detrimental effect on pavements. Also, excessive moisture will cause the loss of bond between 
aggregates and binder, resulting in distresses, such as bottom-up cracking, stripping, and 
raveling. 
 
23 
 
Materials 
As mentioned previously, the properties of materials determine the inherent capability to 
resist damage caused by traffic and environmental effects; thus, they need to be accurately 
characterized. The properties generally include mechanistic properties, such as stiffness and 
Poisson?s ratio, and some other properties, for example, the related properties for low-
temperature cracking: the tensile strength, creep compliance, and coefficient of thermal 
expansion. 
The characterization of material properties differs in asphalt mixture and unbound 
materials. The discussion mainly covers the mechanistic properties for these two components 
since they are key roles in the pavement behavior.  
Asphalt mixture 
Asphalt mixtures are often modeled as elastic materials in elastic layered systems. The 
elastic modulus and the Poisson ratio were used to characterize asphalt mixture (Yoder & 
Witczak, 1975). Sometimes the resilient modulus was used instead of elastic modulus, which 
differed in the testing method. However, neither moduli accounted for the visco-elasticity of 
asphalt mixture. As a result, the dynamic modulus (|E*|) was proposed to characterize asphalt 
mixture so as to compute stress and strain in flexible pavements.  
Elastic modulus 
The elastic modulus, or Young?s modulus, is a measure of the elasticity of a material, 
which is equal to the constant ratio of stress over strain within its elastic range of loading (i.e., a 
load range where the relationship between the stress and strain is linear) (Elastic Modulus, 2007). 
If the elastic modulus of asphalt mixture is higher, it will bend or deform less.  
 
24 
 
Resilient modulus 
The resilient modulus of a material is an estimate of its elastic modulus (Resilient 
Modulus, 2007). The resilient modulus is determined by the tri-axial test, in which a repeated 
axial cycle stress of fixed patterns (i.e., magnitude, load duration, and cycle duration) is applied 
to a cylindrical test specimen. The cyclic load application can simulate actual traffic loading 
more accurately than constant load applications. The resilient modulus can be calculated as the 
ratio of the repeated deviator axial stress to the recoverable axial strain.  
Dynamic modulus 
The dynamic modulus uses a complex number with two components, a viscous part and 
an elastic part, to relate stress to strain for linear visco-elastic materials subjected to 
continuously-applied sinusoidal loading in the frequency domain (Witczak et al., 2002). The 
complex number is defined as the ratio of the amplitude of the sinusoidal stress (?) at any given 
time (t) and the angular load frequency (?), and the amplitude of the sinusoidal strain (?) at the 
same time and frequency (Witczak et al., 2002).  
iEEt te eE ti ti *s i n*c o s*)s i n (s i n*
0
0)(
0
0 ????? ??? ??? ?? ? ?????? ?                  (2-1) 
where,  
?0 = peak (maximum) stress  
?0 = peak (maximum) strain  
? = phase angle  
? = angular velocity  
 t = time  
 i = imaginary component of the complex modulus 
25 
 
This type of characterization for asphalt mixtures is under Level Input 1 in the MEPDG 
to account for the interaction between material properties and environmental conditions. It can be 
determined by the Asphalt Mixture Performance Test (AMPT).  
Unbound materials 
The unbound materials are not as frequency-temperature sensitive as asphalt mixtures. 
The characterization of unbound materials mostly focuses on resilient modulus, Poisson ratio, 
and the coefficient of lateral pressure. The resilient modulus can be measured from 
backcalculation based on FWD data or through the use of correlations with other material 
strength properties, such as California Bearing Ratio (CBR).   
Moreover, only a few design methods, such as the MEPDG, require gradation, Atterberg 
limits, and hydraulic conductivity to predict temperature and moisture conditions within the 
pavement system.  
Failure criteria 
Failure criteria define the indicators related with the pavement failure (i.e., the occurrence 
of major distresses) and their thresholds. In the AASHTO method, the present serviceability 
index (PSI), which indicates the general condition of a pavement, is used as the criterion to judge 
whether the pavement fails or not. The failure criteria for M-E method are described in detail 
below. 
Fatigue cracking 
The failure criterion for fatigue cracking is the allowable number of load repetitions 
before ?significant? cracking (or failure) occurs. In other words, if the number of applied load 
repetitions exceeds the allowable number of load repetitions, significant fatigue cracking is 
expected to occur. To identify and measure cracking, some visually evaluate the level of 
26 
 
cracking severity into low (no or a few cracks), moderate (interconnected cracks), and high 
(severely interconnected cracks) categories (Oregon DOT Distress Survey Manual, 2010). The 
definition of significant cracking (or failure) is often established on agencies? identification 
perspective, engineering experience, management strategy, and funding availability. The Asphalt 
Institute recommended an area greater than 45 percent of the wheel path or an equivalent 20 
percent of the total lane as failure in their transfer functions (Priest and Timm, 2006). The 
MEPDG considered 50 percent fatigue cracking of the total lane at a damage percentage of 100 
percent as failure (NCHRP 1-37A Report, Appendix II-1, 2004). 
Rutting 
The failure criterion for rutting is the allowable number of load repetitions before critical 
rut depth is formed. If the actual number of repetitions exceeds the allowable number of 
repetitions, rutting will cause driving discomfort and even potential safety hazards. The 
evaluation for rutting condition is based on the average depth of rutting, and varies by type of 
road. Usually, rut depth less than 0.2? is good, and rut depth larger than 0.5? is poor. The rut 
depth between 0.2? and 0.5? is fair (Pavement Preservation Manual, 2009). The definition of 
critical rut depth is often established on agencies? engineering experience, management strategy, 
and funding availability. 
IRI 
IRI, as an overall evaluation of pavement performance, can also be taken as a failure 
criterion. The failure criterion for IRI is usually an allowable level, which represents a 
combination of different distresses. The FHWA suggests that a road surface with IRI rating 
below 95 is in good condition, providing a decent ride quality; a road surface with IRI from 95 to 
119 is in fair condition; a road surface with an IRI from 120 to 170 is in mediocre condition; and 
27 
 
a road with IRI above 170 is in poor condition, providing an unacceptable ride quality 
(AASHTO and TRIP 2009, P9). The definition of allowable IRI is often established on agencies? 
engineering experience, management strategy, and funding availability. 
Other criteria 
Besides the two failure criteria above, there are also some other criteria for specific 
distresses, such as low-temperature cracking. The potential of low-temperature cracking for a 
given pavement can be evaluated if the mix stiffness and fracture strength characteristics are 
known and temperature data on-site are available to compute the thermal stress. The pavement 
will crack when the computed thermal stress is greater than the fracture strength. 
Reliability 
The uncertainty and variability associated with design inputs have always been of great 
concern to pavement engineers. Uncertainty exists in any prediction of future situations (i.e., 
traffic and environmental changes). Variability is common because pavement materials are 
constructed with variance at different spatial locations. These may cause less accurate input 
characterization, which is the basis of a pavement design. To guarantee the reliability of design, 
there are two methods to account for uncertainty and variability associated with inputs: 
deterministic and probabilistic (Huang, 1993). In the deterministic method, each design input has 
a fixed value based on engineering judgment and experience. The designer assigns a safety factor 
to each design input to allow more challenging conditions so that the final design can withstand 
the risks beyond the estimated situation. A more realistic approach is the probabilistic method in 
which each of the design factors is assigned a mean and a variance. The variability of a design 
input is determined by characterizing its distribution, and then the reliability of an acceptable 
design can be estimated (Huang, 1993). The reliability of a pavement design can be defined as 
28 
 
the probability that the performance indicator is acceptable. For example, the reliability in the 
MEPDG is the probability that the performance of the pavement predicted for that design will be 
satisfactory over the time period under consideration (Li et al., 2011). It is defined in the design 
as equation (2-2). 
R = P [performance indicator over design period < critical indicator level]           (2-2) 
where, 
P = probability, percent 
The basic elements of pavement design as discussed above are accounted for in the 
pavement design methods which serve the project design. The following will introduce a few 
traditional and modern design methods that use elements of pavement design.   
TRADITIONAL DESIGN METHODS 
California Method 
The California Method was originally developed in the 1940s based on experience 
obtained from test roads and pavement theory. It was adopted by the western states and has been 
modified several times to suit the changes of regional traffic characteristics. The objective of the 
original design method was to avoid plastic deformation and the distortion of the pavement 
surface, but a later modification included the minimization of early fatigue cracking due to traffic 
loads (Garber and Hoel, 2009).   
The factors considered in this method are traffic loads and the strength of construction 
materials. The traffic loads are initially calculated in terms of ESALs and then converted to a 
traffic index (TI) as equation (2-3) (Garber and Hoel, 2009). 
 0 .1 1 9
6 )10E SA L(*9 .0TI ?
                                                       (2-3) 
29 
 
The strength of the subgrade material is determined by the Hveem Stabilometer test in 
terms of the resistance value R. The strength characteristics of each layer material are described 
by a gravel equivalent factor   , which is determined based on the type of material. 
The thicknesses of respective layers are determined by dividing gravel equivalents (GE) 
by their factors   . The GE can be derived by Equation (2-4) depending on the R value. The R 
value is the resistance value of the material of the supporting layer (Garber and Hoel, 2009). For 
subgrade, the R value is normally determined at an exudation pressure of 300 psi. 
R)( 1 0 0*( T I )*0 .0 0 3 2GE ??                                                  (2-4) 
The California Method is simple to use but its applicability is highly subject to the local 
traffic characteristics. Although the method accounts for the strength of supporting materials, it 
does not consider the seasonal effect of the environment. Different types of distresses are 
considered together to evaluate whether a pavement fails. It is not efficient to investigate a single 
type of distress and consider preventive measures.    
Asphalt Institute Method 
The Asphalt Institute Method establishes its own relationships between subgrade 
strength, traffic, and pavement structure. The goal of this method is to determine the minimum 
thickness of the asphalt layer with which the pavement can satisfy the criteria in regard to two 
types of strains: the horizontal tensile strain at the bottom of the asphalt layer, and the vertical 
compressive strain at the surface of subgrade (Garber and Hoel, 2009). The AI method is a good 
example of an early M-E approach to pavement design.  
The basic design factors in this method are traffic characteristics, subgrade engineering 
properties, and base/subbase engineering properties. The annual traffic containing a mix of 
vehicle axle loads is converted to ESALs. In addition, the subgrade engineering properties 
30 
 
should be determined in the laboratory or relative to with other test values (i.e., CBR and R 
value). The base/subbase material should satisfy certain quality requirements in terms of CBR, 
liquid limit, Plasticity Index (PI), particle size distribution, and minimum sand equivalent. 
The minimum thickness for various types of materials is determined by using a computer 
program DAMA or by using the appropriate charts, such as design chart for full-depth asphalt 
concrete shown in Figure 2.6. 
 
Figure 2.6 Design chart for full depth asphalt concrete (Asphalt Institute, 1983) 
The Asphalt Institute method is straightforward and provides a guide for thickness 
determination for various paving materials. It also accounts for the effect of temperature on the 
materials when selecting an appropriate design chart based on mean annual air temperature 
(MAAT) of the local area. More importantly, the Asphalt Institute method is an M-E design 
method which establishes the design criteria based on the mechanics of materials coupled with 
the observed performance. The original fatigue transfer function was based on laboratory 
equation developed by Monismith and Epps (1969). Then, the transfer function was calibrated 
using field data from AASHO Road Test, considering the failure to be 45 percent cracking of 
31 
 
wheelpaths and then further modified to include a correction factor to account for the 
volumetrics of the mixture as suggested by Pell and Cooper (1975). Furthermore, only the 
determination of the soil modulus requires a reliability safety factor to account for the variability 
from design inputs. The design procedures still lack the consideration of variations associated 
with material quality, homogeneity, and construction technique, which might cause a non-
uniform pavement response at different spatial locations. Also, the options of pavement materials 
are limited to several given material alternatives. 
AASHTO Method 
The AASHTO method includes a series of design methods with different versions based 
on the original AASHO Road Test. It has been widely used by state agencies. According to an 
FHWA survey (Asphalt Design Procedure 2007), 63 percent of the state DOTs were using the 
1993 AASHTO method, while 12 percent of them were using the 1972 AASHTO method. In 
addition, 8 percent of the state DOTs were using a combination of AASHTO methods and state 
procedures. The AASHTO method aims to achieve a pavement design that allows a tolerable 
serviceability decrease over the design life. The first version of this design equation directly 
incorporated the empirical observations under the AASHO Road Test local conditions, and was 
modified and extended its applicability to other regions (AASHTO 1972, 1986, 1993).  
The 1993 AASHTO method empirically evaluates the effects of design factors on 
pavement performance, with respect to design time constraints, traffic, reliability, serviceability, 
environmental effect, and effective subgrade modulus. The method adopts the design equation, 
as Equation (2-5) for flexible pavements, or a nomograph (Figure 2.7) to derive a thickness 
design based on the evaluation of those design factors. In either the design equation or the 
32 
 
nomograph, the Structural Number (SN) is usually targeted as the result which can generate the 
thicknesses of layers (AASHTO guide, 1993).   
8 . 0 7l o g M*2 . 3 2
1)( S N
10940 . 4
]1 . 54 . 2? P S I[l o g
0 . 2 01)( S Nl o g*9 . 3 6S*Zl o g W R
5 . 1 9
10
100R18 ??
??
??????
         
(2-5) 
where,          
   = predicted number of 80kN (18,000 lbs) ESALs 
   = standard normal deviate at chosen percent reliability (R) 
   = combined standard error of the traffic prediction and performance prediction 
SN = structural number 
 PSI = difference between the initial design serviceability index,   , and the design 
terminal serviceability index,    
   = the resilient modulus of material used for the lower layer, psi 
 
Figure 2.7 Nomograph (AASHTO guide, 1993) 
Two variables are adopted for design time constraints: analysis period and performance 
period. The analysis period is the length of time over which the whole pavement maintenance 
33 
 
strategy is designed to apply. It includes both the design life of use after initial construction and 
the prolonged life due to planned rehabilitation. The performance period is the length of time 
during which a pavement structure deteriorates from the initial serviceability to terminal 
serviceability before rehabilitation, or the performance time between rehabilitation operations. 
The analysis period is used for evaluating alternative long-term strategies based on the life-cycle 
cost analysis (Huang, 1993).  
The ESAL concept is used to quantify traffic loadings. The ESALs that a pavement will 
encounter over the design life is determined by counting truck traffic, predicting traffic growth, 
and converting this to ESALs (AASHTO Guide, 1993). 
The reliability is determined by the functional classification of the pavement. For 
example, for an Interstate highway in an urban area, the recommended reliability should lie 
between 85 to 99.9 percent. The reliability is related to the ?normal deviate?, which is a statistic 
adopted in the design equation (AASHTO Guide, 1993). 
The change of serviceability represents the capacity of pavement to withstand damage 
experienced between initial construction and terminal failure. The serviceability is evaluated by 
PSI, which is derived based on the original PSR from the AASHO Road Test. The initial PSI is a 
function of pavement classification and construction quality; and the terminal PSI is a design 
index that can be tolerated until rehabilitation, resurfacing, or reconstruction (AASHTO Guide, 
1993).  
The effect of the environment is accounted for by applying an excessive reduction of 
serviceability in the design equation. Since the AASHTO methods are derived based on the two-
year AASHO Road Test, the long-term effect of temperature and moisture on the pavement 
structure cannot be precisely evaluated (AASHTO Guide, 1993).  
34 
 
The subgrade is characterized by one elastic modulus. This modulus would result in the 
same damage as predicted by using the multiple moduli which account for its seasonal (or 
monthly) fluctuation due to temperature, moisture, or load damage. In the Figure 2.8, the 
monthly soil modulus is transferred to relative damage, and then the relative damages are 
averaged. The average relative damage is used to derive the effective soil modulus, which 
accounts for the monthly fluctuation of soil modulus. 
 
Figure 2.8 Chart for estimating effective subgrade resilient modulus (FHWA, 2006) 
The structural number is used to determine the thickness of respective layer. It is a 
function of layer thicknesses, layer coefficients, and drainage coefficients as Equation (2-6) 
shows, and it accounts for all the layers above the layer that MR characterizes (Figure 2.9).  
35 
 
 
Figure 2.9 Pavement structure (Garber and Hoel, 2009) 
For example, if MR in the Equation (2-6) is substituted by the modulus of the base course 
(E2) in Figure 2.9, the SN1 only accounts for the surface layer; if MR is substituted by the 
modulus of the subbase course (E3), the SN2 accounts for the surface and base layer. Therefore, 
the thickness of each layer can be derived by dividing ai from the corresponding ai * Di.     
nnn333222 111n D*m*aD*m*a + D*m*a +D*m*a =SN ?????                (2-6) 
where,  
SNn = structural number for the combined n-layer structure 
ai = layer coefficient for surface, base, or subbase (i = 1, 2, 3, ???, or n) 
mi = dainage coefficient for surface, base, or subbase (i = 1, 2, 3, ???, or n) 
Di = the thickness of surface, base, or subbase (i = 1, 2, 3, ???, or n), in. 
The AASHTO methods first adopted the concept of ESALs to quantify the effect of 
traffic loadings and the concept of PSI to evaluate the serviceability of pavement condition. 
Additionally, the AASHTO methods have specific variables in the design procedure to account 
for the effect of climate and subgrade. However, these methods are empirically-based because 
they are proposed based on the AASHO Road Test. Therefore, the efficiency of the AASHTO 
methods cannot be well-guaranteed in different regions with characteristics of modern traffic, 
climate, and materials. 
 
 
36 
 
MEPDG 
Introduction 
The Mechanistic-Empirical Pavement Design Guide (MEPDG) is a pavement design 
guide associated with its computer software that applies the mechanistic-empirical design to 
analyze input data and to predict the damage accumulation over the design life. It is applicable to 
designs for new or rehabilitated pavements, in any type of flexible, rigid, and semi-rigid 
pavement. The typical predicted distresses for flexible pavement designs are alligator cracking, 
rutting, thermal cracking; and for rigid pavement designs, faulting, cracking, and continuously 
reinforced concrete pavement (CRCP) punch-outs. Design performance can be compared with 
the threshold values, or comparisons of performance may be made for alternate designs with 
varying traffic, structure, and materials (Sharpe, 2004). 
The MEPDG is one of the most sophisticated design methods because it not only 
characterizes design inputs in all-inclusive details, but also incorporates the M-E design structure 
with advanced features in reliability, input hierarchy, and distress analysis. It is not as 
straightforward as the older AASHTO methods, which derive pavement thicknesses directly 
from design equation or the nomograph; it requires an iterative process in which the pavement 
structure configuration is adjusted until the predicted performance of pavement satisfies the 
design criteria. 
Input Characterization 
The MEPDG employs a hierarchical approach for selecting design inputs based on the 
designer?s knowledge of input parameters and importance of project (Li et al., 2011). The input 
Level 1 requires measured data directly, either in-situ or project-specific. This level represents 
the most familiar situation for a specific project. The input Level 2 needs a lower accuracy of 
37 
 
input information. The input parameters could be estimated by correlations or regression 
equations. The input Level 3 is the easiest one to determine just using estimated or default 
values. Most of the inputs in this study were provided at input Level 1, which helps to minimize 
the error in the estimation of design inputs.  
Material characterization 
The MEPDG characterize materials using many properties, which are required by the 
mechanistic model, transfer functions, and the climatic model (NCHRP Project 1-37A Report, 
Part 2, Chapter 2, 2004).  
For asphalt concrete, the mechanistic model requires two basic properties: dynamic 
modulus and Poisson?s ratio. The Poisson?s ratio can be assumed but the dynamic modulus is a 
variable affected by temperature and loading frequency. The dynamic modulus is provided by 
lab data at input Level 1. The properties for transfer functions are tensile strength, creep 
compliance, and coefficient of thermal expansion (NCHRP 1-37A Report, Part 2, Chapter 2, 
2004). In addition, the parameters required for the climatic model are mainly thermal properties 
of HMA and binder viscosity. 
For unbound materials, the resilient modulus and the Poisson?s ratio are used in the 
mechanistic model. The unit weight and the coefficient of thermal expansion are needed if the 
unbound material is considered to be stress-sensitive at input Level 1. No inputs are required by 
transfer functions for flexible pavement designs. In addition, the gradation parameters, plasticity 
index, optimum moisture contents are needed for climatic models (NCHRP Project 1-37A 
Report, Part 2, Chapter 2, 2004). 
 
