Professional Documents
Culture Documents
DETAILS
CONTRIBUTORS
GET THIS BOOK Robert L. Lytton, Xue Luo, Sajib Saha, Yu Chen, Yong Deng, Fan Gu, Meng Ling,
Texas A&M Transportation Institute; National Cooperative Highway Research
Program; Transportation Research Board; National Academies of Sciences,
FIND RELATED TITLES Engineering, and Medicine
SUGGESTED CITATION
Visit the National Academies Press at NAP.edu and login or register to get:
Distribution, posting, or copying of this PDF is strictly prohibited without written permission of the National Academies Press.
(Request Permission) Unless otherwise indicated, all materials in this PDF are copyrighted by the National Academy of Sciences.
NCHRP
Web-Only Document 264:
ACKNOWLEDGMENT
This work was sponsored by the American Association of State Highway and Transportation Officials (AASHTO), in cooperation
with the Federal Highway Administration, and was conducted in the National Cooperative Highway Research Program (NCHRP),
which is administered by the Transportation Research Board (TRB) of the National Academies of Sciences, Engineering, and
Medicine.
COPYRIGHT INFORMATION
Authors herein are responsible for the authenticity of their materials and for obtaining written permissions from publishers or
persons who own the copyright to any previously published or copyrighted material used herein.
Cooperative Research Programs (CRP) grants permission to reproduce material in this publication for classroom and not-for-profit
purposes. Permission is given with the understanding that none of the material will be used to imply TRB, AASHTO, FAA, FHWA,
FMCSA, FRA, FTA, Office of the Assistant Secretary for Research and Technology, PHMSA, or TDC endorsement of a particular
product, method, or practice. It is expected that those reproducing the material in this document for educational and not-for-profit
uses will give appropriate acknowledgment of the source of any reprinted or reproduced material. For other uses of the material,
request permission from CRP.
DISCLAIMER
The opinions and conclusions expressed or implied in this report are those of the researchers who performed the research. They
are not necessarily those of the Transportation Research Board; the National Academies of Sciences, Engineering, and Medicine;
or the program sponsors.
The information contained in this document was taken directly from the submission of the author(s). This material has not been
edited by TRB.
The National Academy of Sciences was established in 1863 by an Act of Congress, signed by President Lincoln, as a private, non-
governmental institution to advise the nation on issues related to science and technology. Members are elected by their peers for
outstanding contributions to research. Dr. Marcia McNutt is president.
The National Academy of Engineering was established in 1964 under the charter of the National Academy of Sciences to bring the
practices of engineering to advising the nation. Members are elected by their peers for extraordinary contributions to engineering.
Dr. John L. Anderson is president.
The National Academy of Medicine (formerly the Institute of Medicine) was established in 1970 under the charter of the National
Academy of Sciences to advise the nation on medical and health issues. Members are elected by their peers for distinguished contributions
to medicine and health. Dr. Victor J. Dzau is president.
The three Academies work together as the National Academies of Sciences, Engineering, and Medicine to provide independent,
objective analysis and advice to the nation and conduct other activities to solve complex problems and inform public policy decisions.
The National Academies also encourage education and research, recognize outstanding contributions to knowledge, and increase
public understanding in matters of science, engineering, and medicine.
Learn more about the National Academies of Sciences, Engineering, and Medicine at www.national-academies.org.
The Transportation Research Board is one of seven major programs of the National Academies of Sciences, Engineering, and Medicine.
The mission of the Transportation Research Board is to increase the benefits that transportation contributes to society by providing
leadership in transportation innovation and progress through research and information exchange, conducted within a setting that
is objective, interdisciplinary, and multimodal. The Board’s varied committees, task forces, and panels annually engage about 7,000
engineers, scientists, and other transportation researchers and practitioners from the public and private sectors and academia, all
of whom contribute their expertise in the public interest. The program is supported by state transportation departments, federal
agencies including the component administrations of the U.S. Department of Transportation, and other organizations and individuals
interested in the development of transportation.
Contents
vi
Appendix A. Annotated Bibliography of Influence of Unbound Layers and Subgrade ............. A-1
Appendix B. Definitions of Model Parameters of Unbound Layer and Subgrade Models ........ B-1
Appendix C. Evaluation and Screening of Unbound Layer and Subgrade Models.................... C-1
Appendix D. Moisture-Sensitive, Stress-Dependent, and Cross-Anisotropic Resilient
Modulus ............................................................................................................................. D-1
Appendix E. Slab-Base Interface Shear Bonding Model.............................................................E-1
Appendix F. Sensitivity Analysis of Modulus of Subgrade Reaction Model .............................. F-1
Appendix G. Determination of Stress State in the Faulting Model ............................................ G-1
Appendix H. Categorization of Traffic Loads State in the Faulting Model ............................... H-1
Appendix I. Subgrade Subroutine for Flexible and Rigid Pavements .......................................... I-1
Appendix J. Unbound Base Course Subroutine for Flexible Pavements..................................... J-1
Appendix K. Unbound Base Course Subroutine for Rigid Pavements ...................................... K-1
Appendix L. Rigid Pavement Structure Model Subroutine .........................................................L-1
Appendix M. Conversion 2-Layer Model for Rigid Pavements Subroutine ............................. M-1
vii
ix
Figure 19. Predicted MR Model Coefficients of Plastic Base Materials from Physical
Properties Using the ANN Approach. .............................................................................. 50
Figure 20. Predicted MR Model Coefficients of Plastic Subgrade Materials from Physical
Properties Using the ANN Approach. .............................................................................. 51
Figure 21. Predicted MR Model Coefficients for Non-plastic Base Materials from
Physical Properties Using the ANN Approach. ................................................................ 52
Figure 22. Predicted MR Model Coefficients for Non-plastic Subgrade Materials from
Physical Properties Using ANN Approach. ...................................................................... 53
Table 15. Prediction Accuracy of SWCC Fitting Parameter Models. .......................................... 54
Figure 23. Comparison of ANN Model Predicted Resilient Moduli against Measured
Values for Base Materials. ................................................................................................ 55
Figure 24. Comparison of ANN Model Predicted Resilient Moduli against Measured
Values for Subgrade Materials.......................................................................................... 55
Figure 25. Comparison of Measured versus Predicted Resilient Moduli Using Regression
Models............................................................................................................................... 56
Table 16. Input Parameters Collected from Literature for Model Validation. ............................. 57
Figure 26. Validation of Measured versus ANN Predicted MR at Various Stress Levels for
Collected Unbound Materials. .......................................................................................... 57
Figure 27. Foundation Models for Rigid Pavement (a) Winkler Model; (b) Pasternak
Model. ............................................................................................................................... 58
Figure 28. Flowchart of Corrected Base Modulus due to Cross Anisotropy. ............................... 61
Figure 29. Illustration of Transformed-Section Method for a Cooperated Concrete Slab
and Base Course System. .................................................................................................. 62
Table 17. Steps of Moment of Inertia Calculation for a Cooperated Slab and Base
System. .............................................................................................................................. 62
Figure 30. Illustration of In-situ Shear Stress in the Base Course on the PCC-Base
Interface Using a Mohr-Coulomb Failure Envelope. ....................................................... 63
Table 18. Collected MR Coefficients and the Simulated Stress Values at the Mid Depth of
Base Layer for LTPP Section 27-4054. ............................................................................ 65
Figure 31. Base Resilient Modulus Convergence with Iteration Number. ................................... 65
Figure 32. Formulation of Friction Angle from Mohr Coulomb Failure Envelope for (a)
Treated Base; (b) Unbound Base. ..................................................................................... 66
Figure 33. Comparison of Calculated Slab-Base Interface Degree of Bonding Ratio with
the BBF Approach for (a) Treated Base; and (b) Unbound Base Layer. .......................... 68
Figure 34. Sensitivity of Slab-Base Degree of Bonding on Wheelpath Fault (mm). ................... 68
Figure 35. Comparison of Modified versus LTPP k-values. ........................................................ 69
Figure 36. (a) Schematic Plot of a Typical Pavement Structure; (b) Axisymmetric Model
of Pavement in ABAQUS. ................................................................................................ 71
Table 19. Selected Range of Input Parameters in ANN Training Data Set. ................................. 72
Figure 37. Illustration of Three-Layered Neural Network Architecture for k-values. ................. 72
Figure 38. Target and Output k-values for Training, Validation, and Overall Data Sets for
1296 Simulation Cases...................................................................................................... 73
Figure 39. Comparison of Calculated versus Predicted Modified k-values. ................................ 73
Figure 40. Modified k-values at 0, 0.3, 0.6, and 1 Degree of Bonding for Selected LTPP
Pavement Sections. ........................................................................................................... 74
Figure 41. Schematic Plot of Mohr’s Circle Showing Dependence of Shear Strength on
Matric Suction. .................................................................................................................. 75
Figure 42. Comparison of Predicted and Measured Shear Strength Model Parameters (a)
c’ and (b) φ’. ..................................................................................................................... 76
Figure 43. Illustration of Three-Layer Neural Network Architecture to Predict c’
Parameter. ......................................................................................................................... 77
Figure 44. Target and Output c’ Values for Training, Validation, and Overall Data Sets
for 432 Subgrade Soils. ..................................................................................................... 78
Figure 45. Comparison of Predicted and Measured Permanent Deformation Model
Parameters (a) 0 , (b) , (c) , (d) m, and (e) n. ........................................................... 81
Figure 46. Schematically Illustration of the Development of Faulting. ....................................... 83
Figure 47. Illustration of Field Faulting Data, including the Critical Faulting Depth. ................. 84
Figure 48. Comparison between Measured and Predicted Faulting Depth. ................................. 86
Table 20. Results of Multiple Regression Analysis for Coefficients in the First Faulting
Model. ............................................................................................................................... 88
Figure 49. Comparison between Measured and Predicted Coefficients in Faulting Model. ........ 89
Figure 50. Comparison between Measured and Predicted Faulting at Inflection Point. .............. 90
Table 21. Results of Multiple Regression Analysis for Faulting at Inflection Point. ................... 90
Figure 51. Mean Critical Faulting Depth with or without Dowels. .............................................. 91
Figure 52. Plastic Deformation Curve before the Inflection Point. .............................................. 92
Figure 53. Comparison between Measured versus Predicted Faulting before Critical
Depth. ................................................................................................................................ 94
Table 22. Results of Multiple Regression Analysis for Coefficients in the Load-Related
Faulting Model. ................................................................................................................. 95
Figure 54. Comparison between Measured and Predicted Coefficients in the Permanent
Deformation Faulting Model. ........................................................................................... 96
Figure 55. The 2D Axisymmetric Model Used in FE Analysis. ................................................... 98
Table 23. Material Properties of Pavement Layers. ...................................................................... 98
Table 24. Example of Base Material Information. ....................................................................... 99
Table 25. Material Parameters of Pavement Layers. .................................................................. 101
Figure 56. (a) Tensile Strain at the Bottom of the Surface and (b) Average Compressive
Strain in the Centerline of the Base under Different Loading Levels............................. 104
Figure 57. Vertical Modulus Contours in the Base Layer under the Loading Level (a)
201 kPa, (b) 566 kPa, (c) 756 kPa, and (d) 1006 kPa (Unit: MPa)................................. 105
Figure 58. Vertical Modulus Contours in the Base Layer at (a) Dry Condition, (b)
Medium Condition, and (c) Wet Condition (unit: MPa)................................................. 105
Figure 59. (a) Tensile Strain at the Bottom of the Surface; (b) Compressive Strain at the
top of the Subgrade the Pavement with Different Anisotropy (1/n) of Base Layers. ..... 106
Figure 60. Vertical Modulus Contours in the Subgrade at Different Loading Levels. ............... 106
Table 26. Base Material Information for Rut Depth Calculation. .............................................. 107
Figure 61. Rut Depth in the Base Layer Using Different Models. ............................................. 107
Figure 62. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure; (b) Rut Depth in the Base at Different Loading Levels..................................... 108
Figure 63. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure; (b) Rut Depth in the Base at Different Thickness of the Asphalt Layer. .......... 109
xi
Figure 64. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure and (b) Rut Depth in the Base at Different Thickness of the Base Layer. ......... 110
Figure 65. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure and (b) Rut Depth in the Base at Different Moisture Conditions of the
Base Layer. ..................................................................................................................... 111
Figure 66. Pavement Structure for LTPP Section 1-3028. ......................................................... 112
Table 27. Basic Information of LTPP Section 1-3028 for Sensitivity Analysis......................... 112
Figure 67. Effect of Use of Dowels on Faulting Based on (a) Full Faulting Model and (b)
Load-Related Faulting Model. ........................................................................................ 113
Figure 68. Effect of Types of Base Layer on Faulting Based on (a) Full Faulting Model
and (b) Load-Related Faulting Model............................................................................. 113
Figure 69. Effect of Thickness of Base Layer on Faulting Based on (a) Full Faulting
Model and (b) Load-Related Faulting Model. ................................................................ 114
Figure 70. Effect of Freeze-Thaw Cycles on Faulting Based on (a) Full Faulting Model
and (b) Load-Related Faulting Model............................................................................. 115
Figure 71. Effect of Number of Days with Temperature Greater than 32°C on Faulting
Based on (a) Full Faulting Model and (b) Load-Related Faulting Model. ..................... 115
Figure 72. Effect of Climatic Zone on Faulting Based on (a) First Faulting Model and (b)
Second Faulting Model. .................................................................................................. 116
Table 28. Selected LTPP Pavement Sections and FWD Backcalculated Modulus Values
for Each Layer................................................................................................................. 117
Table 29. Calculated MR Values at the Mid-depth of Base Layer at Different Moisture
Conditions. ...................................................................................................................... 118
Figure 73. Sensitivity of Degree of Bonding on Subgrade k-value Using (a) ANN Model
and (b) Pavement ME Design Model.............................................................................. 119
Figure 74. Sensitivity of Moisture on Subgrade k-value Using (a) ANN Model and (b)
Pavement ME Design Model. ......................................................................................... 120
Figure 75. PCC Slab-Base Interface Bond Sensitivity on (a) Tensile Stress at Top of Slab;
(b) Tensile Stress at Bottom of Slab; and (c) Deferential Deflection on Transverse
Joints. .............................................................................................................................. 123
Figure 76. Base Layer Moisture Sensitivity on (a) Tensile Stress at Top of Slab; (b)
Tensile Stress at Bottom of Slab; and (c) Deferential Deflection on Transverse
Joints. .............................................................................................................................. 125
xii
CHAPTER 1. INTRODUCTION
INTRODUCTION
This research project proposes and develops enhancements to the Pavement ME Design
procedures for both flexible and rigid pavements, which will better reflect the influence of
subgrade and unbound layers (properties and thicknesses) on the pavement performance. These
enhancements include modifications of the models contained in Pavement ME Design and/or the
development of new models.
RESEARCH SCOPE AND APPROACH
The project required the development and synthesis of several components: (a) evaluation
and screening of unbound layer and subgrade models; (b) development of soil water
characteristics curve models of base and subgrade; (c) development of resilient modulus (MR)
models of base and subgrade; (d) development of modulus of subgrade reaction model; (e)
development of faulting model of base layers; and (f) conduction of performance prediction and
sensitivity analysis.
ORGANIZATION OF THE REPORT
The fifth chapter presents interpretation, appraisal, and application of the findings
discussed in the fourth chapter. More specifically, the sensitivity analysis of the proposed
models will be presented. The pavement performance predicted by the proposed models
will be compared with those in the current version of Pavement ME Design.
The sixth chapter presents conclusions and suggested further research.
The main body of the report overviews the approach and results of this project. More
detailed discussions of the topics in the report are contained in Appendices A through M.
The AASHTOWare Pavement ME Design software and MEPDG are practical tools for
the pavement design and analysis based on ME principles. They predict multiple performance
indicators for flexible and rigid pavements, including the following (1):
Flexible pavements:
o Total rut depth of the asphalt layers, aggregate base, and subgrade.
o Load-related cracking (alligator cracking and longitudinal cracking).
o Thermal cracking.
o Smoothness (International Roughness Index [IRI]).
Rigid pavements:
For each performance indicator above, Pavement ME Design has a distress prediction
model that requires inputs from different layers in a pavement structure. Table 1 shows the inputs
from unbound layers and subgrade to account for the influence of these underlying layers.
MR
Load-related Cracking Percent passing No. 200
Thickness
(alligator and longitudinal Poisson’s ratio
Poisson’s ratio
cracking) Groundwater depth
SWCC
SWCC
Plasticity index
MR MR
Thickness Percent passing No. 200
Thermal Cracking
Poisson’s ratio SWCC
SWCC Poisson’s ratio
MR
MR
Smoothness (IRI) Percent passing No. 200
Thickness
SWCC
Thickness
Transverse Cracking MR Groundwater depth
(JPCP) Erodibility index MR
Loss of friction
MR
Erodibility index
Faulting (JPCP) MR
Thickness
Load transfer efficiency (LTE)
Rigid Pavement
MR
MR
Punchouts (CRCP) Base slab friction
Groundwater depth
Thickness
Base slab friction
MR MR
Crack Width (CRCP)
Thickness Groundwater depth
LTE
MR
Erodibility index
Smoothness (IRI) (JPCP) MR
Base slab friction
Thickness
MR
Smoothness (IRI) (CRCP) Base slab friction MR
Thickness
To find the reasons for these problems, a better understanding of how the
properties/thickness of unbound layers and subgrade affect pavement performance is needed, and
this is detailed next.
INFLUENCE OF UNBOUND LAYERS AND SUBGRADE ON PERFORMANCE OF
FLEXIBLE AND RIGID PAVEMENTS
Table 1 presents the inputs of unbound layers and subgrade required in Pavement ME
Design for predicting the performance of flexible and rigid pavements. However, besides these
parameters, recent studies have identified the pavement performance to be significantly affected
by other characteristics of the underlying layers. According to a comprehensive literature review,
researchers divided the factors into the following categories:
Material properties (e.g., modulus, shear strength).
Material behaviors responding to traffic and environmental (temperature and moisture)
conditions (e.g., permanent deformation and erosion).
Structural characteristics (e.g., thickness of unbound layers).
Behaviors
Indicators
Modulus
Thickness
Permanent
Cross‐ Moisture Shear Strength
Magnitude Deformation
Anisotropy Sensitivity
The amount of
Total rutting
permanent Rutting
Shear strength increases as
Total Rutting
deformation decreases
Total rutting directly affects permanent
significantly with increase
decreases as total rutting; it deformation
increases of the
modulus decreases as of unbound
when thickness of
increases (3, 4) shear strength base course
anisotropic the base layer
increases (6–11) increases
properties are (4)
(12)
used (5)
Use of cross‐ A larger shear
The resistance
Load‐related Cracking
anisotropy of strength
to load‐
(Alligator and
Longitudinal)
susceptibility; thickness of
Thermal
high degree the base layer
cracking is
of moisture possibly helps
accelerated by N/A N/A N/A
causes alleviate the
loss of modulus
decrease of severity of
(14)
the modulus thermal
(15–19) cracking (20)
Permanent
deformation
Cross‐
of unbound
Smoothness (IRI)
anisotropy Change of IRI
IRI decreases base is a
affects total High shear diminishes
with the major
rutting and strength results in with increase
increase of distress
cracking, low IRI values (1, of thickness of
base modulus resulting in
which leads to 11) the base layer
(4) increase of
the change of (4)
surface
IRI (4)
roughness
(9)
Modulus
Shear Permanent
Cross‐ Moisture Erosion
Magnitude Strength Deformation
Anisotropy Sensitivity
Thickness
Transverse Cross‐
Transverse Cracking
High shear of baser
cracking anisotropy
strength layer
would be greatly
prevents directly
(JPCP)
promoted affects
occurrence of N/A N/A affects
with low stress/
transverse amount of
modulus of strain and
cracking (22, transverse
unbound cracking (5,
23) cracking
layers (3, 21) 13)
(3)
Loss of Greater
Increase of
modulus of permanent Faulting
Faulting (JPCP)
shear Development
unbound deformation decreases
Modulus has strength of erosion
base course of unbound with high
N/A a high inhibits the accelerates
lead to base leads to base
sensitivity in development faulting (24–
development higher thickness
change of of faulting 26)
of faulting potential of (3)
matric suction (24)
(24) faulting (27)
that
Increase of
Reduction of represents Potential for
Punchouts (CRCP)
thickness is
modulus of moisture punchouts is
Erosion an
unbound susceptibility; greater when
intensifies effective
base course N/A high degree of shear N/A
punchout (25, method to
causes moisture strength
28, 32) control
punchouts causes decreases
punchouts
(28–30) decrease of (25, 28)
(3)
the modulus
A higher
(14, 19, 31) Unbound Increase of
modulus of Development
(JPCP and
layers with thickness
CRCP)
unbound of erosion
LTE
Cross‐
shear deformation decreases
IRI decreases anisotropy
strength of Erosion of unbound with
with increase affects
base layer aggravates IRI base increase in
in modulus of cracking
diminishes (25, 34, 37) increases base layer
base layer (3) and so IRI
roughness roughness thickness
(5, 13)
(35, 36) (9) (3)
Total rutting increases as
shear
decreases as Use of nonlinear permanent
strength of
modulus anisotropic model deformation of
subgrade
increases (18) of subgrade A higher soil subgrade
increases
affects suction augments (46)
(46)
stress/strain generates a
Resistance to
distribution, and larger modulus
load‐related Lower permanent
(Alligator and
Longitudinal)
then influences of subgrade
Load‐related
cracking would deformation of
Cracking
the inputs in (40–45)
be enhanced subgrade reduces
distress prediction N/A
with increase of the probability of
models (5, 38, 39)
modulus of load‐related
subgrade (2, 3, cracking (39)
18)
Thermal cracking
is related to
shrinkage of
Thermal Cracking
supporting
subgrade soils;
N/A N/A N/A N/A high permanent
deformation
would reduce the
resistance to
thermal cracking
(14)
Use of nonlinear
anisotropic model Decrease of
Smoothness (IRI)
of subgrade shear
Soil suction is a High permanent
IRI has a affects strength of
major factor for deformation
negative relation stress/strain subgrade
prediction of exacerbates the
with modulus of distribution, and results in loss
subgrade roughness of
subgrade (4, 47) then influences of
modulus (41) pavement (9, 14)
the inputs in smoothness
distress prediction (1)
models (5, 38, 39)
Increase of permanent
shear deformation
Increasing modulus of
strength of leads to loss of
subgrade would
subgrade supporting
reduce transverse
raises the layers, which
cracking (3, 21) Cross‐anisotropy
resistance of could cause
affects stress/strain
transverse development
and then influences
cracking (1) of transverse
the inputs in
cracking (1)
distress prediction
Higher shear High
models (5, 13)
Faulting (JPCP)
strength of permanent
Increase in modulus of
subgrade deformation of
subgrade causes a
layer helps subgrade
decrease in faulting
Soil suction is a improve increases the
(48)
major factor resistance to possibility of
for the faulting (27) faulting (1, 49)
prediction of Increase of
modulus of permanent
Punchout is
Punchouts (CRCP)
subgrade deformation of
accelerated
materials; a subgrade
Punchout increases with lower
higher soil makes poorer
with low k‐value of N/A shear
suction LTE; thus leads
subgrade (28, 29) strength of
generates a to
subgrade
larger modulus development
(28)
of subgrade of punchouts
(41) (49)
Loss of LTE
LTE (JPCP and
Increase of
occurs with
LTE is increased by shear
CRCP)
high
high modulus of N/A strength
permanent
subgrade (25) improves LTE
deformation of
(25)
subgrade (28)
Improvement Rutting
Smoothness (IRI)
(JPCP and CRCP)
of shear generated from
Cross‐anisotropy strength of permanent
IRI value diminishes
affects subgrade deformation of
with the increase in
cracking/faulting layer could subgrade is
subgrade modulus (3)
and so IRI (5, 13) increase associated with
smoothness increased
(36) roughness (9)
o Current Pavement ME Design models for the base course of both flexible and rigid
pavements.
o Modulus models (Table 6), particularly those that incorporate the effects of the level
of moisture in addition to the traffic-related stresses. The anisotropy of the base
course is reflected in a separate model for the vertical modulus and the horizontal
modulus.
o Permanent deformation models (Table 7), which are sensitive to the changes of
properties and thickness of the base course, and particularly the ones that predict a
larger range of deformation that is close to the behavior of real unbound layer
materials.
o Shear strength models (Table 8), especially those models that include the effects of
moisture and traffic-related stresses on the shear strength of the base course.
o Erosion and faulting models (Table 9).
o Thickness sensitive models (Table 11). This category refers to the models that have
an influence on the thickness of the base course, which further affects the
performance of the pavement. The moisture-sensitive, stress-dependent, and cross-
anisotropic modulus models; moisture-sensitive shear strength models; stress-
dependent ME permanent deformation models; and ME erosion models could
contribute to this category.
Subgrade models:
o Current Pavement ME Design models for the subgrade of both rigid and flexible
pavements.
o Modulus models (Table 6), particularly those that incorporate the effects of the level
of moisture in addition to the traffic-related stresses on the stiffness of the subgrade.
o Permanent deformation models (Table 7), especially those that predict a larger range
of deformation that is close to the behavior of real soils.
o Shear strength models (Table 8), particularly the ones that include the effects of
moisture and traffic-related stresses on the shear strength of the subgrade. This
property becomes important when attempting to reflect the performance of a
pavement under heavy load, on a moisture-susceptible soil, or where there is poor
drainage or there is slippage between the base course and subgrade on which it rests.
o Foundation models (Table 10) that significantly reduce errors and variations of
critical rigid pavement responses.