 
38 
 
Environmental effects characterization 
The MEPDG adopted the Enhanced Integrated Climate Model (EICM) to characterize 
environmental effects. The EICM is a one-dimensional heat and moisture flow program that 
simulates changes in the behavior and characteristics of pavement and subgrade materials in 
conjunction with climatic conditions. The major output from the EICM is a set of adjustment 
factors for unbound layer materials, which accounts for the effects of environmental conditions. 
Also, three additional outputs of importance are the temperatures at midpoints of each bound 
layer, the temperature profiles within the asphalt for every hour, and the average moisture 
content for each sub-layer (NCHRP 1-37A Report, Part 2, Chapter 3, 2004). These outputs are 
incorporated into the design framework regarding materials characterization, structural response, 
and performance prediction. The EICM enables hourly characterization of temperature, moisture, 
and frost throughout the pavement structure, considerably enhancing the quality of pavement 
design. The source of environmental data comes from more than 800 weather stations of the 
National Climatic Data Center (NCDC) throughout the United States, which allows the user to 
select a given station or generate a climatic data file for local use (Li et al., 2011).  
Traffic characterization 
The MEPDG no longer uses the ESAL concept to evaluate traffic, but adopts the load 
spectra for analyzing traffic directly in a series of load ranges of axles. The inputs accounting for 
axle type, load magnitude, and traffic volume jointly define the number of loading repetitions for 
each axle group classified by axle configuration and magnitude; and the numbers of loading 
repetitions will be analyzed for the damage accumulation.   
 
 
39 
 
Mechanistic Models 
The MEPDG utilizes the JULEA multilayer elastic theory program to calculate 
mechanistic responses of the pavement structure assuming that all materials can be treated as 
linearly elastic (NCHRP 1-37A Report, Part 3, Chapter 3, 2004). Each pavement layer (i.e., 
asphalt, base, and subgrade) is actually divided into thinner sub-layers so that the varying 
properties at different depths can be captured. The JULEA program can predict stress, strain or 
deflection at any point in a particular layer or sub-layer based on material properties, layer 
structure, and loading conditions; meanwhile, it provides a combination of analysis features, 
theoretical rigor, and computational speed (NCHRP 1-37A Report, Part 3, Chapter 3, 2004). The 
load-induced responses are analyzed at critical locations with various depths and the most 
damage-vulnerable points are used to predict pavement distress performance (Schwartz and 
Carvalho, 2007). In addition, the Finite Element Model (FEM) would be applied to the special 
situation that the non-linear behavior of unbound materials is simulated (e.g., Input Level 1 for 
subgrade materials). 
Distress Prediction Models 
Fatigue cracking 
The fatigue cracking model for asphalt concrete in the MEPDG was obtained based upon 
the Asphalt Institute?s equation, which was derived by modifications to constant stress laboratory 
fatigue criteria (Asphalt Institute, 1982). The Asphalt Institute?s equation was calibrated to 
Equation (2-7) using data from 82 LTPP sections including new and rehabilitated pavements 
across the country with different climatic locations and diverse range of site features. The 
calibration process included collecting the LTPP calibration data, running the MEPDG 
simulation using different sets of coefficients, comparing the predicted damage to the measured 
40 
 
cracking, and minimizing the square of errors between predicted damage and the measured 
cracking (NCHRP 1-37A Report, Appendix II-1, 2004). The standard error of model predictions 
was 6.2% (NCHRP 1-37A Report, Appendix II-1, 2004). The local calibration coefficients     
were set to be 1 in the nationally-calibrated models.  
1 . 2 8 1*- ?3 . 9 4 9 2*- ?tff 3f2f1 E*?*?*C*0 . 0 0 4 3 2 =N                                      (2-7) 
where, 
   = number of repetitions of a given magnitude of load to failure  
   ,    ,     = local calibration coefficients 
C = laboratory to field adjustment factor, C=10M and  
0 .6 9 )-VV V(*4 .8 4 =M b e ffa b e ff?  
       = the effective binder content, percent by volume  
    = the air voids, percent by volume 
   = tensile strain at the critical location in the asphalt layer, in./in. 
  = asphalt concrete stiffness at given temperature, psi 
Then, the damage caused by a given magnitude of load will be accumulated in terms of 
the ratio between the actual load repetitions and allowable load repetitions based on the Miner?s 
Law (1945): 
? ijijk Nn =D                                                                (2-8) 
where, 
   = damage for layer k 
    = actual load repetitions for load i within period j 
41 
 
    = allowable load repetitions for load i within period j 
In order to provide in-situ representation for cracking, the MEPDG adopted another 
model, called alligator cracking model, to relate the accumulated damage (  ) to the cracked 
area. Fatigue cracking is measured in terms of cracking area as the percentage of the lane area in 
the MEPDG, and 50% of the total lane area is considered as failure (NCHRP 1-37A Report, 
Appendix II-1, 2004). This model was calibrated to be as followed using the same LTPP 
database. 
)601(*)e1 6000( =FC 1 0 0 ) )*C 2 '* lo g ( D*C2C 1 '*( C 1 k??                                        (2-9) 
- 2 . 8 5 6AC2 )h3 9 . 7 4 8 ( 1-- 2 . 4 0 8 7 4='C ?                                                (2-10) 
'C*-2='C 21                                                               (2-11) 
where, 
   = alligator fatigue cracking, percent 
  =    = 1 
   =damage for layer k, percent 
    = total thickness of asphalt layer, in. 
Rutting 
In the MEPDG, rutting is evaluated for asphalt concrete and unbound materials, 
respectively. For asphalt concrete, the adopted rutting model was initially based on Leahy?s 
model, modified by Ayes, and last by Kaloush (Schwartz and Carvalho, 2007). For unbound 
materials, the model was first derived by Tseng and Lytton, which was modified by Ayres and 
later on by El-Basyouny and Witczak (NCHRP 1-37A Report, Appendix GG-1, 2004). The 
national calibration was performed using the identical LTPP sections used for the fatigue 
42 
 
cracking model calibration. The calibration process included collecting the LTPP calibration 
data, running the MEPDG simulation using different sets of coefficients, selecting the best 
coefficients based on the reasonableness of the results, adjusting functions for confining 
pressure, and minimizing the square of errors between predicted damage and the measured 
cracking. As a result, the predictions produced a SE of 0.055 inch for the asphalt concrete model 
(NCHRP 1-37A Report, Appendix GG-1, 2004). The SE was 0.014 inch for the base/subbase 
model, and 0.056 inch for the subgrade model (NCHRP 1-37A Report, Appendix GG-1, 2004). 
For asphalt rutting prediction in the MEPDG, the asphalt concrete layer is subdivided into 
thinner sublayers, so the total predicted rut depth for the asphalt concrete layer can be achieved 
by: 
? ?n 1i iipAC ?h*)(? =R u t                                                             (2-12) 
where, 
      = rut depth at the asphalt concrete layer, in. 
n = number of sublayers 
(  )  = vertical plastic strain at mid-thickness of sublayer i, in./in. 
    = thickness of sublayer i, in. 
The vertical plastic strain (  ) in the sublayer i can be determined by the nationally-
calibrated model below, which was calibrated using identical LTPP pavement sections used for 
asphalt concrete fatigue cracking model calibration (NCHRP 1-37A Report, Appendix GG-1, 
2004). The resilient strain is derived by the mechanistic model. 
 
 
43 
 
32 *0 . 4 7 9 1*1 . 5 6 0 63 . 3 5 4 1 2-1
r
p N*T*10* = ?????                                      (2-13) 
where, 
   = computed vertical resilient strain at mid-thickness of the sublayer i, in./in. 
  ,   ,    = regression coefficients derived from laboratory testing 
  ,   ,    = local calibration coefficients 
T = temperature, ?F 
N = number of repetitions from a given load group 
The unbound granular materials are also divided into sublayers, and the total rut depth is 
the summation of rut depth appeared in each of sublayers. The rut depth (  ) appeared in any 
given sublayer is computed as: 
iv)N
?(-
r
p11i h*?*e*)??(*k*? =? ?                                                (2-14) 
where, 
  = rut depth or plastic deformation of the sublayer i, in. 
   = local calibration coefficient 
   = regression coefficient determined from the laboratory permanent deformation test 
   = intercept for the sublayer i determined from laboratory repeated load permanent 
deformation tests, in./in. 
   = resilient stain for the sublayer i imposed in laboratory test to obtain material 
properties (i.e.,   ,  , and  ), in./in. 
 ,   = material properties 
N = number of load applications 
44 
 
   = average vertical resilient strain in the sublayer i for a given load and calculated by 
the mechanistic model, in./in. 
   = thickness of the sub-layer i, in. 
It is not computationally feasible to divide the subgrade into sublayers so as to estimate 
the total subgrade rut depth because the subgrade is regarded as a semi-infinite layer in the 
structural response models (NCHRP 1-37A Report, Appendix GG-1, 2004). An adjustment on 
the rutting models is therefore required for computing the plastic strains in a semi-infinite layer. 
The plastic strain at different depths can be computed with a simpler method as Equation (2-15), 
and it will be used to determine the total subgrade rutting (Schwartz and Carvalho, 2007).  
z*-0PP e*)(=(z ) ???
                                                           
(2-15) 
where, 
  ( ) = plastic vertical strain at the depth z (measured from the top of the subgrade), 
in./in. 
    = plastic vertical strain at the top of the subgrade (i.e., z=0), in./in. 
Z = depth measured from the top of the subgrade, in. 
  = regression coefficient 
With regard to determine the value of  , two depths in the subgrade were utilized to fit 
Equation (2-15) and solve a system of two equations. The MEPDG uses the top of the subgrade 
(z=0) and 6 inches below the top (z=6). The plastic strains at these two depths can be derived by 
Equation (2-14) given that the plastic strain is also the ratio of the rut depth over the sublayer 
thickness (i.e.,   /  ).   
The total subgrade (   ) rut depth can be computed by integrating Equation (2-16) over 
the whole subgrade down until the bedrock (Schwartz and Carvalho, 2007). 
45 
 
 
p0
h-
sb )ke-1( =? b ed r o ck ?
?
                                                              
(2-16)
 
where,  
hbedrock = depth to the beckrock, in. 
Roughness  
The MEPDG roughness models were developed for flexible pavements with three types 
of base layers: relatively thick granular base, asphalt-treated base, and cement-treated base 
(NCHRP 1-37A Report, Appendix OO-3, 2004). The IRI model considered over 350 LTPP 
sections in the calibration. The SE was 0.387 m/km. The equation for conventional flexible 
pavement with thick granular base is shown as Equation (2-17). The calibration coefficients 
shown in the MEPDG were specified regarding rutting, fatigue cracking, transverse cracking, 
and site factors (Darter, Titus-Glover, and Von Quintus, 2009). 
IRI = IRI0 + 40 (RD) + 0.400 (FCTotal) + 0.008 (TC) + 0.015 (SF)              (2-17)
 
where,  
RD = average rut depth, in.  
FCTotal = total area of load-related cracking (combined alligator, longitudinal, and 
reflection cracking in the wheel path), percent of wheel path area 
TC = total length of transverse cracks, ft./mile 
SF = site factor 
     SF = Frosth + Swellp * Age1.5                                                              (2-18) 
where, 
Frosth = Ln [(Precip+1) * Fines * (FI + 1)] 
Swellp = Ln [(Precip+1) * Clay * (PI + 1)] 
Fines = Fsand + Silt 
46 
 
Age = pavement age, year 
PI = subgrade soil plasticity index, percent 
Precip = average annual precipitation or rainfall, in 
FI = average annual freezing index, ?F days 
Fsand = amount of fine sand particles (between 0.074 and 0.42 mm) in subgrade, percent 
Silt = amount of silt particles (between 0.074 and 0.002 mm) in subgrade, percent 
Clay = amount of clay particles (less than 0.002 mm) in subgrade, percent 
Local Calibration 
The MEPDG, like all mechanistic-empirical design methods, relies on predictions of 
pavement responses under load correlated to empirical predictions of distress. The distress 
prediction models in the MEPDG were derived based on a national calibration of previously-
developed equations using LTPP data from the entire U.S. Thus, the models reflect a wide range 
of traffic conditions, climates, and structural configurations, rather than a state-specific or local 
design scenario. Furthermore, modern advancements in asphalt technology, such as polymer-
modified materials and warm-mix asphalt were not included in the LTPP sections during the 
MEPDG national calibration. Since state and regional differences, in addFition to material 
characteristics, can strongly influence pavement performance, there is a need for local calibration 
of the transfer functions prior to implementation of the MEPDG.  
The aim of local calibration is to reduce the difference between an observed result and a 
predicted result to a minimum value. The calibration process is performed on pavement sections 
that are characterized in traffic, climate, material, and field performance. The stepwise procedure 
recommended for a local calibration includes (AASHTO?s Guide for the Local Calibration of the 
MEPDG, 2010): 
47 
 
a. Select hierarchical input levels for use in local calibration 
b. Develop experimental design & matrix 
c. Estimate sample size for each distress simulation model 
d. Select roadway segment 
e. Extract and evaluate roadway segment/test section data 
f. Conduct field investigation of test sections to define missing data 
g. Assess bias for the experimental matrix or sampling template 
h. Determine local calibration coefficient to eliminate bias of transfer function 
i. Assess standard error for transfer function 
j. Improve precision of model; Modify coefficients and exponents of transfer functions 
or develop calibration function 
k. Interpretation of results; Decide on adequacy of calibration coefficients 
The recommended procedure by AASHTO provides a systematic guidance for 
performing local calibration of the MEPDG. The methodology of this study is discussed in 
Chapter 4 including local calibration and validation. The following section details local 
calibration efforts conducted by several states.  
LOCAL CALIBRATION CASE STUDIES 
Washington 
The Washington State Department of Transportation (WSDOT) initiated MEPDG local 
calibration for the replacement of the 1993 AASHTO Design Guide by the MEPDG. The 
research utilized a total of 38 weight-in-motion stations with different locations and traffic 
patterns throughout Washington. The climate data recorded by the weather stations in 
Washington were tested, inspected, and judged to be acceptable for calibration efforts (Li et al., 
48 
 
2009). The traffic data were obtained from the WSDOT project design records. The structural 
information and pavement performance were provided by the Washington State Pavement 
Management System (WSPMS). 
The use of the calibration dataset combined two statistical approaches: split-sample 
approach and jackknife testing approach (Li et al., 2009). The split-sample approach is to split 
the sample into two subsets: one for the local calibration and the other for the validation. The 
datasets for local calibration and validation need to be independent of each other to ensure 
validity of analysis. The jackknife approach is to re-compute the statistic estimate leaving out 
one or more observations at a time from the sample set. In the new set of replicates of the 
statistic, an estimate for the bias and an estimate for the variance of the statistic can be 
calculated. 
The local calibration process was implemented by five basic steps: bench testing, model 
analysis, calibration, validation, and iteration (Li et al., 2009). For the bench testing, it aimed to 
evaluate the MEPDG regarding the run-time issues and the reasonableness of model predictions, 
and to identify the needs of local calibration. The reasonableness of the models was checked by 
varying key design parameters of traffic loading, climate, layer thickness, and soil properties; 
and then comparing results with expected pavement performance. 
The local calibration should be performed in an appropriate order among different 
distress models (Li et al., 2009). First, the asphalt mixture fatigue model needed to be calibrated 
before the longitudinal and alligator cracking models. Second, the cracking and rutting models 
needed to be calibrated before calibrating the roughness model. Also, the team conducted a 
sensitivity analysis on the calibration coefficients of the models, which intends to evaluate how 
49 
 
much each calibration coefficient affects the results of pavement distress models (Li et al., 2009). 
The sensitivity was defined as follows: 
i
i
Cd i s t r e s s
CC
d is tr e s sd is tr e s sE i )( )( ???                                             (2-19) 
where, 
           = sensitivity of calibration coefficient Ci for the associated distress condition 
? (distress)= change in estimated distress associated with a change in factor Ci 
? (Ci)= change in the calibration coefficient Ci 
distress = estimated distress using default calibration coefficients 
Ci = default value 
Table 2.1 shows the sensitivity of each calibration coefficient based on typical pavement 
characteristics in Washington State. A positive value implies that the prediction increases as the 
calibration coefficient increases, and a negative value implies that the prediction decreases as the 
coefficient increases. Zero implies that the coefficient has no impact on the model prediction. 
 
 
 
 
 
 
 
 
50 
 
Table 2.1 Sensitivity of the MEPDG calibration coefficients (Li et al., 2009) 
Local calibration 
coefficients 
Sensitivity values Related variables in the  
MEPDG distress models 
AC fatigue   
    -3.3 Effective binder content, air voids, 
AC thickness 
        -40 Tensile strain 
         20 Material stiffness 
Longitudinal cracking   
   -0.2 Fatigue damage, traffic 
          1 Fatigue damage, traffic 
          0 No related variable 
       ? 0 No related variable 
Alligator cracking   
          1 AC thickness 
          0 Fatigue damage, AC thickness 
       ? 0 No related variable 
Rutting   
         0.6 Layer thickness, layer resilient 
strain 
         20.6 Temperature 
         8.9 Number of load repetitions 
IRI   
   Not available Rutting 
   Not available Fatigue cracking 
   Not available Transverse cracking 
   Not available Site factor 
Only new flexible pavement sections were selected in this study. The local calibration 
results were derived by minimizing the root-mean-square error (RMSE) between the MEPDG 
predictions and WSPMS measured data on all selected sections. 
The local calibration was performed based on two representative sections, while the local 
calibration results were validated by using a broad dataset which are independent of the two 
sections. Thirteen flexible pavement stations for validation were selected from three different 
data sources based on traffic level, soil, and climate conditions. Five pavement sections from a 
previous WSDOT study covered high/medium/low traffic with strong subgrade in western 
Washington, medium traffic with strong subgrade in eastern Washington, and medium traffic 
with weak subgrade in eastern Washington. Six representative sections were from WSPMS 
51 
 
accounting for high/low traffic with strong subgrade for both western and eastern Washington, 
low traffic with weak subgrade in eastern Washington, and medium traffic with strong subgrade 
on mountain area. In addition, two sections in Washington were used in the original MEPDG 
national calibration, which was classified as medium traffic with strong subgrade in eastern 
Washington and high traffic with strong subgrade in western Washington (Li et al., 2009). 
For the calibration, an initial set of calibration coefficients was assumed in the MEPDG 
for each of the two selected sections for the local calibration. Then, one of the calibration 
coefficients with a higher sensitivity value was varied and run iteratively to evaluate the RMSE 
(Li et al., 2009). Once no significant improvement for RMSE was made, another coefficient with 
the higher sensitivity among the rest was adjusted in the MEPDG. The results of local calibration 
should be a set of calibration factors that lead to the lowest RMSE between the MEPDG 
predictions and WSPMS observations. Then, the local calibration results were validated by using 
other sections. Table 2.2 below shows the final calibration results of this study. 
 
 
 
 
 
 
 
 
 
 
52 
 
Table 2.2 Final local calibration coefficients (Li et al., 2009) 
Calibration Factor MEPDG Default After Local Calibration 
AC fatigue   
    1 0.96 
    1 0.97 
    1 1.03 
Longitudinal cracking   
   7 6.42 
       3.5     3.596 
   0        0 
       1000     1000 
Alligator cracking   
   1 1.071 
                             1        1 
       6000     6000 
AC rutting   
    1 1.05 
                              1     1.109 
    1      1.1 
Subgrade rutting   
    1        0 
IRI   
        40      N/A 
        0.4      N/A 
   0.008      N/A 
   0.015      N/A 
In this study, it was found that the alligator cracking was significant in WSDOT flexible 
pavements; and the locally-calibrated alligator cracking curve showed the same trend as the 
measured data (Li et al., 2009). The plots below show the measured and predicted alligator 
cracking in western and eastern Washington. 
53 
 
 
Figure 2.10 Measured and predicted alligator cracking in western Washington (Li et al., 
2009) 
 
Figure 2.11 Measured and predicted alligator cracking in eastern Washington (Li et al., 
2009) 
The longitudinal cracking was significant and the locally-calibrated model provided a 
reasonable prediction of longitudinal cracking (Li et al., 2009). The plots below show the 
measured and predicted longitudinal cracking in western and eastern Washington. 
54 
 
 
Figure 2.12 Measured and predicted longitudinal cracking in western Washington (Li et 
al., 2009) 
 
Figure 2.13 Measured and predicted longitudinal cracking in eastern Washington (Li et al., 
2009) 
As for the transverse cracking, WSPMS data showed no transverse cracking in Western 
Washington but a significant amount in eastern Washington (Li et al., 2009). The predicted 
transverse cracking by the nationally-calibrated model matched well with the WSPMS data and 
55 
 
WSDOT observations. The plots below show the measured and predicted transverse cracking in 
western and eastern Washington. 
 
Figure 2.14 Measured and predicted transverse cracking in western Washington (Li et al., 
2009) 
 
Figure 2.15 Measured and predicted transverse cracking in eastern Washington (Li et al., 
2009) 
56 
 
In addition, the rutting predictions of the locally-calibrated model matched well with 
WSPMS data throughout the design life. The plots below show the conformity of the measured 
rutting with predicted rutting in western and eastern Washington. 
 
Figure 2.16 Measured and predicted rutting in western Washington (Li et al., 2009) 
 
Figure 2.17 Measured and predicted rutting in eastern Washington (Li et al., 2009) 
57 
 
For the roughness, the measured values of WSDOT flexible pavements tended to be 
higher than the nationally-calibrated model predictions. However, the difference between the 
MEPDG predictions and measurements can be minimized by the roughness local calibration (Li 
et al., 2009). The plots below show the comparison between the measured and predicted IRI in 
western and eastern Washington. 
 
Figure 2.18 Measured and predicted IRI in western Washington (Li et al., 2009) 
 
Figure 2.19 Measured and predicted IRI in eastern Washington (Li et al., 2009) 
58 
 
North Carolina 
The North Carolina Department of Transportation (NCDOT) conducted a local 
calibration of the MEPDG for asphalt pavements in terms of the rutting model and bottom-up 
fatigue cracking model (Muthadi and Kim, 2008). In terms of the calibration dataset, 30 LTPP 
pavement sections (i.e., 16 new flexible pavement sections and 14 rehabilitated sections) were 
chosen from the LTPP database, and 23 NCDOT pavement sections were selected from the 
NCDOT data system. A total of 53 sections were selected from three geographic areas: 
mountain, piedmont, and coastal regions. The climatic data for these sections was obtained by 
interpolating the data from weather stations nearby based on the local geographic information 
(Muthadi and Kim, 2008).  
An experimental matrix was developed to classify all sections based on their pavement 
types and structures (Muthadi and Kim, 2008). 39 new sections (i.e., 11 sections with thin HMA, 
11 with intermediate HMA, and 17 with thick HMA) and 14 rehabilitated sections (i.e., 8 
sections with intermediate HMA and 6 sections with thick HMA) were selected in this study. 
Approximately 80% of the dataset (including data from 53 sections) were utilized for the 
calibration and the remaining 20% were for the validation.  
The local calibration plan included three steps (Muthadi and Kim, 2008): first, run the 
MEPDG based on the local traffic, climate, and material information using the nationally-
calibrated models. Second, calibrate the coefficients in the models to minimize the bias between 
the predicted and the measured distresses and reduce the standard error, if any exists. Third, 
validate the recommended calibration coefficients by evaluating the accuracy of prediction 
models. The following are major concerns and important findings in this study. 
59 
 
For the rutting model, it was recognized that the appropriate value of calibration 
coefficient    can be solved by using the Microsoft Excel Solver program instead of running the 
MEPDG many times (Muthadi and Kim, 2008). Furthermore,     was varied to reduce the error 
between the predicted and measured granular base rutting; and     was varied for the subgrade 
rutting (Muthadi and Kim, 2008). However, the measured rut depth distributed in each layer (i.e., 
asphalt, granular base, and subgrade) cannot be directly obtained in the field like the total rut 
depth, so the measured rut depth must be estimated. It was assumed that the portion of the 
predicted rut depth distributed in each layer was the same as the portion of the measured rut 
depth distributed in each layer. So the measured granular base rut depth and subgrade rut depth 
can be calculated based on their proportions accounted in the total predicted rut depth (Muthadi 
and Kim, 2008).  
In the local calibration, a Chi-square test was performed to determine whether the 
standard error provided by the local calibration results was significantly different from that 
provided by the national calibration results (Muthadi and Kim, 2008). For the rutting model, the 
standard error provided by local calibration results was not significantly different from that by 
national calibration results (Table 2.3). For the fatigue cracking model, the results of local 
calibration (i.e., R2 = 0.11 and standard error = 3.64) were still considered to be better than the 
results of national calibration (i.e., R2 = 0.03 and standard error = 6.02), but Figure 2.20 revealed 
that neither of them looked satisfactory. One possible reason identified was that the cracks 
outside the wheel path were usually ignored in the measurements by NCDOT, which caused the 
under-estimation of fatigue cracking.  
 
60 
 
 
Table 2.3 Statistical summary of the rutting model local calibration results (Muthadi and 
Kim, 2008) 
 AC  GB  
 National 
calibration 
Local 
calibration 
National 
calibration 
Local 
calibration 
Average Rut 
depth (in) 
0.1178 0.1030 0.0442 0.0344 
SE(in) 0.054 0.047 0.027 0.021 
Bias(in) -0.0149 0 -0.0098 0 
p-value 
 
0.00622, 
<0.05 
0.499, 
>0.05 
0.0002, 
<0.05 
0.5, 
>0.05 
N 111 111 111 111 
 SG  Total  
 National 
calibration 
Local 
calibration 
National 
calibration 
Local 
calibration 
Average Rut 
depth(in) 
0.1551 0.1026 0.3171 0.2399 
SE(in) 0.084 0.056 0.154 0.109 
Bias(in) -0.0525 0 -0.0771 0 
p-value 1.6E-9, 
<0.05 
0.5, 
>0.05 
2.9E-7, 
   <0.05 
0.499, 
>0.05 
N 111      111       111      111 
  
61 
 
 
Figure 2.20 Measured versus predicted distresses: (a) total rut depth predicted by 
nationally-calibrated rutting model (b) total rut depth by locally-calibrated rutting model 
(c) fatigue cracking by nationally-calibrated rutting model (d) fatigue cracking by locally-
calibrated rutting model (Muthadi and Kim, 2008) 
Then, the remaining 20% of the dataset were used to validate the reasonableness of the 
final calibrated models. The results of the validation were shown in Table 2.4. 
Table 2.4 Statistical summary of validation results (Muthadi and Kim, 2008) 
 rutting model fatigue cracking model 
SE 0.145 in 4.86 % 
bias 0.033 in -5.04% 
N 26 32 
chi-square Statistic 36.82 5.66 
degree of freedom 25 31 
p-value 0.0599, >0.05 0.9999, >0.05 
Finally, the recommendation regarding the calibration coefficients was exhibited in Table 
2.5. It was concluded that the rut depth values predicted by the locally-calibrated model matched 
well the observed rut depth values in LTPP sections; the locally-calibrated alligator cracking 
62 
 
model under-predicted the percentage of alligator cracking for most of the sections (Muthadi and 
Kim, 2008). 
Table 2.5 Recommended calibration coefficients (Muthadi and Kim, 2008) 
 calibration 
coefficient 
national 
calibration 
local  
calibration 
Rutting    
AC    -3.4488 -3.41273 
    1.5606 1.5606 
    0.479244 0.479244 
GB     1.673 1.5803 
SG     1.35 1.10491 
Fatigue    
AC    0.00432 0.007566 
    3.9492 3.9492 
    1.281 1.281 
    1 0.437199 
    1 0.150494 
Ohio State 
Mallela et al. composed guidelines of implementing the MEPDG for Ohio. They 
collected input data using information from the LTPP sections. The sensitivity analysis was first 
performed to evaluate how the variation in the output of a model can be apportioned, 
qualitatively or quantitatively, to different sources of variation in the input of a model (Mallela et 
al., 2009). Then, the local calibration of performance prediction models were conducted by 
modifying calibration coefficients. The adequacy of the local calibration was evaluated by 
checking the results of prediction with the measured data. The statistical analysis was provided 
to evaluate the difference between predictions and measured data. 
In this study, the alligator cracking model was not calibrated because the field 
measurement of the alligator cracking was often disturbed by the early generated longitudinal 
cracking during construction. For the HMA transverse cracking, a non-statistical comparison was 
performed between measured and predicted values after the local calibration, and a further 
evaluation was recommended by using data from northern Ohio sites (Mallela et al., 2009). The 
63 
 
nationally-calibrated rutting model consistently over-predicted the rut depth. Considering the fact 
most of the sections used for local calibration have relatively thick HMA layers, the proportion 
of HMA layer rutting (about 17 to 44 percent) in the total rutting was proportionately higher. 
Therefore, they attempted to reduce the proportion of unbound layers in the total rutting by 
adjusting the sub-model coefficients     and    . Also, the MEPDG over-predicted rutting for 
lower magnitudes of measured rutting and under-predicted rutting for the higher magnitudes of 
measured rutting. This requires an adjustment to     and     of the HMA rutting sub-model 
(Mallela et al., 2009). The results of the statistical analysis before local calibration are presented 
in Table 2.6 and Figure 2.21. 
Table 2.6 Hypothesis testing before local calibration of the rutting model (Mallela et al., 
2009) 
Hypothesis DF Parameter 
Estimate 
Std. 
Error 
t Value p-value 95 Percent 
Confidence 
Limits 
H0: Intercept =0 1 0.2178 0.0059 36.8 <0.0001 0.21 0.23 
H0: Slope = 1 1 1.0228 0.0571 13.49 <0.0001 0.65 0.88 
H0: Measured- 
Predicted Rutting 
=0 
101   -52.7 <0.0001   
 
Figure 2.21 Measured versus predicted HMA total rutting before local calibration (Mallela 
et al., 2009) 
64 
 
The total rutting model was locally-calibrated using Equation (2-20) (Mallela et al., 
2009). The results of the statistical analysis after local calibration are presented in Table 2.7. 
S u b g r a d eB a s eACt o t a l R u t*33.0R u t*32.0R u t*0 . 5 1 =R u t ??                   (2-20) 
where, 
Ruttotal= total rut depth, in.  
RutAC= asphalt rut depth, in. 
RutBase= base rut depth, in. 
RutSubgrade= subgrade rut depth, in. 
Table 2.7 Hypothesis testing after local calibration of the rutting model (Mallela et al., 
2009) 
Hypothesis DF Parameter 
Estimate 
Std. 
Error 
t Value p-value 95 Percent 
Confidence 
Limits 
H0: Intercept =0 1 0.083 0.0024 34.4 <0.0001 0.078 0.087 
H0: Slope = 1 1 0.952 0.049 19.4 0.3395 0.855 1.05 
H0: Measured- 
Predicted Rutting 
=0 
101   -5.62 <0.0001   
 
The measured and predicted HMA total rutting are shown in Figure 2.22. It can be seen 
that the number of data points was the same as that in Figure 2.21 but the R2 was 0.63. It was 
found previously in Figure 2.21 that the predicted rutting was generally 0.2 in. higher than the 
measured rutting. However, the R2 was reported to be 0.64, which was even higher than the R2 
value of 0.63 in Figure 2.22. Therefore, the local calibration did improve the prediction but did 
not increase the R2. It suggested that the R2 was not an efficient statistic to evaluate the 
improvement of MEPDG predictions.   
 
65 
 
 
Figure 2.22 Measured versus predicted HMA total rutting after local calibration (Mallela 
et al., 2009) 
It was found that the goodness of fit of the locally-calibrated model was adequate, but the 
model predictions were still with bias. A more comprehensive evaluation of HMA pavement 
mixtures and a larger calibration dataset would be necessary to calibrate the models for ODOT in 
the future (Mallela et al., 2009). Besides, the nationally-calibrated IRI model for HMA provided 
a poor correlation (        ) between the measured and MEPDG predicted values. Through a 
local calibration, the R-square value increased to 0.69 while the SEE was about the same as the 
nationally-calibrated model (Mallela et al., 2009). The bias produced by the locally-calibrated 
model was considered to be acceptable compared with the nationally-calibrated model. 
Minnesota State 
Minnesota utilized the data from the MnROAD project for their local calibration (Hoegh 
et al., 2010). They mainly compared the actual measurement of rutting data from 31 test sections 
located on I-94 highway with the MEPDG predicted rutting data for those sections. It was found 
that the total rut depth predicted by the nationally-calibrated model was far higher than measured 
total rut depth (Hoegh et al., 2010). The predicted base and subgrade rutting was found 
66 
 
unreasonably high as Figure 2.23 shows. However, the HMA rut depth predicted by the 
nationally-calibrated model was quite similar to the measured total rut depth. Therefore, their 
local calibration modified the nationally-calibrated model predictions by subtracting the 
predicted value in the first month. This can reduce the bias between the predicted and measured 
rut depth. The following local calibration of the MEPDG rutting model was proposed as below 
(Hoegh et al., 2010). 
*R u t*R u tR u t =R u t S u b g r a d eB a s eACt o t a l ??                                     (2-21) 
B a s e 1B a s eB a s e R u tR u t =*R u t ?                                               (2-22) 
S u b g r a d e 1S u b g r a d eS u b g r a d e R u tR u t =*R u t ?                                        (2.23) 
where, 
Ruttotal = predicted surface rut depth using the nationally-calibrated model, in. 
RutAC = predicted asphalt rut depth using the nationally-calibrated model, in. 
RutBase*= predicted base rut depth after model local calibration, in. 
RutSubgrade*= predicted subgrade rut depth after model local calibration, in. 
RutBase = predicted base rut depth using the nationally-calibrated model, in. 
RutSubgrade = predicted subgrade rut depth using the nationally-calibrated model, in. 
RutBase1 = predicted base rut depth using the nationally-calibrated model in the first 
month, in. 
RutSubgrade1= predicted subgrade rut depth using the nationally-calibrated model in the 
first month, in. 
67 
 
 
Figure 2.23 Measured and predicted rutting for Section 2 before local calibration (Hoegh et 
al., 2010) 
 
Figure 2.24 Measured and predicted rutting for Section 2 after local calibration (Hoegh et 
al., 2010) 
So the predicted rutting curve after local calibration had a better correlation with 
measured rutting curve for the entire life of pavement and this trend was observed for most of the 
sections (Figure 2.24). It was suggested that the locally-calibrated rutting model was an 
improvement for the Minnesota conditions despite a wide range of local design features, such as 
different traffic mixes, and different pavement ages (Hoegh et al., 2010). 
 
68 
 
SUMMARY 
The local calibration of the MEPDG needs to be performed based on a comprehensive 
understanding of pavement design concepts. Therefore, this chapter gathers the basic knowledge 
from literature to equip the study toward pavement designs. It can help to inform the 
fundamentals of pavement design and illuminate the modern trend of design needs before getting 
into the presentation of the study. The next chapter introduces the dataset of the study including 
input characterization and pavement performance measurements.  
69 
 
CHAPTER 3  
MEPDG INPUT CHARACTERIZATION AND PERFORMANCE DATA COLLECTION 
TEST FACILITY INTRODUCTION 
The National Center for Asphalt Technology (NCAT) at Auburn University has been 
devoted to exploring and advancing asphalt technology to fulfill the pavement industry needs 
since 1986. The main task of NCAT is to perform asphalt material testing and research and to 
investigate real-world pavement performance. 
Test Track 
Research cycle 
The Test Track began in 2000 with 46 sections comprising a 1.7 mile track. In 2003, 23 
of 46 original sections were rebuilt to facilitate another round of research. Of these sections, 
eight were rebuilt and utilized for a structural experiment; fifteen were shallow milled and inlaid 
(i.e., between ? and 4 inches deep), while the other 23 sections remained in place to serve as a 
continuation of the original experiment. In 2006, eleven structural sections were newly built with 
varied thicknesses, and other sections from the past cycle were rehabilitated with shallow milling 
(i.e., less than 4 inches) or left in place. In 2009, the fourth research cycle began. Seventeen of 
the test sections were either reconstructed or rehabilitated, while the remaining 29 were left in 
place to allow for additional traffic loading (West et al., 2012). In this research, the data used 
was mainly from the 2003 and 2006 research cycles.  
Structural study 
NCAT initiated the structural study of pavements in the 2003 research cycle to study 
dynamic response and long-term performance of various pavement structures, and to validate 
and calibrate performance prediction models based on Test Track conditions (Priest, 2013). In 
the 2003 research cycle, eight sections were designed, instrumented, and investigated to be used 
70 
 
for the initial structural study (Timm, 2009). In the 2006 research cycle, five structural sections 
(i.e., N3, N4, N5, N6, and N7) from 2003 were left in-place, and three structural sections (i.e., 
N1, N2, and N8) from 2003 were rebuilt; moreover, three new structural sections (i.e., N9, N10, 
and S11) were added.  
Materials Laboratory 
The NCAT materials laboratory is equipped with state-of-the-art facilities to perform all 
routine mix design and quality control tests for asphalt binders and mixtures. The laboratory 
provided sufficient material characterization for inputs into the MEPDG.       
MEPDG INPUTS CHARACTERIZATION 
Introduction 
A variety of input parameters were characterized to describe the design scenarios based 
on the results of lab testing, field measurement, and theoretical correlation. Level 1 inputs 
required by the MEPDG were developed, when possible, to ensure the highest accuracy of model 
predictions of pavement performance.   
General Information 
From the construction records of the 2003 and 2006 research cycle, the base/subgrade 
construction month for each section was retrieved, as well as the asphalt layer construction 
month (Table 3.1). This information was to provide the project starting point in the MEPDG. 
Once pavement construction finished, traffic loading started for the whole track in the next 
month. At that point, the initial pavement smoothness was measured in terms of IRI (Table 3.2). 
 
 
 
 
71 
 
Table 3.1 Construction and traffic months 
Section Test 
Cycle 
Base/Subgrade 
Construction 
Month 
Asphalt layer 
Construction 
Month 
Traffic 
Opening 
Month 
N1 2003 June July October 
N2 2003 June July October 
N3 2003 June July October 
N4 2003 June July October 
N5 2003 June July October 
N6 2003 June July October 
N7 2003 June July October 
N8 2003 June July October 
N1 2006 August September November 
N2 2006 August September November 
N8 2006 August October November 
N9 2006 August October November 
N10 2006 August October November 
S11 2006 August October November 
 
Table 3.2 Initial IRI values 
Section Test Cycle Initial IRI 
(in/mile) 
N1 2003 57 
N2 2003 56 
N3 2003 35 
N4 2003 48 
N5 2003 59 
N6 2003 52 
N7 2003 44 
N8 2003 42 
N1 2006 115 
N2 2006 106 
N8 2006 95 
N9 2006 105 
N10 2006 65 
S11 2006 67 
 
Traffic 
The truck fleet at the Test Track runs at a target speed of 45 mph, and operates 16 hours 
daily, six days a week for each two-year cycle. Each of the trucks completes about 680 miles per 
day so as to apply 10 million ESAL collectively in two years. Thanks to simple truck patterns 
72 
 
and running schedule, input Level 1 for traffic information were precisely characterized for the 
MEPDG analysis. The details of traffic inputs are explained as below. 
Truck types and axle loads 
In the 2003 research cycle, five of the trucks, termed ?triple trailers?, consisted of a 
steering axle, a tandem axle, and five trailing single axles (Figure 3.1). The sixth truck, termed 
?box trailer?, consisted of a steer axle and two tandems (Figure 3.2). In the 2006 cycle, only five 
triple trailers were used to apply traffic loads. 
 
Figure 3.1 Triple trailer truck 
 
Figure 3.2 Box trailer truck 
The axle weight of trucks is needed by the MEPDG as well. In the 2003 research cycle, 
five triple trailers had the same axle weights distribution. The single and tandem axles weighed 
approximately 21,000 lb and 42,000 lb respectively, while the steer axle weighed around 12,000 
lb. The box trailer had 2 tandem axles that weighed around 34,000 lb. each and a steer axle that 
weighed around 12,000 lb. In the 2006 cycle, the box trailer was no longer used. Five triple 
73 
 
trailers had the same weights of axles as those in the 2003 research cycle; however, one of the 
five had a steer axle that weighed 10,000 lb, which was 2,000 lb less than steer axles of the other 
four. In other words, 80% of the steer axles weighed 12,000 lb and 20% weighed around 10,000 
lb. It is noted that the MEPDG does not differentiate between steer and single axles so all steer 
axles were modeled as single axles with different weights.   
Number of axles per vehicle 
Another required input was the average number of axles for each vehicle classification 
(i.e., FHWA Truck Class 4 through 13). Though the box trailer in the 2003 research cycle could 
be easily modeled as a Class 9 vehicle with one steer and two tandem axles, the triple trailer had 
six single axles, including the steer axle and trailing single axles. It exceeded the maximum 
number of single axles (i.e., five axles) among all the vehicle classes allowed in the MEPDG 
v1.1. In order to represent a triple trailer, two fictitious vehicle classes were used together with 
five single axles and one tandem axle from the Class 13, and the remaining one single axle from 
the Class 12. The same inputs were used to model the triple trailers in the 2006 research cycle. 
The inputs in the MEPDG are shown in Figure 3.3. 
 