10
Nonlinear Stress- k 1
M k 2
Granular Base (51)
dependent Model
R
Nonlinear Stress-
MR k2 k3 k1 d k1 d
Subgrade Soil (52)
MR k2 k4 d k1 k1 d
dependent Model
I
2 k k3
Nonlinear Stress- Granular Base/
dependent Model
M R k1 Pa 1 oct Subgrade Soil
(55)
Pa Pa
I
k k3
2
Nonlinear Stress- Granular Base/
dependent Model
M R k1 Pa 1 oct 1 Subgrade Soil
(15)
Pa Pa
Nonlinear Stress- I 2 J
dependent Model
M R M Pa 1 R 2 Granular Base (56)
Pa Pa
M ba
Moisture-sensitive log R a Granular Base/
b
1 exp ln km S Sopt
M Ropt (57)
Model Subgrade Soil
a
Moisture-sensitive MR k2 k3 k1 d ks ua uw
and Stress- Subgrade Soil (58)
dependent Model MR k2 k4 d k1 ks ua uw
Moisture-sensitive
I 3k 4 oct
2 k k3
Granular Base/
and Stress- M R k1 Pa 1 Subgrade Soil
(59)
dependent Model Pa Pa
Moisture-sensitive
I 3 fhm oct
k2 k3
Granular Base/
and Stress- M R k1 Pa 1 Subgrade Soil
(60)
dependent Model Pa Pa
k2
I
Moisture-sensitive I1 3 f hm 1 oct
oct
k3
3 Granular Base/
and Stress- M R k1 Pa (14)
Pa Pa Subgrade Soil
dependent Model
Moisture-sensitive
MR k1 d wm 2
k
and Stress- Subgrade Soil (41)
dependent Model
11
Moisture-sensitive k2
w m oct
k3
Moisture-sensitive 3k6
k2
oct
k3
k2 k3
Moisture-sensitive 3k4 SVw oct Granular Base/
and Stress- M R k1Pa 1 Subgrade Soil
(62)
dependent Model Pa aP
k2 k3
I
Stress-dependent M RV k1 Pa 1 oct 1
and Cross- Pa Pa Granular Base (63)
H
anisotropic Model M G
n RV ; m VHV
MR MR
k2 k3 k5 k6
oct
M RV k1 Pa oct ; M RH k 4 Pa
Stress-dependent Pa Pa Pa Pa
and Cross- k8
k9 Granular Base (64)
anisotropic Model GVH k 7 Pa oct
Pa Pa
k2 k3
Moisture- I 3 fhm oct
sensitive, Stress- M RV k1 Pa 1
dependent, and Pa Pa Granular Base (31)
Cross-anisotropic MH G
s RV ; r VHV
Model MR MR
k1 1.3577 0.0106 %clay 0.0437wc
k2 0.5193 0.0073P4 0.0095P40
0.0027P200 0.003LL 0.0049wopt
Regression
Models for Stress- k3 1.4258 0.0288P4 0.0303P40
Subgrade Soil (65)
dependent Model 0.0521P200 0.0251 %silt 0.0535LL
Coefficients
0.0672wopt 0.0026 opt
wc
0.0025 s 0.6055
wopt
Regression
Models for ln k1 137.19 13.60ln d 4.35ln A 0.62ln T
Moisture-sensitive
k2 36.14 0.04ofc 3.81ln A 0.22as 0.77ln T Granular Base (31)
and Stress-
dependent Model k3 4.39 0.45ln d 0.01pfc 0.05as 0.15ln T
Coefficients
12
Empirical
Mr 2555 CBR
Regression Model
0.64 Granular Base/
(66)
Subgrade Soil
for MR
Empirical
Regression Model Mr 1155555R Granular Base/
Subgrade Soil
(66)
for MR
Empirical
a Granular Base/
Regression Model M r 30000 i (67)
0.14 Subgrade Soil
for MR
0.64
Empirical
75 Granular Base/
Regression Model M r 2555 (67)
1 0.728 wPI
Subgrade Soil
for MR
Empirical 0.64
292 Granular Base/
Regression Model Mr 2555 1.12 Subgrade Soil
(66)
for MR DCP
Non-stress- 1 N Granular
P
N
dependent ME Base/ (68)
Model r N Subgrade Soil
Non-stress- p Granular
dependent ME aN b Base/ (69)
Model r Subgrade Soil
Non-stress-
Granular
dependent ME Base/ (70)
Model
p 0e N
Subgrade Soil
Non-stress-
dependent ME p z p,z0 ekz Subgrade Soil (71)
Model
Stress- Granular
dependent ME p s 0 e N v Base/ (67)
Model r Subgrade Soil
p k6 oct
log a a a
Stress- r 0 1 Pa 2 Pa
dependent ME Granular Base (69)
Model k6 oct
b0 b1 b2 .log N
Pa Pa
R
p CN b
Stress- 1 R
dependent ME q Granular Base (72)
Model b d c '
q
f
13
D
Stress-
dependent ME p AN f B C
Granular Base (11)
max
d
Model
Stress-
J
I1 K
m
dependent ME p 0e N n Granular Base (73)
2
Model
log 0 0.80978 0.06626Wc 0.003077 106 Er
Regression
Models for
r
Pavement ME log 0.9190 0.03105Wc 0.001806 1.5 106 Er Granular Base (70)
Design Model
Coefficients log 1.78667 1.45062Wc 3.784 1042
2.074 103Wc2 1.05 106 Er
9
e
0.15 e 10 20
0
Regression
r 2
Models for
Pavement ME log 0.61119 0.017638Wc Granular Base (66)
Design Model 1
Coefficients
4.89285
109
1 10
9
ln 0 10.24 0.03 MBV 0.10 pfc
Regression 0.88a A 3.95 ln T
Models for ln 6.74 0.02 MBV 0.04 pfc 0.85aG
Pavement ME Granular Base (74)
Design Model 0.03G 0.13aT
Coefficients ln 10.17 2.75 ln d 0.05 pfc 2.00 aG
1.61ln A 0.34 aT
14
Moisture-
sensitive Model
f c'n a tan' a w tanb Subgrade Soil (79, 80)
15
Empirical Model m
C2 ni Granular Base (87)
Percent erosion damage 100
i 1 Ni
Empirical Model Table-based Erodibility Class Assessment Granular Base (66)
Granular
ME Model D N v Base/ (24)
f %Erosion f0e Subgrade Soil
16
17
18
developed ANN models were validated by comparing a new data set collected from both the
NCHRP 9-23A project and other literature sources to the model predictions.
EQUILIBRIUM SUCTION OF BASE AND SUBGRADE FOR FLEXIBLE AND RIGID
PAVEMENTS
Researchers developed a correlation model to the predict equilibrium suction in base and
subgrade layers based on Thornthwaite moisture index (TMI). TMI is a climatic parameter that
characterizes the annual moisture balance of a specified location based on precipitation,
evaporation, water storage, deficit, and runoff. A TMI contour map of continental United States
was generated in geographic information system (GIS) platform to determine the moisture index
for each specific mapunit collected from the Natural Resources Conservation Service (NRCS)
database. Similarly, pF of all mapunits were determined using a fundamental approach of
oscillating moisture transient equation proposed by Mitchell (93). Finally, a prediction equation
was developed to predict pF from TMI. It showed a good prediction accuracy and had a
relationship with the plasticity index (PI) of soil.
RESILIENT MODULUS OF BASE AND SUBGRADE FOR FLEXIBLE AND RIGID
PAVEMENTS
Researchers developed a modified k-value model to take account of the shear interaction
between the Portland Cement Concrete (PCC) slab and the base course. Formulation of the
modified k-value model contained four steps: (a) correction of the base modulus due to cross-
anisotropy; (b) development of a submodel for slab-base equivalent thickness; (c) development
of a formula for interface shear bonding between the slab and base; and (d) determination of the
k-value using the calculated shear bonding and the deflection patterns of falling weight
deflectometer (FWD). Many rigid pavement structural and strength properties and the
corresponding FWD deflection patterns data were collected from the LTPP database and used to
calculate the modified k-value. These were compared against the k-values using the previously
developed backcalculated best-fit approach. The results showed that the modified k-values
changed significantly due to the consideration of cross-anisotropy and slab-base interface
bonding in the proposed model. Finally, an ANN approach was employed to predict the modified
k-value for various pavement structures, layer moduli, and interface bonding ratios. The FWD
19
deflection pattern for each combination was determined from the finite element (FE) analysis.
The prediction accuracy of the ANN model was also examined by comparing the prediction
results with the calculated modified k-values for the LTPP pavement structures. The comparison
results indicated that the ANN model accurately predicts the modified k-values for the given
pavement structures, layer moduli, and interface bonding ratios.
SHEAR STRENGTH OF BASE AND SUBGRADE LAYERS FOR FLEXIBLE AND
RIGID PAVEMENTS
A new ME permanent deformation model was proposed by the research team for base
and subgrade layers. The advantage of the new model over the Pavement ME Design model is
that it is capable of predicting the permanent deformation behavior at different shear strength
conditions. However, the predictability of this model depends largely on the accuracy of the
coefficients of the model. In this study, repeated and monotonic load triaxial test data were
collected from the literature and the coefficients of permeant deformation were calculated
accordingly. Soil physical properties such as gradation, Atterberg limits, dry density and
moisture content data were collected for base and subgrade materials and regression analysis was
conducted to predict the permanent deformation model coefficients i.e., ϵ0, ρ, β, m and n.
FAULTING OF BASE FOR RIGID PAVEMENTS
A novel ME model was developed to estimate faulting over time. Two stages of the
process of faulting are revealed by the model. To distinguish the two states of faulting, an
inflection point can be directly determined by this model and can indicate the beginning of
erosion for the concrete pavement design. In addition, the faulting depth before the inflection
point is a critical depth due to the permanent deformation of underlying layers. Thus, this first
phase of faulting is principally an effect of the permanent deformation of the supporting base
course. This second ME faulting model predicts faulting before reaching the critical depth using
axle load distributions.
In summary, two faulting prediction models were developed. One is to predict the entire
faulting development over time. This model of the full faulting curve shows that there is an
inflection point in the faulting curve. Before reaching the inflection point, the accumulation of
faulting is caused by the permanent deformation of the supporting layers. After passing the
inflection point, faulting accelerates due to the action of erosion. The second model is to predict
the faulting depth before inflection point with traffic. The proposed models were proven to be
considerably accurate and reliable by using LTPP data. The coefficients in the models are
statistically calibrated with performance-related factors using multiple regression analysis.
20
21
CHAPTER 4. FINDINGS
INTRODUCTION
This project addressed the issues associated with the sensitivity of pavement performance
to base layers and subgrade for flexible and rigid pavements. Therefore, necessary enhancements
to the Pavement ME Design procedures are needed to better reflect the influence of subgrade and
unbound layers (properties and thicknesses) on the performance. These enhancements include
modifications of the models contained in Pavement ME Design and the development of new
models. Following the research plan presented in Chapter 3, this chapter gives detailed
explanations for each aspect in the proposed plan.
SWCC OF BASE AND SUBGRADE FOR FLEXIBLE AND RIGID PAVEMENTS
The SWCC represents a relationship between soil suction and water content (or degree of
saturation). The soil suction due to the presence of water between soil particles is an important
indicator, which is directly related to the strength, volume change, and fluid flow characteristics
of unsaturated soil material (77). The unbound material is usually in an unsaturated condition
when it is exposed in the field. Currently, there are two common ways to determine the SWCC
of unbound material. One way is by conducting laboratory tests (e.g., filter paper test and
pressure plate test) to measure the matric suction at different water contents. This experimental
method is usually time consuming and requires special test equipment and test expertise. An
alternative way is to correlate the suction-water relationship to other unbound material
indicators, such as grain size distribution (GSD), particle diameter, porosity, liquid limit (LL)
and PI, mean annual air temperature (MAAT), etc. There are many prediction methods available,
which can be classified into three categories (94):
Category 1: Direct prediction of SWCC, which is a curve-fitting method to correlate the
soil properties (e.g., GSD and porosity) to the matric suction at the different tested water
contents (95–98).
Category 2: Prediction of SWCC model parameters, which is a statistical method to
correlate the soil properties (e.g., GSD, LL, PI, and MAAT) to the parameters of existing
SWCC models (e.g., Fredlund-Xing equation). These estimated model parameters are
further used to generate the SWCC of unbound material (94, 99–102).
Category 3: Development of a micromechanics-based SWCC model, which uses the
microstructure of unbound material (e.g., porosity distribution) and the water-particle
contact (or capillary rise) model to determine the matric suction (103–105).
Among these approaches, Category 1 usually has a low prediction accuracy, since it
ignores the existing relationship between matric suction and water content. Category 2
recognizes the existing SWCC models and uses an indirect method to predict the SWCC via soil
properties. The prediction accuracy of Category 2 relies on the selected statistical method and the
quantity of the database used in the analysis. However, most of the aforementioned models were
developed based on either a simple regression analysis or a limited number of test results (101).
Compared to Categories 1 and 2, Category 3 is still in a state-of-development status, which is
limited to the soils with large pores (94). Therefore, approaches in Category 2 are more accurate
and feasible to predict the SWCC of unbound material, as long as the issues of statistical method
and quantity of database are solved.
23
Currently, ANN approach becomes a more popular tool for the development of prediction
models. Compared to regression models, the main advantage of the ANN approach is that it can
capture nonlinear and complex scattered relationships between input and output parameters. In
the past, several studies attempted to use the ANN approach to predict the SWCC of soil, which
mainly focused on the Category 1–based method (i.e., direct prediction of SWCC) (106–109).
Currently, there is no Category 2–based ANN model available to predict the SWCC of soil. The
main reason is that the development of Category 2–based ANN model requires a much larger
database of SWCC results. In addition, most of these existing ANN models dealt with
unsaturated soil, but few ANN models can predict the SWCC of unbound granular material that
is commonly used in pavement engineering.
To overcome the aforementioned problems, a Category 2–based ANN approach is
recommended for estimating the SWCC of unbound material. In this project, a large database
collected from the NCHRP 9-23A project is used to develop the ANN models. The Fredlund and
Xing equation is used to estimate the relationship between matric suction and water content, as
shown in Eqs. 4.1 and 4.2 (110):
1
S C (h) x[ ]
h bf cf
{ln[e ( ) ]}
af
(4.1)
where C(h) is a correction factor defined as:
h
ln(1 )
hr
C(h) 1
1.45 x105
ln[1 ( )]
hr (4.2)
where S is the degree of saturation (unit: %); h is the soil matric suction (unit: psi); and af, bf, cf,
and hr are soil fitting parameters. The four fitting parameters are the outputs of ANN models and
are predicted by the input variables.
where f is a transfer function, which normally uses a sigmoidal, Gaussian, or threshold functional
form, and wji are the unknown weight factors. Developing a neural network model specifically
refers to the determination of the weight factors wji in Eq. 4.3. In this project, the output variables
Yj represent the four fitting parameters in the Fredlund-Xing equation (i.e., af, bf, cf, and hr). The
24
input variables Xi are selected from the SWCC-related material indicators, which will be
elaborated in the following section. In general, the development of ANN models includes two
critical steps: 1) data collection, and 2) construction of ANN architecture.
Data Collection
The database used in this project is collected from the NCHRP 9-23A project entitled “A
National Catalog for Subgrade SWCC Default Inputs and Selected Soil Properties for Use with
the ME-PDG” carried out at Arizona State University in 2010 (111). In addition to the measured
SWCC fitting parameters, the database includes material size distribution, soil LL, PI, and
saturated volumetric water content. To better evaluate the influence of soil plasticity, the
collected database is separated into two groups: plastic soil data set (PI>0), and non-plastic soil
data set (PI=0).
Construction of ANN Architecture
As shown in Figure 1a and Figure 1b, two three-layered neural network architectures
consisting of one input layer, one hidden layer, and one output layer were constructed for plastic
soil and non-plastic soil, respectively. The input variables in the ANN models were selected
based on previous studies and the mechanics of unsaturated soil. The SWCC of a soil is greatly
dependent on its pore structure (110, 112, 113). The pore size is related to the height of each
capillary tube using the Young–Laplace equation, which is equivalent to a suction value (77).
The capillary height of the corresponding pore can be expressed in terms of the radius of that
pore. Thus, soil GSD (i.e., percent passing No. 4 sieve and percent passing No. 200 sieve) and ϴ,
and Ѱ, and saturated volumetric content were selected as input parameters. Saturated volumetric
water content is a measure of the total porosity in soil structure. The equilibrium soil suction of
plastic soil is proportional to its specific surface area (98). The PI is a simple indicator that
reflects the specific surface area of soil. The relationship between soil surface area and PI
changes, particularly when the LL is above 30 percent to 40 percent (114). Hence, LL was also
selected as an input parameter in the ANN models. Previous studies investigated the effects of
temperature on soil suction-water retention relationship (115–117). Soil water is composed of
continuous water and isolated pockets of water. When the temperature increases, water flows
from isolated pockets to the continuous phase, which results in a shift in the SWCC. In this
study, the MAAT is identified as one of the inputs for ANN models and is collected from the
NRCS soil data mart.
In Figure 1a, the input variables for plastic soil include percent passing No. 4 sieve,
percent passing No. 200 sieve, LL, PI, saturated volumetric water content (Sat. vol. wc.) and
local MAAT (Unit: °C). Figure 1b shows that the input variables for non-plastic soil are D30, D60,
D90, ϴ, and Ѱ, saturated volumetric water content, and local MAAT (unit: °C). Herein, Ө and Ѱ
are estimated by fitting a power law model (ӨxѰ) to the curve of the cumulative percent passing
versus sieve size in mm. The hidden layer assigns 20 neurons to establish the connection
between the output layer and the input layer.
Both the non-plastic and plastic soil models use the sigmoidal transfer function, as shown
in Eq. 4.4 (118):
1
f Ii
1 exp Ii
(4.4)
25
The development of ANN models is to predict the SWCC parameters (i.e., af, bf, cf, and
hr) in the Fredlund-Xing equation. These parameters were correlated to the soil physical
properties. In this study, the ANN models were programmed using the Matlab software. The
26
database used for training of ANN models consists of 3600 samples of plastic soil and 250
samples of non-plastic soil.
A statistical analysis was performed to determine the root mean squared error (RMSE)
and R associated with the predicted SWCC fitting parameters. The RMSE and R2 are computed
2
SSres
Coefficient of determination: R 1
2
(4.8)
SStot
Figure 2 and Figure 3 compare the predicted SWCC parameters of plastic and non-plastic
soils, respectively, against the measured ones using the training data sets. As shown in Figure 2
and Figure 3, the predicted af, bf, cf, and hr parameters are closely coincident with the measured
results. This indicates that the developed ANN models have the desirable accuracy to predict the
SWCC for plastic and non-plastic soils. The next section compares these ANN models against
the existing regression models for model prediction accuracy.
27
(a) (b)
(c) (d)
Figure 2. Comparison of Measured versus Predicted SWCC Fitting Parameters Using ANN
Model for Plastic Soils.
28
(a) (b)
(c) (d)
Figure 3. Comparison of Measured versus Predicted SWCC Fitting Parameters Using ANN
Model for Non-plastic Soils.
Several regression models have been developed to predict the SWCC fitting parameters.
In this study, a comparison analysis is performed between the developed ANN models and those
of the existing prediction models, including the Zapata model (94) and Perera model (100). Table
12 lists the equations of these regression models.
The predictability of the SWCC fitting parameters using the regression models above are
analyzed in this study. Table 13a and Table 13b list the computed RMSE and R2 values for each
model. The values of R2 and RMSE indicate the model prediction accuracy. The higher R2 value
and the smaller RMSE value represent a higher prediction accuracy for the developed model
(73). As presented in Table 13a and Table 13b, the RMSE values associated with all of the
existing SWCC parameter prediction models are very high, and the corresponding R2 values are
extremely low. This indicates that the Zapata and Perera models have low prediction accuracy of
the SWCC fitting parameters for both plastic and non-plastic soils. Among these two models, the
computed RMSE and R2 values are comparable with each other. This demonstrates that these
models have the same level of prediction accuracy. Compared to these existing models, the
developed ANN models have much higher R2 values and smaller RMSE values. This indicates
29
that the developed ANN models outperform these existing regression models’ prediction
accuracies.
bf 1.421(wPI) 0.3185
a 2.79 14.11log(D20 ) 1.9x106 P2004.34 7log(D30 )
c f
-0 .2 1 5 4{ ln ( w P I )} 0 .7 1 4 5 0.055D100
hr 500 b f 0 .9 3 6 b 3 .8
D90
b {5.39 0.29ln[P200 ( )] 3D00.57 0.021P2001.19 }m10.1
D10
30 20
m1 , m2
[log(D90 ) -log(D60 )] [log(D30 ) log(D10 )]
cf 0.26e0.758c 1.4D10
1
c log(m21.15 ) (1 )
bf
hr 100
where: D10, D20, D30, D60, D90, D100 = material diameter corresponding to 10%, 20%, 30%,
60%, 90%, and 100% passing by weight of material, respectively
30
Although the developed ANN models can accurately predict the SWCC fitting
parameters, it is still necessary to examine the prediction accuracy of the SWCC. In the
laboratory, these fitting parameters (i.e., af, bf, cf, and hr) in the Fredlund-Xing equation are
calculated by fitting the SWCC with experimentally available suction-saturation data points. A
nonlinear least squared regression analysis is performed to determine the best fit parameters. In
this project, the predicted SWCC fitting parameters are input into the Fredlund-Xing equation to
estimate the soil suction at various water contents. The estimated soil suction values are then
compared against the measured ones to investigate the model prediction accuracy.
To evaluate the suction predictability over the full range of the saturation level, the matric
suction values are calculated at three different degrees of saturation (i.e., 10 percent, 40 percent,
and 80 percent). The predicted matric suction values using different models were compared with
experimental data at the same saturation level. Figure 4 shows the plots of measured versus
predicted matric suction for plastic soils. Figure 4a, Figure 4b, and Figure 4c correspond to
comparisons made by experimental data with the Zapata model, Perera model, and ANN model,
respectively. Compared to the two existing models, the developed ANN model has the least
RMSE value and the highest R2 value. This indicates that the developed ANN model
outperforms other models to accurately estimate the matric suction of plastic soils at various
degrees of saturation. The obtained R2 value of 0.95 demonstrates that the developed ANN
model can accurately predict the SWCC of plastic soil.
Figure 5 presents the measured versus predicted suction for non-plastic soils. As seen
from Figure 5a and Figure 5b, the predicted matric suction values using the existing models
deviate significantly from the measured values. This indicates that the existing models yield
inaccurate matric suction of non-plastic soils. As shown in Figure 5c, the developed ANN model
has an R2 value of 0.91, which significantly improves the prediction accuracy of the matric
suction for the non-plastic soils. Researchers concluded from Figure 4 and Figure 5 that the
developed ANN models are capable of accurately predicting the SWCC of both plastic and non-
plastic soils.
31
32
The validation of the prediction accuracy of the developed ANN models involves two
steps: 1) the validation through the collected data from NCHRP 9-23A database, and 2) the
validation via the independent data from other literature sources.
At first, a new data set of 500 plastic soils and 33 non-plastic soils were selected from the
NCHRP 9-23A database. The selected soil physical and climatic properties were input into the
developed ANN models. Figure 6 and Figure 7 compare the measured SWCC fitting parameters
to those predicted by the ANN models for the selected plastic and non-plastic soils, respectively.
The ANN model predictions have relatively small RMSE and high R2 values for both cases. The
determined R2 values for plastic soils are in the range from 0.44 to 0.87, non-plastic soils show
R2 ranging from 0.49 to 0.87. This indicates that the model predicted SWCC fitting parameters
match well with the measured results. The model predicted fitting parameters are then input into
the Fredlund-Xing equation to estimate the matric suction values at the selected degrees of
saturation.
33
(a) (b)
(c) (d)
Figure 6. Validation of Measured versus Predicted SWCC Fitting Parameters Using ANN
Model for Plastic Soils.
34
(a) (b)
(c) (d)
Figure 7. Validation of Measured versus Predicted SWCC Fitting Parameters Using ANN
Model for Non-plastic Soils.
Figure 8a and Figure 8b plot the measured versus predicted matric suction values at
10 percent, 40 percent, and 80 percent saturation level for plastic and non-plastic soils,
respectively. Results show that the predicted suction values fit well with the measured ones. Both
plastic and non-plastic soil data sets show a R2 value greater than 0.90. This validates that the
developed ANN models provide a desirable prediction accuracy of the SWCC.
35
To further validate the prediction accuracy of the developed ANN models, the
independent data sources (i.e., two plastic and two non-plastic soil data) were collected from the
literature (18, 119, 120). All these soil physical and climatic properties are shown in Table 14a
and Table 14b, which are used as inputs of the developed ANN models.
Table 14. Input Parameters Collected from Literature for Model Validation.
(a) Plastic soil
Reference Soil source #4 sieve #200 sieve LL PI Sat. vol. wc* MAAT(°C)
(119) Louisiana 100 45 23 12 5.4 18.5
(18) Red Lake Falls 100 93.9 29 10 8.17 3.9
Note: *Sat.vol.wc = Saturated volumetric water content
Figure 9a and Figure 9b compare the measured and predicted SWCCs for plastic and
non-plastic soils, respectively. As illustrated, the SWCC curves generated by using the predicted
parameters from the predicted ANN model produce good fits with the experimental curves. In
Figure 9a, the model predicted matric suction values of plastic soils (i.e., Red Lake Falls and
Louisiana soils) are generally coincident with the experimental results. A small deviation is
observed in the residual region for the Louisiana soil. This is due primarily to the effect of the cf
model, which has a lower R2 than the af and bf parameters. Similarly, Figure 9b presents that the
predicted matric suction values of non-plastic soils (i.e., Torpsbruk and Mississippi soils) were in
good agreement with the measured results. Considering the SWCC curve generated from
experiment involves the high level of variability, the difference between predicted and
experimental curves were sufficiently accurate.
36
Figure 9. Comparison of Measured versus Predicted SWCC Curves for Unbound Materials.
EQUILIBRIUM SUCTION OF BASE AND SUBGRADE FOR FLEXIBLE AND RIGID
PAVEMENTS
In Pavement ME Design, the control moisture level is determined from the depth of
ground water table. This approach has limitations when there is no water table data available and
moisture level is controlled by other factors. Researchers proposed a new approach stating that
when the water table is deeper than the moisture level underneath the pavement surface is
controlled by the equilibrium suction and its variations.
A relationship has been developed in this study to predict the equilibrium suction from
readily available parameters (i.e., TMI, PI, and dry suction value at surface ]) based on the
data collected from the continental United States. The TMI (121) is a moisture index that reflects
the aridity or humidity of the soil and climate. This index is calculated using actual data such as
the collective effects of precipitation, evapotranspiration, soil water storage, etc. The value of
TMI can be found from the TMI map of the United States (Figure 10) (121).
37
However, since the development of the original TMI map in 1948, various studies have
been performed to improve the water balance approach and simplify the calculation process.
TMI values vary depending on the methods adopted for calculation and the time span of data
used in computation (122). Olaiz et al. (123) performed a comparison study of the TMI contour
maps developed by Thornthwaite et al. (121, 124–126) and concluded that the method developed
by Witczak et al. (126) matched most closely with the original moisture balance approach of the
Thornthwaite method. Witczak’s equations were used in this study to calculate the TMI value in
the continental United States. Detailed description of the development of the TMI map in the
GIS platform is presented in the next section.
The spatial climatic data sets (precipitation and temperature) that were used to generate a
GIS-based map of TMI were collected from Oregon State University’s Parameter-elevation
Regressions on Independent Slopes Model (PRISM) Climate Group, named for the PRISM
climate mapping system (127–129). In situ measurements of temperature and precipitation from
5852 weather stations across the continental United States are obtained from the National
Oceanic and Atmospheric Administration climate database and ingested into the PRISM
statistical mapping system. The PRISM products use a weighted regression scheme to account
for complex climate regimes associated with orography, rain shadows, temperature inversions,
slope aspect, coastal proximity, and other factors and develop climatological normal (from 1981
to 2010) at 30-arcsec (800 meters) resolution. Figure 11 shows the average yearly precipitation
(P) and temperature map of the United States in GIS platform.
38
(a)
(b)
Figure 11. Average Annual (a) Precipitation; (b) Mean Temperature GIS Map (1981 to
2010).
The potential evapotranspiration was calculated empirically using the mean monthly
temperature (ti) of the site (130). A yearly heat index (Hy) is determined first, as shown in Eq. 4.9:
12
H y (0.2ti )1.54
i 1 (4.9)
where, H y is the yearly heat index.
ti is the mean monthly temperature (i denotes the given month; from 1 to 12).
39
The monthly pe values determined using Eq. 4.10 represents a quantity of 30-day month
i
with 12-hour daylight duration. To consider the variations in daylight due to latitude and varying
days in each month, a day length correction factor is introduced (131) to obtain the corrected
potential evapotranspiration in Eq. 4.11:
d i ni
PEi pei ( )
30 (4.11)
where P E is the monthly corrected potential evapotranspiration (unit: cm).
i
Figure 12 depicts the calculated average yearly potential evapotranspiration map of the
United States on GIS.
Figure 12. Average Annual Potential Evapotranspiration GIS Map (1981 to 2010).
40
Finally, the TMI value is calculated following Eq. 4.13 according to Witczak et al. (126)
as shown in Figure 13:
P
TMI 75( 1) 10
PEy
(4.13)
The terms udry and u w et represent soil suction at the surface corresponding to an air dry
state and soil field capacity, respectively. udry is controlled by vegetation (4.5 pF) and by bare
soil (5.7 pF). The term u w et corresponds to 3 pF suction level (132). The soil water
characteristics curve prediction model developed by Saha et al. (133) was used in this study to
determine the volumetric water content at specific suction levels. Finally, the equilibrium suction
value is calculated for the continental United States following Eq. 4.14 and shown in Figure 14.
41
Next, the developed TMI and equilibrium suction map of United States is divided into
72332 map units. The mapunit shapefile of the United States is collected from the U.S.
Department of Agriculture (USDA) NRCS STATSGO2 database. A zonal distribution analysis is
performed in ArcGIS to determine the average TMI and pF value for each mapunit. The
following section describes the relationship between TMI and equilibrium suction for various
AASHTO soil classes.
All the NRCS mapunits are divided according to the AASHTO soil classes, and Figure
15 plots the equilibrium suction versus TMI values. For each soil class, the equilibrium suction
value decreases with an increase of TMI. A polynomial trend line is fitted for each case to
express the relationship between the equilibrium suction and TMI. The A-3 soil class has the
lowest equilibrium suction range whereas the A-7-6 class shows the highest range. In general,
the equilibrium suction increases with the increase of fine material and especially with clay fines.