Figure 3.3 Number of axles per vehicle 
 
74 
 
Axle configuration 
The truck axle configurations were specified by Powell and Rosenthal (2011), and the 
average value of trucks used in the 2003 and 2006 research cycle were entered into the MEPDG. 
The average axle width was 8.5 ft, the dual tire spacing was 13.5 in, and the tire pressure was 
approximately 100 psi. Other traffic inputs (i.e., lateral traffic wander) were assumed to be 
routine design values, and they were left as the defaults provided by the MEPDG. 
AADTT and traffic distribution  
The volume of traffic on highways is often evaluated depending on the truck traffic 
because it is not necessary to consider the minor damage on pavement caused by vehicles with 
less than six tires (i.e., motorcycles, passenger cars, buses, and other vehicles). All the heavily-
loaded vehicles used at the Test Track were regarded as the truck traffic. The annual average 
daily truck traffic (AADTT) is a major input regarding traffic information in the MEPDG. 
According to traffic records, the daily truck passes was determined to be 1,155 trucks per day on 
average for the 2003 research cycle and 1,541 trucks per day for the 2006 cycle.  
The heavily-loaded vehicles were simulated in the MEPDG by utilizing existing vehicle 
classes, but adjusting the number of truck passes or AADTT. For example, a triple trailer was 
equivalently treated as a Class 13 vehicle and a Class 12 vehicle; as a result, the number of daily 
truck passes describing the amount of triple trailers was doubled. In the 2003 research cycle, 5/6 
of the total truck passes was completed by triple trailers, so 962.5 out of 1,155 were double 
counted as 1,925. The remainder of the truck passes by the box trailer, 193, was added by the 
doubled value of 1,925 to reach a final AADTT of 2,118 per day. As for the 2006 research cycle, 
the box trailer was not used, so the AADTT of 1,541 was doubled to 3,082 per day. For the 
vehicle class distribution, in the 2003 research cycle, the Class 9 comprised 9%, and the Class 12 
75 
 
together with the Class 13 equally comprised 91%; for the 2006 research cycle, the Class 12 
represented 50% and the Class 13 represented 50%. 
The MEPDG also allows users to specify the average hourly distribution of traffic on a 
daily basis. The hourly traffic distribution for the Test Track is shown in Figure 3.4 (for the 2003 
research cycle) and Figure 3.5 (for the 2006 research cycle). The fluctuation of volume over the 
hours was the result of shift changes, truck refueling, driver breaks, and truck maintenance stops. 
The monthly traffic applied was consistent over research cycles, so the monthly distribution 
factors were all 1.0 and there was no annual traffic growth. 
 
Figure 3.4 Hourly traffic distribution for the 2003 research cycle 
0. 0 0. 0 0. 0 0. 0
2. 5
6. 5 6. 5
3. 3
6. 5 6. 5
3. 3
6. 4 6. 3
3. 4
6. 3 6. 4
3. 2
6. 4 6. 4
3. 3
6. 4 6. 5
3. 9
0. 0
0
1
2
3
4
5
6
7
8
9
10
1:
00 
AM
2:
00 
AM
3:
00 
AM
4:
00 
AM
5:
00 
AM
6:
00 
AM
7:
00 
AM
8:
00 
AM
9:
00 
AM
10:
00 
AM
11:
00 
AM
12:
00 
PM
1:
00 
PM
2:
00 
PM
3:
00 
PM
4:
00 
PM
5:
00 
PM
6:
00 
PM
7:
00 
PM
8:
00 
PM
9:
00 
PM
10:
00 
PM
11:
00 
PM
12:
00 
AM
H o u r  o f D ay
Per
cen
tag
e 
o
f 
D
ai
l
y 
Vo
l
u
me
76 
 
 
Figure 3.5 Hourly traffic distribution for the 2006 research cycle 
Climate 
Introduction 
The climatic data required in the MEPDG is used by the Enhanced Integrated Climate 
Model (EICM) to calculate changes in the temperature and moisture profile throughout the 
pavement cross section. The climatic input for the MEPDG is actually a file that contains a 
recorded history of temperature, rainfall, wind speed, humidity, and sunlight conditions for a 
specific area. There are two ways to prepare the climatic inputs for the MEPDG, either by 
selecting a climatic data file for representative areas or by preparing a new climatic data file 
based on a local weather station. The latter was adopted in this study because the Test Track has 
an on-site weather station (Figure 3.6), which is responsible for collecting environmental 
information on an hourly basis. The Test Track is at a geographic coordinate of 32?59?N, -
85?30?W, and an elevation of 600ft. The next section will cover the method to prepare a climate 
file for a particular condition. 
0. 0 0. 0 0. 0 0. 0 0. 0
6. 6
6. 8
3. 5
6. 8 6. 7
3. 6
6. 8
5. 1
3. 5
6. 8 6. 9
3. 6
6. 9 7. 0
3. 5
7. 0
5. 5
3. 4
0. 0
0. 0
1. 0
2. 0
3. 0
4. 0
5. 0
6. 0
7. 0
8. 0
9. 0
10. 0
1:
00 
AM
2:
00 
AM
3:
00 
AM
4:
00 
AM
5:
00 
AM
6:
00 
AM
7:
00 
AM
8:
00 
AM
9:
00 
AM
10:
00 
AM
11:
00 
AM
12:
00 
PM
1:
00 
PM
2:
00 
PM
3:
00 
PM
4:
00 
PM
5:
00 
PM
6:
00 
PM
7:
00 
PM
8:
00 
PM
9:
00 
PM
10:
00 
PM
11:
00 
PM
12:
00 
AM
H o u r  o f D ay
Per
cen
tag
e 
o
f 
D
ai
l
y 
Vo
l
u
me
77 
 
 
Figure 3.6 Test Track on-site weather station 
Generating the climate file for the Test Track  
It is noted that two formats of files function in the MEPDG; one is the ICM file and the 
other is the hourly climatic database file. The ICM file was generated by the MEPDG calculation 
based on an hourly climatic database file. In fact, the hourly climatic database file was either 
given for those representative areas or can be self-developed. The required form of the hourly 
climatic database file is shown in Figure 3.7. Each row of the data file contains the elements 
explained in parentheses. 
78 
 
 
Figure 3.7 Hourly climatic database file 
The procedure of preparing an hourly climatic database file is detailed in three steps as 
follows: 
1. Extract useful elements from the station data file. Microsoft Excel can help to generate 
climatic data in an hourly order.  
2. Check the raw data to see whether some hourly records were missing or replicated. 
This temporary collection error may be the result of a battery problem. If one hour?s data was 
missing, the data record of the same month, day and hour in the previous or the next year was 
used to supplement the missing data. If one hour?s data replicated, the data collected at the first 
time were kept as valid in the file. The data file can be finally exported to a CSV file and copied 
into a HCD file by Microsoft Notepad for later use.    
3. Fill in basic information into the station.dat file which outlines a basic description of 
weather station such as station number, location, and first date in file. The station.dat file was 
written in the format defined by the MEPDG.  
79 
 
4. Place the ?station.dat? file into a folder named ?default? under the directory of the 
MEPDG and copy the HCD file into the ?hcd? folder; then, upload these files into the climate 
module in MEPDG by a ?generate? command. Finally, the climatic data for the NCAT Test 
Track weather station was generated for climatic inputs.   
Structure 
When it comes to the pavement structure, the MEPDG requires detailed inputs in terms 
of the layer thickness, material types, and structural capacity. The investigated pavement sections 
were discussed below. 
Cross sections 
The cross sections of the investigated test sections in the 2003 and 2006 research cycle 
are shown in Figures 3.8 and 3.9. 
 
Figure 3.8 Cross section of structural sections in the 2003 research cycle (Timm and Priest, 
2006) 
80 
 
In Figure 3.8, eight structural sections (i.e., N1 through N8) in the 2003 research cycle 
were designed in pairs to compare HMA mixtures difference, but all sections had the same 
subgrade soil and 6-inch crushed aggregate base. The shared subgrade soil, commonly termed 
the ?Track soil? was excavated from the west curve of the Test Track, which can be classified as 
an AASHTO A-4(0) soil (Timm and Priest, 2006). Sections N1 through N6 were built in such a 
pairwise manner that a pair had similar HMA thicknesses; one contained modified asphalt binder 
while the other contained unmodified binder. Section N7 was topped with a 1 inch wearing layer 
made with stone matrix asphalt (SMA). Section N8 was topped with a 1 inch SMA as well, but 
had a 2-inch rich bottom PG 67-22 layer with an additional 0.5% binder (Timm and Priest, 2006).     
 
Figure 3.9 Cross section of structural sections in the 2006 research cycle (Timm, 2009) 
For the 2006 research cycle, the structural experiment was expanded to eleven sections 
including Florida, Oklahoma, Missouri, and additional Alabama sections. Among eleven 
structural sections, five of the original eight structural sections (i.e., N3 through N7) were left in 
place from the 2003 cycle. It is noted that section N5 was milled and inlaid 2 inches in an 
0.0
2.0
4.0
6.0
8.0
10 . 0
12 . 0
14 . 0
16 . 0
18 . 0
20 . 0
22 . 0
24 . 0
N1 N2 N3 N4 N5 N6 N7 N8 N9 N10 S11
As 
Bui
lt
 
T
hi
ckne
ss,
 
in.
PG 67-22 PG 76-22 PG 76-22 (SMA) PG 76-28 (SMA)
PG 76-28 PG 64-22 PG 64-22 (2% Air Voids ) PG 70-22
Lim eroc k  Bas e Granit e Bas e T y pe 5 Bas e T rac k  Soil Seale Subgrade
Flori d a
(ne w )
A lab a m a  &  FHW A
(l e f t in-pla ce )
Okla h o m a
(ne w )
FHW A
M issou ri
(ne w )
A lab a m a
(ne w )
81 
 
attempt to control extensive top-down cracking throughout the section. Three of the sections 
from the 2003 research cycle (i.e., N1, N2, and N8) were reconstructed with new pavement 
structure designs and three new sections were added in the structural study (i.e., N9, N10, and 
S11). As Figure 3.9 shows, among all sections only N8 and N9 were newly built with the Seale 
subgrade material, which was classified as an AASHTO A-7-6 soil (Taylor and Timm, 2009). 
The other sections continued to use the Test Track soil as subgrade material. 
Section N1 and N2 both used 10 inches of limerock, rather than crushed aggregate, as the 
base material. N1 had 3 lifts of HMA made with unmodified PG 67-22 binder, while N2 differed 
in the upper 2 lifts made with modified 76-22 binder. Five sections (i.e., N3 through N7) were 
left with the same structure in the 2003 research cycle. Section N8 had 10 inches of HMA, 
including a top lift of PG 76-28 SMA, a lift of PG 76-28 HMA, and a lift of PG 64-22 HMA, and 
base HMA made with PG 64-22 binder designed with 2% air voids. Section N9 was topped with 
SMA with PG 76-28 binder, followed by a lift of PG 76-28 HMA and two lifts of unmodified 
HMA, with a PG 64-22 HMA base lift designed to 2% air voids. Section N10 used 4 inches of 
Missouri Type 5 base, which is a dolomitic limestone base material. The upper 8-inch HMA 
layers were made by one lift of PG 64-22 HMA and one lift of PG 70-22 HMA. Section S11 also 
consisted 8 inches of HMA including the upper lift made with modified PG 76-22 binder over 
bottom lifts made with unmodified PG 67-22 binder. 
Thickness input 
The measured thickness for each layer was recorded during construction, and these 
values were input into the MEPDG. The measured thicknesses of structural sections for both the 
2003 and 2006 research cycles are shown in Table 3.3 and Table 3.4.  
82 
 
When entering lift thicknesses into the MEPDG, there are a few requirements for users. 
First of all, the minimum value for lift thickness is one inch. There were two sections (i.e., N1 
and N5 in 2003 research cycle) with lifts thinner than one inch. For this case, the thickness of the 
top lift was specified as one inch, and then the thickness increase of the top lift was deducted 
from the lift below; therefore, those lifts thinner than one inch were entered without changing 
overall HMA thickness. 
Another requirement is that a layer of pavement can only have 4 asphalt lifts at most. 
There were three test sections placed with an asphalt layer of more than 4 lifts (i.e., N3 and N4 in 
the 2003 research cycle, and N9 in the 2006 research cycle). For these sections, the fourth and 
fifth lifts were combined together and a thickness equal to the sum was assigned. In these test 
sections with five lifts, the combined two lifts shared the same mix design, so the accuracy of 
structure characterization did not decrease by making this adjustment.  
 
 
 
 
 
 
 
 
 
 
 
83 
 
Table 3.3 Surveyed layer thicknesses of the 2003 sections 
Test Section &Year Layer ID Layer Thickness (in) Input Thickness (in) 
N1 2003 HMA-1 0.6 1.0 
N1 2003 HMA-2 2.1 1.7 
N1 2003 HMA-3 2.2 2.2 
N1 2003 GB 6.0 6.0 
N2 2003 HMA-1 1.1 1.1 
N2 2003 HMA-2 2.0 2.0 
N2 2003 HMA-3 1.8 1.8 
N2 2003 GB 6.0 6.0 
N3 2003 HMA-1 1.2 1.2 
N3 2003 HMA-2 1.8 1.8 
N3 2003 HMA-3 2.7 2.7 
N3 2003 HMA-4 2.1 3.4 
N3 2003 HMA-5 1.3 - 
N3 2003 GB 6.0 6.0 
N4 2003 HMA-1 1.0 1.0 
N4 2003 HMA-2 1.7 1.7 
N4 2003 HMA-3 2.3 2.3 
N4 2003 HMA-4 1.8 3.8 
N4 2003 HMA-5 2.0 - 
N4 2003 GB 6.0 6.0 
N5 2003 HMA-1 0.9 1.0 
N5 2003 HMA-2 2.2 2.1 
N5 2003 HMA-3 2.0 1.8 
N5 2003 HMA-4 1.8 2.0 
N5 2003 GB 6.0 6.0 
N6 2003 HMA-1 1.1 1.1 
N6 2003 HMA-2 2.3 2.3 
N6 2003 HMA-3 2.2 1.8 
N6 2003 HMA-4 1.6 2.0 
N6 2003 GB 6.0 6.0 
N7 2003 HMA-1 1.0 1.0 
N7 2003 HMA-2 2.3 2.3 
N7 2003 HMA-3 2.1 1.8 
N7 2003 HMA-4 1.7 2.0 
N7 2003 GB 6.0 6.0 
N8 2003 HMA-1 1.1 1.1 
N8 2003 HMA-2 2.1 2.1 
N8 2003 HMA-3 1.9 1.8 
N8 2003 HMA-4 1.9 2.0 
N8 2003 GB 6.0 6.0 
 
 
 
 
84 
 
Table 3.3 Surveyed layer thicknesses of the 2006 sections 
Test Section &Year Layer ID Layer Thickness (in) Input Thickness (in) 
N1 2006 HMA-1 2.2 2.2 
N1 2006 HMA-2 1.9 1.9 
N1 2006 HMA-3 3.3 3.3 
N1 2006 GB 10.0 10.0 
N2 2006 HMA-1 2.0 2.0 
N2 2006 HMA-2 2.0 2.0 
N2 2006 HMA-3 3.1 3.1 
N2 2006 GB 10.0 10.0 
N8 2006 HMA-1 2.3 2.3 
N8 2006 HMA-2 2.9 3.0 
N8 2006 HMA-3 2.8 2.8 
N8 2006 HMA-4 1.9 2.0 
N8 2006 GB 6.4 6.4 
N9 2006 HMA-1 2.0 2.0 
N9 2006 HMA-2 3.5 3.5 
N9 2006 HMA-3 3.1 3.1 
N9 2006 HMA-4 2.6 5.8 
N9 2006 HMA-5 3.2 - 
N9 2006 GB 8.4 8.4 
N10 2006 HMA-1 1.0 1.0 
N10 2006 HMA-2 3.4 3.4 
N10 2006 HMA-3 2.2 2.2 
N10 2006 GB 5.0 5.0 
S11 2006 HMA-1 1.0 1.0 
S11 2006 HMA-2 2.1 2.1 
S11 2006 HMA-3 2.2 2.2 
S11 2006 HMA-4 2.3 2.3 
S11 2006 GB 6.1 6.1 
 
Material properties 
After specifying the layer thickness of the pavement, the material properties were input to 
the MEPDG. Because sufficient information was collected, the input Level 1 was used for most 
material properties. 
Asphalt mixtures 
At input Level 1, the MEPDG requires users to characterize asphalt material regarding 
dynamic modulus (E*), complex binder modulus, volumetric properties, and general properties. 
85 
 
E* test data 
The input for the dynamic modulus (E*) is a minimum of 3*3 matrix (temperature vs. 
frequencies) for each asphalt mixture or HMA lift. These data were obtained by the dynamic 
modulus test in the laboratory. For the 2003 research cycle, the dynamic modulus testing was 
performed in an unconfined state by Purdue University (Timm and Priest, 2006). The range of 
test temperatures starts at a minimum temperature between -12.2  (10 ) and -6.6   (20 ), 
and ends at a maximum temperature between 51.6  (125 ) and 57.2  (135 ). For the 2006 
cycle, the testing was performed at the NCAT laboratory under the guidance of AASHTO TP 62-
07 (Robbins, 2013). The range of temperatures tested were from a low temperature of 4.4  
(40 ) to a high temperature of 37.7  (100 ). The testing temperature range for the 2006 cycle 
was determined based on the climatic conditions for Alabama pavements. However, the 
temperature range for the 2006 cycle was not as wide as that for the 2003 cycle. To keep 
consistency of test temperature ranges, the raw data from the laboratory was used to extrapolate 
to the extreme temperatures of -9.4  (15 ) and 54.4  (130 ) using non-linear regression 
equations of the following form (Timm et al., 2009): 
ceT e m p e r a t u r Fr e q u e n c ybaE* ???                                              (3-1) 
where,  
E* = the mixture dynamic modulus, psi 
Temperature = test temperature in the lab,   
Frequency = test frequency in the lab, Hz 
a, b, c = regression constants 
The models for predicting E* at the extreme temperatures generally have high    values 
so it is reasonable to perform extrapolation on raw data. Table 3.5 and Table 3.6 show the 
86 
 
regression model constant and corresponding    value for each test lift mixture in the 2003 and 
2006 research cycle.  
Table 3.5 E* regression coefficients for the 2003 research cycle (Davis, 2009) 
Mixture (Section-Lift Number) a b c R2 
1 (N1-1, N4-1 & N5-1) 3,776,660 0.9705 0.1601 0.99 
2 (N1-2, N1-3, N4-2, N4-3, N4-4, N4-5, 
N5-2, N5-3, N5-4 & N5-5) 
4,370,389 0.9696 0.1619 0.99 
3 (N2-1, N3-1 & N6-1) 3,750,543 0.9708 0.1517 0.99 
4 (N2-2, N2-3, N3-2, N3-3, N3-4, N3-5, 
N6-2, N6-3, N6-4, N7-2, N7-3, N7-4, 
N8-2, & N8-3) 
4,967,657 0.9668 0.1447 0.85 
5 (N7-1 & N8-1) 4,546,113 0.9712 0.1464 0.98 
6 (N8-4) 4,636,938 0.9740 0.1313 0.97 
 
 
Table 3.6 E* regression coefficient for the 2006 research cycle (Davis, 2009) 
Mixture (Section-Lift Number) a b c R2 
1 (N1-1 & N1-2) 3,776,660 0.9705 0.1601 0.99 
2 (N1-3, N2-3, S11-3 & S11-4) 4,370,389 0.9696 0.1619 0.99 
3 (N2-1 & N2-2) 3,750,543 0.9708 0.1517 0.99 
4 (N8-1 & N9-1) 4,967,657 0.9668 0.1447 0.85 
5 (N8-2 & N9-2) 4,546,113 0.9712 0.1464 0.98 
6 (N8-3, N9-3 & N9-4) 4,636,938 0.9740 0.1313 0.97 
7 (N8-4 & N9-5) 3,342,206 0.9692 0.1734 0.99 
8A (N10-1) 4,555,840 0.9683 0.1484 0.99 
8B (N10-2) 5,992,918 0.9704 0.1546 0.97 
9 (N10-3) 4,152,065 0.9710 0.1542 0.98 
28A (S11-1) 3,112,030 0.9699 0.1646 0.98 
28B (S11-2) 4,702,712 0.9726 0.1635 0.98 
 
The goodness-of-fit is further illustrated by the graph comparing the measured and 
predicted E* data in the Figure 3.10. 
87 
 
 
Figure 3.10 Measured vs. predicted E* data (Davis, 2009) 
Another stipulation of the MEPDG is that E* values are not allowed to exceed 5,000,000 
psi. Only two lifts had E* values that were greater than this value, which means they were too 
stiff under test conditions. The two values were not considered in the data. Once the data were 
input into the software, master curves were automatically generated, and the reasonableness 
could be easily checked. Figure 3.11 shows a sample E* master curve generated for N1-1 lift in 
the 2003 research cycle. In some cases, certain values were contradictory to common sense. For 
example, E* values did not increase with a decrease in temperature at certain test frequencies. 
These data were not input because they cannot generate an accurate master curve. The E* value 
for each lift mixture in the 2003 and 2006 research cycle are shown in Appendix A. 
0
500000
1000000
1500000
2000000
2500000
3000000
0 500000 1000000 1500000 2000000 2500000 3000000
Measure d E*
Pr
edict
ed E*
Mi x  1
Mi x  2
Mi x  3
Mi x  4
Mi x  5
Mi x  6
Mi x  7
Mi x  8A
Mi x  8B
Mi x  9
Mi x  28A
Mi x  28B
88 
 
 
Figure 3.11 The MEPDG E* master curve for N1-1 lift in the 2003 research cycle (Davis, 
2009) 
G* test data 
The binder characterization at input Level 1 requires shear modulus (G*) and phase angle 
(?), which are measured in the dynamic shear rheometer (DSR) test. This test can be performed 
in three phases of binder aging: no aging, short-term aging, and long-term aging. The MEPDG 
requires the characterization of short-term aged binder (i.e., through the Rolling Thin Film Oven 
test). 
Although G* and ? were provided as inputs, they had a minimal effect on the results of 
the analysis. This assumption was tested by running a baseline case for one of the test sections, 
and then changing only the G* or the phase angle values (i.e., doubling values). There was very 
little difference found between the predicted performances (in terms of rutting, bottom-up fatigue 
cracking, and IRI) of the two cases. The reason for this result was that the MEPDG does not use 
G* and ? to construct the master curves to characterize the mixture performance if the E* test 
data are provided. In addition, although the G* and ? were used in aging calculations, the binder 
only functioned in a short term (two years) so binder aging had a minimal effect on the distress 
predictions. Therefore, the G* and ? values under representative testing conditions were used in 
the MEPDG, and these values are shown in Appendix B.      
10000
100000
1000000
10000000
-5 -4 -3 -2 -1 0 1 2 3 4 5
Log Re duced Ti me (Sec)
E* 
(psi)
14 ?F
40 ?F
70 ?F
10 0?F
13 0?F
Master Cu rve HMA La y er 1
89 
 
Mixture volumetric determination 
The MEPDG requires some general information for HMA lifts to capture their volumetric 
compositions, and mechanical and thermal properties. Some of these inputs were entered as the 
default values because they are typical, including the master curve reference temperature (70 ), 
Poisson ratio (0.35), thermal conductivity (0.67 BTU/hr?ft? ), and heat capacity (0.23 
BTU/lb? ). However, the as-built volumetric properties were provided by construction and lab 
reports for each mix type, regarding in-place air voids, mixture unit weight, and effective binder 
content. A sample of construction records is presented in Figure 3.12 for lift N1-1 of the 2006 
research cycle. 
 