42
5.5
St. Deviation,σ = 0.18
5.5 St. Deviation,σ = 0.25
Ue = 3.582e-3E-04(TMI)
Ue = 3.527e-3E-04(TMI)
5 5
4.5 4.5
4 4
3.5 3.5
3 3
‐60 ‐20 20 60 100 ‐60 ‐20 20 60 100
TMI TMI
(a) AASHTO soil type: A-1 (b) AASHTO soil type: A-2
5.5 St. Deviation,σ = 0.09 5.5 St. Deviation,σ = 0.17
Ue= 3.621e-3E-04(TMI)
Equilibrium Suction (pF)
4.5 4.5
4 4
3.5 3.5
3 3
‐40 0 40 80 ‐60 ‐20 20 60 100
TMI TMI
(c) AASHTO soil type: A-3 (d) AASHTO soil type: A-4
5.5 St. Deviation,σ = 0.26 5.5 St. Deviation,σ = 0.34
Ue = 3.706e-9E-05(TMI) Ue = 3.786e-3E-04(TMI)
Equilibrium Suction (pF)
5 5
4.5 4.5
4 4
3.5 3.5
3 3
‐60 ‐20 20 60 100 ‐60 ‐20 20 60 100
TMI TMI
(e) AASHTO soil type: A-6 (f) AASHTO soil type: A-7-6
Figure 15. TMI versus Equilibrium Suction (pF) (cm) for (a) A-1; (b) A-2; (c) A-3; (d) A-4;
(e) A-6; and (f) A-7-6 Soil.
43
A linear regression analysis was performed to predict the equilibrium suction value from
the TMI and PI. Eqs. 4.15 and 4.16 express the relationship for vegetative and bare soil,
respectively. Figure 16 shows the plot of the calculated versus predicted equilibrium suction
using Eqs. 4.15 and 4.16. An R2 value of 0.8 indicates that the regression equations have very
good prediction accuracy:
ue (pF) [Vegetative Soil] = 0.0199*PI-0.0047*TMI+3.3976 (4.15)
ue (pF) [Bare Soil] = 0.0214*PI-0.00607*TMI+3.5318 (4.16)
Unbound base and subgrade materials beneath the pavement surface are characterized by
the MR in the Pavement ME Design (67), which directly affects the design and analysis of
flexible pavement structures. The advantage of MR over other material properties is that it
represents the response to dynamic loading coming from vehicle movement. It is defined as the
ratio of repeated cyclic loading ( c y c ) divided by recoverable strain ( r ) as shown in Eq. 4.16:
cyc
MR
r (4.16)
The MR can be estimated in three different approaches: 1) conduct lab tests with
compacted soil specimens (e.g., repeated load triaxial test), 2) follow a back calculation method
from an in situ device (e.g., FWD), and 3) predict MR model coefficients from soil physical
properties. The first and second approaches are expensive, time consuming, and require expert
labor. An MR model based on the third approach was used in this study to characterize the
unbound and subgrade material under different conditions.
Various models have been developed to predict the MR of unbound base and subgrade
materials. The most widely used generalized model developed in the NCHRP Project 1-28A, is
shown in Eq. 4.17 (67):
44
I1 k2 oct
M R k 1 Pa ( ) ( 1) k 3
Pa Pa (4.17)
where I 1 is first invariant of the stress tensor, oct is octahedral shear stress, Pa is atmospheric
pressure, k 1 , k 2 , and k 3 are regression coefficients.
However, several studies have shown that the MR of unbound granular material is not
only stress dependent but also moisture dependent (134). The AASHTO employed a moisture-
dependent MR model, which is later adopted by the Pavement ME Design:
MR ba
log a
M Ropt b
1 exp[ln k m (S Sopt )]
a (4.18)
where MR is resilient modulus at a given degree of saturation, M R o p t is resilient modulus at a
reference condition, a and b are minimum and maximum of log(MR/MROPT), respectively, km is
regression coefficient, (S S o p t ) is variation of degree of saturation, expressed as a decimal.
Other researchers (19, 61, 135) also proposed different moisture-dependent MR models
for both unbound aggregates and subgrade. However, previous studies also demonstrated the
importance of the matric suction in the MR model. The Pavement ME Design incorporated the
suction in the MR model through NCHRP Project 9-23A (111):
wc * matric _ suction oct
M R k1 Pa ( ) k2 ( 1) k3
Pa Pa
(4.19)
where is bulk stress, wc is water content.
There have been several other studies recommending the use of soil suction in modeling
MR (17, 18, 40, 58, 136). The matric suction can be different for the same variation of degree of
saturation (i.e., S-Sopt), from one material to another (66). Therefore, the degree of saturation and
soil suction alone cannot accurately estimate the change of MR due to moisture. In order to
consider the moisture and suction dependency of resilient modulus characteristics, a new model
is proposed by Lytton (60):
I1 3 fhm k2 oct k3
E y k1Pa ( ) ( )
Pa Pa (4.20)
where I 1 is first invariant of the stress tensor, Pa is atmospheric pressure, is volumetric water
content, h m is matric suction in the base material, f is saturation factor, 1 < f < 1/θ, oct is
octahedral shear stress, k 1 , k 2 , and k 3 are regression coefficients.
In this model, I 1 and oct vary with the stress state, and h m is related to the moisture
content of unbound aggregates. Appendix D provides more details on this model. Three steps are
followed to determine the coefficients k 1 , k 2 , and k 3 using Eq. 4.20:
45
1. Matric suction hm in Eq. 4.21 is estimated first at the tested specimen water content. A
separate ANN model is used to predict the coefficients of the SWCC parameters in the
Fredlund-Xing equation as described in subsequent sections.
2. Volumetric water content, θ, and saturation factor, f, are calculated for each base and
subgrade material from the collected moisture content, MDD, and specific gravity data.
The saturation factor is an indicator of degree of saturation (60). It is multiplied with θ
and hm to express the stress on soil mineral skeleton in the transition zone. The transition
zone occurs between the air entry point suction value and the unsaturation point suction
value. The saturation corresponding to the air entry point and the unsaturation point were
considered as 100 percent and 85 percent, respectively, in this study. Eq. 4.21 is followed
to calculate the saturation factor, f in the transition zone:
S 85 1
f 1 ( 1)
15 (4.21)
where S is degree of saturation (unit: %), is volumetric water content (unit: decimal); it is
considered to be 2 pF and 3.5 pF at the air entry point and the unsaturation point, respectively.
3. Coefficients k1, k2, and k3 are calculated from collected MR data with different
combinations of confining pressure and deviator stresses. A Matlab code was generated
to fit the curve for each base and subgrade material.
Researchers have developed many correlation models to predict k1, k2, and k3 from soil
gradation, index, and strength properties (137–140). However, most of these models are either
confined to a limited number of data sets or have a poor prediction accuracy. As a result, an
ANN model is developed in this study to predict the coefficients from soil properties. The main
advantage of a neural network model over nonlinear regression models are that it can capture
complex nonlinear scattered relationships between input and output parameters and train the
model based on the evaluation of error function. The application of the ANN approach for
predicting MR of subgrade soil has been studied by a few researchers (141, 142). Soil index
properties (i.e., material percent passing No. 200 sieve, PI, OMC, MDD, variation in compaction
water content) and stress variables (i.e., confining pressure, bulk stress, deviator stress) were
used as input variables to predict MR (141). A different neural network analysis was performed
with nine different soil sources in Georgia (142). A similar study has been performed for ln(k1),
k2, and k3 coefficients by Nazzal and Mohammad (138). Input variables for the correlation
models consisted of material percent passing No. 200 sieve, LL, PI, MDD, test dry density
(TDD), OMC, TMC, and clay%. However, previous ANN approaches have had some limitations.
The resilient modulus values were predicted either as model output for a single stress
combination or using only the locally available soil data as input (143). To overcome these
limitations, an ANN model is developed in this project using a large database of unbound base
and subgrade materials collected from the LTPP. Moreover, the resilient modulus equation
coefficients (k1, k2, and k3) were predicted as output to cover any stress combination.
Matric suction of unbound granular base and subgrade material was required to
accurately estimate the resilient modulus in an unsaturated condition as shown in Eq. 4.20.
However, the quantity of available SWCC data of unbound base materials from the available
46
literature was not sufficient to train an ANN model. Hence the Soil Survey Geographic Database
was used for data collection from USDA-NRCS. The database included soil physical and
climatic properties as well as water content at specific suction pressures. Collected suction-
saturation data points for each soil were fitted with the Fredlund-Xing equation.
A Matlab code was generated to fit the Fredlund-Xing equation with measured data
points and to calculate the SWCC fitting parameters. Out of a total 34,237 data sets on soils
collected from the NRCS, soils that satisfy gravel characteristics according to Unified Soil
Classification System (more than 50 percent material retained on the No. 200 sieve is defined as
coarse grained material and more than 50 percent of the coarse grained material retained on the
No. 4 sieve) were separated out to train and validate the ANN models for the SWCC fitting
parameters. The 2598 plastic (PI>0) soil data and 311 non-plastic (PI=0) soil data were collected
and used to train the neural network to predict the SWCC fitting parameters for unbound
materials. The ANN models for the fitting parameters (i.e., af, bf, cf, and hr) of the SWCC in the
Fredlund-Xing equation are detailed above. The hidden layer in the ANN model assigned 20
neurons to establish the connection between input and output layers. The 80 percent of the total
data set was used for training and 20 percent for validation of ANN models. Water contents at
0.1, 0.33, and 15 bars (1 bar = 100 kPa) were available in the USDA-NRCS database. The
predicted SWCC parameters were input into the Fredlund-Xing equation to estimate the water
content at the same suction levels. The estimated water contents were compared against the
measured values to investigate the prediction accuracy.
Figure 17a and Figure 17b show the plots of the measured versus the predicted degrees of
saturation for plastic and non-plastic soils, respectively. The developed ANN models had a very
good prediction accuracy generating a R2 value of 0.91 for both plastic and non-plastic soils.
Hence the same trained models were used to predict the SWCC fitting parameters for base and
subgrade materials collected from the LTPP database. Predicted af, bf, cf, and hr parameters of the
LTPP materials were input into the Fredlund-Xing equation to estimate matric suction of the
specimen at the tested water content.
Figure 17. Comparison of Measured versus Predicted Saturation (%) at 0.1, 0.33, and 15
Bars Suction Level for Unbound Granular Base Materials Using ANN Model.
47
A neural network model specifically adjusts the weight factors wji and bias bj in Eq. 4.22
based on the minimum error function. In pavement engineering, the ANN approach is usually
used to develop prediction models based on a large number of data collected from experiments
and numerical analysis. ANN models have been successfully developed to predict the crack
growth function (e.g., reflective cracking and top-down cracking) in asphalt concrete (144, 145).
The ANN approach is also used to predict the geogrid-reinforced flexible pavement performance
(118). In general, the development of ANN models includes two critical steps: 1) data collection;
and 2) construction of ANN architecture.
Data Collection
A large collection of unbound base materials was available in the LTPP InfoPave
database. Out of total 3010 unbound base materials, 717 materials were selected that pass gravel
specifications. Researchers separated out 217 plastic (PI>0) and 500 non-plastic (PI=0) base
materials to develop two different sets of ANN models for k1, k2, and k3. In this way, the
subgrade ANN models for k1, k2, and k3 were developed based on non-plastic subgrade data from
180 LTPP sections and plastic subgrade data from 400 LTPP sections. MR tests were conducted
on 15 different combinations of confining pressure and nominal maximum axial stress level. In
addition to the MR test data, physical properties of base materials such as gradation, Atterberg
limits, moisture content, density, and specific gravity were collected from the LTPP database.
Construction of ANN Architecture
Two three-layered ANN models, one for plastic and another for non-plastic soil were
constructed as shown in Figure 18a and Figure 18b, respectively. Three layers of the ANN model
consisted of a input layer, a hidden layer, and an output layer. Soil physical properties (i.e.,
percent of material passing 3/8 in. sieve, percent of material passing No. 200 sieve, PL, PI, OMC,
MDD, and TMC) were introduced as input parameters for plastic soils. Input parameters for non-
plastic soils included the percent of material passing the 3/8 in. sieve, percent of material passing
the No. 200 sieve, ϴ, Ѱ, OMC, MDD, and TMC. The hidden layer assigned 10 neurons to
establish the connection between the output layer and the input layer. The number of neurons
was selected based on the number of data points. Too many hidden neurons allow the network to
memorize instead of generalizing the training set (146). Both the non-plastic and plastic soil
48
models used the same sigmoidal transfer function as that for the SWCC ANN models, 80 percent
of the data set was used for training and 20 percent of the data set for validation.
Figure 18. Illustration of Three-Layered Neural Network Architecture (a) Plastic; (b) Non-
plastic Soil.
The ANN model was developed to predict the MR model coefficients (i.e., k1, k2, and k3)
from base physical properties. The coefficients of the MR model were correlated with base
material physical properties and used to estimate the MR. Figure 19–Figure 20 show the
comparison between measured and predicted MR model coefficients for plastic soils.
The predicted MR coefficients show high R2 and low RMSE, which indicate a good
prediction accuracy. The R2 values for MR coefficients of plastic base and subgrade materials
were in the range of 0.66 to 0.92. Similarly, Figure 21–Figure 22 plotted the comparisons
between the measured and predicted MR model coefficients for non-plastic soils.
49
The R2 value for k1, k2, and k3 coefficients of non-plastic base and subgrade materials
were in the range of 0.51–0.97. Note that a high level of variability was involved in estimated
coefficients due to variations in cyclic load triaxial test results and seasonal changes in selected
physical properties (147). Hence, the predicted k1, k2, and k3 values need to be scrutinized in the
validation process. If the validation data set provides comparable results, then the predicted
coefficients can be sufficiently accurate to use in MR model equation.
(a) k1 (b) k2
(c) k3
Figure 19. Predicted MR Model Coefficients of Plastic Base Materials from Physical
Properties Using the ANN Approach.
50
(a) k1 (b) k2
(c) k3
Figure 20. Predicted MR Model Coefficients of Plastic Subgrade Materials from Physical
Properties Using the ANN Approach.
51
(a) k1 (b) k2
(c) k3
Figure 21. Predicted MR Model Coefficients for Non-plastic Base Materials from Physical
Properties Using the ANN Approach.
52
(a) k1 (b) k2
(c) k3
Figure 22. Predicted MR Model Coefficients for Non-plastic Subgrade Materials from
Physical Properties Using ANN Approach.
Several regression models have been developed to correlate the soil physical properties
with MR model coefficients. In this study, the correlation models (137, 139, 148) were selected
and compared against measured values to investigate the accuracy of the MR coefficients. They
were named as Yau model (137), Malla model (139), and Soliman model (148), respectively. All
these three regression models used the generalized MR constitutive equation shown in Eq. 4.17.
Yau and Quintus (137) proposed seven different sets of regression equations for crushed
stone materials, crushed gravel, uncrushed gravel, sand, coarse grained soil-aggregate mixture,
fine grained soil-aggregate mixture, and fine grained soil. The physical properties used to
correlate with the generalized MR model coefficients were material percent passing No. 3/8 in.
sieve, No. 4 sieve, No. 40 sieve, No. 200 sieve, percentage of silt (% silt), percentage of clay (%
clay), LL, PI, OMC, MDD, TMC, and TDD.
Malla and Joshi (139) studied the correlation between the MR model coefficients and the
gradation and compaction parameters of base materials. Regression analysis was performed on
MR model coefficients for uncrushed gravel and crushed limestone. The variables were selected
based on the literature and the logical influence of gradation, compaction, and moisture content
parameters on the regression constants (k1, k2, and k3). These variables included OMC, TMC,
53
MDD, material percent passing No. 4 sieve, material percent passing No. 40 sieve, and material
percent passing No. 200 sieve.
Soliman and Shalaby (148) presented regression equations of the generalized MR model
coefficients for six AASHTO soil types (A-1-b, A-3, A-2-4, A-4, A-6, and A-7-6). Data used in
the regression analysis were collected from 19 states in New England and the nearby regions in
the United States and two provinces in Canada. The soil properties that were considered in the
regression analysis include: OMC, TMC, MDD, LL, PI, TDD, percentage of coarse sand (%
CSAND), percentage of fine sand (% FSAND), percentage of silt (% silt), percentage of clay (%
clay), material percent passing 3 in. sieve, 2 in. sieve, 1 ½ in. sieve, 1 in. sieve, ¾ in. sieve, ½ in.
sieve, 3/8 in. sieve, No. 4 sieve, No. 10 sieve, No. 40 sieve, No. 80 sieve, and No. 200 sieve.
The accuracy of the prediction of the MR model coefficients by the existing correlation models
were analyzed in this study. Table 15 listed all the predicted MR coefficients using Yau, Soliman,
Malla, and ANN models.
Compared to the ANN model, the estimated RMSE values from the three regression
models were high, and the R2 values were extremely low. High RMSE and low R2 values
indicated a poor prediction accuracy. The R2 values for k1, k2, and k3 from all the three regression
models were less than zero, while R2 values from ANN model were in the range of 0.63 to 0.76.
The negative R2-values indicated the slope of the best fit line relating the predicted to the
measured coefficients was negative. Therefore, the developed ANN models had a significant
improvement in predicting the MR model coefficients from base physical and compaction
properties.
The accuracy of MR values using the predicted coefficients were examined in the next
section. Predicted coefficients from the three regression models were input into Eq. 4.17, and the
ANN model coefficients were input into Eq. 4.20 to calculate MR values, which were compared
to the measured values. Figure 23–Figure 24 showed the comparisons between measured and
estimated MR values using the ANN predicted coefficients for base and subgrade materials.
54
Figure 23. Comparison of ANN Model Predicted Resilient Moduli against Measured Values
for Base Materials.
Figure 24. Comparison of ANN Model Predicted Resilient Moduli against Measured Values
for Subgrade Materials.
Clearly the MR values estimated using the ANN coefficients matched well with the test
data. The R2 value estimated for both plastic and non-plastic base materials were 0.91, which
demonstrate a high accuracy. Figure 25a, Figure 25b, and Figure 25c presented the results from
Yau, Soliman, and Malla models, respectively. The R2 value estimated from Yau, Soliman, and
Malla models were 0.32, −1.53, and −1.62e06, respectively, which indicate that the regression
models failed to predict the coefficients. The negative R2 value from Soilman and Malla model
indicate that the best fit line relating the predicted to the measured values had a negative slope.
55
Figure 25. Comparison of Measured versus Predicted Resilient Moduli Using Regression
Models.
To validate the developed ANN models for the MR model coefficients, test data were
collected from the Texas A&M Transportation Institute (TTI) (74, 82) and University of Illinois
Urbana-Champaign (UIUC) (149). Epps et al. (74) conducted triaxial repeated load tests to
evaluate the laboratory performance of unbound granular materials. In addition to the MR test
results at different combinations of confining pressure and deviator stresses, aggregate physical,
strength, and moisture properties were also collected. Titus-Glover and Fernando (82) performed
the compressive creep and recovery tests for collected unbound base materials in the TTI
laboratory. Tutumluer et al. conducted MR tests at the 15 various stress states. Since no suction
values were provided by UIUC researchers, the suction is estimated using the developed ANN
models from the base physical properties. Table 16 listed all the collected physical properties
used as the ANN inputs for the validation data set.
56
Table 16. Input Parameters Collected from Literature for Model Validation.
#3/8 #200
Data Sources Soil source PL PI MDD OMC TMC
sieve sieve
E-02-2-3-2 45.9 12.36 10 5 136.2 7.2 7.2
Epps et al. (74)
E-08-2-1-6 49.2 6.3 9 5 140.4 6.5 6.5
Limestone
Titus-Glover 64.71 17.61 12.8 7.9 127.4 9.045 7.045
(−2% OMC)
and Fernando
Limestone
(82) 64.71 17.61 12.8 7.9 127.4 9.045 9.045
(OMC)
Dolomite FC 4%
69.9 4 N/A N/A 133 10.5 10.5
Tutumluer et al. (OMC)
(149) Dolomite FC 8%
71.1 8 N/A N/A 133 8.48 8.48
(OMC)
The collected base properties were input into the trained ANN model to predict k1, k2, and
k3. MR values were calculated using the predicted coefficients from the ANN models and
compared with the test results. Figure 26 showed the comparison between measured and
estimated MR values using the ANN predicted coefficients.
The MR values estimated using the predicted coefficients from the ANN models produced
a good fit with the test results. Figure 26 depicted an R2 value of 0.8 between measured and
predicted data set. Hence, the developed ANN model can accurately estimate the MR model
coefficients that are used in pavement design and analysis.
Figure 26. Validation of Measured versus ANN Predicted MR at Various Stress Levels for
Collected Unbound Materials.
MODULUS OF SUBGRADE REACTION FOR RIGID PAVEMENTS
57
The k-value can be measured either from a field plate load test conducted on top of the
subgrade (150, 151) or correlation with other load bearing capacity tests (e.g., consolidation test,
triaxial test, and California Bearing Ratio test) (152). The k-value from the plate load test is
calculated using the Winkler model, which assumes an elastic plate resting on a liquid
foundation (88). This model considers the soil behavior as a series of linear elastic springs, as
shown in Figure 27a. Many researchers have used the model to characterize soil-foundation
interaction (153–157). In the Winkler model, the submodel for the elastic plate is well
established, but the submodel of the foundation is particularly in need of modification.
Researchers attempted to improve the Winkler model by adding other forms of interaction
among the spring elements (89, 91, 158–160). The Pasternak model shown in Figure 27b allows
the transverse connection in the supporting media and introduces shear interaction on slab-base
interface (161, 162).
(a) (b)
Figure 27. Foundation Models for Rigid Pavement (a) Winkler Model; (b) Pasternak
Model.
The problem of beams on foundations was studied using the FE method and compared
the k-values with an analytical model (163, 164). In the FE analysis, they considered the
complex phenomena such as the realistic contact stresses between slab and foundation, and the
nonlinear stress-dependent elastoplastic behavior of the soil foundation.
Recently, the concept of backcalculating the modulus values was proposed using the
deflection basin generated by a FWD (165). Hall (166) derived the method of determining of the
dynamic k-value of the subgrade from the FWD deflection basin. Subsequently, the
backcalculated best-fit (BBF) approach was adopted by the LTPP program to determine k-values
for rigid pavements (167).
For the rigid pavement analysis in the Pavement ME Design, the load bearing capacity of
the subgrade foundation is characterized by an effective dynamic k-value. The actual rigid
pavement structure is transformed into an equivalent structure with a k-value, which represents
the compressibility of all layers beneath the concrete slab. The subbase and subgrade materials in
the design guide have been characterized by the Winkler model, which is easy for
implementation but neglects the shear interaction with the supporting media. In the BBF
approach, the degree of bonding on the slab-base interface is assumed to be either 1 for
completely bonded or 0 for non-bonded situation. However, none of these assumptions are
realistic for base courses (168). Shear restraints exist between the slab and base layer because of
rough interfaces. This causes an interlocking of the slab and base, which has a significant effect
on the structural performance of pavement (24, 25, 169).
In addition, no cross anisotropy is considered in the Pavement ME Design while
calculating the k-value. Recent studies have shown that granular base materials exhibit stress-
dependent and cross-anisotropic behavior, which indicates the directional dependency
58
(anisotropy) of the base moduli. Newly acquired triaxial testing devices were introduced by Abu-
Osei et al. and Tutumluer and Seyhan (13, 170) for determining anisotropic resilient properties of
granular materials in the laboratory. Tutumluer et al. and Park and Lytton (171, 172) found that
the use of nonlinear anisotropic base modulus significantly affects the stress distribution in the
base layer and diminishes the horizontal tensile stresses in the bottom half of the base layer. The
performance of pavement was predicted by Masad and Masad and Oh et al. (5, 39) using the
nonlinear cross-anisotropic model and showed good agreement with the field measurements.
Hence, there was a need to develop a modified k-value model, which involves both corrected
base modulus due to cross anisotropy and the slab-base interface bond.
The development of modified k-value model is elaborated as below, which contains four
submodels:
Cross-anisotropic modulus submodel for the base layer.
Slab-base equivalent thickness submodel.
Slab-base interface shear bonding submodel.
Modified k-value submodel.
An FE program, WinJulea, is used in this submodel to determine the stress values at the
mid-depth of the base layer. The estimated stress values are input in the generalized MR model,
shown in Eq. 4.24, to calculate the vertical modulus in the base layer (15):
M Rv k1Pa ( )k2 ( oct 1)k3
Pa Pa (4.24)
59
M RH
n
M RV (4.25)
where M RVis the resilient modulus in the vertical direction.
M R H is the resilient modulus in the horizontal direction.
n is the modulus ratio.
The resilient moduli in the vertical and horizontal direction follows a constant ratio, n.
The corrected base modulus is the golden mean of the vertical and horizontal modulus, as shown
in Eq. 4.26:
M (M V
M H
)1 / 2 M V
( n )1 / 2
R R R R (4.26)
60
This indicates that the base modulus is finalized whenever the difference between the
input and calculated modulus is less than 1 percent of the calculated modulus. If the difference
between calculated and input MR > 1 percent of M R ( C a lc u la te d ) , the calculated modulus is input
again in the FE program, and the iteration process is continued until the desired criterion is
reached.
Slab-Base Equivalent Thickness Submodel
The formulation of the equivalent thickness submodel includes three steps. First, a
cooperating slab and base system is introduced using the transformed-section method (Figure 29).
61
Second, the moment of inertia of the pavement cross section is calculated based on the
transformed section as follows:
I tr I s la b I b a s e Ai d i 2
(4.28)
where Itr is the moment of inertia of the transformed pavement section.
I s la b is the moment of inertia of the slab.
I b a se is the moment of inertia of the transformed base.
Ai is the area of the slab and the transformed area of the base course.
͞di is centroidal distance to each of the areas.
δ is the interface shear bonding.
Table 17 shows the calculation steps of the moment of inertia for a typical cooperated
slab and base system.
Table 17. Steps of Moment of Inertia Calculation for a Cooperated Slab and Base System.
Body Area(A) y Ay z I C .G . d I N . A.
bhs hs / 2 b hs / 2
Ai y i 1 bhs 3 I C . G . A1 d 1 2
2
(1) h
z s 1
Ai 12 2
1 Eb 3 h hb z I C . G . A 2 d 2
2
(2) Eb h h / 2 Eb
(hs hb / 2)b( )hb
s b
b( )hb b( )hb s 2
2
Es Es 12 Es
A i A1 A 2 A i y i A1 y 1 A2 y 2
Third, the thickness of the equivalent section is calculated using the moment of inertia of
the transformed section. The transformed section that consists of concrete and base is converted
into an equivalent cross section composed of only concrete. The thickness of the equivalent
section is estimated by considering the same moment of inertia for both sections:
62
I tr *12
heq 3
b (4.29)
Clearly, the thickness of the equivalent cross section depends on the degree of bonding,
, in the slab-base interface. A formulation of slab-base interface shear bonding is presented in
the next section.
Slab-Base Interface Shear Bonding Model
The interface shear bonding, , in Eq. 4.30 is expressed by the ratio of in situ shear
stress, ( z x ) 2 , in the base course on the slab-base interface, and the shear stress, v 2 , in the base
f
Figure 30 shows a schematic figure of the developed shear stress in the PCC-base
interface. The parameters v 1 and v 2 are shear stress on the interface in the slab and base layer,
respectively, when full shear is transferred. The parameter ( z x ) 2 is the shear stress in the base
f
course on the interface for in situ conditions, which is limited by the shear strength, sf , of the
base course on the failure plane. When ( zx )2 f is greater than v 2 , the interface is considered as
fully bonded. Depending on the ratio of the ( zx )2 f and v 2 , the partial bonding condition is
defined in the PCC-base interface.
Figure 30. Illustration of In Situ Shear Stress in the Base Course on the PCC-Base
Interface Using a Mohr-Coulomb Failure Envelope.
63
ah 2
3 2 s 2 2
(c tan ) cos sin[tan 1 ( ( hs a ) )]
1 3( hs hs a 2 )
3
(1 2 )
[ ]
2 ( hs 2 a 2 ) 2 h 2 a 2 1/ 2 h
s s
hb
3P ahs 2 hb ( hs 2 z )
2 ( hs 2 a 2 )5/ 2 h ( z hs )
s
2 (4.31)
Modified Subgrade k-value Submodel
The modified k-value submodel is developed using the calculated heq and δ values from
the previous submodels. The determination of the modified k-value is divided into the following
steps:
The deflection patterns generated by FWD are used to determine the modified k-values.
The FWD sensor deflections (0 cm, 30.48 cm [12 inch], 60.96 cm [24 inch], 91.44 cm
[36 inch], and 121.92 cm [48 inch] away from the loading point) are obtained, and the
basin area (BA) is calculated as:
SS
BA [ D0 2( D1 D2 ....... Dj 1 ) Dj ]
2 D0 (4.32)
where SS is the FWD sensor spacing (30.48 cm); Dj is the sensor deflection (j=0 to 4).
The effective relative stiffness length is calculated as follows:
le a b ( B A ) c ( B A ) 2
(4.33)
where a, b, and c are the coefficients obtained from the field correlation, a=0.992, b=−0.2891,
c=0.0284 (180).
Finally, the subgrade k-value is formulated in Eq. 4.34. The concept of equivalent slab
thickness is applied from Eq. 4.29 to consider the effects of both the slab and base layer
on the pavement subgrade k-value:
E s heq 3
k (4.34)
12(1 2 )le 4
where E s is the elastic modulus of the PCC; is the Poisson’s ratio of PCC.
The pavement data used in this study to estimate the modified k-value are collected from
the LTPP database. A total of 505 sets of rigid pavement section data are collected, including
both JPCP and CRCP. In addition to the backcalculated layer moduli, the database includes layer
thicknesses and FWD deflection patterns. In the LTPP, the FWD test uses three loading
64
sequences with a target load of 40 kN (9000 lb), 53.38 kN (12000 lb), and 71.17 kN (16000 lb).
In this study, only the deflection basins generated from 53.83 kN load are selected for analysis.
To avoid the discrepancy in measured deflection patterns, the deflection basin tests that were
performed along the mid-lane path are selected for the analysis. The mid-lane test locations are
designated as J1 and C1 for JPCP and CRCP pavements, respectively.
Correction of Base Modulus
As shown in Eq. 4.25, the resilient moduli of base material in the horizontal and vertical
directions follow a constant ratio. Based on the existing laboratory results, the n value is in the
range of 0.3–0.5. In this study, the n value is assumed as 0.4 for all base materials collected from
the LTPP pavement sections. Following the flow chart in Figure 28, the modulus of the base
layer is corrected for all pavement sections.
The procedure to determine the corrected base modulus is illustrated for an example
LTPP section 27-4054 (State code-SHRP ID). The pavement section 27-4054 consists of a
0.24 m (9.4-in.) PCC layer, a 0.15 m (6-in.) unbound base layer, and a semi-infinite silty clay
subgrade layer. Herein, the backcalculated modulus value is collected from the LTPP database
and used as the initial input at each layer. Table 18 lists the value of collected MR coefficients k1,
k2, and k3 for the base layer and the determined vertical and horizontal stress values after first
iteration. Figure 31 shows the plot of mean base modulus against iteration number. As seen, final
value is achieved after four iterations when it meets the convergence criteria shown in Eq. 4.27.
Table 18. Collected MR Coefficients and the Simulated Stress Values at the Mid-Depth of
Base Layer for LTPP Section 27-4054.
k1 (kPa) k2 k3 σ1 (kpa) σ3 (kPa)
998.5 0.55 −0.023 16.75 9.38
The interface shear bonding ratio is calculated for the collected LTPP pavement sections.
From the collected 505 pavement sections, 267 sections have a treated base layer beneath the
PCC surface layer and 238 pavements section are constructed using unbound base course.
65
(a) (b)
Figure 32. Formulation of Friction Angle from Mohr-Coulomb Failure Envelope for (a)
Treated Base; (b) Unbound Base.
As shown in Eq. 4.31, the expression for the degree of bonding, , is based on the shear
strength parameters c and .
In the LTPP database, the available experimental data are not sufficient to determine the
shear strength parameters. Unconfined compressive strength test data are available for cement
and lime treated base materials, whereas triaxial shear strength tests are conducted for unbound
base materials only at 34.47 kPa (5 psi) confining pressure. Hence, empirical models are used in
this study to estimate the c and parameters from base strength and saturation properties.
Existing studies have investigated the relationship between the cohesion parameter, c,
and the physical and strength properties of stabilized base materials. Balmer and Clough et al.
(176, 177) reported that cement treatment increased cohesion while internal friction angle
remained constant. Thompson (178) stated that addition of lime to unbound material yields a
substantial increase in cohesion and minor improvement in internal friction angle. This study
uses the relationship suggested by Park and Lytton (172) to determine the cohesion of treated
soils based on unconfined compressive strength:
c 9 .3 0 .2 9 2 c
(4.35)
where c is the unconfined compressive strength, unit: psi.
Then the friction angle, , is determined from c and the collected unconfined
compressive strength test data (Figure 32a). Unconfined compressive strength test data are
collected for 55 treated base materials from the LTPP database. Degree of bonding, , on the
slab-base interfaces is calculated from Eq. 4.31 using the estimated shear strength parameters.
66
Similarly, many researchers had analyzed the shear strength test data and identified the
influence of several physical properties on the shear strength parameters of unbound aggregate
materials. Density and degree of saturation are the two most important parameters affecting the
shear strength parameters of unbound base (179). The effect of relative density and degree of
saturation on the shear strength parameters was examined by Theyse (180) and proposed the
following relationship for cohesion:
c 0.0107 e12.12 RD e 2.30 S (4.36)
where RD is the relative density, unit: %.
S is the degree of saturation, unit: %.
c is the cohesion, unit: kPa.
Density and water content data are collected from the LTPP database and used to
calculate the RD and S in Eq. 4.36. Figure 32b presents the equation to estimate the friction angle
. Shear strength and compressive strength test data are collected from 88 granular unbound
base courses, which are further used to calculate in Eq. 4.31.
The black and green bars in Figure 33a denote the degree of bonding on slab-base
interface using the proposed shear bonding submodel and BBF approach (181), respectively.
Using the proposed bonding submodel, approximately 50 percent of the pavement sections with
treated base material is estimated to be fully bonded ( =1) with the PCC slab layer and rest of
the pavements have partially bonded slab-base interface. However, no partial bonding condition
is adopted in the BBF approach. The bonding condition in the BBF approach is assumed based
on the PCC modulus (167). The slab-base interface is considered as fully bonded if the PCC
modulus is greater than 26890 MPa (3900 ksi) whereas a non-bonded interface is applied for
lower PCC modulus values. Figure 33b compares the estimated degree of bonding using the
proposed bonding submodel with the BBF approach for unbound base course. Most pavement
sections with unbound granular base layer are estimated as partially bonded with the PCC layer
whereas a full bonding condition is assigned in the BBF approach.
(a)
67
(b)
Figure 33. Comparison of Calculated Slab-Base Interface Degree of Bonding Ratio with the
BBF Approach for (a) Treated Base; and (b) Unbound Base Layer.
The degree of interface bonding has a significant impact on erodibility prediction of rigid
pavements (168, 169, 182, 183). In the process of erosion, the concrete slab deforms under
repeated traffic loading, which causes mechanical shear stress between the slab and base. As a
result, adequate interface bonding can significantly reduce erosion. Erosion is addressed through
the faulting distress model (66). A better interfacial bonding model can infer better sensitivity to
measured faulting. Figure 34 presents the correlation between the measured wheelpath faulting
and the calculated degree of bonding. The wheelpath faulting data are collected for JPCP
pavement sections from the LTPP database.
68
The calculated degree of bonding shows better sensitivity to the observed faulting data.
The faulting value decreases with a higher degree of bonding in the slab-base interface. However,
the assigned interface bonding in the BBF approach shows no sensitivity to observed faulting.
The developed interface bonding model can better predict the interface condition for rigid
pavements. The modified k-values are estimated in the next section using the calculated interface
degree of bonding.
Estimation of Modified k-value
The modified k-values are calculated for the collected LTPP pavement sections following
the steps in modified k-value model that were outlined previously.
Figure 35 plots the modified k-values versus BBF k-values for the same pavement
structure collected from the LTPP. As presented, the BBF approach has higher k-values than the
modified k-value model. This discrepancy has occurred due to the different interface bonding
ratios. The interface between the slab and base is considered as fully bonded in the BBF
approach for most of the pavement sections whereas most of them are partially bonded in the
modified k-value model.
An ANN model is developed in the next section to determine the modified k-values for
various combinations of pavement structure and layer modulus. A wide range of pavement
structural properties, which include layer thicknesses and material strength properties, resilient
moduli, and slab-base interface bonding ratios, are input in the ANN model. The development of
the ANN model is elaborated as follows.
69
n
y j f w ji xi b j
i 1 (4.37)
where f is a transfer function, which normally uses a sigmoidal, Gaussian, or threshold functional
form.
wji and bj are the unknown weight factors and bias term, respectively.
The ANN model specifically adjusts the weight factors wji and bias bj in Eq. 4.37 based
on the minimum error function. In pavement engineering, the ANN approach is usually used to
develop prediction models based on many data collected from experiments and numerical
analysis. The ANN models have been successfully developed to predict the crack growth
function (e.g., reflective cracking and top-down cracking) in asphalt concrete (144, 145). The
ANN approach was also used to predict the geogrid-reinforced flexible pavement performance
(118). In general, the development of ANN models includes two critical steps: 1) deflection
basin estimation; and 2) construction of ANN architecture. These two steps are described below.
FE Model Development
The deflection basin of a specific pavement structure for FWD loading is estimated using
FE program ABAQUS. Researchers developed a total of 1296 simulation cases with different
combinations of pavement layer thicknesses, layer modulus, and PCC-base interface bond.
Pavement structural responses are calculated under FWD loading using the FE software
ABAQUS. Figure 36 shows a typical rigid pavement structure and the corresponding FE model.
In this study, a large variety of pavement structures are modeled with different combinations of
PCC, base thickness, base, and subgrade modulus. To properly represent the effect of the
strength of the base course and its level of erosion on the interface, it is necessary to accurately
characterize the slab-base interface. A Coloumb interface model is placed between the bottom of
the concrete and the top of the base course. The interface model has a normal and a tangential
load transfer behavior at each nodal point. By adjusting the friction coefficient and the elastic
slip distance, it is possible to represent various shear strength levels of the base course on the
interface, including zero strength. Varying the strength of the base course interface will generate
different deflection patterns. These deflection patterns are used to generate the different levels of
the subgrade k-values that are needed to represent the effect of lowered shear strength and
erosion of the surface of the base course on the foundational support. The FWD sensor
deflections (0, 30.48 cm, 60.96 cm, 91.44 cm, and 121.92 cm away from the loading point) are
obtained from the FEM analysis, and the modified k-value is calculated using Eqs. 4.32, 4.33,
and 4.34.
70
(a) (b)
Figure 36. (a) Schematic Plot of a Typical Pavement Structure; (b) Axisymmetric Model of
Pavement in ABAQUS.
Construction of ANN Architecture
Researchers constructed a three-layered ANN model in this study (Figure 37). Pavement
structural properties (i.e., PCC slab and base thicknesses), strength properties (i.e., PCC, base,
and subgrade moduli), and PCC-base interface bonding ratio were introduced as input
parameters for the ANN model. Table 19 summarizes the range of the input parameters in the
ANN model. The hidden layer is assigned 20 neurons to establish the connection between the
output layer and the input layer. The number of neurons is selected based on the number of data
points (146). The developed ANN model uses the sigmoidal transfer function, as shown in
Eq. 4.38 (118):
1
f Ii
1 exp Ii
(4.38)
where Ii is the input quantity.
φ is a positive scaling constant.
The parameter φ controls the steepness between the two asymptotic values 0 and 1. The
ANN model determines these weight factors wji through the two major functions: training and
validating. The training data set is used to determine the trial weight factors, wji, and bias term,
bj,. and the validating data set is employed to examine the accuracy of the model prediction.
71
Table 19. Selected Range of Input Parameters in ANN Training Data Set.
Input parameters Level Input values
PCC thickness (mm) 3 178, 254, and 348
Base thickness (mm) 3 101.6, 203.2, and 254
PCC modulus (MPa) 3 14,420, 41,400, and 82,737
Base modulus (MPa) 4 69, 690, 6894, and 25,000
Subgrade modulus (MPa) 3 34.5, 282, and 551
PCC-base interface bonding 4 0, 0.3, 0.6, and 1
In this study, 80 percent of the data set is used for training and 20 percent of the data set
is for validation. The training algorithm uses the Levenberg-Marquardt back propagation method
to minimize the MSE. The gradient descent weight function is employed as a learning algorithm
to adjust the weight factors wji.
Figure 38 shows the comparison between output and target k-values from the ANN
model. The output k-values shows high R2 for the training and the validation data set, which
indicates a good prediction accuracy.
To validate the developed ANN model, the ANN predicted k-values are compared with
the values calculated with the modified k-value model. Figure 39 presents the comparison of
calculated and ANN predicted k-values for LTPP pavement sections at the calculated degree of
bonding.
The ANN model shows high prediction accuracy with an R2 value of 0.92. This indicates
that the developed ANN model is capable of accurately predicting the modified k-value at any
given degree of bonding.
A sensitivity analysis for the degree of bonding is performed for selected pavement
sections and shown in Figure 40. Figure 40 illustrates the modified k-values increase with the
increase of degree of bonding for most pavement sections. However, for pavement sections
8-0214 and 29-500, the k-values have a decreasing trend with the increase of bonding ratio. The
reason behind that is the increase of slab-base bonding ratio causes a higher thickness of
equivalent transformed section and basin area. The k-value increases with the increase of
equivalent thickness, but it continues to decrease when the basin area increases. More sensitivity
analyses are presented in Appendix F.
72
Figure 38. Target and Output k-values for Training, Validation, and Overall Data Sets for
1296 Simulation Cases.
73
Figure 40. Modified k-values at 0, 0.3, 0.6, and 1 Degree of Bonding for Selected LTPP
Pavement Sections.
SHEAR STRENGTH OF BASE AND SUBGRADE FOR FLEXIBLE AND RIGID
PAVEMENTS
The Pavement ME Design mainly considers the elastic behavior of unbound base and
subgrade materials, but little attention has been paid to their shear strength. Previous studies have
presented that the shear strength of underlying layer materials is related to the pavement
performance through the following aspects:
Influence on accumulation of permanent deformation in flexible pavements.
Influence on extent of erosion in rigid pavements.
Influence on degree of bonding between the concrete slab and the base course in rigid
pavements.
In flexible pavements, the shear strength directly affects the amount of total rutting (179,
184, 185). For example, the dominant factor in determining permanent deformation is the
relationship between the shear strength of the soil and the applied shear stress. The ratio of the
shear strength to shear stress performs well in limiting the permanent deformation of granular
materials against shear failure (179, 184).
Shear strength affects the extent of erosion in the base course of rigid pavements, which
is the key factor in the development of faulting in JPCP and punchouts in CRCP. Soil Erodibility
(186) depicted that a small change in the shear strength has a considerable effect of the degree of
erosion in the lower range of shear strength. Furthermore, in rigid pavements, the shear strength
is directly related to the degree of bonding between the concrete slab and base course and the
degree of bonding has a significant impact on the performance of rigid pavements (168, 169, 182,
183).
In light of the critical role of shear strength in performance prediction, researchers
incorporated in the model formulation of both flexible and rigid pavements. The general shear
strength model is defined according to the Mohr-Coulomb failure envelope, which is determined
from triaxial tests on laboratory molded specimens:
74
c n ta n
(4.39)
where is the shear stress.
c is the total cohesion.
n is the normal stress on the failure plane.
is the angle of internal friction.
Furthermore, the impact of moisture variations on the shear strength is considered in this
study. As the water content increases by a small amount, the shear strength decreases
significantly depending on the magnitude of normal stress (187). Such a reduction accelerates
shear failure and intensifies rutting in flexible pavements and erosion in rigid pavements.
Researchers have developed a moisture-sensitive shear strength model for unbound base
materials (60):
c ' ( n fh m ) ta n ' (4.40)
where c ' is the effective cohesion.
' is the effective friction angle.
is the volumetric water content.
f is the saturation factor.
hm is the matric suction.
Figure 41 illustrates Eq. 4.40, which represents the dependence of the shear strength of
unbound materials on the matric suction. To make the moisture-sensitive shear strength model
more applicable in the pavement design, researchers developed prediction models for the shear
strength parameters c and φ’. In this way, the shear strength of unbound layers and subgrade can
be estimated using common soil properties in the absence of triaxial test data.
Figure 41. Schematic Plot of Mohr’s Circle Showing Dependence of Shear Strength on
Matric Suction.
To develop the prediction models for unbound base materials, researchers collected the
measured shear strength parameters of different types of base materials from Epps et al.,
75
Tutumluer et al., and Chow et al. (74, 189, 190). Additional data were collected to obtain soil
physical properties such as gradation, Atterberg limit, optimum and saturated moisture content,
and the SWCC of the various materials tested. Finally, regression analysis was conducted to
determine the relationships between the shear strength parameters (c’ and φ’) and the collected
physical properties:
c ' 0.221* PI 4.6 * sat (%) 455.62 * Gs 1262.75
' 0.0272 * PI 0.638 * MC opt (%) 1.487 * sat (%) 69.92
(4.41)
The goodness of fitting by the developed regression equations is shown in Figure 42 for
the collected base materials. R2 value of 0.81 and 0.87 for c’ and φ’ parameters, respectively,
indicate that the regression equations have a good prediction accuracy.
150 70
2
R = 0.81 65 R 2 = 0.87
60
55
100
50
45
40
50
35
30
25
0 20
0 50 100 150 20 25 30 35 40 45 50 55 60 65 70
Figure 42. Comparison of Predicted and Measured Shear Strength Model Parameters
(a) c’ and (b) φ’.
Similar prediction models have been developed for the shear strength parameters of
subgrade. Due to insufficient triaxial test data for subgrade, unconfined compressive strength test
data were collected from LTPP database. The friction angle had been found to be solely
dependent on the PI by Holtz and Kovacs (188). They developed an empirical correlation
between φ’ and PI for normally consolidated soils as shown in Eq. 4.42:
' 0 .0 0 1 4 * P I 2 0 .2 8 * P I 3 5 .8 7 (4.42)
The matric suction, friction angle, porosity, and gradation of the subgrade material affect
the cohesive shear strength, and this is confirmed with subsequent studies in the mechanics of
unsaturated soils (e.g., 60, 74). In this study, an ANN model has been developed to predict the
cohesion parameter from subgrade physical and strength properties. Figure 43 shows the
architecture of the ANN model. MDD, soil porosity, effective friction angle, and suction value at
OMC are selected as input parameters. The hidden layer is consisted of 10 neurons and the c’
76
parameter is the output of the developed model. The cohesive strength parameters were
calculated from the collected 432 unconfined compressive strength test samples.
Figure 44 shows the prediction accuracy of the developed ANN model for training and
validation data sets. The obtained R2 value 0.98 and 0.97 for training and validation data set
demonstrate that the developed ANN model can accurately predict the c’ parameter.
77
Figure 44. Target and Output c’ Values for Training, Validation, and Overall Data Sets for
432 Subgrade Soils.
PERMANENT DEFORMATION OF BASE AND SUBGRADE FOR FLEXIBLE AND
RIGID PAVEMENTS
Permanent deformation is a major form of distress in flexible and rigid pavement. In the
Pavement ME Design, the total permanent deformation is the sum of the individual layer
permanent deformations (i.e., surface layer, unbound layers, and subgrade layer). The permanent
deformation model for unbound base and subgrade layers used in the Pavement ME Design is
shown in Eq. 4.43:
( )
s ( 0 )e N v h
r (4.43)
where δ is the permanent deformation for the layer.
ϵr is the resilient strain imposed in the laboratory test.
ϵv is the average vertical resilient strain in the layer.
78
However, the rutting model used in the Pavement ME Design is less sensitive to the
modulus and the thickness of the unbound layers (4). As a result, a different permanent
deformation model is proposed by Tseng and Lytton (70), shown in Eq. 4.44:
( )
0e N
h
(4.44)
The Tseng-Lytton model was improved, and a new ME permanent deformation model for
unbound materials was proposed, which is capable of predicting the permanent deformation
behavior at different stress states using the single-stage test protocol (73). The formulation of the
model is given as follows:
J 2 m I1 K n
p ( N ) 0e( / N ) ( ) ( ) ( pa ) ( m n )
pa pa (4.45)
2 sin '
3 (3 sin ') (4.46)
c '.6 cos '
K
3 (3 sin ') (4.47)
where J2 is the second invariant of the deviatoric stress tensor.
I1 is the first invariant of the stress tensor.
ε0, ρ, β, m, and n are model coefficients.
c’ and φ’ are effective cohesion and friction angle, respectively.
In this model, the two terms, J2 and (αI1 + K), are incorporated into the Tseng-Lytton
model, which is used to reflect the influence of a stress state on the permanent deformation of
unbound materials.
Researchers have conducted many laboratory tests and validated the accuracy of the
model proposed in Eq. 4.45. They concluded that the proposed model matches well with the
measured permanent deformation curves. Gu et al. (118) studied the sensitivity of the proposed
model and demonstrated that it improved the sensitivity of unbound layer significantly. Hence
the model proposed by Gu and Zhang is adopted is this study to predict permanent deformation
of base and subgrade soils.
The model coefficients (ϵ0, ρ, β, m, and n) in Eq. 4.45 are determined by two approaches:
By fitting the model to the permanent strain measured in the repeated load permanent
deformation test.
Using prediction models developed from a set of physical properties such as the MDD,
percent fines content, aggregate gradation, etc.
79
Regression Models for Permanent Deformation Model Coefficients for Unbound Base
Layers
In order to develop the prediction models of the coefficients of Eq. 4.45, researchers used
the measurements from the repeated load triaxial test on 108 different types of base materials
collected from various sources (74, 189–192). In addition to triaxial test data, soil physical
properties such as gradation, Atterberg limit, MDD, and OMC data were also collected for the
base samples. Finally, regression analysis was conducted to determine the relationships between
the permanent deformation model coefficients (ϵ0, ρ, β, m, and n) and the collected physical
properties:
log 0 0.017 * P4 0.00967 * MDD 0.99 (4.48)
log( ) 0.0574 * P4 0.0937 * MC 5.0756 (4.49)
0.01483* P4 0.00813* MDD 0.00136 * sat (%) (4.50)
m 0.2153* 0 0.396 (4.51)
n 0.22993* m 0.746 (4.52)
where P4 is the percent of material passing No. 4 sieve (unit: %).
MDD is the maximum dry density (unit: lb/ft3).
MC is the test moisture content (unit: %).
Sat(%) is the degree of saturation (unit: %).
Figure 45 shows the goodness of fitting by the developed regression equations for the
collected base materials. The predicted coefficients showed a good match with the measured
values.
80
(a) (b)
(c) (d)
(e)
81
82
An ME model was developed to estimate the growth of faulting with time. An inflection
point in the field faulting data (shown in Figure 47) is found to differentiate two stages of
faulting. The inflection point, as a critical faulting depth, can be directly determined by this
model and is an indicator of the beginning of erosion for the concrete pavement design. The
faulting before reaching the critical depth results in the accumulated deformation of the
underlying layers (Figure 47). Water infiltrates through the joint and fills the void caused by the
83
permanent deformation. Beyond this inflection point, faulting accelerates because this infiltrated
water being driven by the moving traffic scours the surface of the base course and causes erosion.
In summary, two faulting prediction models were developed. One is to predict the entire
faulting development over time. This model of the full faulting curve shows that there is an
inflection point in the faulting curve. Before reaching the inflection point, the accumulation of
faulting is caused by the permanent deformation of the supporting layers. After passing the
inflection point, faulting accelerates due to the action of erosion. The second model is to predict
the faulting depth below the inflection point with traffic. The proposed models were proven to be
considerably accurate and reliable by using LTPP data. The coefficients in the models are
statistically calibrated with performance-related factors using multiple regression analysis.
Figure 47. Illustration of Field Faulting Data, Including the Critical Faulting Depth.
The ME model was developed to characterize the faulting depth over time. The
mathematical form of the model, according to the faulting depth curve as shown in Figure 47, is
expressed as follows:
N
f e [ln( )]1/ e
N N0 (4.53)
where f is the faulting depth.
N is the number of days after pavement construction date.
N 0 is the number of days when faulting initiates.
N is the number of days to failure due to erosion.
e and e are model coefficients.
A model was developed to predict the critical faulting depth in rigid pavements using the
field faulting data by collecting from the LTPP database. Figure 47 indicates this concept of
critical faulting depth. The critical faulting depth is defined as the inflection point in the faulting
84
depth at which erosion begins. The point of inflection is where the curvature of the faulting depth
curve changes from negative to positive, as shown in Figure 46. To determine the inflection
point linking those two curved segments, take the second derivative of Eq. 4.53 and set it equal
to zero to yield:
e 1
( )
e
N ip N e N0
(4.54)
where N ip is the number of days at the inflection point at which the acceleration of faulting
begins.
The critical faulting depth, f ip , which occurs at the inflection point, is expressed as:
e 1 1/
fip e ( ) e
e (4.55)
The data for the full faulting prediction model were collected from the field data from the
LTPP database, which provides a large amount of research quality pavement performance
information. The data of each jointed concrete pavement test site were analyzed and compared
between the measured and the model predicted faulting data.
The field data collected from the LTPP database required data cleansing since they have
limitations such as small data sets, poor data quality, and bias. Firstly, the minimum number of
faulting points for each LTPP section are four points that are required to calculate the model
coefficients ( e , e , N 0 , and N ). Individual data points that were clearly outliers were
eliminated. Consequently, a total of around 130 LTPP pavement sections are obtained to perform
the analysis of the faulting prediction model.
Based on the collected eligible LTPP faulting data, the analysis with nonlinear
optimization was performed in each pavement section. The four unknown coefficients, e , e ,
N 0 , and N , are determined using the SOLVER function in Excel®.
The comparison between the measured and predicted faulting depths is made to test how
well the developed model fits the data sets. Figure 48 plots the predicted faulting to observed
data, which are total analysis results from the collected data sets of around 130 LTPP sections.
The majority of scatter falls on the line of equality and the value of R2 in this plot is 0.95,
indicating remarkable accuracy and reliability of the developed model.
85
To implement the proposed model, model coefficients were calibrated with performance-
related factors by using multiple regression analysis.
Variable Selection
Both categorical and numerical variables are involved in the calibration of model
coefficients. The categorical variables include climatic zones (Dry-Freeze [DF], Dry Non-Freeze
[DNF], Wet-Freeze [WF], and Wet Non-Freeze [WNF]), use of load transfer device (dowel), and
use of drainage features. Because the support provided by the base course is the key to the
progression of faulting, the numerical variables consist of aggregate gradation in the base layer
and thickness of slab and base layer. The moisture infiltration also governs the development of
faulting. Thus, the moisture content and saturation are also taken into consideration.
Furthermore, some environmental information is incorporated into the model calibration,
including the annual average number of days with high/low temperatures and with rainfall (194).
Based on the collected performance-related data, a stepwise method is adopted for
variable selection. This is a method of fitting regression models by adding or subtracting
predictive variables at the various steps in either direction. The popular statistical software
package R is used to perform the stepwise regression to select the predominant predictive
variables. The p-value is the result from the student’s t-test to indicate the level of significance of
the variables. If the p-value is less than 0.05, the corresponding variable is of significance for the
response variable at a 95 percent confidence level. Thus, the select predictive variables are
required to meet the criterion that the p-value is less than 0.05 to ensure a 95 percent confidence
level.