Figure 3.12 Construction and lab record for N1-1 lift in 2006 research cycle 
Based on this record, the in situ air voids was obtained by subtracting the average section 
compaction from 100%. For example, the air voids for N1-1 lift in the 2006 research cycle was 
90 
 
5.4% by subtracting the section compaction rate 94.6% from 100%. The unit weight was 
calculated using the QC Gmm value and the average section compaction. First, the Gmm of 2.499 
was multiplied by 94.6% to get a bulk specific gravity of the compacted mixture (Gmb) of 2.364; 
then, this value was multiplied by 62.4 lb/ft3 to yield a mixture unit weight of 147.5 lb/ft3. The 
compacted mixture (Gmb), percent aggregate in the mixture (Ps), and the bulk specific gravity of 
the aggregate blend (Gsb) were used to find the voids in mineral aggregate (VMA) of the mixture 
as equation (3-2). 
?????? ??? sb smbG PGV M A 100                                                      (3-2) 
The Gsb was calculated by the mass percent in total (fA) and bulk specific gravity (GA) of 
each mixture component. The equation used is shown as below: 
B
B
A
A
sb G
fGfG ??1                                                          (3-3) 
Then, the volume of effective binder was calculated by subtracting air voids from VMA. 
For example, the Ps and Gsb were 95.1% and 2.669, respectively. Using those values with the 
Gmb of 2.364, the VMA was calculated as 15.78% by equation (3-2), with effective volumetric 
binder content as 10.38%. The same computation was performed on other lifts of test sections 
and the results of volumetric inputs are shown in Appendix C. As mentioned earlier in the 
section ?Thickness Input?, for those test sections consisted of five HMA lifts, a weighted average 
was taken of the volumetric properties of the bottom two lifts for input as the ?fourth? layer. 
Unbound materials 
The unbound layers, specifically the base and subgrade layers were characterized at the 
input Level 3 in the MEPDG. The reason that the input Level 1 or 2 was not used was due to a 
software running error in the analysis in the version 1.1 of MEPDG. The level 3 inputs for 
91 
 
unbound layers are material type, gradation, Poisson ratio, coefficient of lateral earth pressure, 
and a representative resilient modulus (MR).  
2003 test sections 
For the 2003 research cycle structural sections, only two types of unbound material were 
used: the crushed aggregate base and the Test Track soil subgrade. In the MEPDG, the ?crushed 
stone? was selected as the base material, and a typical value of 0.4 was adopted for the Poisson 
ratio of this material. The coefficient of lateral earth pressure was specified as 0.3572 for each 
material to calculate horizontal stresses (Taylor, 2008). For the Test Track soil, an A-4 material 
was used with a Poisson ratio of 0.45. The gradations for the crushed aggregate base and the 
Track soil are shown in Table 3.7.   
Table 3.7 Unbound materials gradations for the 2003 research cycle (Timm and Priest, 
2006) 
 Percent passing (%) 
Sieve size (mm) Crushed aggregate 
base 
Test Track soil 
subgrade 
37.5 100 100 
25 96 83 
19 90 81 
12.5 80 78 
9.5 76 75 
4.75 59 71 
2.36 49 68 
1.18 40 66 
0.6 32 64 
0.3 23 61 
0.15 15 56 
0.075 10 48 
For input Level 3, the representative MR values were needed for different kinds of 
unbound materials. The MR values were obtained from the backcalculation of falling weight 
deflectometer (FWD) deflection data. The backcalculation was performed by the software 
EVERCALC 5.0 developed by Washington State DOT. It was found that the software accuracy 
92 
 
depends on the number of pavement layers being analyzed, and the optimal number of pavement 
layers for backcalculation was three. For the 2003 sections, the three-layer structure consisted of 
HMA as the top layer, the granular base, and ?new fill? material (Test Track soil) combined as 
the second layer, and the remainder of the subgrade from the 2000 research cycle being the third 
layer (Timm and Priest, 2006). The new fill material was used to fill the gap depth between the 
granular base and the existing subgrade from the 2000 research cycle. Therefore, a level road 
surface for each test section can be ensured once the granular base and HMA lifts had been 
placed. The new fill material was excavated from the same embankment as the subgrade material 
which was already in place for those sections. So the same Poisson ratio, gradation, and material 
classification were used for the new fill. 
 
Figure 3.13 Cross sections comparison 
However, in order to provide a three-layer structure for backcalculation, there was only 
one MR value for the granular base and new fill material collectively, and a different MR value 
for the remainder of the subgrade material. In the MEPDG, since good predictions are based on 
good characterization of pavement structure, it is better to specify a four-layer structure as built 
in reality than three-layer structure as considered in backcalculation (Figure 3.13). To rectify the 
93 
 
issue, three layers, instead of two layers, were assumed to be underneath the last HMA lift in the 
MEPDG: one 6-inch granular base layer, one layer of new fill material using the thickness 
designated in Table 3.8, and finally the subgrade with an assumed infinite thickness. The 
backcalculated modulus found for the combined base and new fill was taken as representative 
modulus for each of those two layers, while backcalculated subgrade modulus was the input for 
the last Test Track soil subgrade layer. The backcalculated moduli values are shown in Table 3.9. 
Although it may be unusual that the subgrade moduli are significantly higher than the base 
moduli, this trend is consistent with other findings regarding the Test Track materials (Timm and 
Priest, 2006; Taylor and Timm, 2009). 
Table 3.8 New fill depths for the 2003 research cycle (Taylor and Timm, 2009) 
Test section depth of  
new fill (in) 
N1 19 
N2 19 
N3 15 
N4 15 
N5 17 
N6 17 
N7 17 
N8 17 
 
Table 3.9 Unbound materials moduli for the 2003 research cycle (Timm and Priest, 2006) 
Test section Granular base & 
new fill MR (psi) 
Subgrade MR 
(psi) 
N1 8,219 26,958 
N2 10,061 27,482 
N3 13,083 32,516 
N4 11,717 33,889 
N5 7,852 30,564 
N6 11,153 34,682 
N7 12,243 33,817 
N8 10,258 30,859 
 
 
 
94 
 
2006 test sections 
For the 2006 research cycle structural sections, five different unbound materials were 
used as base or subgrade layers: the Test Track soil discussed previously, the crushed aggregate 
base used in the 2003 structural sections, a poor quality subgrade material for sections N8 and 
N9, a Florida limerock base material for sections N1 and N2, and a dolomitic limestone base 
material (termed the Type 5 base) for section N10. For the Test Track soil and the granular base 
material used in the previous test cycle, the same material selections and gradations were used as 
previously discussed. The poor quality subgrade soil used for sections N8 and N9, termed the 
Seale subgrade, was classified as an AASHTO A-7-6 soil within the MEPDG. The Florida 
limestone and Type 5 base materials were both classified as crushed stone within the MEPDG. 
The Poisson ratio was commonly assumed to be 0.40 for all three materials. The gradations used 
in the MEPDG for the three new materials are shown in Table 3.10. 
Table 3.10 Unbound materials gradations 2006 research cycle (Taylor and Timm, 2009) 
 Percent Passing (%) 
Sieve Size (mm) Seale Subgrade Florida Limestone Type 5 Base 
37.5 100 100 100 
25 100 100 99 
19 100 100 97 
12.5 100 88 92 
9.5 100 81 88 
4.75 100 61 79 
2.36 100 44 71 
1.18 99 32 64 
0.6 98 26 58 
0.3 92 23 49 
0.15 82 21 36 
0.075 58 19 25 
 
A different procedure was explained to calculate the representative MR values for the 
base and subgrade materials used in the 2006 test cycle. The in-situ stress state of each material 
was monitored by dynamic pavement response measurements, namely vertical pressure, made 
95 
 
throughout the 2006 test cycle. Figure 3.13 shows an example of these measurements made over 
one year for section N1.  
 
Figure 3.14 Vertical pressures in N1 subgrade in the 2006 research cycle(Davis, 2009) 
As seen in Figure 3.14, the pressures for each axle type increase during warmer months 
when the HMA is softer, and the pressures correspondingly decrease during cooler months. 
Using the relative frequency of each axle type (14.3% steer axles, 14.3% tandem axles, and 71.4% 
single axles), a weighted average vertical stress due to traffic loading was used to calculate the 
representative MR of each unbound material. It was combined with the overburden pressure to 
obtain the total average vertical stress (?1). The overburden pressure existed due to the weight of 
overlying material, which can be computed from the overlying layer thicknesses and unit weights. 
Finally, the corresponding horizontal stresses (?3) were calculated by multiplying the vertical 
stress by the lateral earth pressure coefficient 0.3572 (Taylor, 2008). The vertical and horizontal 
stresses were then used in Equation (3-4) to estimate a representative MR for each unbound 
0
5
10
15
20
25
30
10/10/2006 11/29/2006
1/18/2007
3/9/2007
4/28/2007 6/17/2007
8/6/2007
9/25/2007
11/14/2007
Date
Ver
tical Subg
rade P
ress
ure (p
si)
St eer Ax le
Single Ax le
T andem  Ax le
W eighted Av erage
96 
 
material (Taylor and Timm, 2009). The MR input for MEPDG was finally calculated and shown 
in Table 3.12. 
3k
a
d2
k
aa1R p
?*p?*pkM ??
?
???
?
???
?
???
?
??                                          (3-4) 
where, 
            MR = the resilient modulus, psi 
            k1, k2, and k3 = regression coefficients (Table 3.11) 
            pa = the atmospheric pressure (14.6 psi) 
            ??= the bulk stress (?1+2*?3), psi 
            ?d = the deviator stress (?1 ? ?3), psi 
Table 3.11 Regression coefficients for unbound materials (Taylor and Timm, 2009) 
Material Type k1 k2 k3 
Track Soil 1095.43 0.5930 -0.4727 
Crushed Granite 581.08 0.8529 -0.1870 
Florida Limestone 717.04 1.2338 -0.5645 
Type 5 643.69 1.0318 -0.2833 
Seale Subgrade 225.09 0.3598 -0.7751 
 
Table 3.12 Unbound materials moduli for the 2006 research cycle (Davis, 2009) 
Test Section Granular Base  
MR (psi) 
Subgrade MR  
(psi) 
N1 23,597 29,701 
N2 25,605 29,638 
N8 26,020 9,801 
N9 24,155 14,804 
N10 16,752 28,174 
S11 12,530 28,873 
PERFORMANCE MONITORING 
Introduction 
Performance data were observed and recorded weekly in regard to fatigue cracking, 
rutting, and IRI over the research cycle. It was usually scheduled on every Monday when trucks 
97 
 
were paused to facilitate performance testing and any needed track or truck maintenance (Priest, 
2005).  
Fatigue Cracking 
Inspection method 
Fatigue cracking was monitored using a crack mapping method. First, the cracks were 
identified by its alligator-like appearance. Only interconnected areas of cracks were considered 
as alligator cracking. Either longitudinal or transverse cracking was not accounted as alligator 
cracking. The identified cracking were carefully inspected and marked. Second, the marked 
pavement was photographed using a digital camera. The picture was transferred to a digital map 
(Figure 3.15). Then, the percentage of cracking area divided by the total area was used to 
measure the severity level of fatigue cracking. It is important to note that fatigue cracking was 
only measured over the middle 150 feet length area of each test section; the first and last 25 feet 
were considered as a ?transition zones? between two sections, and the distresses may be caused 
by the transition of materials. In addition, the fatigue cracking only refers to ?bottom-up? 
cracking that initiates from the bottom of asphalt layer in the study. It was verified by cutting 
cores during trafficking phase (i.e., the first two years of each research cycle) or cutting trenches 
after the trafficking phase. 
98 
 
 
Figure 3.15 Sample crack map from section N6 at the end of the 2006 research cycle 
Rutting and IRI 
Inspection method 
For the 2003 research cycle, the rut depth measurement was made periodically with a 
dipstick profiler (Figure 3.16), and the roughness measurements was performed occasionally by 
an Automatic Road Analyzer (ARAN) Inertial Profiler (Figure 3.17) owned by NCAT. This 
profiler can measure the small wavelengths in the longitudinal profile in the pavement surface at 
high speeds for each wheel path, which can be converted into IRI or rut depth measurements 
through models. For the 2006 research cycle, NCAT used the ARAN van to measure rut depths 
and IRI three times per month. 
0
2
4
6
8
10
12
25 45 65 85 105 125 145 165
L o n g i t u d i n a l  D i s t a n c e  f r o m  F a r  T r a n s v e r s e  J o i n t  ( f t )
T
r
a
n
s
v
e
r
s
e
 O
ff
s
e
t 
(
ft
)
99 
 
 
Figure 3.16 Dipstick profiler 
 
Figure 3.17 ARAN inertial profiler 
SUMMARY 
This chapter covers the generation of MEPDG inputs regarding traffic, climate, and 
material properties. The dataset was prepared at the highest input level of MEPDG when 
possible, mostly at input Level 1, which provided the highest confidence for prediction accuracy. 
Also, because of the detailed dataset, the local calibration can be performed to the most specific 
local condition. The next chapter will discuss the methodology of local calibration and validation 
in this study.  
  
100 
 
CHAPTER 4  
THE METHODOLOGY OF LOCAL CALIBRATION AND VALIDATION 
INTRODUCTION 
Calibration 
Calibration is defined as a process to improve the goodness of fit or accuracy of 
prediction models in the MEPDG. The goodness of fit indicates how well the model predictions 
correspond with field-measured values. It can be graphically assessed by evaluating similarity of 
prediction and measurement trend lines.  
Figure 4.1 and Figure 4.2 provide an example to illustrate the concept of goodness of fit. 
The fatigue cracking at a pavement is investigated over 20 months in this example. The solid line 
is the trend line for predicted fatigue cracking, and the dashed line is for measured fatigue 
cracking. Figure 4.1 shows a correspondence between the predicted and measured values, which 
suggests predictions are reasonable. However, Figure 4.2 presents a disparity between the 
predicted and measured trend lines, which indicates the predictions are not reliable. In this case, 
the model is overpredicting the actual cracking progression. 
101 
 
 
Figure 4.1 Reasonable goodness of fit for model predictions 
 
Figure 4.2 Unreasonable goodness of fit for model predictions 
 
0
10
20
30
40
50
60
70
80
90
100
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Fati
gu
e c
rac
kin
g val
ue
s, %
 of l
an
e ar
ea
 
Pavement age, years 
Measured Values Predicted Values
0
10
20
30
40
50
60
70
80
90
100
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Fati
gu
e c
rac
kin
g val
ue
s, %
 of l
an
e ar
ea
 
Pavement age, years 
Measured Values Predicted Values
102 
 
Validation 
Validation is usually performed after the calibration process, and it aims to verify the 
goodness of fit and precision of the calibrated models. The data used for validation must be 
independent from those used for the calibration so as to maintain the objectivity of validation. 
The results of validation determine whether the calibrated models can be applied to broader 
scenarios. 
LOCAL CALIBRATION 
Goal of Calibration 
The local calibration was designed to gain an improvement in the MEPDG performance 
predictions by adjusting the calibration coefficients built in prediction models (i.e., transfer 
functions) to reflect actual, locally-measured, field performance. The calibrated models with 
best-fit calibration coefficients could fairly predict the performance of a local pavement, which 
would contribute to an efficient pavement design. The final result of calibration in this study is 
the best-fit set of calibration coefficients for the fatigue cracking, rutting, and IRI models.        
Calibration Methods 
It was found in the literature review that two approaches are commonly used in the local 
calibration. One was to calibrate prediction models by varying its calibration coefficients (Li et 
al., 2009; Muthadi and Kim, 2008; Schram and Abdelrahman, 2010; Banerjee et al., 2009). The 
other was to adjust the results of the model predictions directly (Hoegh et al., 2010). For 
example, the calibration was performed by subtracting a certain constant value from the 
prediction results of the nationally-calibrated models. 
The first approach (Figure 4.3) included several steps: build project templates with 
section inputs, assign trial calibration coefficients, run the MEPDG, evaluate predictions, and 
103 
 
repeat the process if necessary. In order to minimize the number of trials, a sensitivity analysis 
was performed to obtain a better sense of how the calibration coefficients affect the predictions.  
 
Figure 4.3 Trial and error calibration steps 
The second approach was to perform arithmetic adjustment on the predicted values. This 
approach was useful for the case when a consistent bias was found between predictions and 
measured values throughout design life (Figure 4.4). It minimized the bias by subtracting a 
constant value for predicted values. This constant value was determined by ?Solver? in Microsoft 
Excel which can help you find an optimal (maximum or minimum) value for a formula. 
However, this adjustment shifted the prediction curve without changing its shape, and this 
approach would not be effective if the prediction curve from the model is not in a similar shape 
with the curve of measured values (Figure 4.5).  
104 
 
 
Figure 4.4 Curves in similar shape 
 
 
Figure 4.5 Curves not in a similar shape 
 
0
2
4
6
8
10
12
14
16
18
1-Jan 1-Feb 1-Mar 1-Apr 1-May 1-Jun 1-Jul 1-Aug 1-Sep 1-Oct 1-Nov 1-Dec
Rut
 De
pth,
 m
m 
Date 
measured rutting predicted rutting predicted rutting after adjustment
5 mm (0.2 in) 
0
2
4
6
8
10
12
14
16
18
1-Jan 1-Feb 1-Mar 1-Apr 1-May 1-Jun 1-Jul 1-Aug 1-Sep 1-Oct 1-Nov 1-Dec
Rut
 De
pth,
 m
m 
Date 
measured rutting predicted rutting predicted rutting after adjustment
5 mm (0.2 in) 
105 
 
Methodology of Study 
The local calibration adopted in this study was a trial-and-error approach. Before 
applying ?trial? calibration coefficients, the project templates in the MEPDG (Figure 4.6) were 
set up describing the local condition. The design scenarios (i.e., traffic, climate, and structure) in 
research cycles were carefully entered into the MEPDG, and they were saved as project 
templates for all of the investigated sections. Since the investigated sections sustained the same 
traffic loadings and the same local climate effects, the major difference between project 
templates was only structural configuration. In total, there were nineteen project templates 
generated: eight sections in the 2003 research cycle for calibration and eleven sections in the 
2006 research cycle for validation. In fact, the project templates in the same research cycle only 
differed in those inputs describing structure. As introduced in Chapter 3, the traffic inputs and 
HMA layer properties inputs were entered at input Level 1, and some unbound material 
properties were entered at input Level 3 or assumed as default values.  
106 
 
 
Figure 4.6 Project templates in the MEPDG 
By running the project templates, the MEPDG provided the predictions of various 
distresses for each month during the design life. Then, these predicted values were compared 
against the measured values during a research cycle. For rutting and IRI, the data recorded on a 
weekly basis was averaged to monthly readings. For fatigue cracking, the surfaces of sections 
were inspected weekly, yet the measured cracking data were recorded every several months, 
which provided cracking development versus time plots. Thus, only a few months? predicted 
fatigue cracking values by the MEPDG were able to be compared against the available measured 
values. For a specific performance indicator (i.e., fatigue cracking, or rutting, or IRI), the 
difference between predicted and measured values was used to evaluate the accuracy of 
predictions, and whether the calibration coefficients were acceptable. If no other calibration 
coefficients could be found that could reduce the sum of squared errors (SSE) more than a 
107 
 
certain trial set of calibration coefficients, this trial calibration coefficients was assumed to be the 
best-fit set of calibration coefficients, and the iteration process was ended. 
In statistics, SSE is often used as a quantitative indictor to evaluate the deviation of the 
fitted values from the actual values. The SSE reflects the accuracy of the model, so it was 
adopted in this study to quantify the difference between predicted and measured distress values. 
The MEPDG predicts the development of distresses (i.e., fatigue cracking, rutting, and IRI) on a 
monthly basis, so the residual (i.e., the difference between the predicted and measured distress) 
was calculated for each month, and then they were squared and summed up to evaluate MEPDG 
predictions. Considering all investigated sections, the SSE (Equation 4-1) is regarded as the final 
indicator of the MEPDG prediction accuracy.  
? ?? ?? 8 1n 1 2 v a l u e )m e a s u r e d- v a l u e( p r e d i c t e d AmSSE                                   (4-1) 
where, 
m = the mth measurement (A is the total number of monthly measurements over the 
investigated cycle) 
n = the nth investigated section (because eight sections involved in the calibration, the 
maximum of n was eight) 
VALIDATION 
Goal of Validation 
The validation was to verify the reasonableness and robustness of the calibrated 
prediction models. If the calibrated prediction model was validated to be sufficiently accurate, 
the calibrated model can be considered by local agencies for future adoption. 
  