86
The correlation between the model coefficients and the selected performance-related data
is investigated using multiple linear regression analysis. The model parameter of was found
to be a constant value of 2e4. Due to the variability of performance-related LTPP data, the model
coefficient, , is also incorporated into the model calibration. Table 20 presents the results of
multiple linear regression analysis for the coefficients in the model. From Table 20, the p-values
of the predictive variables are less than 0.05, indicating that prediction models of each coefficient
are significant at a 95 percent confidence level. The coefficient models are expressed as:
ln( e ) 0.952394 0.816222 dowel 0.164161 e 0.000173
N 0 0.024705 intensedays 0.012736
days32C 0.081787 basethick 0.451114
drainage (4.56)
N 2e 4 (4.57)
ln( N 0 ) 9.003655 0.019214 days32C 0.009733 FT
(4.58)
0.141653 basethick 0.631479 bound
ln( e ) 0.85277 1.605784 DNF 0.994204 WF 1.275704
WNF -1.474294 dowel -1.379156 bound 0.033711 (4.59)
days32C 0.149092 basethick 0.016701 wetdays
where dowel is a dummy variable for the use of dowels (No dowel = 0, dowel = 1).
intensedays is the annual average number of days with precipitation greater than 12.7 mm.
basethick is the base course thickness, in.
drainage is dummy variable for the use of drainage features (the section subsurface drainage
information can be collected from the LTPP table of SECTOIN_DRAINGE. The code of
DRAINAGE_TYPE is used to denote the dummy variable drainage. drainage = 0 if code is no
subsurface drainage; drainage = 1 if code is any other drainage features such as longitudinal or
transverse drains, drainage blanket, etc.).
FT is the freeze-thaw cycle that is number of days in the period when the air temperature goes
from less than 0°C to greater than 0°C.
days32C is the annual average number of days with the maximum temperature greater than
32°C.
wetdays is the annual average number of days with precipitation greater than 0.25 mm.
bound is a dummy variable (bound = 1 for bound base layer; bound = 0 for unbound base layer).
WF is a dummy variable (wet-freeze zone = 1, else = 0).
WNF is a dummy variable (wet non-freeze zone = 1, else = 0).
DNF is a dummy variable (dry non-freeze zone = 1, else = 0). Note that in the DF zone the
dummy variables of WF , WNF , and DNF should be equal to zero.
87
Table 20. Results of Multiple Regression Analysis for Coefficients in the First Faulting
Model.
Parameter Standard
Objectives Variable t-value p-value
Estimate Error
Intercept 9.00365 0.45718 19.69383 7.11E-28
days32C 0.01921 0.00389 4.93505 6.70E-06
Prediction
FT -0.00973 0.00194 -5.01109 5.08E-06
model of ln
basethick -0.14165 0.05244 -2.70138 0.008964
Bound -0.63148 0.17016 -3.71107 0.000454
Parameter Standard
Objectives Variable t-value p-value
Estimate Error
Intercept 0.85277 0.66645 1.27956 0.205976
dowel -1.47429 0.27515 -5.35824 1.64E-06
Bound -1.37916 0.28814 -4.78633 1.28E-05
days32C 0.03371 0.00708 4.76000 1.41E-05
Prediction
basethick -0.14909 0.07012 -2.12621 0.037907
model of ln
DNF -1.60578 0.47300 -3.39488 0.001268
WF -0.99420 0.29543 -3.36525 0.001386
WNF -1.27570 0.38121 -3.34647 0.001467
wetdays 0.01670 0.00523 3.19625 0.002289
Parameter Standard
Objectives Variable t-value p-value
Estimate Error
Intercept 0.95239 0.31918 2.98391 0.004185
dowel -0.81622 0.21215 -3.84747 0.000304
rho -0.16416 0.03220 -5.09822 4.07E-06
Prediction No 0.00017 0.00005 3.78797 0.000368
model of ln intensedays 0.02471 0.01168 2.11502 0.038812
days32C -0.01274 0.00461 -2.76011 0.007758
basethick -0.08179 0.04858 -1.68352 0.097744
Drainage -0.45111 0.25342 -1.78008 0.080394
Figure 49 shows the results of the coefficient models by comparing the faulting
parameters predicted by Eqs. 4.56, 4.58–4.59 with the raw parameters. To achieve a valid
calibration model, 80 percent of selected LTPP sections were randomly taken for multiple
regression and the rest were taken for validation. Figure 49 shows satisfactory and accurate
correlations between faulting parameters and pavement performance-related factors.
88
(a) N0
(b)
(c)
Figure 49. Comparison between Measured and Predicted Coefficients in Faulting Model.
Calibration Model of the Inflection Point Faulting
Due to the importance of the inflection point in the proposed model, multiple regression
analysis is also performed to calibrate the faulting depth. Table 21 lists the regression results for
faulting at the inflection point. The faulting depth at the inflection point is expressed:
89
Figure 50 presents the comparison between measured and predicted faulting depth at the
inflection point. The multiple regression of Eq. 4.60 is used to determine the predicted faulting
depth. The regression model of Eq. 4.60 has the capability to accurately estimate the faulting
depth with an R2 value of 0.89 in the validation plot of Figure 50.
Figure 50. Comparison between Measured and Predicted Faulting at Inflection Point.
Table 21. Results of Multiple Regression Analysis for Faulting at Inflection Point.
Parameter Standard
Objectives Variable t-value p-value
Estimate Error
Intercept 0.58284 0.77211 0.75487 0.453382
Drainage -1.51550 0.44948 -3.37166 0.001335
dowel -1.55997 0.20638 -7.55860 3.37E-10
Prediction
Bound -1.57098 0.23781 -6.60602 1.34E-08
model of ln
basethick -0.25955 0.07067 -3.67270 0.000525
wetdays 0.01400 0.00536 2.61277 0.011419
days32C 0.02676 0.00462 5.78941 3.02E-07
Figure 51 presents the mean critical faulting depth with or without dowels. The mean
critical faulting depth is about 1.2 mm when dowels are used but roughly 2.4 mm when no
dowels are used. Thus, Figure 50 shows that the use of dowels results in lower critical faulting
depth.
90
The faulting depth accumulated before the inflection point (Figure 52) is related primarily
to the permanent deformation of the underlying layers under traffic loads. To characterize the
stress-dependent deformation behavior of unbound aggregates, a new ME faulting model is
proposed to summarize faulting depths at different stress levels and expressed as:
k ( ) J 2i m I1i K n
f ( Ni ) f e Ni N0
( ) ( )
i 1 Pa Pa (4.61)
I1 x y z (4.62)
1
J 2 [( x y )2 ( x z )2 ( y z )2 ] xy2 yz2 zx2 (4.63)
6
2sin
3(3 sin ) (4.64)
6c cos
K
3(3 sin ) (4.65)
91
The variables, as known in the model including predictive traffic variables ( Ni ), power
coefficients ( m and n ), stress state terms ( J 2 and I1), cohesion c , and friction angle , are
required to perform nonlinear optimization analysis to determine the unknown coefficients ( ,
, N0 , and f ). Therefore, the determination of the variables is required by taking following
steps:
1. The power coefficients of m and n are determined by Eqs 4.66 and 4.67 from the
developed permanent deformation model.
2. The cohesion c and friction angle are obtained by Eqs. 4.68 and 4.69 from the
developed shear strength model.
3. The stress state terms of J 2 and I1 are determined from the analytical elastic method
that is elaborated in Appendix G.
4. The predictive traffic variable of N i represents the cumulative number of axles at the
axle load level i. To determine the traffic variable, the load distributions are required to
be rearranged into a tabulation of axle load levels by number of axles. Based on this
established tabulation, the calculated faulting at various axle load levels were added up to
the total faulting depth:
92
The faulting model that incorporates the effects of a load spectrum on the permanent
deformation of the base course requires a simple, consistent, and reasonably accurate method of
estimating the state of stress in the middle of the base course. The method adopted for this
faulting model is from Timoshenko (195) for estimating the stress state in a half space in which
the only material property that is needed is the Poisson’s ratio. The tire load for both single and
dual tires is represented as a point load on the concrete pavement surface. The use of these
equations permits the superposition of the stresses caused by both dual tires. Appendix G
introduces the calculations of stress in the base course for single and dual tires.
The traffic input for the faulting model is the number of axles corresponding to different
load levels. The weigh-in-motion (WIM) data can be used to determine the traffic input. The
WIM tables present the number of axles by types of axles within vehicle classes at various load
levels. The WIM data can be directly collected from the LTPP database. If the data are
unavailable in the database, it can be estimated with the annual average daily truck traffic
(AADTT) and axle load distributions. The axle distributions shown in WIM tables are required
to be categorized into the groups (single and dual tire) to permit the determination of the stress
state. Appendix H elaborates on the categories of axle distribution and the approach to estimate
the number of axles with AADTT.
93
The validation of the model for the load-related faulting is based on the LTPP field
faulting data. First, the field data were collected for validation of the first faulting model to
characterize the entire life of faulting and determine the critical faulting depth at the inflection
point. Then the field data before reaching the determined critical depth were retained for the
validation of the second faulting model. The coefficients ( , , N0 , and f ) in the second
load-related faulting model were determined by nonlinear optimization analysis.
Figure 52 presents the comparison between the measured and predicted faulting before
reaching the critical faulting depth at the inflection point. Figure 53 shows that the overwhelming
majority of the scatter falls close to the identity line. Moreover, the value of R2 is 0.97 indicating
a high degree of accuracy of the proposed second faulting model.
Figure 53. Comparison between Measured versus Predicted Faulting before Critical Depth.
The calibration coefficients for the second load-related faulting model are similar to the
first faulting model. Similarly, both the categorical and numerical variables are included in the
calibration coefficients for the second faulting model and were collected from the LTPP
database. The stepwise multiple regression analysis was performed to investigate the significant
predictive variables with a p-value below 0.05 at a 95 percent confidence level. Table 22
presents the results of the multiple regression analysis for the coefficients in the second load-
related faulting model. The model parameter of was solved to a constant value of 2e7. The
regression equations for the coefficients are as follows:
94
where days 0 C is the annual average number of days with the temperature lower than 0°C.
FI is the calculated freezing index for year (it can be collected from LTPP table of
CLM_VWS_TEMP_ANNUAL).
Table 22. Results of Multiple Regression Analysis for Coefficients in the Load-Related
Faulting Model.
Parameter Standard
Objectives Variable t-value p-value
Estimate Error
Intercept -429.471 592.975 -0.72426 0.473858
FI -15.7221 2.424347 -6.48507 2.03E-07
Prediction
model of FT -180.181 26.09081 -6.90593 5.88E-08
days0C 196.1607 29.10503 6.739752 9.59E-08
N
dowel 831.6622 235.89 3.525637 0.001231
days32C 23.58644 4.154377 5.677491 2.26E-06
Parameter Standard
Objectives Variable t-value p-value
Estimate Error
Intercept 0.028917 0.028854 1.002181 0.322766
FI -0.00063 8.91E-05 -7.04431 2.47E-08
FT -0.00693 0.001018 -6.80987 5.09E-08
Prediction
DNF 0.016353 0.031002 0.527485 0.601006
model of β
WF -0.03123 0.012386 -2.52151 0.016122
WNF 0.02095 0.019958 1.049707 0.300661
days0C 0.007663 0.001129 6.787173 5.46E-08
Parameter Standard
Objectives Variable t-value p-value
Estimate Error
Intercept 0.184024 0.019227 9.571042 2.51E-10
Prediction basethick -0.00905 0.003044 -2.97453 0.005982
model of f Bound -0.06024 0.010982 -5.48506 7.38E-06
dowel -0.02143 0.011367 -1.88516 0.069825
95
Figure 54 presents the results of the comparison between the coefficients predicted by
Eqs. 4.70–4.73. The correlations between those coefficients and pavement performance-related
factors are shown in Figure 54 together with satisfactory R2 values.
Figure 54. Comparison between Measured and Predicted Coefficients in the Permanent
Deformation Faulting Model.
96
This project developed several models as described in previous chapters for unbound
granular materials and subgrade, which are intended to enhance the sensitivity of predicted
pavement performance to these underlying layers. The sensitivity of these enhanced models is
evaluated in this chapter for both flexible and rigid pavements. The common performance
indicators contained in the Pavement ME Design that will be evaluated here include fatigue
cracking, rutting, and IRI for flexible pavements, and faulting for rigid pavements. The
sensitivity analysis will demonstrate whether the predicted performance changes with the change
of traffic load, operational conditions, material properties, and thickness of the structural layers.
In addition, the proposed enhanced models are compared with the corresponding models in the
Pavement ME Design, which shows the resulting difference in the prediction of pavement
performance.
MOISTURE-SENSITIVE, STRESS-DEPENDENT, AND CROSS-ANISOTROPIC
RESILIENT MODULUS
Unbound granular materials are generally applied in the bases and subbases to support
pavement surfaces and transfer vehicle loads to the subgrades. The stiffness of the bases and
subbases in the pavement design are described using MR, which is the ratio of axial stress and
recoverable axial strain (196). However, in most pavement designs, the bases and subbases are
treated as linear elastic and isotropic materials using one MR and Poisson’s ratio, which do not
represent realistic properties of unbound granular materials. Researchers have shown that the MR
of unbound granular materials depends upon the confining pressure (or sum of the principle
stresses) and the octahedral shear stress. Predicted results using models (50, 51, 54, 197–201)
that consider the effects of bulk and shear stresses have shown good to experimental results. In
addition to the effects of stress conditions of materials, characterization of the MR can also be
achieved with the density, moisture content, and gradation (4, 196).
The anisotropy of unbound granular materials is mainly due to the preferred orientation
and the compact force. Researchers generally believe that the modulus of unbound granular
materials in the vertical direction is greater than that in the horizontal direction. The ignorance of
this property will cause overestimation of tensile stresses in base layers, which affect further
pavement performance models. Studies with the anisotropic base and subbase layers showed
better predictions with field measurements compared with isotropic models (202).
This section presents a pavement structure model with stress-dependent, moisture-
sensitive, and anisotropic base layers using the commercial FE software COMSOL (Error!
Reference source not found.). The responses of the pavement under different loading levels and
the effects of some base material properties will also be shown.
FE MODEL OF PAVEMENT STRUCTURES
97
FWD tests (205). Figure 55 and Table 23 show the configuration of the model built in the
software.
Nonlinear, Anisotropic,
Base -
Moisture-Sensitive
Nonlinear, Moisture-Sensitive,
Subgrade v=0.4
Isotropic
The surface material is modeled as viscoelastic and isotropic. Due to the incomplete
implementation of the constitutive model for viscoelastic materials as Eq. 5.1 (206, 207) in
COMSOL, a method using the Partial Differential Equation Module coupling with the Linear
Elastic Solid Module is applied and can successfully model linear viscoelastic materials (208).
For solid-like viscoelastic materials such as asphalt concrete, a generalized Maxwell model is
applied to represent the relaxation modulus of the material in terms of Prony series as in Eq. 5.2:
ij ekkij 2eij
(5.1)
where , are Lame’s constant.
ij , eij are stress and strain components.
ekk is the first invariant of stress tensor.
ij is Kronecker delta.
98
n
E (t) E Ei exp( t/ i )
i 1 (5.2)
where E (t) is relaxation modulus.
E is long-term modulus.
n is the number of Prony series.
i is relaxation time.
where E z is vertical modulus of the material (assuming z axis is the vertical axis).
k1 , k2 , k3 are regression coefficients from laboratory tests.
Pa is atmosphere pressure.
I1 is the first invariant of stress tensor.
oct is the octahedral shear stress.
is volumetric water content.
f is saturation factor.
hm is matric suction.
Table 24 shows an example of the base information.
99
in which:
n E x / Ez
(5.6)
m Gzx / Ez
(5.7)
0 1 xy (5.8)
100
The approach explained in the Pavement ME Design Guide to predict fatigue life of the
pavement is based on the calculation of damage at either the surface for top-down cracking or the
bottom for bottom-up cracking (67). The final form to predict the number of load repetitions to
failure is from the Asphalt Institute (AI) model (210) with the national field calibrated model.
Since the original form of AI model was proposed to only predict load repetitions in the phase of
crack initiation (211) and the difference between national and local conditions, the prediction
model for top-down cracking is limited in its accuracy and variability (202). The performance of
fatigue cracking in this section is evaluated using bottom-up fatigue cracking model in Pavement
ME Design. For top-down cracking, alternative models can be found in recent research (194,
212):
1 1
N f 0.00432k1C ( )3.9492 ( )1.281
t E (5.10)
101
To predict the permanent deformation of the pavement, the Pavement ME Design Guide
applies models corresponding to the material type and computes the accumulated plastic strain at
each sublayer. For the unbound granular materials, the proposed model (67) with the calibrated
coefficient to predict the permanent deformation is presented as Eq. 5.14:
a (N) GB ( 0 )e( /N) v h
r (5.14)
in which:
log 0.61119 0.017638Wc
(5.15)
9
0 e a1 Erb e ( /10 ) a9 Erb
1 9
(5.16)
r 2
a1 Erb1
C0 ln( ) (5.17)
a9 Erb9
C0 1/
109 ( ) (5.18)
1 109
102
v is the average vertical strain in the layer/sublayer as obtained from the primary response
model.
h is the thickness of layer/sublayer, in; a1 =0.15; a9 =20.0; b1 =0; b9 =0.
100.679 0.744
0
(5.25)
0.026P200 0.029 d 2.647 (5.26)
103
Before the sensitivity analysis, several cases are run to check the nonlinearity and
anisotropy of the base layer in the pavement model. Different from most of current simulations
that treat the subgrade as a linear elastic material, the pavement models with moisture-sensitive
subgrade are shown in the following three sections.
Based on the Eq. 3.2, the nonlinearity of the base material reflects on its sensitivity to the
stress state and the moisture condition of the base layer. In this case, four loading levels 201 kPa,
566 kPa, 755 kPa, and 1006 kPa and two moisture conditions are applied to determine the
pavement responses.
Figure 56 and Figure 57 show the critical responses and modulus contours at different
loading levels. As the loading increases, the strains at critical points increase accordingly. The
modulus in the base course increases with the loading level and decreases with the distance from
the loading area, which matches with its stress-dependent property and previous research (213,
214).
600 1200
Strain, µɛ
Strain, µɛ
500 1000
201 kPa 201 kPa
400 800
566 kPa 566 kPa
300 600
755 kPa 755 kPa
200 400
1006 kPa 1006 kPa
100 200
0 0
(a) (b)
Figure 56. (a) Tensile Strain at the Bottom of the Surface and (b) Average Compressive
Strain in the Centerline of the Base under Different Loading Levels.
104
(a) (b)
(c) (d)
Figure 57. Vertical Modulus Contours in the Base Layer under the Loading Level
(a) 201 kPa, (b) 566 kPa, (c) 756 kPa, and (d) 1006 kPa (Unit: MPa).
Figure 58 shows the modulus contours at different moisture conditions while other
factors are the same. The modulus in the base course is higher in the dryer condition, which
reflects that the moisture softens the base material.
(a) (b)
(c)
Figure 58. Vertical Modulus Contours in the Base Layer at (a) Dry Condition, (b) Medium
Condition, and (c) Wet Condition (unit: MPa).
The anisotropy of the granular base material reflects on the ratio of vertical modulus to
the horizontal modulus, which is the value of 1/n in the Eq. 5.6. Figure 59 shows that as the
value of 1/n increases, the tensile strain at the bottom of the surface and the compressive strain at
the top of subgrade show an ascending trend, which matches the conclusion of a previous study
(215).
105
340 760
Strain, µɛ
Strain, µɛ
335 720
330 0.67 680 0.67
325 1 640 1
320 2.22 600 2.22
315 560
310 520
(a) (b)
Figure 59. (a) Tensile Strain at the Bottom of the Surface; (b) Compressive Strain at the
top of the Subgrade the Pavement with Different Anisotropy (1/n) of Base Layers.
From the cases above, the FE model applied in the study shows differences with the
loading level and the value of n, which demonstrates that the model proposed in this project is
sensitive in terms of both the stress-dependent and anisotropic characteristics of the base.
In most of the current simulation of pavement structures, the subgrade was considered as
linear and isotropic. However, the properties of the subgrade should also be characterized and
modeled like base materials. It is necessary to study the effects of the stress-dependent and
moisture-dependent MR to the pavement performance. In this section, the MR of the subgrade is
modeled as in Eq. 3.2, and the coefficients are obtained and calculated from data in the LTPP
database. Table 24 and Table 25 show the basic information of the subgrade. Figure 60 shows
that 1) with the increase of the loading level, the modulus of the subgrade increases accordingly;
and 2) with the increase of the depth, the modulus first increases and then decreases.
Vertical Modulus, MPa
0 566 kPa
Depth, cm
‐50 30 40 50 60 70
755 kPa
‐100
1006 kPa
‐150
‐200
‐250
‐300
‐350
‐400
Figure 60. Vertical Modulus Contours in the Subgrade at Different Loading Levels.
Next, the models were applied to study the effects of the pavement structure, properties,
and moisture conditions of the base and subgrade materials to the pavement performance. The
models for the MR of the unbound base layer and the permanent deformation (rut depth) in the
106
unbound base layer proposed by the Pavement ME Design Guide and researchers were
compared.
The validation of the proposed model and comparison between the proposed model and
the model in the Pavement ME Design were presented in previous studies (213, 214). The
proposed model by researchers predicts the MR change of unbound materials due to moisture
change more accurately since it considers the influence of the moisture change on the stress state
of the material.
Eqs. 5.15 and 5.20 present the rut depth that is the permanent deformation in the base
layer under repeated loads. The material properties required in these two models are obtained
from LTPP database and presented in Table 26.
Figure 61 shows the rut depth in the base layer using two models when other factors such
as the pavement structure, material properties, loading level, and number of load repetitions are
the same. When using the model proposed by researchers, the rut depth is greater.
0.4
Total Rut Depth, in
0.3 Model in
Pavement ME
Design
0.2
New Model
0.1
Figure 61. Rut Depth in the Base Layer Using Different Models.
107
The sensitivity analyses were performed on the FE models with the proposed
base/subgrade models to determine the effects of the load level, pavement structure, material
properties, and moisture conditions on the pavement performance, in which the rutting model is
the proposed rutting model.
Figure 62 presents the pavement performance under different loading levels as computed
with the COMSOL using models developed by researchers. For the fatigue cracking, the loading
repetitions are calculated based on the Pavement ME Design Guide criteria. The number of load
repetitions decreases dramatically with the increase of the loading level. The total rut depth in the
base layer is calculated using the permanent deformation model proposed by researchers when
the loading repetition number is 100,000. The rut depth increases with the increase of the loading
level.
4.00E+05
3.50E+05
3.00E+05 566 kPa (9 kips)
2.50E+05
2.00E+05 755 kPa (12 kips)
N
1.50E+05 1006 kPa (16
1.00E+05 kips)
5.00E+04
0.00E+00
(a)
4 566 kPa (9 kips)
Total Rut Depth, in
3.5
3 755 kPa (12 kips)
2.5
2 1066 kPa (16
1.5 kips)
1
0.5
0
(b)
Figure 62. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure; (b) Rut Depth in the Base at Different Loading Levels.
108
Figure 63 presents the pavement performance when using stresses and strains as
computed with COMSOL and material models developed by researchers and when the thickness
of the asphalt layer varies. Based on the Eqs. 5.10–11, the thickness has opposite effects on the
terms k’1 and the tensile strain at the critical location. Hence, an increase of the thickness of the
asphalt layer delays the top-down cracking when the thickness is small. In terms of bottom-up
cracking and the total rut depth in the base layer, the increase of the asphalt layer improves the
pavement performance.
1.40E+06
1.20E+06
4‐in AC
1.00E+06
8.00E+05 6‐in AC
N
6.00E+05
8‐in AC
4.00E+05
2.00E+05
0.00E+00
(a)
3.5
Total Rut Depth, in
2.5 4‐in AC
2
6‐in AC
1.5
1 8‐in AC
0.5
(b)
Figure 63. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure; (b) Rut Depth in the Base at Different Thickness of the Asphalt Layer.
Figure 64 presents the pavement performance when using stresses and strains as
computed with COMSOL and material models developed by researchers and when the thickness
109
of the base layer varies. As the thickness of the base layer increases, the tensile strain at the
bottom of the asphalt layer decreases, which reduces the load repetitions to the fatigue cracking.
For the rut depth in the base layer, an increase of the base layer thickness reduces the rut depth in
the base layer.
4.00E+05
3.80E+05
6‐in Base
3.60E+05 10‐in Base
N
3.40E+05 15‐in Base
3.20E+05
3.00E+05
(a)
0.7
Total Rut Depth, in
0.6
0.5
6‐in Base
0.4
10‐in Base
0.3
0.2 15‐in Base
0.1
0
(b)
Figure 64. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure and (b) Rut Depth in the Base at Different Thickness of the Base Layer.
Figure 65 presents the pavement performance when stresses and strains computed with
COMSOL and the material models developed by researchers and when the moisture of the base
layer varies. As the moisture content of base layer increases, the modulus of the base decreases,
which reduces the load repetitions to the fatigue cracking. For the rut depth, when the moisture
content of the base layer changes from 8.76 percent to 10.70 percent, the total rut depth also
increases.
110
4.50E+05
4.00E+05
3.50E+05 Dry Condition
3.00E+05
2.50E+05 Optimum
N 2.00E+05 Condition
1.50E+05
Wet Condition
1.00E+05
5.00E+04
0.00E+00
(a)
0.6
Total Rut Depth, in
0.5 Dry Condition
0.4
Medium
0.3 Condition
0.2 Wet
Condition
0.1
(b)
Figure 65. Pavement Performance Including (a) Load Repetitions to the Fatigue Cracking
Failure and (b) Rut Depth in the Base at Different Moisture Conditions of the Base Layer.
The sensitivity analysis was performed on both the proposed first (or full) and second (or
load-related) faulting models to assess the effect of various relevant performance-related
variables on the development of faulting. Because the effect on the development of faulting not
only depended on the designated variable but also on the interactions with other variables, the
sets of selected variables should be values at their most normal scales when varying the
designated variable. For simplicity, a LTPP pavement section of 1-3028 was used for conducting
sensitivity analysis of the proposed faulting models. Figure 66 shows the pavement structure for
the LTPP section 1-3028. Table 27 presents the basic information of the LTPP section 1-3028 for
sensitivity analysis.
111
Table 27. Basic Information of LTPP Section 1-3028 for Sensitivity Analysis.
Information Value
State Alabama
County Jefferson
Climatic zone WNF
Thickness of surface 10.2"
Thickness of base layer 7"
Type of base layer Unbound
Use of dowel No
Freeze-thaw cycles 60.88
Annual number of wet days 124
Annual number of days temperature greater than 32°C 52
Annual number of days temperature lower than 0°C 63
Use of Dowel
The effect of whether to use dowels in jointed concrete pavement on the development of
faulting was investigated. The results of the investigation based on both faulting models are
shown in Figure 67 indicating use of dowels in jointed concrete pavement always results in
lower faulting based on both proposed faulting models. The effect of the use of dowels on
faulting is to greatly diminish faulting.
112
(a) (b)
Figure 67. Effect of Use of Dowels on Faulting Based on (a) Full Faulting Model and (b)
Load-Related Faulting Model.
The effect of the type of base layer (bound or unbound) on the development of faulting
was studied. From Figure 68a, even though the faulting begins earlier when bound base layer
was applied, the ensuing faulting depth grows faster and greater when unbound baser layer was
applied. For the second faulting model, Figure 68b shows that selection of bound base layer
diminishes faulting depth.
(a) (b)
Figure 68. Effect of Types of Base Layer on Faulting Based on (a) Full Faulting Model and
(b) Load-Related Faulting Model.
113
Figure 69 shows the sensitivity of the thickness of the base layer on the development of
faulting based on both the first and the second faulting models. From Figure 69a, it seems that
the effect of the thickness of the base layer has slight impact on development of faulting. The
thicker base layer can reduce the faulting depth even if it causes earlier occurrence of faulting.
Obviously, in Figure 69b, an increase in thickness of base results in a decrease in the
development of faulting.
(a) (b)
Figure 69. Effect of Thickness of Base Layer on Faulting Based on (a) Full Faulting Model
and (b) Load-Related Faulting Model.
Freeze-Thaw Cycles
To illustrate the effect of the annual average freeze-thaw cycles on faulting, three
different values of this variable are used in the sensitivity analysis. Figure 70 shows that by
varying the freeze-thaw cycles in the two faulting models. A larger freeze-thaw cycle results in
more faulting. This is consistent with the fact that a major cause of faulting is temperature
variations. The greater the freeze-thaw cycles, the larger temperature variations that can severely
curve slabs upward or downward and produce more faulting.
114
(a) (b)
Figure 70. Effect of Freeze-Thaw Cycles on Faulting Based on (a) Full Faulting Model and
(b) Load-Related Faulting Model.
The effect of the annual average number of days with temperature greater than 32°C
(days32C) on faulting was studied. Figure 71b shows that the greater days32C results in less
faulting based on the load-related faulting model, which indicates the deformation of underlying
layers. In Figure 71a, for the full model, the initial segments of the three curves prior to the
critical inflection point basically tally with the results in Figure 71b. After passing the critical
inflection point where erosion begins to dominate the rate of faulting, the order of the faulting
magnitude reverses the order in Figure 71b. Thus, prior to reaching the critical inflection point,
the greater the number of days with days32C, the less the pre-critical faulting. After passing the
critical point where erosion is the principal cause of accelerating faulting, the order of faulting
magnitude reverses.
(a) (b)
Figure 71. Effect of Number of Days with Temperature Greater than 32°C on Faulting
Based on (a) Full Faulting Model and (b) Load-Related Faulting Model.
115
Figure 72 shows the development of faulting in different climate zones. The faulting
develops faster and greatest in the WF climatic zone and accumulates slowest and lowest in the
DF climate zone from Figure 72a but in WNF climate zone from Figure 72b. This difference
comes from the unbalanced data sets regression, which indicates section sizes for four different
climatic zones were unbalanced for first and second faulting model regressions.
(a) (b)
Figure 72. Effect of Climatic Zone on Faulting Based on (a) First Faulting Model and
(b) Second Faulting Model.
Researchers developed a moisture and suction dependent MR model, as noted in Eq. 4.20.
Similarly, to improve the slab-base interface bond sensitivity in rigid pavement performance, an
ANN model was developed in Chapter 4. To examine the sensitivity of the moisture and degree
of bonding on pavement performance, the proposed models were applied in eight LTPP
pavement sections listed in Table 28.
116
Table 28. Selected LTPP Pavement Sections and FWD Backcalculated Modulus Values for
Each Layer.
Backcalculated values
Slab Base
Climate State SHRP
State thickness thickness Slab Base Subgrade δ
zone code ID modulus modulus modulus
(in.) (in.)
(psi) (psi) (psi)
Wet- Minnesota 27 4034 10 3.6 5342000 98000 11000 0.52
Freeze Kentucky 21 4025 9.8 6 5693000 195000 25000 0.9
Alabama 01 0606 10.3 6.3 7798000 195000 8000 0.5
Wet-
North
Nonfreeze 37 5037 7.8 15.1 4875000 20000 14000 0.11
Carolina
Colorado 08 7776 10.7 15.3 4147000 100000 27000 0.48
Dry-
North
Freeze 38 3006 8.5 3.8 10000000 199000 50000 0.37
Dakota
New
Dry- 35 3010 7.9 6.9 7171000 64000 22000 0.22
Mexico
Nonfreeze
Arizona 04 0214 8.3 6.1 7087000 98000 25000 0.32
Each pavement section in Table 29 consisted of one surface, one base, and a subgrade
layer. The modulus values for each layer are calculated using FWD backcalculation procedure.
To illustrate the effect of moisture in subgrade k-value and therefore, pavement performance, MR
values were calculated first at three different moisture conditions. Table 29 lists the MR values
for each pavement structure at three selected moisture conditions: 1) equilibrium volumetric
water content + 10 percent; 2) equilibrium volumetric water content; and 3) equilibrium
volumetric water content −10 percent. Equilibrium suction and the corresponding volumetric
water content were calculated using Eq. 4.15 and the SWCC equation developed by Fredlund
and Xing as shown in Eqs. 4.1 and 4.2. The four fitting parameters (af, bf, cf, and hr) of the
SWCC equation were predicted from the ANN model developed by researchers.
117
Table 29. Calculated MR Values at the Mid-depth of Base Layer at Different Moisture
Conditions.
State SHR Saha et al. (133) -hm Saha et al. (216) MR
δ θ f
Code P ID af bf cf hr (kPa) k1 k2 k3 (MPa)
0.0030
0 2621 1 37.2
6
27 4034 0.52 4.91 2.62 1.65 3000 0.0034 1931 1 689.3 0.66 -0.03 33.9
0.0037
1 1635 1 32.8
4
0 0.0418 5976 1 880.2
21 4025 0.9 5.86 0.34 1.74 2999 0.0465 3845 1 945.55 0.67 -0.29 707.5
1 0.0511 2498 1 569.2
0 0.1265 9386 1 1397.2
01 0606 0.5 6.71 1.01 0.07 2998 0.1406 1750 2.74 913.7 0.73 -0.03 930.3
1 0.1546 41 5.81 115.8
0 0.0606 732 1 126.3
37 5037 0.11 7.57 0.98 1.08 2999 0.067 533 1 431.43 0.92 -0.23 105.7
1 0.0741 403 1 91.5
0 0.072 3832 1 167.2
-
08 7776 0.48 1.06 1.01 0.69 2999 0.08 1858 1 983.52 0.207 147.6
0.027
1 0.088 975 1 132.4
0 0.038 3778 1 184.2
38 3006 0.37 1.00 1.01 0.79 2999 0.042 1961 1 544.43 0.65 -0.08 131.1
1 0.046 1100 1 98.5
0 0.012 3189 1 114.3
-
35 3010 0.22 5.30 3.35 1.05 2998 0.013 2093 1 859.5 0.73 92.5
0.025
1 0.015 1565 1 84.8
0 0.066 1995 1 162.9
04 0214 0.32 5.05 0.12 2.18 2999 0.073 533 1 900.14 0.509 0.047 92.3
1 0.08 124 1 54.7
The coefficients of the proposed MR model were predicted using the ANN model shown
in Figure 17. To compare the sensitivity of moisture on subgrade k-value, the MR values were
calculated at different moisture conditions using the proposed model and the Pavement ME
Design model. After that, the subgrade k-values were calculated using the developed ANN
model shown in Figure 36 and Pavement ME Design approach. Table 29 also list three different
bonding conditions for each pavement structure: 1) no bond, 2) partially bonded, and 3) fully
bonded. Each bonding condition was applied as an input in the developed ANN model and
compared with the predicted k-values from Pavement ME Design. Figure 73 presents the
sensitivity of degree of bonding on subgrade k-value using the developed ANN model and the
Pavement ME Design model. Figure 74 illustrates the effect of moisture on subgrade k-value.
118
8.E+07
ANN predicted k-value (N/m3)
38.44%
No bond Partial bond Full bond
18.4%
6.E+07
‐0.22%
‐0.33%
2.87%
13.99%
0.46%
9.82%
2.97%
1.07%
4.E+07
13.82%
1.88%
2.74%
2.66%
10.18%
3.07%
2.E+07
0.E+00
8.E+07
Full bond
38.07%
Pavement ME k-value (N/m3)
No bond
6.E+07
‐0.34%
2.83%
13.63%
2.76%
4.E+07
13.74%
2.51%
9.98%
2.E+07
0.E+00
Figure 73. Sensitivity of Degree of Bonding on Subgrade k-value Using (a) ANN Model and
(b) Pavement ME Design Model.
119
8.E+07 Eq. vol. wc. + 10%
ANN predicted k-value (N/m3)
Eq. vol. wc.
Eq. vol. wc. - 10%
6.37%
‐0.0037%
1.78%
2.73%
0.83%
6.E+07
3.18%
0.12%
0.40%
4.89%
2.23%
‐2.75%
‐5.34%
4.E+07
0.08%
0.30%
14.237%
20.42%
2.E+07
0.E+00
Eq. vol. wc. + 10%
Eq. vol. wc.
Eq. vol. wc. - 10%
2.3E‐04%
1.5E‐04%
6.E+07
0.27%
1.20%
1.16%
0.65%
‐0.07%
‐0.03%
1.2E‐05%
2.5E‐05%
4.E+07
3.50%
‐0.37%
1.68%
‐0.68%
2.E+07
‐1.09%
‐0.49%
0.E+00
Figure 74. Sensitivity of Moisture on Subgrade k-value Using (a) ANN Model and
(b) Pavement ME Design Model.
Figure 73 shows that both ANN model and the Pavement ME Design have similar
sensitivity of degree of bonding on k-value. But, as shown in Figure 73a, ME design model has
no partial bonding condition whereas ANN model can predict k-value at any bonding conditions.
In both cases, degree of bonding has higher sensitivity when the base-slab modulus ration is
relatively higher. LTPP section 21-4025 and 08-7776 have a higher modulus ratio compared to
the other sections and therefore shows a significant change in k-value due to the change in
degree of bonding.
120
Figure 74 compares the sensitivity of moisture on k-value using the ANN model and
Pavement ME Design model. Pavement ME Design has almost no sensitivity of moisture on
subgrade k-value. But the proposed ANN model shows relatively higher sensitivity.
Although the combination of moisture-sensitive MR model and the developed ANN
model show large sensitivity to moisture and slab-base degree of bonding on k-value, it is still
necessary to evaluate the effect of moisture and degree of bonding on pavement performance.
The next section describes the prediction of the fatigue cracking (top-down and bottom-up) and
faulting performance for various bonding and moisture conditions and compares with the
predicted performance using Pavement ME Design model.
121
2.96%
Tensile stress at surface (MPa)
0.89%
1.54%
2.46%
0.23%
‐2.04%
No bond
8.11%
3.01%
0.64%
6.94%
No bond
2.01%
1.2 1.2
12.41%
7.40%
3.45%
‐0.13%
0.43%
3.85%
4.27%
0.96%
1.8%
17.49%
0.9 0.9
2.65%
7.55%
0.6 0.6
0.3 0.3
0 0
‐0.11%
2 Full bond 2
3.18%
2.68%
2.74%
Full bond
1.66%
0.29%
8.76%
Tensile stress at slab bottom (MPa)
9.02%
7.53%
Partial bond
3.63%
3.41%
2.41%
No bond
1.6 No bond 1.6
‐0.83%
6.44%
4.27%
1.91%
7.2%
2.01%
1.34%
0.99%
5.2%
21.28%
14.26%
1.2 1.2
9.31%
0.8 0.8
0.4 0.4
0 0
122
0.4 0.4
Differential slab deflection (mm) Full bond Full bond
4.15%
3.83%
3.46%
2.46%
2.20%
1.66%
0.90%
0.77%
0.37%
1.16%
0.73%
0.87%
13.21%
12.65%
1.60%
4.51%
0.73%
4.18%
0.07%
1.49%
1.16%
1.28%
7.99%
6.16%
0.2 0.2
0.1 0.1
0 0
Figure 75. PCC Slab-Base Interface Bond Sensitivity on (a) Tensile Stress at Top of Slab;
(b) Tensile Stress at Bottom of Slab; and (c) Deferential Deflection on Transverse Joints.
The developed ANN model has larger sensitivity of tensile stress and differential
deflection due to change in degree of bonding at slab-base interface. For both ANN model and
Pavement ME Design model, fully bonded condition shows the lowest tensile stress and
differential deflection whereas no bonding between slab and base develop largest tensile stress
and deflection. As seen, Pavement ME Design model can only calculate tensile stress and
deflection at two extreme bonding conditions, but the developed ANN model has the capability
of predict tensile stress and deflection at partially bonded condition as well.
Figure 76 shows the sensitivity of moisture on tensile stress and differential deflection
using the proposed moisture and suction dependent MR model and Pavement ME Design model.
123
0.33%
0.12%
0.12%
0.0005%
0.0002%
0%
0%
Eq. vol. wc. - 10% Eq. vol. wc. - 10%
0.48%
0%
0%
0.33%
1.2 1.2
1.96%
1.10%
0.00046%
‐0.02%
0.22%
‐0.0002%
3.33%
‐0.15%
0.50%
0.9 0.9
0%
0%
0%
0%
1.10%
0.59%
0%
0%
0%
0.6 0.6
0.3 0.3
0 0
0.001%
‐0.0001%
0.0002%
0.01%
Eq. vol. wc. - 10% Eq. vol. wc. - 10%
0.16%
0.16%
0.15%
0.15%
2 2
0.59%
0.42%
0%
0%
2.19%
1.23%
0.001%
0.001%
‐0.03%
‐0.02%
1.6 1.6 ‐0.01%
‐0.03%
0.015%
0.011%
0.03%
0.03%
0.97%
1.69%
0.20%
1.29%
0.10%
0.70%
1.2 1.2
0.8 0.8
0.4 0.4
0 0
124
‐0.02%
‐0.02%
0.49%
0.25%
‐0.04%
‐0.1%
0.04%
0.03%
‐0.45%
‐0.97%
0.03%
0.06%
1.03%
1.82%
0.06%
0.06%
0.17%
0.12%
0.24%
0.76%
1.38%
0.04%
0.31%
0.71%
0.21%
0.37%
0%
0%
0%
0%
0%
0%
0.2 0.2
0.1 0.1
0 0
Figure 76. Base Layer Moisture Sensitivity on (a) Tensile Stress at Top of Slab; (b) Tensile
Stress at Bottom of Slab; and (c) Deferential Deflection on Transverse Joints.
As shown in Figure 76, the Pavement ME Design model shows very low sensitivity of
tensile stress at slab top and bottom and differential deflection due to the change in moisture at
middle of base layer. However, the proposed MR model and the corresponding k-value from
ANN model shows higher sensitivity in tensile stress and differential deflection for change in
moisture. Tensile stress at slab top and bottom increases with the increase of moisture in base
layer. In summary, the models developed in this project are sensitive and capable of identifying
various pavement responses including stress and strain at the bottom and top of asphalt layer, and
distresses including cracking and rutting in asphalt pavement and faulting in concrete pavement
with different base and subgrade properties under different moisture and temperature conditions,
while in the Pavement ME Design such obvious differences cannot be observed.
125
CHAPTER 6. SUMMARY
CONCLUSIONS
This project proposed several enhancements to the Pavement ME Design with the
purpose of increasing the sensitivity of pavement performance to base layers and subgrade.
These enhancements include: a) ANN models to predict the SWCC by using the soil physical
properties and climatic parameters; b) ANN models to predict the stress and moisture-dependent
MR; c) a modified modulus of subgrade reaction (k) model that considered the cross anisotropy
of base material and the shear interaction between the PCC slab and the base course; d) a new
faulting model for predicting the faulting depth at joints along the wheel path in jointed concrete
pavements over time by using the LTPP data; e) subroutines written in the C# language to
implement all of the enhancements described above (Appendix I to Appendix M); and f)
characteristic equations that relate the equilibrium soil moisture suction beneath a pavement to
the TMI for each of the AASHTO Soil Classes of base course and subgrade soils.
This last item is essential to make the Pavement ME Design sensitive to moisture
throughout the United States. The present version of Pavement Design accounts for the presence
of moisture in the pavement layers only in those cases where the water table elevation is within
approximately 33 ft (10 m) of the pavement surface. When the water table is deeper than 10 m as
is usually the case in large portions of the United States, the pF beneath the pavement is
controlled by the long-term local climatic moisture balance between rainfall and
evapotranspiration from the soil and vegetation. The TMI provides a numerical determination of
the local moisture balance. The pF that develops at depths below the soil surface also depends
upon the soil water retention capacities of the subgrade soil in the form of its SWCC. Using the
relationship between the TMI and the SWCC of local soils and the soil map of the United States
that is provided by USDA-NRCS, researchers have developed equations that predict the pF
beneath a pavement at any location within the United States. This development permits the
determination of the equilibrium soil moisture suction within each layer of a pavement around
which the suction with vary seasonally. As noted in Chapter 4, all the properties of the unbound
base layer and subgrade beneath a pavement are characterized by their dependence upon the
suction and its variations. When it is implemented within the Pavement ME Design software, this
development will provide the moisture sensitivity in design, which was one of the major
objectives of this project.
Soil Water Characteristics Curve of Base and Subgrade for Flexible and Rigid
Pavements
Two three-layered neural network architectures consisting of one input layer, one hidden
layer, and one output layer are constructed for plastic and non-plastic soil. The input
variables for plastic soil include the material percent passing the No. 4 sieve, material
percent passing the No. 200 sieve, LL, PI, saturated volumetric water content, and local
MAAT. The input variables for non-plastic soil are particle diameter corresponding to
30 percent, 60 percent, and 90 percent passing of material (D30, D60, and D90), ϴ, Ѱ,
saturated volumetric water content, and local MAAT. The hidden layer assigns 20
neurons. A total of 3600 plastic soil and 250 non-plastic soil data collected from the
NCHRP 9-23A project were used to develop the ANN models.
126
Compared to the existing prediction models, such as the Zapata and Perera models, the
developed ANN models have the highest accuracy (e.g., smallest RMSE and highest R2
values) to predict the SWCC fitting parameters in the Fredlund-Xing equation. The
developed ANN models can accurately estimate the matric suction of soil at any given
saturation level. The obtained R2 values are 0.95 and 0.91 for plastic and non-plastic
soils, respectively.
The prediction accuracy of the developed ANN models is validated by two data sources
(i.e., test data from the NCHRP 9-23A database and independent data from other
literatures). The comparison of model predicted matric suction values to the measured
ones validates that the developed ANN models are capable of accurately predicting the
SWCCs for both plastic and non-plastic soils.
Resilient Modulus of Base and Subgrade for Flexible and Rigid Pavements
Two three-layered ANN models were developed for plastic and non-plastic base
materials. Input variables for plastic and non-plastic soil included material percent
passing 3/8 in. sieve, material percent passing No. 200 sieve, plastic limit, PI, MDD,
OMC, and TMC. The output variables are the three coefficients k1, k2, and k3 of MR
model.
The developed ANN models showed a higher prediction accuracy compared to the three
regression models selected from the existing literature. The ANN models can accurately
estimate the MR of base materials at any stress level. The obtained R2 values are 0.91 and
0.90 for plastic and non-plastic base materials, respectively.
The ANN models do not provide any insight into the complex relationship between the
MR model coefficients and the base physical properties. Thus, it is not recommended to
use as a prediction tool for the values that are out of the range of training data set.
Appendix D contains the range of each variable used. In this study, researchers collected
a large data set of 779 base materials from the LTPP database and hence provided a wide
range of input properties. A sensitivity analysis should be conducted in a future study to
evaluate the influence of each input parameter on MR model coefficients.
The developed ANN models were validated using the MR test data from different sources.
The R2 value between the measured and predicted validation MR values was 0.8.
A slab-base interface shear bonding submodel was developed based on the shear strength
properties of the base course, c and . Depending on the degree of bonding in the slab-
base interface, the equivalent section, and the altered deflection basin, which further
affected the modified k-value.
The estimated degree of bonding values was compared with the previously developed
BBF approach bonding condition. In the BBF approach, the slab-base interface is only
considered as a non-bonded or fully bonded condition. However, this study found that
most of the treated base layers were either in fully bonded or partially bonded condition
and the unbound base layers were mostly partially bonded.
Modified k-values were compared with the BBF k-values. Significant changes in the k-
values were observed due to the base modulus and interface bonding corrections.
127
Modified k-values were compared with the BBF k-values. Significant changes in the
k-values were observed due to the base modulus and interface bonding corrections. The
BBF approach has higher k-value than the modified model due to the difference in
interface bonding ratios. The interface bonding between slab and base is considered fully
bonded for most of the cases while modified k-value model considered partially bonded
condition.
A three-layered ANN model was constructed to predict the modified k-value, which
included one input layer, one hidden layer, and an output layer. The FWD deflection
basins were computed using FE program. The developed ANN model was validated by
comparing the prediction results with the calculated modified k-values from the LTPP
pavement sections. The obtained R2 value of 0.92 indicated that the developed models
had a desirable accuracy in the prediction of the modified k-value. A sensitivity analysis
was conducted to evaluate the effect of the degree of bonding on k-value. The results
showed that, in general, a higher degree of bonding produces a higher modified k-value.
Two faulting prediction models were developed to estimate faulting over time and with
axle load distributions. One is to predict the entire faulting development over time. The
other is to predict the faulting depth prior to the inflection point in the faulting curve with
traffic (axle load distributions).
Field faulting data plot graphically as an S-shape curve with negative curvature prior to
the inflection point and positive curvature after the inflection point. The inflection point
is the critical point in the development of faulting. Prior to the inflection point faulting is
controlled by the permanent deformation in the supporting base course. After the
inflection point, water that has infiltrated through the joint in the concrete pavement and
filled the void created by the permanent deformation is driven by the passing traffic to
scour the surface of the base course. This erosion accelerates the rate of faulting, because
the critical inflection point signals the beginning of rapidly deteriorating erosion. The
structural design of concrete pavements should use the critical inflection point as a design
criterion. Erosion can be controlled if the predicted faulting depth can be kept below the
critical faulting depth.
The load-related faulting model depends on different stress states and axle load
distributions. The reasonable and consistent stress terms in the model can be determined
by using elastic analytical equations. The axle load distributions are determined by the
WIM data collected from LTPP data or estimated with AADTT.
Because of its potential importance for jointed concrete pavement design, a separate
model was developed to predict the critical faulting depth.
To better implement the proposed models, the coefficients in the models were statistically
calibrated with performance-related factors using a stepwise method to select the relevant
variables and a generalized linear model to perform multiple regression analysis. All
three predictive models fit the field data very well, and the models of the coefficients of
the full faulting and load-related faulting models have high R2-values and can be
expected to provide exceptionally reliable predictions of the development of faulting with
time and with accumulating traffic loads.
128
Based on the sensitivity analysis performed on the pavement models and the comparison
of models proposed by researchers and the Pavement ME Design Guide, several findings are
presented as follows:
As an unbound granular material, the nonlinearity and the anisotropy of the pavement
materials should be considered in the pavement design and analysis. With such
properties, the modulus of the base and subgrade are sensitive to the moisture content, the
loading level, and the pavement structure, which affect the pavement responses and
performance.
The proposed model for MR of the base and subgrade eliminates the steps to transfer the
modulus for different moisture conditions as in the Pavement ME Design Guide. The
modulus can be estimated based on the suction/moisture, depth, stress state, etc.
The proposed rutting model is more involved with the stress state and material properties.
The calculated rut depth is greater than the one using the current method in the Pavement
ME Design Guide.
The increase of the loading level reduces the load repetitions to the fatigue cracking
failure and increases the rut depth in the base layer.
The increase of the asphalt layer thickness increases the load repetitions to the fatigue
cracking failure and reduces the rut depth in the base layer.
The increase of the base layer thickness reduces the load repetitions to the fatigue
cracking failure and reduces the rut depth in the base layer.
As the moisture content increases in the base layer, the load repetitions to the fatigue
cracking failure reduces. The rut depth in the base layer shows an increasing trend with
the increasing moisture content.
Some factors can reduce the development of faulting, including the use of dowel in
jointed concrete pavement, the selection of stabilized (bound) base course and thicker
base course. On the contrary, the freeze-thaw cycle is a favorable impact on the
development of faulting. The more freeze-thaw cycles result in larger faulting. This is
consistent with the fact that a major cause of faulting is temperature variations.
Similarity, the faulting develops faster and greater in the WF climatic zone. Additionally,
the greater of days32C results in less faulting but after the critical inflection point, the
order of faulting magnitude reverses.
The ANN model for k-value was used to evaluate the sensitivity of moisture and degree
of bonding using the proposed models. The sensitivity of modified k-values was found to
be improved significantly to moisture and degree of bonding compared to the exiting
Pavement ME Design k-values. The sensitivity of moisture increased for all selected
pavement sections due to the inclusion of suction effect in the MR model. However, the
effect of degree of bonding is quite same for modified k-value and Pavement ME Design
k-value when there is either fully bonded or no bond condition in slab-base interface. But
the developed ANN model can predict k-value for partially bonded condition as well.
To study the effect of moisture and degree of bonding on pavement performance, tensile
stress at top and bottom of slab and differential deflection across the transverse joint were
evaluated. The sensitivity of moisture and degree of bonding were calculated using the
proposed MR and modified k-value model and compared with the results from Pavement
ME Design MR and k-values. The MR values using the proposed model and the
129
corresponding modified k-values from ANN model showed much higher sensitivity on
calculated stress and deflections compared to the results from Pavement ME Design
models.
There are several items of future work that emerge from the work that has been
accomplished in this project including the following:
Replace the models that are currently in the Pavement ME Design software by those that
have been developed in this project. All these models are currently incorporated in
separate subroutines that can be added to the current version of Pavement ME Design
software.
Incorporate the TMI- AASHTO Soil Class – Equilibrium Soil Moisture Suction
relationships that have been developed in this project into the Pavement ME Design
software.
Develop a mechanistic method of predicting the post-critical faulting of jointed concrete
pavements (after the inflection point) to include the effects of scour of the surface of the
base course by water trapped in the void beneath to concrete slab, which is propelled by
the passage of approaching and receding traffic. The model that was developed in this
project predicts the faulting in the wheelpath. What is needed in addition is the faulting
that develops in the corners and along the edges of a concrete pavement where the curling
and warping of the pavement is typically larger than what occurs in the wheelpath.
Develop the properties of stabilized base course materials as they are controlled and
affected by the soil moisture suction, and incorporate those relationships into the
Pavement ME Design software.
Re-calibrate the prediction equations of the IRI for both the asphalt-surfaced and
concrete-surfaced pavements after incorporating the models that have been developed in
this project.
130
REFERENCES
1. American Association of State Highway and Transportation Officials (AASHTO). 2008.
Mechanistic-Empirical Pavement Design Guide—A Manual of Practice. Interim Edition,
Washington, DC.
2. Schwartz, C.W., R. Li, S. Kim, H. Ceylan, and K. Gopalakrishnan. 2011. Sensitivity
Evaluation of MEPDG Performance Prediction. Final Report of NCHRP Project 1-47,
University of Maryland and Iowa State University. Available at:
http://onlinepubs.trb.org/onlinepubs/nchrp/docs/NCHRP01-47_FR.pdf.
3. Shahji, S. 2006. Sensitivity Analysis of AASHTO’s 2002 Flexible and Rigid Pavement
Design Methods. Master of Sc. in the Dept. of Civil & Environmental Engineering—
College of Engineering and Computer Science—University of Central Florida. Orlando, FL.
4. Masad, S.A., and D.N. Little. 2004. Sensitivity Analysis of Flexible Pavement Response and
AASHTO 2002 Design Guide to Properties of Unbound Layers. Research Report ICAR
504-1. International Center for Aggregates Research, Austin, TX.
5. Masad, S., D. Little, and E. Masad. 2006. “Analysis of Flexible Pavement Response and
Performance Using Isotropic and Anisotropic Material Properties.” Journal of
Transportation Engineering-ASCE, Vol. 132, No. 4, pp. 342–349.
6. Brown, S.F. 1996. “Soil Mechanics in Pavement Engineering.” Geotechnique, Vol. 46, No.
3, pp. 383–426.
7. Theyse, H.L., F.T.H. Legge, P.C. Pretorius, and H. Wolff. 2007. “A Yield Strength Model
for Partially Saturated Unbound Granular Material.” Road Materials and Pavement Design,
Vol. 8, No. 3, pp. 423–448. doi: Doi 10.3166/Rmpd.8.423-448
8. Núñez, W.P., R. Malysz, J.A. Ceratti, and W.Y.Y. Gehling. 2004 “Shear Strength of
Permanent Deformation of Unbound Aggregates Used in Brazilian Pavements.” In Proc. Of
the 6th International Symposium on Pavements Unbound, pp. 23–31
9. Zhou, F., E.G. Fernando, and T. Scullion. 2010. Development, Calibration, and Validation
of Performance Prediction Models for the Texas M-E Flexible Pavement Design System.
Research Report FHWA/TX-10/0-5798-2. Texas Transportation Institute, College Station,
TX.