108 
 
Methods of Validation (Criteria)  
The prediction results were obtained from the MEPDG using a set of data independent 
from the calibration. The method of validation was to use the statistical hypothesis test (i.e., 
Student?s t-test) to examine whether the calibration results were acceptable. The null hypothesis 
was that the predicted values and measured values were not significantly different. The ? level 
was assumed to be 0.05 so the p-value was compared with 0.05 to evaluated whether the null 
hypothesis was rejected.  If the p-value was less than 0.05, the null hypothesis was rejected and 
the calibration results were considered to not be satisfactory. The results of the Student?s t-test 
can provide a statistical inference whether the calibration can make a difference on the prediction 
models.  
Methodology of Study 
As discussed earlier, eleven sections in the 2006 research cycle were used for validation, 
and the project template was generated for each of them. The predicted values of distresses were 
obtained from the MEPDG using the best-fit coefficients, and the measured values were obtained 
from the Test Track in-field performance. The Student?s t-test was performed using Excel on two 
sets of data (i.e predicted and measured) to see if the null hypothesis was rejected.  
  
109 
 
CHAPTER 5  
A METHOD OF RUNNING THE MEPDG AUTOMATICALLY 
BACKGROUND 
The local calibration of the MEPDG requires iteration to achieve the best-fit calibration 
coefficients. The performance predictions were achieved through sophisticated models in the 
MEPDG, so it took considerable time to perform the necessary calculations. For example, a 64-
bit system with an Intel Core i7 CPU and 16 GB RAM would take nearly one minute to run a 
two-year lifetime pavement design. In this study, eight structural sections (or project templates) 
in the 2003 research cycle were considered in the local calibration. The software operation and 
results analysis were very time consuming. Therefore, the investment of time and human labor 
was an important concern in this study. 
In this study, one trial set of calibration coefficients took around 18 minutes (calculated 
as 1 min for each project template multiplied by 8 sections plus 10 minutes used for calculating 
SSE) to determine whether they could improve the accuracy of a specific distress model. The 
calculation of SSE usually took at least ten minutes to perform manually for opening files, 
reading prediction results, and calculating the SSE values. In fact, it also needed considerable 
time to enter the calibration coefficients in the MEPDG. During this study, 66 sets of calibration 
coefficients were tried for the fatigue cracking model; 106 sets of calibration coefficients were 
tried for the rutting model. Therefore, the demand rose to run the MEPDG automatically with 
little human intervention as possible. 
The calibration process in this study included a series of actions: creating project 
templates, selecting ?trial? calibration coefficients, running project analysis, evaluating 
predictions (i.e., calculating SSE), and repeating this process as necessary. The human labor 
involved in these steps was finished by mouse-clicking and keyboard-typing. To facilitate 
110 
 
automation of MEPDG operation, these actions could be recorded by using Visual Basic (VB) 
script language and replayed by VB programs. The following discussion will focus on the 
method of how to run the MEPDG automatically.  
METHOD DESCRIPTION 
Software Introduction 
The software ?Quick Macro v6.60? was adopted, which could record, replay, and edit 
macros that specify a series of repetitive actions (i.e., mouse-clicking and keyboard-typing). By 
using this software, human labor was replaced by an execution of programmed macros.  
Basic Steps 
There were only two steps run by macro: project analysis and results evaluation. 
Generating the project templates was not included in the macros. The reason was that too much 
information needed to be input into the MEPDG non-repetitively, so the macro programming 
cost more effort than doing so manually. Figure 5.1 shows the programmed macros including 
assigning ?trial? calibration coefficients, running project analysis, and evaluating predictions. 
111 
 
 
Figure 5.1 VB script flow chart 
First, the trial calibration coefficients were composed into an Excel file. The Excel file 
contained the test section information, the project template file name, and the location path. 
Second, the ?trial? calibration coefficients were transferred to the calibration coefficients input 
textboxes in the MEPDG. Next, eight project templates were compiled into the batch file and 
ready for a ?batch? run. Third, the MEPDG predictions were extracted from the output file and 
employed for the SSE computation. The result of SSE for one specific set of calibration 
coefficients was shown in a message box in Quick Macro (Figure 5.2).  
112 
 
 
Figure 5.2 SSE result in a message box 
Technical Issues 
Several technical issues affect the running of macros. These issues were categorized into 
project analysis delays, foreground/background shift, and excel overrunning error. 
Software running delay 
Although a two-year project analysis takes nearly one minute to run, the VB script 
commands were not supposed to initiate the next action (e.g. extracting distress data from the 
MEPDG output files) before the project analysis was completed. To account for the running 
delay, the waiting time was set between two actions to coordinate the running pace and it was 
determined to be one minute in this study. In fact, the longer the design life of the pavement, the 
longer the waiting time needs to be. The appropriate length of the waiting time is case-specific, 
and it usually depends on the design scenario of the project and the running speed of the 
computer.  
 
113 
 
Foreground/background shift 
In the MEPDG, when a new set of calibration coefficients were run, a dialog box will pop 
up to ask whether you want to reject new calibration coefficients (Figure 5.3). When the dialog 
box popped up, the focus of the system (i.e., the activated window or menu or textbox) was on 
this dialog box, in other words, this dialog box was activated as a foreground application. The 
macro needed to select ?No? to move on. However, through considerable times of macro running, 
the focus of system sometimes shifted from this dialog box to some other applications in the 
computer system. This could cause erratic running, even wrongly ending the macro. The only 
remedy was to reinstall the system which brought much inconvenience. An improvement on the 
macro scripts may be helpful to solve this problem.  
 
Figure 5.3 Dialog box pop-up 
Excel overrunning error 
When the macro program finished running, there were numerous MEPDG output files 
opened in Excel. Before initiating another ?trial? set of calibration coefficients, it was necessary 
to shut down those opened Excel files manually. Otherwise, the MEPDG would have a problem 
to open Excel after project analysis. If many Excel files are shut down by clicking the ?close? 
button in the upper right corner, an error warning (Figure 5.4) popped up when you ran Excel the 
next time. It should be noted that it was better to remove the opened files from the list of disabled 
items as the warning below said, or else, the macro usually had a running problem.   
114 
 
 
Figure 5.4 Excel error warning 
SUGGESTION 
It saves labor and time to employ macros; however, there were still issues that needed to 
be solved. The biggest issue was to ensure smooth running of macros, which determined the 
efficiency of this method. It would be helpful to develop a similar but more robust program in 
the future.  
 
 
 
 
 
 
 
 
 
 
 
 
115 
 
CHAPTER 6  
RESULTS AND DISCUSSIONS 
INTRODUCTION 
The MEPDG uses a mechanistic model to predict the pavement response based on design 
inputs, and applies transfer functions to correlate the pavement response to various distresses 
over the design life. In the study, three distresses: fatigue cracking, rutting, and IRI were 
evaluated since sufficient data were available for them from the Test Track. The study was to 
first examine the MEPDG predictions provided by nationally-calibrated models. Then, the 
MEPDG predictions were compared with actual Test Track measurement to evaluate whether 
they reflected the field performance. If the accuracy of predictions was not satisfactory, the local 
calibration of MEPDG would be applied using the Test Track data from the 2003 research cycle. 
The calibration was performed by adjusting the calibration coefficients to reduce the bias or 
standard error between measured and predicted distress data.  
Statistics were computed to measure the prediction performance. The standard error 
shows how much variation or dispersion the data points have from their average. The SSE 
evaluates how accurate the predictions replicate the measured data. Minimizing the SSE results 
in maximum model fit. 
RESULTS OF CALIBRATION 
Fatigue Cracking 
Figure 6.1 shows that the nationally-calibrated model predictions appear reasonably 
consistent with measured values for fatigue cracking.  
116 
 
 
Figure 6.1 Model predictions vs. measured values for fatigue cracking 
The statistics of the nationally-calibrated model predictions are shown in Table 6.1. 
Results for section N8 were not included is because though cracking was observed in this section 
extensive forensic investigation had determined the cracking resulted from the layer slippage not 
predicted by the MEPDG (Willis and Timm, 2007). 
Table 6.1 Statistical analysis for the 2003 fatigue cracking predictions 
 nationally-calibrated model 
standard error 2.84% 
SSE 452.28 
number of data points 28 
p-value 0.38 
*The p-value is derived from the Student?s t-test on measured fatigue cracking and nationally-calibrated model 
predictions. 
The local calibration tried 66 sets of calibration coefficients in which each coefficient 
was adjusted from the MEPDG default. The adjustment was performed on three coefficients 
individually and simultaneously with a range of +/- 0.2 from defaults in which the SSE did not 
sharply increase. The optimization process is presented in Figure 6.2. The red line sets a SSE 
level of 452.28, which is calculated based on nationally-calibrated model predictions. It can be 
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
Pre
dic
ted
 Fati
gu
e C
rac
kin
g, %
 of
 to
tal
 lan
e ar
ea
 
Measured Fatigue Cracking, % of total lane area 
N1 - 03
N2 - 03
N3 - 03
N4 - 03
N5 - 03
N6 - 03
N7 - 03
117 
 
seen that there was no appropriate calibration coefficients found that could significantly improve 
the model prediction accuracy. 
 
Figure 6.2 Fatigue cracking model prediction optimization 
Rutting 
Figure 6.3 shows that the nationally-calibrated model over-predicted the rut depths for all 
the 2003 sections.  
(a) National calibration                                  (b) Local calibration 
Figure 6.3 Model predictions vs. measured values for rutting 
1
10
100
1000
10000
100000
1000000
1 6 11 16 21 26 31 36 41 46 51 56 61 66
SSE
 
Trial Number 
452.28 
118 
 
The local calibration using the Test Track data improved the accuracy of the rutting 
model. The locally-calibrated rutting model predicted much less rutting (Figure 6.3 (b)). Using 
the locally-calibrated model may allow designers to select a more economical pavement design 
(i.e., thinner pavement and/or lower cost materials). 
From Table 6.2, the standard error was reduced by 55%, and the SSE decreased by a 
factor of 58 after the local calibration. The Student?s t-test showed that the MEPDG predictions 
were no longer statistically different from the measured rut depths after performing the local 
calibration. 
Table 6.2 Statistical analysis for the 2003 rutting predictions 
 nationally-
calibrated 
model 
locally-
calibrated 
model 
standard error 0.20 mm 0.09 mm 
SSE 6644.36 114.05 
number of data points 157 157 
p-value 8.79E-78 0.37 
*The p-value is derived from the Student?s t-test on measured rut depths and nationally-calibrated/locally-calibrated 
model predictions. 
The local calibration tried 106 sets of calibration coefficients in which each coefficient 
was varied from the MEPDG default. The adjustment was performed on five coefficients 
individually and simultaneously. The trial coefficient value was within a range (i.e., between 
0.05 and 1.2) in which the SSE did not sharply increase. The optimization process is presented in 
Figure 6.4. The red line sets a SSE level of 6644.36, which is calculated based on nationally-
calibrated model predictions. 
119 
 
 
Figure 6.4 Rutting model prediction optimization 
It was found that the SSE decreased to be 114.05 when the trial calibration coefficients 
were   =   =   = 1 in Equation (2-13),   = 0.05 in Equation (2-14) for granular base, and   = 
0.05 in Equation (2-14) for subgrade.  
32 *0 . 4 7 9 1*1 . 5 6 0 63 . 3 5 4 1 2-1
r
p N*T*10* = ?????                                           (2-13) 
iv)N
?(-
r
p11i h*?*e*)??(*k*? =? ?                                                (2-14) 
IRI 
In Figure 6.5, it can be seen that the nationally-calibrated model predictions reasonably 
matched the measured values. Using the locally-calibrated rutting model also improved the IRI 
prediction. However, it still can be seen that the data points were mostly below the line of 
equality after adjusting the calibration coefficients. No additional adjustments could improve the 
accuracy of IRI predictions. N8 was not included because it was eliminated from the fatigue 
cracking analysis which strongly affected its IRI predictions. 
1
10
100
1000
10000
100000
1000000
10000000
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
10
1
10
6
SSE
 
Trial Number 
6644.36 
114.05 
120 
 
(a) National calibration                                                 (b) Local calibration 
Figure 6.5 Model predictions vs. measured values for IRI 
From Table 6.3, the standard error decreased by 17%, and the SSE decreased by 51% 
after adjusting the rutting model while using the same calibration coefficients for the IRI 
prediction. In addition, the Student?s t-test showed that the MEPDG predictions were no longer 
statistically different from the measured IRI after adjusting the rutting model. 
Table 6.3 Statistical analysis for the 2003 IRI predictions 
 IRI predictions using 
nationally-calibrated 
model 
IRI predictions after 
adjusting the rutting 
model 
standard error 0.99 in./mile 0.82 in./mile 
SSE 28706.39 14090.60 
number of data points 143 143 
p-value 6.65E-11 0.10 
*The p-value is derived from the Student?s t-test on measured IRI and nationally-calibrated/locally-calibrated model 
predictions. 
RESULTS OF VALIDATION 
Fatigue Cracking 
Since the nationally-calibrated fatigue model was found to reasonably predict 
performance of the 2003 Test Track sections, the model was then validated using the 2006 
sections. In Figure 6.6, it can be seen that these data points scattered over both sides of the 
121 
 
equality line. The nationally-calibrated model poorly predicted the fatigue cracking for the 2006 
sections.  
 
Figure 6.6 Model predictions vs. measured values for fatigue cracking 
N1 and N2 in the 2006 research cycle were not included in the validation analysis 
because the cracking in those sections were identified as top-down cracking. Sections N3 
through N7 were left in place from the 2003 research cycle and sustained additional two-year 
traffic; The MEPDG does not allow users to input two sets of traffic characteristics (i.e., one for 
the 2003 research and the other for the 2006 research cycle). The statistics of the nationally-
calibrated model predictions are shown in Table 6.4.  
Table 6.4 Statistical analysis for the 2006 fatigue cracking predictions 
 nationally-calibrated model 
standard error 3.30% 
SSE 1119.56 
number of data points 10 
p-value 0.23 
*The p-value is derived from the Student?s t-test on measured fatigue cracking and nationally-calibrated model 
predictions. 
 
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
pre
dic
ted
 fati
gu
e c
rac
kin
g, 
% o
f to
tal
 lan
e ar
ea
 
measured fatigue cracking, % of total lane area 
N8 - 06
N9 - 06
N10 - 06
S11 - 06
122 
 
Rutting 
In Figure 6.7, it can be seen that the nationally-calibrated model over-predicted the rut 
depths for all Test Track sections. Using the adjusted coefficients from the local calibration 
based on the 2003 sections, the rutting predictions for most of the 2006 sections were closer to 
the equality line. However, the rut depths in N1 and N2 were still over-predicted, which implied 
that the local calibration coefficients were not suitable for these two sections. Only these two 
sections were built on a thick limerock base over strong subgrade material, which resulted in 
minor actual rutting. The reason for the prediction inaccuracy was believed to be due to unique 
material characteristics of the limerock base which differ from the granular based used in the 
other test sections. It is recommended to initiate section-specific calibration only using N1 and 
N2 data if higher prediction accuracy was needed. 
 
(a) National calibration                                    (b) Local calibration 
Figure 6.7 Model predictions vs. measured values for rutting 
From Table 6.5, the standard error was dropped down by 12%, and the SSE decreased by 
a factor of 4 after the local calibration. Although the best-fit calibration coefficients improved the 
accuracy of predictions, the predicted rut depths were still statistically different from the 
measured values.  
123 
 
Table 6.5 Statistical analysis for the 2006 rutting predictions 
 nationally-
calibrated 
model 
locally-
calibrated 
model 
standard error 1.96 mm 1.73 mm 
SSE 8422.68 2020.98 
number of data points 136 136 
p-value  1.04E-42 4.72E-6 
*The p-value is derived from the Student?s t-test on measured rut depths and nationally-calibrated/locally-calibrated 
model predictions. 
N1 and N2 used limestone base combined with unusually strong subgrade. If they were 
excluded, the improvement of prediction accuracy was more significant using the locally best-fit 
calibration coefficients. In Figure 6.8, it can be seen that the predicted vs. measured rutting plots 
without N1 and N2 are much better. In fact, the best-fit calibration coefficients significantly 
improved the accuracy of model predictions. 
(a) National calibration                                       (b) Local calibration 
Figure 6.8 Model predictions vs. measured values for rutting (excluding N1 and N2) 
 
From Table 6.6, the standard error decreased by 17%, and the SSE diminished 
approximately by a factor of 6 after the local calibration. In addition, the Student?s t-test showed 
that the MEPDG predictions were no longer statistically different from the measured rut depths 
after adjusting the rutting model.  
124 
 
Table 6.6 Statistical analysis for the 2006 rutting predictions (excluding N1 and N2) 
 nationally-
calibrated 
model 
locally-
calibrated 
model 
standard error 1.82 mm 1.51 mm 
SSE 2864.79 429.73 
number of data points 91 91 
p-value 1.51E-20 0.42 
*The p-value is derived from the Student?s t-test on measured rut depths and nationally-calibrated/locally-calibrated 
model predictions. 
IRI 
Figure 6.9 shows that the IRI values shifted slightly after adopting the best-fit coefficients 
of rutting model. The data points of N8 and N10 were distributed with a gentle slope, which 
suggested the increase of predicted IRI did not progress at the same rate as the measured values. 
N1and N2 were not included because they were eliminated from the fatigue cracking analysis 
which strongly affected their IRI predictions. 
                          
(a) National calibration                                       (b) Local calibration 
Figure 6.9 Model predictions vs. measured values for IRI 
From Table 6.7, the standard error barely changed, and the SSE increased significantly by 
29% after calibrating the rutting model. The MEPDG predictions were statistically different from 
the measured IRI after calibrating the rutting model. It is recommended to perform local 
calibration on the IRI model using the 2006 sections in the future. 
125 
 
Table 6.7 Statistical analysis for the 2006 IRI predictions 
 predictions using 
nationally-calibrated 
model 
predictions after 
adjusting the rutting 
model 
standard error 3.99 in./mile 3.98 in./mile 
SSE 26653.26 34442.98 
number of data points 88 88 
p-value  0.47 0.003 
*The p-value is derived from the Student?s t-test on measured IRI and nationally-calibrated/locally-calibrated model 
predictions. 
SUMMARY 
The local calibration of this study was performed using the data from the 2003 research 
cycle. The nationally-calibrated fatigue cracking model performed fairly well. It was not able to 
be locally-calibrated for better prediction accuracy. The rutting prediction model in the MEPDG 
was able to be calibrated to field measured values. By adopting the best-fit values (as Table 6.8 
shows), the rutting prediction model could provide reasonable estimation of rut depth 
development during service life. The IRI prediction model was significantly improved after 
adopting the local calibration of the rutting model so it did not need to be locally-calibrated. 
Table 6.8 Best-fit calibration coefficients for three models in the MEPDG 
 
 
 
 
 
 
 
 
 
Calibration coefficients Local calibration values 
Fatigue Cracking  
?f1 1 
?f2 1 
?f3 1 
AC Rutting  
?r1 1 
?r2 1 
?r3 1 
Granular Rutting  
?s1 0.05 
Subgrade Rutting  
?s1 0.05 
IRI  
C1 40 
C2 0.4 
C3 0.008 
C4 0.015 
126 
 
The validation was conducted using the data from 2006 research cycle. Five sections left 
in place from the 2003 research cycle were not used due to their different design scenarios. The 
validation for the fatigue cracking model was performed on nationally-calibrated model since the 
fatigue cracking model could not be locally calibrated. The model poorly predicted the fatigue 
cracking so it is recommended to perform local calibration using the 2006 sections in the future. 
In terms of the rutting model, the validation indicated that the local calibration did greatly 
improve the accuracy of rutting prediction. In addition, the IRI model was validated to have no 
remarkable improvement on the prediction accuracy after the adoption of local calibration for the 
rutting model.  
 