10. Gabr, A.R., and D.A. Cameron. 2012. “Permanent Strain Modeling of Recycled Concrete
Aggregate for Unbound Pavement Construction.” Journal of Materials in Civil Engineering,
Vol. 25, No. 10, pp. 1394–1402.
11. Chow, L.C., D. Mishra, and E. Tutumluer. 2014. “Framework for Development of an
Improved Unbound Aggregate Base Rutting Model for Mechanistic–Empirical Pavement
Design.” In Transportation Research Record: Journal of the Transportation Research
Board, No. 2401, pp. 11–21.
12. Cerni, G., F. Cardone, A. Virgili, and S. Camilli. 2012. “Characterisation of Permanent
Deformation Behaviour of Unbound Granular Materials under Repeated Triaxial Loading.”
Construction and Building Materials, Vol. 28, No. 1, pp. 79–87. doi:
http://dx.doi.org/10.1016/j.conbuildmat.2011.07.066.
13. Adu-Osei, A., D.N. Little, and R.L. Lytton. 2001. “Cross-Anisotropic Characterization of
Unbound Granular Materials.” In Transportation Research Record: Journal of the
Transportation Research Board, No. 1757, pp. 82–91.
131
14. Sahin, H., F. Gu, Y. Tong, R. Luo, and R.L. Lytton. 2013. “Unsaturated Soil Mechanics in
the Design and Performance of Pavements.” Advances in Unsaturated Soils, pp. 87–99.
15. Witczak, M.W. 2003. Harmonized Test Methods for Laboratory Determination of Resilient
Modulus for Flexible Pavement Design. Final Report of NCHRP Project 1-28A,
Transportation Research Board, Washington, D.C.
16. Butalia, T.S., J. Huang, D.G. Kim, and F. Croft. 2003. “Effect of Moisture Content and Pore
Water Pressure Buildup on Resilient Modulus of Cohesive Soils in Ohio.” ASTM Special
Technical Publication, No. 1437, pp. 70–84.
17. Wolfe, W., and T. Butalia. 2004. Continued Monitoring of SHRP Pavement Instrumentation
Including Soil Suction and Relationship with Resilient Modulus. Report No. FHWA/
OH-2004/007. U.S. Department of Transportation, Federal Highway Administration,
Washington, DC.
18. Gupta, S., A. Ranaivoson, T. Edil, C. Benson, and A. Sawangsuriya. 2007. Pavement Design
Using Unsaturated Soil Technology. Report No. MN/RC-2007-11. Final Research Report
submitted to Minnesota Department of Transportation, University of Minnesota,
Minneapolis, MN.
19. Cary, C.E., and C.E. Zapata. 2011. “Resilient Modulus for Unsaturated Unbound Materials.”
Road Materials and Pavement Design, Vol. 12, No. 3, pp. 615–638.
20. Carpenter, S.H., and R.L. Lytton. 1977. Thermal Pavement Cracking in West Texas: Final
Report. Research Report 18-4F. Texas Transportation Institute, Texas A&M University,
College Station, TX.
21. Hansen, W., and E.A. Jensen. 2001. Transverse Crack Propagation of JPCP as Related to
PCC Toughness. Research Report: RC-1404. Michigan Department of Transportation,
Lansing, MI.
22. Cleveland, G.S., J.W. Button, and R.L. Lytton. 2002. Geosynthetics in Flexible and Rigid
Pavement Overlay Systems to Reduce Reflection Cracking. Research Report 1777-1. Texas
Transportation Institute, The Texas A&M University System, College Station, TX.
23. Lytton, R.L., F.L. Tsai, S.I. Lee, R. Luo, S. Hu, and F. Zhou. 2010. NCHRP Report 669:
Models for Predicting Reflection Cracking of Hot Mix Asphalt Overlays. Transportation
Research Board, Washington, DC.
24. Jung, Y.S., and D.G. Zollinger. 2011. “New Laboratory-Based Mechanistic-Empirical Model
for Faulting in Jointed Concrete Pavement.” In Transportation Research Record: Journal of
the Transportation Research Board, No. 2226, pp. 60–70.
25. Jeong, J.H., and D.G. Zollinger. 2001. “Characterization of Stiffness Parameters in Design of
Continuously Reinforced and Jointed Pavements.” In Transportation Research Record:
Journal of the Transportation Research Board, No. 1778, pp. 54–63.
26. Jung, Y.S., and D.G. Zollinger. 2010. Advancement of Erosion Testing, Modeling, and
Design of Concrete Pavement Subbase Layers. PhD diss., Texas A&M University, College
Station, TX. Retrieved from http://hdl.handle.net/1969.1/ETD-TAMU-2010-08-8415.
27. Bakhsh, K.N., and D. Zollinger. 2014. “Faulting Prediction Model for Design of Concrete
Pavement Structures.” Pavement Materials, Structures, and Performance, pp. 327–342. doi:
10.1061/9780784413418.033
28. Jung, Y.S., D.G. Zollinger, and B.M. Ehsanul. 2012. “Improved Mechanistic-Empirical
Continuously Reinforced Concrete Pavement Design Approach with Modified Punchout
Model.” In Transportation Research Record: Journal of the Transportation Research Board,
No. 2305, pp. 32–42.
132
29. Vandenbossche, J.M., S. Nassiri, L.C. Ramirez, and J.A. Sherwood. 2012. “Evaluating the
Continuously Reinforced Concrete Pavement Performance Models of the Mechanistic-
Empirical Pavement Design Guide.” Road Materials and Pavement Design, Vol. 13, No. 2,
pp. 235–248.
30. Rao, C., and M.I. Darter. 2013. “Enhancements to Punchout Prediction Model in
Mechanistic-Empirical Pavement Design Guide Procedure.” In Transportation Research
Record, No. 2367, pp. 132–141.
31. Gu, F., X. Luo, Y. Zhang, H. Sahin, and R.L. Lytton. 2015. “Modeling of Moisture-
Sensitive and Stress-Dependent Nonlinear Cross-Anisotropic Behavior of Unbound
Aggregates.”
32. Ren, D.Y., L. Houben, and L. Rens. 2013. “Cracking Behavior of Continuously Reinforced
Concrete Pavements in Belgium Characterization of Current Design Concept.” In
Transportation Research Record: Journal of the Transportation Research Board, No. 2367,
pp. 97–106.
33. Jung, Y.S., D.G. Zollinger, M. Won, and A.J. Wimsatt. 2009. Subbase and Subgrade
Performance Investigation for Concrete Pavement. Research Report 6037-1. Texas
Transportation Institute, Texas A&M University, College Station, TX.
34. Jung, Y.S., D.G. Zollinger, and A.J. Wimsatt. 2010. “Test Method and Model Development
of Subbase Erosion for Concrete Pavement Design.” In Transportation Research Record:
Journal of the Transportation Research Board, No. 2154, pp. 22–31.
35. Byrum, C.R., and R.W. Perera. 2005. “The Effect of Faulting on IRI Values for Jointed
Concrete Pavements.” Paper presented at the Proc. 8th International Conference on Concrete
Pavements, International Society for Concrete Pavements, Bridgeville, PA, and Purdue
University, West Lafayette, IN.
36. Bakhsh, K.N. 2014. Design Methodology for Subgrades and Bases under Concrete Roads
and Parking Lots. PhD diss., Texas A&M University, College Station, TX.
37. Jung, Y., D.G. Zollinger, B.H. Cho, M. Won, and A.J. Wimsatt. 2010. Subbase and
Subgrade Performance Investigation and Design Guidelines for Concrete Pavement.
Research Report No. FHWA/TX-12/0-6037-2. Texas Transportation Institute, The Texas
A&M University System, College Station, TX.
38. Yu, S.Y., and P. Dakoulas. 1993. “General Stress-Dependent Elastic-Moduli for Cross-
Anisotropic Soils.” Journal of Geotechnical Engineering-ASCE, Vol. 119, No. 10, pp.
1568–1586. doi: 10.1061/(ASCE)0733-9410(1993)119:10(1568)
39. Oh, J.H., R.L. Lytton, and E.G. Fernando. 2006. “Modeling of Pavement Response Using
Nonlinear Cross-Anisotropy Approach.” Journal of Transportation Engineering-ASCE, Vol.
132, No. 6, pp. 458–468.
40. Khoury, N.N., and M.M. Zaman. 2004. “Correlation between Resilient Modulus, Moisture
Variation, and Soil Suction for Subgrade Soils.” In Transportation Research Record:
Journal of the Transportation Research Board, No. 1874, pp. 99–107.
41. Yang, S.-R., W.-H. Huang, and Y.-T. Tai. 2005. “Variation of Resilient Modulus with Soil
Suction for Compacted Subgrade Soils.” In Transportation Research Record: Journal of the
Transportation Research Board, No. 1913, No. 1, pp. 99–106.
42. Sawangsuriya, A., T.B. Edil, and P.J. Bosscher. 2008. “Modulus-Suction-Moisture
Relationship for Compacted Soils.” Canadian Geotechnical Journal, Vol. 45, No. 7, pp.
973–983.
133
43. Sawangsuriya, A., T.B. Edil, and C.H. Benson. 2009. “Effect of Suction on Resilient
Modulus of Compacted Fine-Grained Subgrade Soils.” In Transportation Research Record:
Journal of the Transportation Research Board, No. 2101, pp. 82–87. Doi 10.3141/2101-10
44. Khoury, C., G.A. Miller, and Y.N. Abousleiman. 2010. “Effect of Suction Hysteresis on
Resilient Modulus of Fine-grained Cohesionless Soil.” 8th Int. Conf. Bearing Capacity
Roads, Railways, and Airfields, Univ. of Illinois–Urbana Champaign, Champaign, IL, pp.
71–78.
45. Han, Z., and S.K. Vanapalli. 2015. “Model for predicting the resilient modulus of
unsaturated subgrade soil using the soil-water characteristic curve.” Canadian Geotechnical
Journal, Vol. 52, No. 10, pp.1605–1619.
46. Li, Q., H.J. Lee, and S.Y. Lee. 2011. “Permanent Deformation Model Based on Shear
Properties of Asphalt Mixtures Development and Calibration.” In Transportation Research
Record: Journal of the Transportation Research Board, No. 2210, pp. 81–89.
47. Vaillancourt, M., L. Houy, D. Perraton, and D. Breysse. 2014. “Variability of Subgrade Soil
Rigidity and its Effects on the Roughness of Flexible Pavements: A Probabilistic Approach.”
Materials and Structures, pp. 1–10.
48. Velasquez, R., K. Hoegh, I. Yut, N. Funk, G. Cochran, M. Marasteanu, and L. Khazanovich.
2009. “Implementation of the MEPDG for New and Rehabilitated Pavement Structures for
Design of Concrete and Asphalt Pavements in Minnesota.” MnDOT Report No. MN/RC
2009-06. University of Minnesota—Twin Cities, St. Paul.
49. Huang, Y.H. 1993. Pavement Analysis and Design: Prentice Hall, Englewood Cliffs, NJ.
50. Seed, H.B., F.G. Mitry, C.L. Monismith, and C.K. Chan. 1967. NCHRP Report 35:
Prediction of Flexible Pavement Deflections from Laboratory Repeated-Load Tests. HRB,
National Research Council, Washington, D.C.
51. Hicks, R.G., and C.L. Monismith. 1971. “Factors Influencing the Resilient Properties of
Granular Materials.” In Highway Research Record, No. 345, pp. 15–31.
52. Thompson, M.R., and R.L. Robnett. 1979. “Resilient Properties of Subgrade Soils.”
Transportation Engineering Journal, ASCE, Vol. 105, No. TE1, pp. 71–89.
53. Drumm, E.C. 1990. “Estimation of Subgrade Resilient Modulus from Standard Tests.”
Journal of Geotechnical Engineering, Vol. 116, No 5, pp. 774–789.
54. Uzan, J. 1985. “Characterization of Granular Material.” In Transportation Research Record,
No. 1022, pp. 52–59.
55. Witczak, M.W. and J. Uzan. 1988. The University Airport Pavement Design System Report
I of V: Granular Material Characterization. Department of Civil Engineering, University of
Maryland, Silver Spring, MD.
56. Lade, P.V., and R.D. Nelson. 1987. “Modeling the Elastic Behavior of Granular Materials.”
International Journal for Numerical and Analytical Methods in Geomechanics, Volume II,
pp. 521–542.
57. American Association of State Highway and Transportation Officials (AASHTO). 1993.
Guide for Design of Pavement Structure, Volume I & II. Highway Sub-committee on Design,
Washington, D.C.
58. Oloo, S.Y., and D.G. Fredlund. 1998. “The Application of Unsaturated Soil Mechanics
Theory to the Design of Pavements.” In Proc. 5th Intl. Conf. on the Bearing Capacity of
Roads and Airfields, Trondheim, Norway, pp. 1419–1428.
134
59. Lytton, R.L., J. Uzan, E.G. Fernando, R. Roque, D. Hiltunen, and S.M. Stoffels. 1993.
SHRP A-357: Development and Validation of Performance Prediction Models and
Specifications for Asphalt Binders and Paving Mixes. National Research Council,
Washington, D.C.
60. Lytton, R.L. 1995. “Foundations and Pavements on Unsaturated Soils.” Paper presented at
the Proceedings of the First International Conference on Unsaturated Soil, Paris.
61. Liang, R. Y., S. Rabab’ah, and M. Khasawneh. 2008. “Predicting moisture-dependent
resilient modulus of cohesive soils using soil suction concept.” Journal of Transportation
Engineering, Vol. 134, No. 1, pp. 34–40.
62. Oh, J.H., and E.G. Fernando. 2008. Develop of Thickness Design Tables Based on the M-E
PDG. Research Report No. BDH10-1. Texas Transportation Institute, The Texas A&M
University System, College Station, TX.
63. Al-Qadi, I.L., H. Wang, and E. Tutumluer. 2010. “Dynamic Analysis of Thin Asphalt
Pavements by Using Cross-Anisotropic Stress-Dependent Properties for Granular Layer.”
In Transportation Research Record: Journal of the Transportation Research Board, No.
2154, pp. 156–163.
64. Tutumluer, E., and M.R. Thompson. 1997. “Anisotropic Modelling of Granular Bases in
Flexible Pavements.” Transportation Research Record: Journal of the Transportation
Research Board, No. 1577, pp. 18–26.
65. Yau, A., and H.L. Von Quintus. 2002. Study of LTPP Laboratory Resilient Modulus Test
Data and Response Characteristics. Report No. FHWA-RD-02-051. Federal Highway
Administration Research, U.S. Department of Transportation.
66. ARA, Inc, 2004. Guide for Mechanistic-Empirical Design of New and Rehabilitated
Pavement Structures. Final Report of NCHRP Project 1-37A. Transportation Research
Board, Washington, D.C.
67. AASHTO, AAoSHaTOec. Mechanistic-empirical pavement design guide: A manual of
practice. AASHTO Designation: MEPDG-1, Washington, D.C., 2008
68. Kenis, W.J. 1977. Predictive Design Procedures, VESYS User’s Manual. Final Report, No.
FHWA-RD-77-154. Federal Highway Administration, McLean, VA.
69. Uzan, J. 2004. “Permanent Deformation in Flexible Pavements.” Journal of Transportation
Engineering, Vol. 130, No. 1, pp. 6–13.
70. Tseng, K.H., and R.L. Lytton. 1989. “Prediction of Permanent Deformation in Flexible
Pavement Materials.” In Implication of Aggregates in the Design, Construction, and
Performance of Flexible Pavement, ASTHMA STP 1016, ASTM, Philadelphia, PA., pp.
154–172.
71. Ayres, M. and M.W. Witczak. 1998. “AYMA: A Mechanical Probabilistic System to
Evaluate Flexible Pavement Performance.” Transportation Research Record 1629, pp.
137–148.
72. Korkiala-Tanttu, L. 2009. “Verification of Rutting Calculation for Unbound Road
Materials.” In Proceedings of the Institution of Civil Engineers-Transport, Vol. 162, No. 2,
pp. 107–114. doi: 10.1680/tran.2009.162.2.107
73. Gu, F., Zhang, Y.Q., Droddy, C.V., Luo, R. and Lytton, R.L. 2015. “Development of a New
Mechanistic-Empirical Rutting Model for Unbound Granular Material.” Journal of
Materials in Civil Engineering, Vol. 28, No. (8).
135
74. Epps, J., S. Sebesta, B. Hewes, H. Sahin, R. Luo, J. Button, R.L. Lytton, C.A. Herrera, R.
Hatcher, and F. Gu. 2014. Development of a Specification for Flexible Base Construction.
Research Report No. FHWA/TX-13/0-6621. Texas A&M Transportation Institute, College
Station, TX.
75. Lambe, T.W., and R.V. Whitman. 1969. Soil Mechanics. Wiley, New York.
76. Abramento, M., and S. Carvalho. 1989. “Goetechnical Parameters for Study of Slope
Instabilization at Serra do Mar-Brazillan Southeast.” In Proceedings of the 12th
International Conference on Soil Mechanics and Foundation Engineering, Rio de Janeiro,
Vol. 3, pp. 1599–1602.
77. Fredlund, D.G., and H. Rahardjo. 1993. Soil Mechanics for Unsaturated Soils. John Wiley
& Sons, Inc., New York, NY.
78. Öberg, A.L., and G. Sällfors. 1997. “Determination of Shear Strength Parameters of
Unsaturated Silts and Sands based on the Water Retention Curve.” Geotechnical Testing
Journal, Vol. 20, No. 1, pp. 40–48.
79. Fredlund, D.G., A. Xing, M.D. Fredlund, and S.L. Barbour. 1996. “The Relationship of the
Unsaturated Soil Shear to the Soil-Water Characteristic Curve.” Canadian Geotechnical
Journal, Vol. 33, No. 3, pp. 440–448.
80. Vanapalli, S.K., D.G. Fredlund, D.E. Pufahl, and A.W. Clifton. 1996. “Model for the
Prediction of Shear Strength with Respect to Soil Suction.” Canadian Geotechnical Journal,
Vol. 33, No. 3, pp. 379–392.
81. Rassam, D.W., and F.J. Cook. 2002. “Predicting the Shear Strength Envelope of
Unsaturated Soils.” Geotechnical Testing Journal, Vol. 28, pp. 215–220.
82. Titus-Glover, L., and E.G. Fernando. 1995. Evaluation of Pavement Base and Subgrade
Material Properties and Test Procedures. Research Report 1335-2. Texas Transportation
Institute, The Texas A&M University System, College Station, TX.
83. Rauhut, J.B., R.L. Lytton, and M.I. Darter. 1982. Pavement Damage Functions for Cost
Allocation, Vol. 1, Damage Functions and Load Equivalency Factors. Report No.
FHWA/RD-82/126. Federal Highway Administration, Washington, DC.
84. Markow, M.J., and B.D. Brademeyer. 1984. EAROMAR Version 2. Final Technical
Report FHWA/RD-82/086. Federal Highway Administration, U.S. Department of
Transportation.
85. Larralde, J. 1984. Structural Analysis of Ridge Pavements with Pumping. PhD diss.,
Purdue University, West Lafayette, IN.
86. Van Wijk, A.J. 1985. Rigid Pavement Pumping: (1) Subbase Erosion and (2) Economic
Modeling. Joint Highway Research Project File 5-10. School of Civil Engineering, Purdue
University, West Lafayette, IN.
87. PCA. 2008. Pavement—Soil Cement. Portland Cement Association, Skokie, IL.
http://www.cement.org/pavements/pv_sc.asp. Accessed: October 1, 2008.
88. Winkler, E. 1867. Die Lehre von der Elasticitaet und Festigkeit. Prag, Dominicus.
89. Filonenko-Borodich, M.M. 1940. “Some Approximate Theories of the Elastic Foundation.”
Uchenyie Zapiski Moskovskogo Gosudarstvennogo Universiteta. Mekhanica, No. 46, pp.
3–18 (in Russian).
90. Hetenyi, M. 1950. “A General Solution for the Bending of Beams on an Elastic Foundation
of Arbitrary Continuity.” Journal of Applied Physics, Vol .21, pp. 55–58.
136
91. Pasternak, P.L. 1954. “On a New Method of Analysis of an Elastic Foundation by Means
of Two Foundation Constants.” Gosudarstrennoe Izdatel′stro Literaturi po Stroitel′stru i
Arkhitekture, Moscow, USSR (in Russian).
92. Kerr, A. 1965. “A Study of a New Foundation Model.” Acta Mechanica, Vol. 1, No. 2, pp.
135–147. doi: 10.1007/BF01174308
93. Mitchell, P. W. 1979. The structural analysis of footings on expansive soils. Kenneth W.G.
Smith and Associates Pty. Ltd. Research Report No. 1, Newton.
94. Zapata, C. E., Houston, W. N., and Walsh, K. D. (2003). “Soil-water characteristic curve
variability.” Advances in Unsaturated Geotechnics, 99, 84-124.
95. Gupta, S., and Larson, W. 1979. “Estimating soil-water retention characteristics from
particle size distribution, organic matter percent, and bulk density.” Water Resources
Research, 6 (15), 1633–1635.
96. Hutson, J., and Cass, A. 1987. “A retentivity function for use in soil water simulation
models.” Soil Science Journal, Vol. 1, No. 38, pp. 105-113.
97. Reddi, L., and Poduri, R. 1997. “Use of liquid limit state to generalize water retention
properties of fine-grained soils.” Geotechnique, 47(5), 1043– 1049.
98. Zapata, C. 1999. Uncertainty in Soil-Water Characteristic Curve and Impacts on
Unsaturated Shear Strength Predictions. Tempe, Arizona State University.
99. Williams, J., Prebble, R., Williams, W., and Hignett, C. 1983. “The influence of texture,
structure and clay mineralogy on the soil moisture characteristic.” Australian Journal of
Soil Research, 1(21), 15–32.
100. Perera, Y. Y., Zapata, C. E., Houston, W. N., and Houston, S. L. 2005. “Prediction of the
soil water characteristic curve based on grain-size-distribution and index properties.”
Geotechnical Special Publication, 49-60.
101. Torres-Hernandez, G. 2011. Estimating the Soil–Water Characteristic Curve Using Grain
Size Analysis and Plasticity Index. Tempe: Arizona State University.
102. Sahin, H., Gu, F., and Lytton, R. L. 2015. “Development of soil-water characteristic curve
for flexible base materials using the methylene blue test.” Journal of Materials in Civil
Engineering, Vol. 27, No. 5.
103. Mishra, S., Parker, L., and Singhal, N. 1989. “Estimation of soil hydraulic properties and
their uncertainty from particle size distribution data.” Journal of Hydrology, 108, 1-18.
104. Basile, A., and D'Urso, G. 1997. “Experimental corrections of simplified methods for
predicting water retention curves in clay-loamy soils from particle-size determination.”
Soil Technology, 10(3), 261 -272.
105. Fredlund, M., Fredlund, D., and Wilson, G. 1997. Prediction of the soil-water characteristic
curve from grain-size distribution and volume-mass properties. Proceedings of the Third
Brazilian Symposium on Unsaturated Soils.
106. Pachepsky, Y.A., Timlin, D. and Varallyay, G.Y. 1996. “Artificial neural networks to
estimate soil water retention from easily measurable data.” Soil Science Society of
America Journal, 60(3), 727-733.
107. Johari, A., and Javadi, A. A. 2010. “Prediction of soil-water characteristic curve using
neural network.” Proceedings of the Fifth International Conference on Unsaturated Soils, 1,
pp. 461-466. Barcelona.
108. Haghverdi, A., Cornelis, W.M. and Ghahraman, B. 2012. “A pseudo-continuous neural
network approach for developing water retention pedotransfer functions with limited data.”
Journal of Hydrology, 442, 46-54.
137
109. Jain, S.K., Singh, V.P. and Van Genuchten, M.T. 2004. “Analysis of soil water retention
data using artificial neural networks.” Journal of Hydrologic Engineering, Vol. 9, No. 5, pp
415-420.
110. Fredlund, D. G., and Xing, A. 1994. “Equations for the Soil–Water Characteristic Curve.”
Canadian Geotechnical Journal, 31(3), 521-532.
111. Zapata, C.E. 2010. NCHRP Web-Only Document 153: A National Database of Subgrade
Soil-Water Characteristic Curves and Selected Soil Properties for Use with the MEPDG.
Transportation Research Board of the National Academies, Washington, D.C.
112. Sillers, W.S., Fredlund, D.G., and Zakerzadeh, N. 2001. “Mathematical attributes of some
soil-water characteristic curve models.” Geotechnical and Geological Engineering, 19(34),
243–283.
113. Rahardjo, H., Satyanaga, A., D’Amorec, G.A.R., and Leong, E.C. 2012. “Soil–water
characteristic curves of gap-graded soils.” Engineering Geology, 125(27), 102–107.
114. Aubertin, M., Mbonimpa, M., Bussière, B. and Chapuis, R.P. 2003. “A model to predict
the water retention curve from basic geotechnical properties.” Canadian Geotechnical
Journal, 40(6), 1104-1122.
115. Liu, H., and Dane, J. 1993. “Reconciliation between measured and theoretical temperature
effects on soil water retention curves.” Soil Sci Soc Am J, 57:1202-7.
116. Wu, W., Dandy, G. C., and Maier, H. R. 2014. “Protocol for developing ANN models and
its application to the assessment of the quality of the ANN model development process in
drinking water quality modelling.” Environmental Modelling & Software, 54, 108–127.
117. El-Keshky, M. 2011. Temperature Effect on the Soil Water Retention Characteristic. MS
thesis, Arizona State University.
118. Gu, F., X. Luo, Y. Zhang, Y. Chen, R. Luo, and R. L. Lytton. 2017. “Prediction of geogrid-
reinforced flexible pavement performance using artificial neural network approach.” Road
Materials and Pavement Design, pp. 1-17.
119. Ruttanaporamakul, P. 2012. Resilient moduli properties of compacted unsaturated
subgrade materials. Arlington: The University of Texas at Arlington.
120. Salour, F., Erlingsson, S., and Zapata, C. 2014. “Modelling Resilient Modulus Seasonal
Variation of Silty Sand Subgrade Soils with Matric Suction Control.” Canadian
Geotechnical Journal, 51(12), 1413-1422.
121. Thornthwaite, C.W. 1948. “An Approach Toward a Rational Classification of Climate.”
Geographical Review, pp. 55–94.
122. Sun, X, Li, J. and Zhou, A.N. 2017. “Evaluation and comparison of methods for
calculating Thornthwaite Moisture Index.” Journal of Australian Geomechanics, Vol. 52.
No. 2, pp. 61-75.
123. Olaiz, A.H., Singhar, S.H., Vann, J.D., and Houston, S.L. 2017. “Comparison and
Applications of the Thornthwaite Moisture Index Using GIS.” Proceeding of the PanAm
Unsaturated Soils, 280-289.
124. Thornthwaite, C. W., and Mather, C. W. 1955. “The Water Balance.” Publications in
Climatology 8, 1-104.
125. Willmott, C. J., and Feddema, J. J. 1992. “A More Rational Climate Moisture Index.”
Professional Geographer 44 (1), 84-87.
138
126. Witczak, M.W., Zapata, C.E., and Houston, W.N. 2006. Models Incorporated into the
Current Enhanced Integrated Climatic Model for Used in Version 1.0 of the MEPDG.
NCHRP 9-23 Project Report, Arizona State University, Tempe, Arizona.
127. Daly, C., Neilson, R.P., and Phillips, D.L. 1994. “A Statistical-Topographic Model for
Mapping Climatological Precipitation over Mountainous Terrain.” Journal of Appl.
Meteorology, 33, 140-158.
128. Daly, C., Gibson, W.P., Taylor, G.H., Johnson, G.L., and Pasteris, P. 2002. “A knowledge-
based approach to the statistical mapping of climate.” Climate Research, 22, 99-113.
129. Daly, C., Halbleib, M., Smith, J.I., Gibson, W.P., Doggett, M.K., Taylor, G.H., Curtis, J.,
and Pasteris, P.A. 2008. “Physiographically-sensitive mapping of temperature and
precipitation across the conterminous United States.” International Journal of Climatology,
28, 2031-2064.