 
 
  
127 
 
CHAPTER 7 
CONCLUSION AND RECOMMENDATION 
This study mainly focuses on calibrating the fatigue cracking, rutting, and IRI models to 
the Test Track condition. The MEPDG inputs, in terms of traffic, climate, and material properties, 
were characterized at highest level of details to represent the Test Track design scenario. The 
eight 2003 sections (i.e., N1 through N8) were investigated for nationally-calibrated model 
evaluation and local calibration. In the calibration process, 66 sets of calibration coefficients, in 
which each coefficient was varied from the MEPDG default, were tried for the fatigue cracking 
model; 106 sets of calibration coefficients were tried for the rutting model. The eleven 2006 
sections were investigated for the validation of local calibration results. Four of them were used 
for the fatigue cracking model, and six of them were used for the rutting model. Statistical 
analysis was applied to evaluate the model predictability. Based on the study presented in the 
previous chapters, conclusions and recommendations were drawn as follows: 
1. For the 2003 sections, the predicted fatigue cracking from the nationally-calibrated model 
reasonably matched the measured values. No local calibration coefficients were found 
that further improved the accuracy of the model. 
2. For the 2006 sections, used for validation purposes, the default fatigue cracking model 
did not provide good predictions that matched the measured values. It is recommended to 
perform local calibration on the 2006 sections in the future.  
3. For the 2003 sections, the nationally-calibrated model significantly over-predicted rut 
depths. The standard error was 0.20 mm. The best-fit calibration coefficients found were 
  =   =   = 1 in the AC rutting model,   = 0.05 in the unbound rutting model for 
granular base, and   = 0.05 in the unbound rutting model for subgrade, which could 
128 
 
significantly improve the prediction accuracy. Using the best-fit calibration coefficients, 
the standard error was 0.09 mm.    
4. For the 2006 sections, again used for validation purposes, the best-fit calibration 
coefficients were able to improve rutting model predictions but the predictions still did 
not match the measured values very well. If N1 and N2 were excluded from dataset for 
their uncommon material use, the standard error decreased from 1.82 mm to 1.51 mm. It 
suggested that the best-fit coefficients for the rutting model are a choice for pavement 
designers.   
5. For the 2003 sections, the nationally-calibrated IRI model fairly predicted the trend of 
measured values. The standard error was 0.99 in./mile. Although no local calibration 
coefficients were found for the IRI model, the adoption of the rutting calibration 
coefficients improved the predictions of the IRI model. The standard error decreased to 
0.82 in./mile. 
6. For the 2006 sections, the adoption of best-fit calibration coefficients for the rutting 
model was not able to improve the accuracy of the IRI model. The standard error changed 
from 3.99 in./ mile to 3.98 in./mile. It was recommended to perform local calibration on 
the IRI model using the 2006 sections in the future. 
7. The VB-based macro program significantly reduced the labor intensity in trying different 
sets of calibration coefficients in the MEPDG. Although the reliability of the program 
was not completely guaranteed, the program can be recommended for further 
development in the local calibration studies. 
8. The level of improvement on the prediction accuracy tends to be section-specific since 
the local calibration results depend on the calibration data. The local calibration using 
129 
 
various sections would result in the prediction accuracy improvement over all sections, 
rather than for one particular section or a group of sections that have similar pavement 
structures. Therefore, the local calibration of MEPDG could be tried using smaller groups 
of sections to seek a further enhancement of prediction accuracy for different categories 
of pavement structures.  
9. Owing to the limited amount of sections at the Test Track, the calibration results would 
not cover the all material use in local design projects in the Southeastern states. Therefore, 
the calibration coefficients still need to be validated for pavement sections with rare local 
materials before applying them into pavement design practice. 
10. The identification of cracking types is important in the local calibration of fatigue 
cracking model. The cracking caused by slippage between asphalt layers can be possibly 
regarded as bottom-up cracking from pavement surface survey. Coring is necessary to 
identify where the cracks initialized. If the cracking that happened was not fatigue 
cracking, the sections should not be included in the calibration dataset. 
11. The local calibration for the IRI model needs to be performed after the calibration for the 
fatigue cracking and rutting models. The IRI model provides results based on the fatigue 
cracking and rutting predictions.       
 
 
 
 
 
 
130 
 
REFERENCES 
1. Equivalent Single Axle Load 2009, Pavement Interactive, accessed Jul 10th 2013,  
<http://www.pavementinteractive.org/article/equivalent-single-axle-load/> 
2. Yang H. Huang. Pavement Analysis and Design, 1st edition, Prentice-Hall, Inc., Englewood 
Cliffs, New Jersey, 1993. 
3. Charles W. Schwartz, and Regis L. Carvalho. Implementation of NCHRP 1-37A Design 
Guide, Final Project, Volume 2: Evaluation of Mechanistic-Empirical Design Procedure, 
Department of Civil and Environmental Engineering, The University of Maryland, Prepared 
for Maryland State Highway Administration, Lutherville, MD, February, 2007. 
4. A. I. M. Claussen , J. M. Edwards, P. Sommer, and P. Uge. ?Asphalt Pavement Design ? The 
Shell Method?, Proceedings of the 4th International Conference on the Structural Design of 
Asphalt Pavements, pp. 39-74, 1977. 
5. Shook J. F., F. N. Finn, M. W. Witczak, and C. L. Monismith. ?Thickness Design of Asphalt 
Pavements ? The Asphalt Institute Method?, Proceedings of the 5th International Conference 
On the Structural Design of Asphalt Pavements, pp. 17-44, 1982. 
6. M. A.Miner, ?Cumulative Damage in Fatigue?, J. Applied Mechanics, 12, A159-A164, 1945. 
7. Qiang Li, Danny X. Xiao, Kelvin C. P. Wang, Kevin D. Hall, and Yanjun Qiu. ?Mechanistic-
Empirical Pavement Design Guide (MEPDG): a bird?s-eye view?, Journal of Modern 
Transportation, Volume 19, Page 114-133, June 2011. 
8. ?MEPDG Overview & National Perspective?, CEMEX, Federal Highway Administration, 
88th Annual TRB meeting, January 2009. 
9. Tommy E. Nantung. ?Implementing the Mechanistic-Empirical Pavement Design Guide for 
Cost Savings in Indiana?, TR News Journal, Number 271, page 34-36, November-December 
2010. 
131 
 
10. Jianhua Li, Linda M. Pierce, and Jeff Uhlmeyer. ?Calibration of Flexible Pavement in 
Mechanistic-Empirical Pavement Design Guide for Washington State?, Transportation 
Research Record: Journal of the Transportation Research Board. No. 2095, Transportation 
Research Board of the National Academies. Washington, D. C., pp. 73-83, 2009. 
11. Naresh R. Muthadi, and Y. Richard Kim. ?Local Calibration of Mechanistic-Empirical 
Pavement Design Guide for Flexible Pavement Design?, Transportation Research Record: 
Journal of the Transportation Research Board. No. 2087, Transportation Research Board of 
the National Academies. Washington, D. C., pp. 131-141, 2008. 
12. Kyle Heogh, Lev Khazanovich, and Maureen Jensen. ?Local Calibration of Mechanistic- 
Empirical Pavement Design Guide Rutting Model, Minnesota Road Research Project Test 
Sections?, Transportation Research Record: Journal of the Transportation Research Board. 
No. 2180, Transportation Research Board of the National Academies. Washington, D. C., pp. 
130-141, 2010. 
13. David H. Timm, and Angela L. Priest. NCAT Report 06-01, ?Material Properties of the 2003 
NCAT Test Track Structural Study?, 2006. 
14. Fatigue cracking 2009, Pavement Interactive, accessed April 17th 2013,  
<http://www.pavementinteractive.org/article/fatigue-cracking/> 
15. Rutting 2008, Pavement Interactive, accessed April 17th 2013, 
<http://www.pavementinteractive.org/article/rutting/> 
16. longitudinal cracking 2008, Pavement Interactive, accessed April 17th 2013, 
<http://www.pavementinteractive.org/article/longitudinal-cracking/> 
17. Transverse cracking 2006, Pavement Interactive, accessed April 17th 2013, 
<http://www.asphaltwa.com/2010/09/18/transverse-thermal-cracking/> 
132 
 
18. Roughness 2007, Pavement Interactive, accessed March 5th 2013,  
<http://www.pavementinteractive.org/article/roughness/> 
19. Loads, WSDOT Pavement Guide, accessed March 5th, 2013 from 
<http://classes.engr.oregonstate.edu/cce/winter2012/ce492/> 
20. NCHRP 1-37A Report, Guide for Mechanistic-Empirical Pavement Design of New and 
Rehabilitated Pavement Structures, ARA, Inc., Part 2, Chapter 4, Champaign, IL, 2004. 
21. Khaled Ksaibati, and Ryan Erickson, ?Evaluation of Low Temperature Cracking in Asphalt 
Pavement Mixes?, the University of Wyoming, October, 1998. 
22. Yoder E. J., and M. W. Witczak, ?Principles of Pavement Design?, John Wiley & Sons, Inc., 
New York, 1975. 
23. Richard D. Barksdale, Jorge Alba, N. Paul Khosla, Ricahrd Kim, Phil C. Lambe, and M.S. 
Rahman, ?Laboratory Determination of Resilient Modulus for Flexible Pavement Design?, 
Georgia Tech Project E20-634, June, 1997.   
24. Elastic Modulus 2007, Pavement Interactive, accessed April 23th, 2013, 
< http://www.pavementinteractive.org/article/elastic-modulus/> 
25. Resilient Modulus 2007, Pavement Interactive, accessed April 23th, 2013, 
< http://www.pavementinteractive.org/article/resilient-modulus/> 
26. M.W. Witczak, K. Kaloush, T. Pellinen, M. El-Basyouny, and H. Von Quintus, NCHRP 
Report 465, Simple Performance Test for Superpave Mix Design. Transportation Research 
Board, National Research Council, Washington, D.C., 2002. 
27. Pavement Distress Survey Manual, Pavement Services Unit, Oregon Department of 
Transportation, Jun. 2010. 
133 
 
28. Angela L. Priest, and David H. Timm, ?Methodology and Calibration of Fatigue Transfer 
Functions for Mechanistic-Empirical Flexible Pavement Design?, NCAT Report 06-03, 
December 2006. 
29. NCHRP 1-37A Report, Guide for Mechanistic-Empirical Pavement Design of New and 
Rehabilitated Pavement Structures, ARA, Inc., Appendix II-1, Champaign, IL, 2004. 
30. Pavement Preservation Manual, Part 2: pavement condition data, Office of Asset 
Management, Utah Department of Transportation, 2009. 
31. AASHTO and TRIP 2009, Rough Roads Ahead, e-book, accessed March 5th, 2013 from 
books.google.com 
32. Qiang Li, Danny X. Xiao, Kelvin C.P. Wang, Kevin D. Hall, and Yanjun Qiu, ?Mechanistic-
empirical pavement design guide: a bird?s-eye view?, Journal of Modern Transportation, 
Volume 19, June 2, pp. 114-133, January 2011. 
33. Nicholas J. Garber, and Lester A. Hoel, Traffic and Highway Engineering, 4th edition, 
Cengage Learning, Toronto, Canada, 2009. 
34. Asphalt Institute. Asphalt Overlays for Highways and Street Rehabilitation, Manual Series 
No. 17 (MS-17), College Park, Maryland, June, 1983. 
35. AASHTO Guide for Design of Pavement Structures. American Association of State Highway 
and Transportation Officials, Washington D.C., 1993. 
36. Asphalt Design Procedure 2007, accessed March 11th 2013, 
<https://engineering.purdue.edu/NCSC/documents/presentations/FHWA%20National%20up
date> 
37. AASHTO Interim Guide for Design of Pavements Structures, American Association of State 
Highway and Transportation Officials, Washington, D.C., 1972.  
134 
 
38. AASHTO Guide for Design of Pavements Structures, American Association of State 
Highway and Transportation Officials, Washington, D.C., 1986.  
39. AASHTO Guide for Design of Pavements Structures, American Association of State 
Highway and Transportation Officials, Washington, D.C., 1993. 
40. FHWA Report, ?Seasonal Variations in the Moduli of Unbound Pavement Layers?, Chapter 
2, July 2006. 
41. Cary W. Sharpe, Memorandum, ?Distribution of the Recommended Mechanistic-Empirical 
Pavement Design Guide (NCHRP Project 1-37A)?, American Association of State Highway 
and Transportation Officials, 2004. 
42. NCHRP 1-37A Report, Guide for Mechanistic-Empirical Pavement Design of New and 
Rehabilitated Pavement Structures, Part 2, Chapter 3, ARA, Inc., Champaign, IL, 2004. 
43. Angela L. Priest, ?Calibration of Fatigue Transfer Functions for Mechanistic-Empirical 
Flexible Pavement Design?, Unpublished Master?s thesis, Auburn University, 2005. 
44. NCHRP 1-37A Report, Guide for Mechanistic-Empirical Pavement Design of New and 
Rehabilitated Pavement Structures, Part 3, Chapter 3, ARA, Inc., Champaign, IL, 2004. 
45. Asphalt Institute, Research and Development of the Asphalt Institute?s Thickness Design 
Manual (MS-1), 9th edition, Research Report 82-2, 1982.  
46. Jagannath Mallela, Leslie Titus Glover, Michael I. Darter, Harold Von Quintus, Alex Gotlif, 
Mark Stanley, and Suri Sadasivam, ARA, Inc., Guidelines for Implementing NCHRP 1-37A 
M-E Design Procedures in Ohio: Volume 1- Summary of Findings, Implementation Plan, and 
Next Steps, Champaign, IL, 2009. 
47. Kyle Hoegh, Lev Khazanovich, and Maureen Jensen, ?Local Calibration of Mechanistic-
Empirical Pavement Design Guide Rutting Model: Minnesota Road Research Project Test 
135 
 
Sections?, Transportation Research Record: Journal of the Transportation Research Board. 
No. 2180, Transportation Research Board of the National Academies. Washington, D. C., pp. 
130-141, 2010. 
48. Randy West, David Timm, Richard Willis, Buzz Powell, Nam Tran, Don Watson, Maryam 
Sakhaeifar, Ray Brown, Mary Robbins, Adriana Vargas Nordcbeck, Fabricio Leiva 
Villacorta, Xiaolong Guo, and Jason Nelson. NCAT Report 12-10, ?Phase IV NCAT 
Pavement Test Track Findings?, 2012 
49. Angela Priest, Auburn University, ?NCAT Test Track Structural Study?, accessed March 
18th, 2013 
<http://www.webpages.uidaho.edu/bayomy/trb/afd60/TRB2005/Committee%20Meeting%20
Presentations/NCAT%20Test%20Track.pdf> 
50. David H. Timm, and Angela L. Priest, NCAT Report 06-01, ?Material Properties of the 2003 
NCAT Test Track Structural Study?, 2006 
51. David H. Timm, NCAT Report 09-01, ?Design , Construction, and Instrumentation of the 
2006 Test Track Structural Study?, 2009 
52. R. Buzz Powell, and Bob Rosenthal, ?Reference Information Describing the NCAT Fleet?, 
Opelika, 2011 
53. Adam J. Taylor, and David H. Timm, NCAT Report 09-06, ?Mechanistic Characterization of 
Resilient Moduli for Unbound Pavement Layer Materials?, 2009 
54. David H. Timm, Nam Tran, Adam Taylor, Mary M. Robbins, and Buzz Powell, NCAT 
Report 09-05, ?Evaluation of Mixture Performance and Structural Capacity of Pavements 
Using Shell Thiopave, Phase I: Mix Design, Laboratory Performance Evaluation and 
Structural Pavement Analysis and Design?, 2009 
136 
 
55. Mary R. Robbins, ?New Method for Predicting Critical Tensile Strains in an M-E 
Framework?, Doctoral Dissertation, Auburn University, 2013 
56. Kendra Davis (2009), MEPDG Local Calibration, Unpublished raw data. 
57. Adam J. Taylor, ?Mechanistic Characterization of Resilient Moduli for Unbound Pavement 
Layer Materials?, Unpublished Master?s thesis, Auburn University, 2008 
58. Michael I. Darter, Leslie Titus-Glover, and H. Von Quintus, ?Implementation of the 
Mechanistic-Empirical Pavement Design Guide in Utah: Validation, Calibration, and 
Development of the UDOT MEPDG User?s Guide?, Applied Research Associates, Inc., 
October 2009. 
59. AASHTO, ?Guide for the Local Calibration of the Mechanistic-Empirical Pavement Design 
Guide?, Publication Code: LCG-1, Washington, DC, 2010 
60. Scott A. Schram, and Magdy Abdelrahman, ?Integration of Mechanistic-Empirical Pavement 
Design Guide Distresses with Local Performance Indices?, Transportation Research Record: 
Journal of the Transportation Research Board. No. 2153, Transportation Research Board of 
the National Academies. Washington, D. C., pp. 13-23, 2010. 
61. Ambarish Banerjee, Jose P. Aguiar-Moya, and Jorge A. Prozzi, ?Calibration of Mechanistic-
Empirical Pavement Design Guide Permanent Deformation Models ? Texas Experience with 
Long-Term Pavement Performance?, Transportation Research Record: Journal of the 
Transportation Research Board. No. 2094, Transportation Research Board of the National 
Academies. Washington, D. C., pp. 12-20, 2009. 
62. James R. Willis, and David H. Timm, ?A Forensic Investigation of Debonding in a Rich-
Bottom Pavement,? Transportation Research Record: Journal of the Transportation Research 
Board. No. 2040, Washington, D. C., 2007, pp. 107-114. 
137 
 
APPENDIX A 
Each asphalt concrete mix was designated with a number. The following tables show the 
mix number for each lift of the 2003 and 2006 test sections. Also, the E* data for each mix was 
presented. 
Table A-1 Mix number for each lift of the 2003 Test Track sections 
Test section Lift number Mix number 
N1 1 1 
N1 2 2 
N1 3 2 
N2 1 3 
N2 2 4 
N2 3 4 
N3 1 3 
N3 2 4 
N3 3 4 
N3 4 4 
N3 5 4 
N4 1 1 
N4 2 2 
N4 3 2 
N4 4 2 
N4 5 2 
N5 1 1 
N5 2 2 
N5 3 2 
N5 4 2 
N6 1 3 
N6 2 4 
N6 3 4 
N6 4 4 
N7 1 5 
N7 2 4 
N7 3 4 
N7 4 4 
N8 1 5 
N8 2 4 
N8 3 4 
N8 4 6 
 
 
 
 
138 
 
Table A-2 E* data for mixes in the 2003 research cycle (Timm and Priest, 2006) 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
1 14 0.1 2083358 
1 14 0.5 2468142 
1 14 1 2629424 
1 14 5 2985601 
1 14 10 3162982 
1 14 25 3344061 
1 40 0.1 974980 
1 40 0.5 1241885 
1 40 1 1362085 
1 40 5 1656076 
1 40 10 1793318 
1 40 25 2000360 
1 70 0.1 298959 
1 70 0.5 436418 
1 70 1 511874 
1 70 5 732658 
1 70 10 867760 
1 70 25 1029622 
1 100 0.1 85463 
1 100 0.5 124986 
1 100 1 147902 
1 100 5 245658 
1 100 10 310272 
1 100 25 380289 
1 130 0.1 41408 
1 130 0.5 55332 
1 130 1 63164 
1 130 5 99387 
1 130 10 124043 
1 130 25 152833 
2 14 0.1 2277563 
2 14 0.5 2725367 
2 14 1 2907715 
2 14 5 3304176 
2 14 10 3502587 
2 14 25 3689759 
2 40 0.1 1153195 
2 40 0.5 1474779 
2 40 1 1623189 
2 40 5 1988503 
2 40 10 2148479 
2 40 25 2404942 
139 
 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
2 70 0.1 394829 
2 70 0.5 565357 
2 70 1 653141 
2 70 5 913955 
2 70 10 1061821 
2 70 25 1243734 
2 100 0.1 151383 
2 100 0.5 214801 
2 100 1 254106 
2 100 5 404945 
2 100 10 507342 
2 100 25 614706 
2 130 0.1 65267 
2 130 0.5 83832 
2 130 1 95036 
2 130 5 142898 
2 130 10 174009 
2 130 25 218137 
3 14 0.1 2029222 
3 14 0.5 2417379 
3 14 1 2581816 
3 14 5 2944301 
3 14 10 3101703 
3 14 25 3271941 
3 40 0.1 791507 
3 40 0.5 1031399 
3 40 1 1139742 
3 40 5 1432827 
3 40 10 1564304 
3 40 25 1723628 
3 70 0.1 288879 
3 70 0.5 430798 
3 70 1 508140 
3 70 5 737843 
3 70 10 859820 
3 70 25 1031689 
3 100 0.1 87639 
3 100 0.5 123391 
3 100 1 146234 
3 100 5 235505 
3 100 10 295514 
3 100 25 382066 
 
140 
 
    
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
3 130 0.1 31256 
3 130 0.5 39958 
3 130 1 45397 
3 130 5 64941 
3 130 10 77414 
3 130 25 97647 
4 14 0.1 2298485 
4 14 0.5 2724243 
4 14 1 2897164 
4 14 5 3289201 
4 14 10 3421801 
4 14 25 3617276 
4 40 0.1 1072481 
4 40 0.5 1395625 
4 40 1 1543418 
4 40 5 1920335 
4 40 10 2087564 
4 40 25 2327093 
4 70 0.1 378875 
4 70 0.5 547046 
4 70 1 643423 
4 70 5 930961 
4 70 10 1064722 
4 70 25 1310597 
4 100 0.1 139490 
4 100 0.5 197650 
4 100 1 233293 
4 100 5 359657 
4 100 10 445592 
4 100 25 574095 
4 130 0.1 63055 
4 130 0.5 79553 
4 130 1 89815 
4 130 5 135103 
4 130 10 167845 
4 130 25 215345 
5 14 0.1 2501030 
5 14 0.5 3002969 
5 14 1 3219256 
5 14 5 3670070 
5 14 10 3851331 
5 14 25 4038067 
141 
 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
5 40 0.1 1271183 
5 40 0.5 1639905 
5 40 1 1818265 
5 40 5 2239708 
5 40 10 2427604 
5 40 25 2700312 
5 70 0.1 378222 
5 70 0.5 526886 
5 70 1 602668 
5 70 5 840348 
5 70 10 962434 
5 70 25 1150910 
5 100 0.1 136843 
5 100 0.5 192356 
5 100 1 227383 
5 100 5 354291 
5 100 10 434497 
5 100 25 536096 
5 130 0.1 54679 
5 130 0.5 71504 
5 130 1 82128 
5 130 5 132746 
5 130 10 166975 
5 130 25 203415 
6 14 0.1 2761276 
6 14 0.5 3333933 
6 14 1 3219220 
6 14 5 3974323 
6 14 10 4214904 
6 14 25 4448777 
6 40 0.1 1157219 
6 40 0.5 1588235 
6 40 1 1796618 
6 40 5 2290544 
6 40 10 2514047 
6 40 25 2828271 
6 70 0.1 366474 
6 70 0.5 537292 
6 70 1 639000 
6 70 5 954565 
6 70 10 1119401 
6 70 25 1364623 
    
142 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table A-3 Mix number for each lift of the 2006 Test Track sections 
Test section Lift 
number 
Mix 
number 
N1 1 1 
N1 2 1 
N1 3 2 
N2 1 3 
N2 2 3 
N2 3 2 
N8 1 4 
N8 2 5 
N8 3 6 
N8 4 7 
N9 1 4 
N9 2 5 
N9 3 6 
N9 4 6 
N9 5 7 
N10 1 8A 
N10 2 8B 
N10 3 9 
S11 1 28A 
S11 2 28B 
S11 3 2 
S11 4 2 
 
 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
6 100 0.1 127959 
6 100 0.5 177671 
6 100 1 209543 
6 100 5 324123 
6 100 10 403205 
6 100 25 533630 
6 130 0.1 52540 
6 130 0.5 66536 
6 130 1 75637 
6 130 5 116139 
6 130 10 142935 
6 130 25 180318 
    
143 
 
Table A-4 E* data for mixes in the 2006 research cycle 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
1 15 0.5 2157627 
1 15 1 2410856 
1 15 2 2693804 
1 15 5 3119436 
1 15 10 3485547 
1 15 20 3894627 
1 40 0.5 1046350 
1 40 1 1173742 
1 40 2 1313558 
1 40 5 1501575 
1 40 10 1649804 
1 40 20 1805284 
1 70 0.5 331943 
1 70 1 399192 
1 70 2 477657 
1 70 5 597217 
1 70 10 701644 
1 70 20 821300 
1 100 0.5 101212 
1 100 1 123625 
1 100 2 157849 
1 100 5 214849 
1 100 10 268658 
1 100 20 334312 
1 130 0.5 69097 
1 130 1 77207 
1 130 2 86268 
1 130 5 99899 
1 130 10 111624 
1 130 20 124724 
2 15 0.5 2458627 
2 15 1 2750572 
2 15 2 3077182 
2 15 5 3569202 
2 15 10 3993019 
2 15 20 4467161 
    
    
    
    
    
144 
 
    
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
2 40 0.5 1157594 
2 40 1 1304856 
2 40 2 1459611 
2 40 5 1676636 
2 40 10 1848167 
2 40 20 2021584 
2 70 0.5 362401 
2 70 1 438207 
2 70 2 526632 
2 70 5 662194 
2 70 10 779674 
2 70 20 910982 
2 100 0.5 101227 
2 100 1 123137 
2 100 2 155480 
2 100 5 210643 
2 100 10 263824 
2 100 20 330154 
2 130 0.5 70623 
2 130 1 79009 
2 130 2 88391 
2 130 5 102524 
2 130 10 114698 
2 130 20 128318 
3 15 0.5 2164720 
3 15 1 2404672 
3 15 2 2671222 
3 15 5 3069457 
3 15 10 3409696 
3 15 20 3787650 
3 40 0.5 1055004 
3 40 1 1178044 
3 40 2 1307756 
3 40 5 1483590 
3 40 10 1622101 
3 40 20 1763900 
3 70 0.5 349879 
3 70 1 417370 
3 70 2 494772 
3 70 5 613171 
3 70 10 715519 
3 70 20 829761 
145 
 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
3 100 0.5 99167 
3 100 1 118974 
3 100 2 149824 
3 100 5 201119 
3 100 10 249852 
3 100 20 309317 
3 130 0.5 71687 
3 130 1 79633 
3 130 2 88461 
3 130 5 101649 
3 130 10 112916 
3 130 20 125432 
4 15 0.5 2703490 
4 15 1 2994983 
4 15 2 3317905 
4 15 5 3798828 
4 15 10 4208421 
4 15 20 4662178 
4 40 0.5 1184842 
4 40 1 1321351 
4 40 2 1464910 
4 40 5 1663524 
4 40 10 1818250 
4 40 20 1972919 
4 70 0.5 321868 
4 70 1 385365 
4 70 2 460930 
4 70 5 575335 
4 70 10 675179 
4 70 20 785611 
4 100 0.5 97097 
4 100 1 119862 
4 100 2 152214 
4 100 5 205185 
4 100 10 256073 
4 100 20 317038 
4 130 0.5 55862 
4 130 1 61886 
4 130 2 68558 
4 130 5 78495 
4 130 10 86959 
4 130 20 96335 
    
146 
 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
5 15 0.5 2649429 
5 15 1 2932408 
5 15 2 3245610 
5 15 5 3711559 
5 15 10 4107980 
5 15 20 4546743 
5 40 0.5 1311068 
5 40 1 1458970 
5 40 2 1610788 
5 40 5 1816343 
5 40 10 1975921 
5 40 20 2135970 
5 70 0.5 422966 
5 70 1 505348 
5 70 2 600746 
5 70 5 745893 
5 70 10 870516 
5 70 20 1007939 
5 100 0.5 124707 
5 100 1 153417 
5 100 2 195620 
5 100 5 265165 
5 100 10 331194 
5 100 20 413611 
5 130 0.5 91896 
5 130 1 101711 
5 130 2 112575 
5 130 5 128736 
5 130 10 142486 
5 130 20 157705 
6 15 0.5 2853544 
6 15 1 3125467 
6 15 2 3423302 
6 15 5 3861014 
6 15 10 4228941 
6 15 20 4631929 
6 40 0.5 1525071 
6 40 1 1674363 
6 40 2 1826363 
6 40 5 2023856 
6 40 10 2175227 
6 40 20 2324616 
    
147 
 
    
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
6 70 0.5 566179 
6 70 1 669494 
6 70 2 781656 
6 70 5 945936 
6 70 10 1083093 
6 70 20 1229629 
6 100 0.5 164473 
6 100 1 202618 
6 100 2 255218 
6 100 5 338421 
6 100 10 415678 
6 100 20 507729 
6 130 0.5 138663 
6 130 1 151876 
6 130 2 166349 
6 130 5 187619 
6 130 10 205497 
6 130 20 225080 
7 15 0.5 1852345 
7 15 1 2088949 
7 15 2 2355775 
7 15 5 2761508 
7 15 10 3114241 
7 15 20 3512029 
7 40 0.5 862249 
7 40 1 983597 
7 40 2 1112197 
7 40 5 1289095 
7 40 10 1427703 
7 40 20 1565150 
7 70 0.5 249030 
7 70 1 311686 
7 70 2 385075 
7 70 5 499075 
7 70 10 599924 
7 70 20 712618 
7 100 0.5 61936 
7 100 1 79263 
7 100 2 102812 
7 100 5 146749 
7 100 10 190434 
7 100 20 245404 
148 
 
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
7 130 0.5 50456 
7 130 1 56901 
7 130 2 64169 
7 130 5 75221 
7 130 10 84829 
7 130 20 95664 
8A 15 0.5 2533493 
8A 15 1 2807917 
8A 15 2 3112066 
8A 15 5 3565268 
8A 15 10 3951453 
8A 15 20 4379469 
8A 40 0.5 1158174 
8A 40 1 1292866 
8A 40 2 1432586 
8A 40 5 1622391 
8A 40 10 1768444 
8A 40 20 1914739 
8A 70 0.5 318793 
8A 70 1 386332 
8A 70 2 466538 
8A 70 5 588273 
8A 70 10 691588 
8A 70 20 809649 
8A 100 0.5 92713 
8A 100 1 117891 
8A 100 2 153005 
8A 100 5 212238 
8A 100 10 268803 
8A 100 20 337889 
8A 130 0.5 61988 
8A 130 1 68702 
8A 130 2 76144 
8A 130 5 87233 
8A 130 10 96682 
8A 130 20 107154 
8B 15 0.01 1872962 
8B 15 0.1 2673802 
8B 15 0.5 3429180 
8B 15 1 3817066 
8B 15 5 4895428 
    
    
149 
 
    
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
8B 40 0.01 868486 
8B 40 0.1 1325886 
8B 40 0.5 1703564 
8B 40 1 1872146 
8B 40 5 2268873 
8B 70 0.01 171773 
8B 70 0.1 385317 
8B 70 0.5 615540 
8B 70 1 732199 
8B 70 5 1066414 
8B 115 0.01 20494 
8B 115 0.1 42032 
8B 115 0.5 75492 
8B 115 1 98776 
8B 115 5 192078 
8B 130 0.01 58961 
8B 130 0.1 84171 
8B 130 0.5 107950 
8B 130 1 120161 
8B 130 5 154108 
9 15 0.5 2399518 
9 15 1 2670238 
9 15 2 2971501 
9 15 5 3422533 
9 15 10 3808672 
9 15 20 4238377 
9 40 0.5 1181658 
9 40 1 1322744 
9 40 2 1468724 
9 40 5 1669420 
9 40 10 1824538 
9 40 20 1982448 
9 70 0.5 371840 
9 70 1 449218 
9 70 2 539721 
9 70 5 675912 
9 70 10 792885 
9 70 20 923382 
    
    
    
    
150 
 
    
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
9 100 0.5 110254 
9 100 1 139298 
9 100 2 179593 
9 100 5 246419 
9 100 10 307842 
9 100 20 382392 
9 130 0.5 81342 
9 130 1 90519 
9 130 2 100732 
9 130 5 116022 
9 130 10 129112 
9 130 20 143678 
28A 15 0.5 1756400 
28A 15 1 1968603 
28A 15 2 2206444 
28A 15 5 2565513 
28A 15 10 2875471 
28A 15 20 3222877 
28A 40 0.5 835804 
28A 40 1 944340 
28A 40 2 1058002 
28A 40 5 1218220 
28A 40 10 1344113 
28A 40 20 1472277 
28A 70 0.5 259521 
28A 70 1 313185 
28A 70 2 377436 
28A 70 5 476981 
28A 70 10 563375 
28A 70 20 660985 
28A 100 0.5 77489 
28A 100 1 94120 
28A 100 2 119100 
28A 100 5 163573 
28A 100 10 205132 
28A 100 20 259472 
28A 130 0.5 52461 
28A 130 1 58799 
28A 130 2 65903 
28A 130 5 76628 
28A 130 10 85886 
28A 130 20 96262 
151 
 
 
*It should be noted that the underlined E* values were not obtained from the lab test but predicted by using the 
Equation (3-1) shown earlier. 
 
 
 
 
 
 
 
 
 
 
  
Mix 
number 
Test temperature 
(?F) 
Test 
frequency 
(Hz) 
E* (psi) 
28B 15 0.01 1458885 
28B 15 0.1 2126002 
28B 15 0.5 2766145 
28B 15 1 3098177 
28B 15 5 4031044 
28B 15 10 4514906 
28B 40 0.01 1438870 
28B 40 0.1 1595850 
28B 40 0.5 1977347 
28B 40 1 2141433 
28B 40 5 1094986 
28B 40 10 726107 
28B 70 0.01 595186 
28B 70 0.1 701402 
28B 70 0.5 990559 
28B 70 1 1141978 
28B 70 5 385317 
28B 70 10 171483 
28B 130 0.01 59497 
28B 130 0.1 86704 
28B 130 0.5 112811 
28B 130 1 126352 
28B 130 5 164397 
28B 130 10 184130 
152 
 
APPENDIX B 
The following table shows the G* data of asphalt binders in the 2003 and 2006 test 
sections.  
Table B-1 G* data of asphalt binders in the 2003 and 2006 test sections 
Test section & Year Layer 
number 
Binder 
type 
Test 
temp. 
(?F) 
G* (Pa) Phase 
angle (?) 
N1, N4 & N5 2003 All PG 76-22 70 1,881,000 57.98 
N1, N4 & N5 2003 All PG 76-22 100 134,200 60.55 
N1, N4 & N5 2003 All PG 76-22 130 22,420 58.39 
N2, N3, & N6 2003 All PG 67-22 70 2,156,000 55.80 
N2, N3, & N6 2003 All PG 67-22 100 241,800 61.59 
N2, N3, & N6 2003 All PG 67-22 130 26,710 70.67 
N7 & N8 2003 1 PG 76-22 70 1,881,000 57.98 
N7 & N8 2003 1 PG 76-22 100 134,200 60.55 
N7 & N8 2003 1 PG 76-22 130 22,420 58.39 
N7 & N8 2003 2, 3 & 4 PG 67-22 70 2,156,000 55.80 
N7 & N8 2003 2, 3 & 4 PG 67-22 100 241,800 61.59 
N7 & N8 2003 2, 3 & 4 PG 67-22 130 26,710 70.67 
N1 2006 1 & 2 PG 67-22 70 2,053,000 55.59 
N1 2006 1 & 2 PG 67-22 100 225900 62.21 
N1 2006 1 & 2 PG 67-22 130 28460 69.96 
N1 & N2 2006 3 PG 67-22 70 2,156,000 55.80 
N1 & N2 2006 3 PG 67-22 100 241,800 61.59 
N1 & N2 2006 3 PG 67-22 130 26,710 70.67 
N2 2006 1 & 2 PG 76-22 70 2,177,000 55.88 
N2 2006 1 & 2 PG 76-22 100 267,400 59.63 
N2 2006 1 & 2 PG 76-22 130 37,560 65.30 
N8 2006 & N9 2006 1 & 2 PG 76-28 70 2,672,000 54.25 
N8 2006 & N9 2006 1 & 2 PG 76-28 100 344,800 57.77 
N8 2006 & N9 2006 1 & 2 PG 76-28 130 34,410 59.23 
N8 2006 & N9 2006 3 PG 64-22 70 2,109,000 56.23 
N8 2006 & N9 2006 3 PG 64-22 100 228,700 62.42 
N8 2006 & N9 2006 3 PG 64-22 130 29,080 73.34 
N8 2006 & N9 2006 4 PG 64-22 70 768,800 57.77 
N8 2006 & N9 2006 4 PG 64-22 100 183,600 62.92 
N8 2006 & N9 2006 4 PG 64-22 130 25,920 68.94 
N10 2006 1 & 2 PG 70-22 70 2,413,000 60.11 
N10 2006 1 & 2 PG 70-22 100 223,600 65.73 
N10 2006 1 & 2 PG 70-22 130 21,580 70.63 
N10 2006 3 PG 64-22 70 2,066,000 56.25 
N10 2006 3 PG 64-22 100 220,100 62.21 
N10 2006 3 PG 64-22 130 24,310 70.56 
153 
 
Test section & Year Layer 
number 
Binder 
type 
Test 
temp. 
(?F) 
G* (Pa) Phase 
angle (?) 
S11 2006 1 & 2 PG 76-22 70 1,881,000 57.98 
S11 2006 1 & 2 PG 76-22 100 134,200 60.55 
S11 2006 1 & 2 PG 76-22 130 22,420 58.39 
S11 2006 3 & 4 PG 67-22 70 2,156,000 55.80 
S11 2006 3 & 4 PG 67-22 100 241,800 61.59 
S11 2006 3 & 4 PG 67-22 130 26,710 70.67 
 
 
  
154 
 
APPENDIX C 
The following table shows the as-built volumetric properties of asphalt mixes in the 2003 
and 2006 test sections.  
Table C-1 As-built volumetric properties for the 2003 and 2006 Test Track sections 
Test section & 
Year 
Lift number Air voids (%) Unit weight 
(pcf) 
Effective 
binder content 
(%) 
N1 2003 1 7.2 143.7 13.4 
 2 7.2 149.1 9.7 
 3 7.0 150.1 9.2 
N2 2003 1 7.1 144.2 13.1 
 2 6.1 151.8 9.2 
 3 5.9 151.2 9.9 
N3 2003 1 7.2 144.1 13.1 
 2 6.7 150.8 9.1 
 3 6.3 150.6 9.8 
 4 7.0 149.2 9.7 
 5 5.4 150.7 10.7 
N4 2003 1 6.6 145.4 13.0 
 2 7.1 149.0 9.7 
 3 6.8 149.3 9.9 
 4 7.2 148.1 10.5 
 5 7.3 148.7 9.8 
N5 2003 1 6.7 145.2 13.0 
 2 7.1 149.0 9.7 
 3 7.2 148.7 9.9 
 4 6.8 148.7 10.5 
N6 2003 1 6.3 145.0 13.6 
 2 5.9 151.3 9.9 
 3 6.6 149.3 10.2 
 4 4.0 153.2 11.1 
N7 2003 1 6.9 141.6 13.7 
 2 5.7 151.6 9.9 
 3 6.7 149.0 10.2 
 4 5.0 151.6 11.0 
N8 2003 1 6.9 141.6 13.8 
 2 7.0 149.5 9.8 
 3 7.0 148.6 10.2 
 4 6.7 148.9 10.9 
N1 2006 1 5.4 147.5 10.4 
 2 7.8 143.1 10.6 
 3 7.9 147.5 9.7 
     
155 
 
     
Test section & 
Year 
Lift number Air voids (%) Unit weight 
(pcf) 
Effective 
binder content 
(%) 
N2 2006 1 5.0 148.1 10.3 
 2 5.8 145.5 11.3 
 3 5.1 152.0 10.1 
N8 2006 1 8.2 137.3 10.9 
 2 6.4 145.8 7.8 
 3 7.1 145.1 7.1 
 4 2.8 147.0 11.2 
N9 2006 1 7.0 139.1 11.1 
 2 7.1 144.7 7.7 
 3 4.9 148.5 7.5 
 4 6.1 146.9 6.9 
 5 5.6 142.8 10.7 
N10 2006 1 8.7 139.9 10.8 
 2 7.5 143.9 8.8 
 3 6.7 144.7 9.6 
S11 2006 1 6.8 143.3 14.2 
 2 5.8 150.4 10.9 
 3 7.4 148.4 10.0 
 4 8.2 147.3 9.8