130. Wilm, H.G., Thornthwaite, C.W., Colman, E.A., Cummings, N.W., Croft, A.R., Gisborne,
H.T., and Harding, S.T. 1944. “Report of the Committee on Transpiration and Evaporation
1943-44.” Eos. Transactions American Geophysical Union. 683-693.
131. McKeen, R. G., and L. D. Johnson. 1990. “Climate Controlled Soil Design Parameters for
Mat Foundations.” Journal of Geotechnical Engineering (ASCE) 116 (7), 1073-1094.
132. Gay, D. A. 1994. Development of a predictive model on pavement roughness on expansive
clay. Ph.D. dissertation, Texas A&M University, Texas.
133. Saha, S., Gu, F., Luo, X., and Lytton, R. L. 2018. “Prediction of soil-water characteristics
curve for unbound material using Fredlund-Xing equation based ANN approach.” Journal
of materials in civil engineering, Vol. 30, No. 5.
134. Lekarp, F., U. Isacsson, and A. Dawson. 2000. “State of the art. I: Resilient response of
unbound aggregates.” Journal of Transportation Engineering, Vol. 66, No. 1, pp. 66–75.
135. Heath, A. C., J. M. Pestana, J. T. Harvey, and M. O. Bejerano. 2004. “Normalizing
behavior of unsaturated granular pavement materials.” Journal of Geotechnical and
Geoenvironmental Engineering, Vol. 130, No. 9, pp. 896–904.
136. Edil, T., C. Benson, and A. Sawangsuriya. 2006. Interim Report to University of
Minnesota: Resilient Behavior of Unsaturated Subgrade Soils. University of Wisconsin–
Madison.
137. Yau, A., and H. L. Quintus. 2004. “Predicting Elastic Response Characteristics of Unbound
Materials and Soils.” Transportation Research Record: Journal of the Transportation
Research Board, No. 1874, pp. 47-56.
138. Nazzal, M. D., and L. N. Mohammad. 2010. “Estimation of resilient modulus of subgrade
soils for design of pavement structures.” Journal of Materials in Civil Engineering, Vol. 22,
No. 7, pp 726-734.
139. Malla, R. B., and S. Joshi. 2007. “Resilient modulus prediction models based on analysis of
LTPP data for subgrade soils and experimental verification.” Journal of Transportation
Engineering, Vol. 133, No. 9, pp 491-504.
140. Gu, F., H. Sahin, X. Luo, R. Luo, and R. L. Lytton. 2014. “Estimation of resilient modulus
of unbound aggregates using performance-related base course properties.” Journal of
Materials in Civil Engineering, Vol. 27, No. 6, p. 04014188.
141. Coleri, E., M. Guler, A. Gungor, and J. Harvey. 2010. “Prediction of Subgrade Resilient
Modulus Using Genetic Algorithm and Curve Shifting Methodology: Alternative to
Nonlinear Constitutive Models.” Transportation Research Record: Journal of the
Transportation Research Board, No. 2170, pp. 64-73.
139
142. Kim, S.H., J. Yang, and J. H. Jeong. 2014. “Prediction of subgrade resilient modulus using
artificial neural network.” KSCE Journal of Civil Engineering, Vol. 18, No.5, pp.1372-1379.
143. Nazzal, M. D., and O. Tatari. 2013. “Evaluating the use of neural networks and genetic
algorithms for prediction of subgrade resilient modulus.” International Journal of Pavement
Engineering, Vol. 14, No. 4, pp. 364-373.
144. Ceylan, H., K. Gopalakrishnan, and R. L. Lytton. 2011. “Neural networks modeling of stress
growth in asphalt overlays due to load and thermal effects during reflection cracking.”
Journal of Materials in Civil Engineering, Vol. 23, No. 3, pp 221-229.
145. Ling, M., X. Luo, S. Hu, F. Gu, and R. L. Lytton. 2017. “Numerical Modeling and Artificial
Neural Network for Predicting J-Integral of Top-Down Cracking in Asphalt Pavement.”
Transportation Research Record: Journal of the Transportation Research Board, No. 2631,
pp. 83-95.
146. Lawrence, M., and A. Petterson. 1993. Brainmaker Professional: Neural Network Simulation
Software User’s Guide and Reference Manual. California Scientific Software, Nevada City,
Calif.
147. Ji, R., N. Siddiki, T. Nantung and D. Kim. 2014. “Evaluation of resilient modulus of
subgrade and base materials in Indiana and its implementation in MEPDG.” The Scientific
World Journal.
148. Soliman, H. and A. Shalaby. 2016. “Validation of Long-Term Pavement Performance
Prediction Models for Resilient Modulus of Unbound Granular Materials.” Transportation
Research Record: Journal of the Transportation Research Board, No. 2578, pp. 29-37.
149. Tutumluer, E., D. Mishra, and A. Butt. 2009. Final Report, Illinois Center for Transportation
(ICT) R27-1 Project: Characterization of Illinois Aggregates for Subgrade Replacement and
Subbase, University of Illinois Urbana-Champaign, Urbana, IL.
150. Rao, K. 2000. Dynamic soil tests and applications. A. H. Wheeler & Co. Ltd, New Delhi,
First Edition.
151. MnDOT Pavement Design Manual. 2007. Minnesota Department of Transport, Chapter 3.
152. Ziaie-Moayed, R., and Janbaz, M. 2009. “Effective parameters on modulus of subgrade
reaction in clayey soils.” Journal of Applied Sciences, 9, 4006-4012.
153. Biot, M. 1922. “Bending of an infinite beam on an elastic foundation.” Journal of Applied
Mathematics and Mechanics, 2(3), 165-184.
154. Terzaghi, K.V. 1955. “Evaluation of coefficient of subgrade reaction.” Geotechnique, 5,
297-326.
155. Vesic, A. B. 1961. “Beams on elastic subgrade and the Winkler’s hypothesis.” Proceedings
of the 5th International Conference on Soil Mechanics and Foundation Engineering, 1,
845-850.
156. Horvath, J. S. 1983. “New subgrade model applied to mat foundations.” Journal of
Geotechnical Engineering, 109(12), 1567-1587.
157. Vallabhan, C. V. G., and Das, Y. C. 1988. “An improved model for beams on elastic
foundations.” Proceedings of the ASME/PVP Conference, Pittsburgh, Pennsylvania, 19-23.
158. Hetényi, M. 1971. Beams on elastic foundation: theory with applications in the fields of civil
and mechanical engineering, University of Michigan, Ann Arbor.
159. Vlasov, V. Z. and Leont'ev, N. N. 1966. Beams, plates and shells on elastic foundations.
Israel Program for Scientific Translations, Jerusalem.
140
160. Vallabhan, C.V.G., and Das, Y.C. 1989. “A refined model for beams on elastic
foundations.” International Journal of Solids and Structures, 27, 629-637.
161. Shi, X., Fwa, T., and Tan, S. 1993. “Warping stresses in concrete pavements on Pasternak
foundation.” Journal of transportation engineering, 119(6), 905-913.
162. Shi, X., Tan, S., and Fwa, T. 1994. “Rectangular thick plate with free edges on Pasternak
foundation.” Journal of engineering mechanics, 120(5), 971-988.
163. Iancu-Bogdan, T. and Vasile, M. 2010. The Modified Vlasov Foundation Model: An
Attractive Approach for Beams Resting on Elastic Supports. EJGE, Vol.15.
164. Farouk, H., and Farouk, M. 2014. “Calculation of subgrade reaction modulus considering
the footing-soil system rigidity.” Proc., 2nd Int. Conf. on Vulnerability and Risk Analysis
and Management (ICVRAM).
165. Hoffman, M. S., and Thompson, M. R. 1981. Mechanistic Interpretation of Nondestruction
Pavement Testing Deflection, Civil Engineering Studies, Transportation Engineering Series
No. 32, Illinois Cooperative Highway and Transportation Research Series No. 190,
University of Illinois, Urbana, Illinois.
166. Hall, K. T. 1991. Performance, Evaluation, and Rehabilitation of Asphalt-Overlaid Concrete
Pavement, Ph.D. Dissertation, University of Illinois, Urbana, Illinois.
167. Federal Highway Administration. 2001. Backcalculation of Layer Parameters for LTPP Test
Sections- Slab on Elastic Solid and Slab on Dense-Liquid Foundation Analysis of Rigid
Pavements. Publication No. FHWA-RD-00-086, Washington, D.C.
168. Tarr, S. M., Okamoto, P. A., Sheehan, M. J., and Packard, R. G. 1999. “Bond Interaction
Between Concrete Pavement and Lean Concrete Base.” Transportation Research Record:
Journal of the Transportation Research Board, No. 1668, pp. 9–17.
169. Bari, M. E., Zollinger, D.G., and Jung, Y.S. 2013. Modeling and Calibration of Concrete
Slab Interfacial Effects. TRB 92nd Annual Meeting Compendium of Papers, Transportation
Research Board of the National Academies, Washington, D.C.
170. Tutumluer, E., and Seyhan, U. 1999. “Laboratory Determination of Anisotropic Aggregate
Resilient Moduli Using an Innovative Test Device.” Transportation Research Record:
Journal of the Transportation Research Board, No. 1687, pp. 13-21.
171. Tutumluer, E., Little, D. N., and Kim, S.-H. 2003. “Validated Model for Predicting Field
Performance of Aggregate Base Courses.” Transportation Research Record: Journal of the
Transportation Research Board, No. 1837, pp. 41-49.
172. Park, S. W., and Lytton, R. L. 2004. “Effect of stress-dependent modulus and Poisson’s
ratio on structural responses in thin asphalt pavements.” Journal of Transportation
Engineering, 130(3), 387-394.
173. Boussinesq, J. 1885. Application des potentiels à l'étude de l'équilibre et du mouvement des
solides élastiques: Gauthier-Villars.
174. Bari, M. E., and Zollinger, D. G. 2016. “New concepts for the assessment of concrete slab
interfacial effects in pavement design and analysis.” International Journal of Pavement
Engineering, 17(3).
175. Kim, S. H. 2004. Prediction of level of anisotropy of aggregates based on gradation, shape,
form, and textural properties. Ph.D. dissertation, Texas A&M Univ., College Station, Texas.
176. Balmer, G. G. 1958. Shear strength and elastic properties of soil-cement mixtures under
triaxial loading. Retrieved from Portland Cement Association Research and Development
Laboratories.
141
177. Clough, G. W., Sitar, N., Bachus, R. C., and Rad, N. S. 1981. “Cemented sands under static
loading.” Journal of Geotechnical and Geoenvironmental Engineering, ASCE,
107(GT6), 799-817.
178. Thompson, M. R. 1966. “Shear strength and elastic properties of lime-soil mixtures.”
Highway Research Record, Washington, D.C., 139, 1-14.
179. Maree, J. H. 1978. Ontwerpparameters vir klipslag in plaveisels. M Eng thesis. University of
Pretoria, South Africa.
180. Theyse, H. 2000. Laboratory design models for materials suited to labour-intensive
construction. Pretoria: CSIR Transportek. Contract Report CR-99/038.
181. Khazanovich, L., Tayabji, S. D., and Darter, M. I. 2001. Backcalculation of layer parameters
for LTPP test sections, Volume I: Slab on elastic solid and slab on dense-liquid foundation
analysis of rigid pavements, Report No. FHWA-RD-00-086, ERES Consultants, Inc.,
Columbia, Maryland.
182. Croney, D. 1977. The design and performance of road pavements. Transport and road
research laboratory, HMSO, U.K.
183. Delatte, N. J., Jr., Wade, D. M., and Fowler, D. W. 2000. “Laboratory and field testing of
concrete bond development for expedited bonded concrete overlays.” ACI Materials
Journal, 97(3), 272-280.
184. Theyse, H. L., De Beer, M., and Rust, F. C. 1996. “Overview of South African Mechanistic
Pavement Design Method.” Transportation Research Record: Journal of the Transportation
Research Board, No. 1539, pp. 6-17.
185. Fernando, E. G., J. Oh, D. Ryu, and S. Nazarian. 2008. Consideration of Regional
Variations in Climatic and soil Conditions in the Modified Triaxial Design Method.
Research Report 0-4519-2, Texas Transportation Institute, Texas A&M University System,
College Station, Texas.
186. Soil Erodibility. 2014. Retrieved March 2014, from
http://web2.uwindsor.ca/courses/earth_science/hudec/nigeria/Erodibility.htm.
187. Oloo, S.Y. 1994. A bearing capacity approach to the design of low-volume traffic roads.
Ph.D. dissertation, University of Saskatchewan, Saskatoon, Canada.
188. Holtz, R.D., and Kovacs. W.D. 1981. An Introduction to Geotechnical Engineering.
PrenticeHall, Englewood Cliffs, NJ.
189. Tutumluer, E., Mishra, D., and Butt, A.A. 2009. Characterization of Illinois Aggregate for
Subgrade Replacement and Subbase. Technical Report FHWA-ICT-09-060. Illinois Center
for Transportation, Urbana– Champaign, pp. 1–179.
190. Chow, L.C., Mishra, D. and Turumluer, E. 2014. Aggregate base course material testing and
rutting model development. Final Report NCDOT Project 2013-18, FHWA/NC/2013-18.
191. Cetin, A., Kaya, Z., Cetin, B., and Aydilek, A. H. 2014. “Influence of compaction method
on mechanical properties of unbound granular materials.” Road Materials and Pavement
Design, Vol. 15, No. 1, pp. 220–235.
192. Soliman, H., and Shalaby, A. 2015. “Permanent Deformation Behavior of Unbound
Granular Base Materials with Varying Moisture and Fines Content.” Transportation
Geotechnics, Vol. 4, pp. 1–12.
193. Shahin, M. Y., and Kohn, S. D. 1981. Pavement maintenance management for roads and
parking lots. Construction Engineering Research LAB (ARMY) CHAMPAIGN IL.
142
194. Ling, M., Luo, X., Chen, Y., Gu, F., and Lytton, R. L. 2018. “Mechanistic-empirical
models for top-down cracking initiation of asphalt pavements.” International Journal of
Pavement Engineering, 1-10.
195. Timoshenko, S., and Goodier, J. 1951. Theory of elasticity. New York, 412, 108.
196. Kalcheff, I. V., and Hicks, R. G. 1973. “A Test Procedure for Determining the Resilient
Properties of Granular Materials.” Journal of Testing and Evaluation, Vol. 1, No. 6, pp.
472-479.
197. May, R. W., and Witczak, M.W. 1981. “Effective Granular Modulus to Model Pavement
Responses.” Transportation Research Record 810, pp. 1-9.
198. Monismith, C. L., Seed, H. B., Mitry, F. G., and Chan, C. 1967. “Predictions of Pavement
Deflections from Laboratory Tests.” Second International Conference on the Structural
Design of Asphalt Pavements. University of Michigan, Ann Arbor.
199. Brown, S. F., and Pell, P. S. 1967. “An Experimental Investigation of the Stresses, Strains
and Deflections in A Layered Pavement Structure Subjected to Dynamic Loads.” in
Proceedings, Second International Conference on the Structural Design Asphalt
Pavements, University of Michigan.
200. Elliott, R. P., and Lourdesnathan, D. 1989. “Improved Characterization Model for Granular
Bases.” Transportation Research Record 1227, pp. 128-133.
201. Boyce, H. R. 1980. “A Nonlinear Model for the Elastic Behaviour of Granular Materials
under Repeated Loading.” in Proceedings of the International Symposium. Soils under
Cyclic and Transient Loading, Balkema PubI, pp. 285-294.
202. Wang, H., and Al-Qadi, I. L. 2012. “Importance of Nonlinear Anisotropic Modeling of
Granular Base for Predicting Maximum Viscoelastic Pavement Responses under Moving
Vehicular Loading.” Journal of Engineering Mechanics, Vol. 139, No. 1, pp. 29-38.
203. COMSOL Multiphysics Reference Manual, version 5.3", COMSOL, Inc,
www.comsol.com
204. Gu, F., Luo, X., Luo, R., Lytton, R. L., Hajj, E. Y., and Siddharthan, R. V. 2016.
“Numerical Modeling of Geogrid-Reinforced Flexible Pavement and Corresponding
Validation Using Large-Scale Tank Test.” Construction and Building Materials, Vol. 122,
pp. 214-230.
205. Li, M., Wang, H., Xu, G., and Xie, P. 2017. “Finite element modeling and parametric
analysis of viscoelastic and nonlinear pavement responses under dynamic FWD loading.”
Construction and Building Materials, 141, 23-35.
206. Mojtaba, HY., and Lee-Sullivan, P. 2011. “Modeling Linear Viscoelasticity in Glassy
Polymers Using Standard Rheological Models.” Proceedings of the COMSOL Conference,
Boston, MA, USA.
207. Findley, W. N., Lai, J. S., Onaran, K., and Christensen, R. M. 1977. “Creep and Relaxation
of Nonlinear Viscoelastic Materials with An Introduction to Linear
Viscoelasticity.” Journal of Applied Mechanics, Vol. 44, pp. 364.
208. Zhang, Y., Birgisson, B., and Lytton, R.L. 2015. “Weak Form Equation–Based Finite-
Element Modeling of Viscoelastic Asphalt Mixtures.” Journal of Materials in Civil
Engineering, Vol. 28, No. 2, pp. 0401511.
209. Zhang, Y., Gu, F., Luo, X., Birgisson, B., and Lytton, R. L. 2018. “Modeling Stress-
Dependent Anisotropic Elastoplastic Unbound Granular Base in Flexible
Pavements.” Transportation Research Record: Journal of the Transportation Research
Board, No. 2672, pp. 46-56. 0361198118758318.
143
210. Asphalt Institute. (1981). Thickness Design-Asphalt Pavements for Highways and Streets
(No. 1). Asphalt Institute.
211. O. NEW. 2004. Guide for Mechanistic-Empirical Design, Washington, D.C.
212. Roque, R., Zou, J., Kim, Y. R., Baek, C., Thirunavukkarasu, S., Underwood, B. S., and
Guddati, M. N. 2010. NCHRP Web-Only Document 162: Top-Down Cracking of Hot-Mix
Asphalt Layers: Models for Initiation and Propagation. Transportation Research Board of
the National Academies, Washington, D.C.
213. Gu, F. 2015. Characterization and Performance Prediction of Unbound Granular Bases
with and Without Geogrids in Flexible Pavements. PhD diss., Texas A&M University,
College Station, TX.
214. Gu, F., Luo, X., Zhang, Y., Lytton, R., and Sahin, H. 2016. “Modeling of unsaturated
granular materials in flexible pavements.” In E3S Web of Conferences (Vol. 9, p. 20002).
EDP Sciences.
215. Kim, S. H. 2005. Determination of aggregate physical properties and its effects on cross-
anisotropic behavior of unbound aggregate materials. PhD diss., Texas A&M University,
College Station, TX.
216. Saha, S., Gu, F., Luo, X. and Lytton, R.L., 2018. “Use of an Artificial Neural Network
Approach for the Prediction of Resilient Modulus for Unbound Granular Material.”
Transportation Research Record: Journal of the Transportation Research Board, No. 2672,
pp. 23-33. 0361198118756881.
144
3) Moisture sensitivity of modulus of unbound base course on total rutting and load-related
cracking (5, 6, 7, 8, 9)
Many researchers demonstrate that matric suction is a better prediction factor than moisture
content as a fundamental stress state variable. Modulus of unbound base course has a high
sensitivity in the change of matric suction and therefore, the modulus equation has been modified
by entering the suction term into the effective stress or as external stress tensor.
7) Modulus of unbound base course on smoothness (International Roughness Index [IRI]) (13)
The change of smoothness (IRI) decreases with the increase of base modulus.
8) Shear strength of unbound base course on total rutting (14, 15, 16, 17, 18)
Shear strength plays an important role in affecting total rutting accumulation in unbound
aggregate materials. The unbound base materials with low shear strength might correspond to a
significantly high shear stress ratio value, which positively affects the total rutting.
A-1
10) Shear strength of unbound base course on smoothness (IRI) (18, 19)
IRI value depends upon the development of distresses, such as rutting, fatigue cracking, and
thermal cracking. A higher shear strength of unbound base leads to a higher resistance to rutting
and cracking, hence reducing the IRI value.
1) Modulus of unbound base course on transverse cracking (jointed plain concrete pavement
[JPCP]) (1, 22)
The reduction of the base support would increase the transverse crack deterioration rate. The
sensitivity analysis in this literature shows that transverse cracking decreases from 0.8 percent to
0.7 percent with the increase in modulus of base layer from 1,250,000 to 1,500,000 psi.
A-2
4) Modulus of unbound base course on load transfer efficiency (LTE) (JPCP and CRCP) (27,
28)
LTE is a major factor affecting deflections and denoted as unloaded deflection over loaded
deflection. A higher modulus of unbound base layer could minimize the deflection imposed by
applied loading, thus improving the LTE.
6) Shear strength of unbound base course on transverse cracking (JPCP) (29, 30)
Higher shear stresses due to deformation from compaction and/or trafficking caused immediate
reflective cracking at all transverse joints in the eastbound direction of the tested pavements.
9) Shear strength of unbound base course on LTE (JPCP and CRCP) (23)
The equation to determine the shear stress is provided in this literature, which shows that low
LTE contributes to the high shear stress. It is inferred that high shear strength results in high
LTE.
10) Shear strength of unbound base course on smoothness (IRI) (31, 32)
The bond between the slab and subbase is dependent on the shear strength of unbound base layer
and affects the erosion damage, which is the main cause of faulting. Faulting could minimize the
smoothness of surface pavement so that the increase of shear strength of base layer diminishes
the roughness on pavement surface.
A-3
12) Erosion of unbound base course on punchouts (CRCP) (24, 27, 33)
If the development of erosion of unbound base course takes place, the punchout life of pavement
is decreasing. Sensitivity analysis of subbase erosion indicates that the punchout is aggravated
with high erodibility.
13) Erosion of unbound base course on LTE (JPCP and CRCP) (27, 34)
Erosion mechanism is due to the joint deterioration process that is reflected in lower relative
stiffness, which causes the low LTE. Therefore, the LTE is promoted by the reduction of the
erosion.
14) Erosion of unbound base course on smoothness (IRI) (27, 34, 35)
The erosion of unbound base course promoted the probability of faulting and punchouts, which
also aggravates the smoothness of the surface layer.
A-4
6) Moisture sensitivity of modulus of subgrade on total rutting and load-related cracking (41,
42, 43, 44, 45, 46)
Many researchers have pointed out that soil suction is a major factor for the prediction of
modulus of cohesive subgrade materials. Much research has been conducted to correlate
modulus of subgrade with soil suction. A higher soil suction generates a larger modulus of
subgrade.
A-5
A-6
A-7
A-8
M R k1 3k2
where M R is the resilient modulus, 3 is the confining pressure, and k1 and k 2 are regression
coefficients.
M R k1 k2
where M R is the resilient modulus, is the bulk stress, and k1 and k 2 are regression
coefficients.
M R k 2 k 3 k1 d k1 d
M R k 2 k 4 d k1 k1 d
where M R is the resilient modulus; d is the deviatoric shear stress; and k1 , k 2 , k3 , and k 4
are regression coefficients.
a' b' d
MR
d
where M R is the resilient modulus, d is the deviatoric shear stress, and a ' and b ' are
regression coefficients.
M R k1 k 2 dk3
where M R is the resilient modulus; d is the deviatoric shear stress; is the bulk stress; and
k1 , k 2 , and k3 are regression coefficients.
k k
I1 oct
2 3
M R k1Pa
Pa Pa
where M R is the resilient modulus; I 1 is the first invariant of stress tensor; oct is the
octahedral shear stress; Pa is the atmospheric pressure; and k1 , k 2 , and k3 are regression
coefficients.
k k
I
2 3
M R k1Pa 1 oct 1
Pa Pa
where M R is the resilient modulus; I 1 is the first invariant of stress tensor; oct is the
octahedral shear stress; Pa is the atmospheric pressure; and k1 , k 2 , and k3 are regression
coefficients.
B-1
I1 2 J2
M R M Pa R
Pa Pa
where M R is the resilient modulus; I 1 is the first invariant of stress tensor; J 2 is the second
invariant of shear stress tensor; Pa is the atmospheric pressure; R is a function of the Poisson’s
ratio; and M and are model coefficients.
MR ba
log a
b
km S Sopt
M Ropt
1 exp ln
a
where M R is the resilient modulus at a given degree of saturation; M Ropt is the resilient
MR
modulus at reference condition; a is the minimum of log ; b is the maximum of
M Ropt
M
log R ; k m is the regression parameter; and S Sopt is the variation of degree of
M
Ropt
saturation expressed in decimal.
M R k 2 k 3 k1 d k s u a u w
M R k 2 k 4 d k 1 k s u a u w
where M R is the resilient modulus; d is the deviatoric shear stress; u a is the air pressure; uw
is the pore water pressure; and k1 , k 2 , k3 , k 4 , and k s are regression coefficients.
k k
I 3k4 oct
2 3
M R k1Pa 1
Pa Pa
where M R is the resilient modulus; I 1 is the first invariant of stress tensor; oct is the
octahedral shear stress; Pa is the atmospheric pressure; and k1 , k 2 , k3 , and k 4 are regression
coefficients.
k2 k3
I 3 fhm oct
M R k1Pa 1
Pa Pa
where M R is the resilient modulus; I 1 is the first invariant of stress tensor; oct is the
octahedral shear stress; Pa is the atmospheric pressure; is the volumetric water content; f is
the saturation factor; hm is the matric suction; and k1 , k 2 , and k3 are regression coefficients.
B-2
k2
I1
I1 3 f hm 3 oct k3
M R k1Pa oct
Pa Pa
where M R is the resilient modulus; I 1 is the first invariant of stress tensor; oct is the
octahedral shear stress; Pa is the atmospheric pressure; is the volumetric water content; f is
the saturation factor; hm is the matric suction; and are the Henkel pore water pressure
coefficients; and k1 , k 2 , and k3 are regression coefficients.
MR k1 d w m 2
k
where M R is the resilient modulus; d is the deviatoric shear stress; w is the Bishop’s
effective stress coefficient; m is the matric suction; and k1 and k 2 are regression coefficients.
k2 k3
w m oct
M R k1Pa 1
Pa Pa
where M R is the resilient modulus; is the bulk stress; oct is the octahedral shear stress; Pa
is the atmospheric pressure; w is the Bishop’s effective stress coefficient; m is the matric
suction; and k1 , k 2 , and k3 are regression coefficients.
k' k k4
3 u w sat oct mo m
2 4
M R k P net
'
1 a 1 1
Pa aP Pa
where MR is the resilient modulus; net is the net bulk stress; oct is the octahedral shear stress;
Pa is the atmospheric pressure; u w sat is the build-up of pore water pressure under saturated
conditions; m 0 is the initial matric soil suction; m is the relative change of matrix soil
suction with respect to m 0 ; and k 1' , k 2' , k 3' , and k 4' are regression coefficients.
k2 k3
3k6 oct
kus pa a w
M R k1 pa b k7
pa pa
where M R is the resilient modulus; b is the bulk stress; oct is the octahedral shear stress; Pa
is the atmospheric pressure; u a is the air pressure; uw is the pore water pressure; is the
normalized volumetric water content; is the fitting parameter; and k1 , k 2 , k3 , k 6 , k 7 , and
k us are regression coefficients.
B-3
k2 k3
3k4 SVw oct
M R k1Pa 1
Pa Pa
where MR is the resilient modulus; is the bulk stress; oct is the octahedral shear stress; P a is
the atmospheric pressure; Vw is the volumetric water content; S is the soil suction; and k1 , k2 ,
k3 , and k4 are regression coefficients.
k2 k3
I
M k1Pa 1 oct 1
V
R
Pa Pa
H
M G
s RV ; r VHV
MR MR
where M RV is the resilient modulus in the vertical direction; I1 is the first invariant of stress
tensor; oct is the octahedral shear stress; P a is the atmospheric pressure; k1 , k2 , and k3 are
regression coefficients; M RH is the resilient modulus in the horizontal direction; GVH is the
shear modulus in the horizontal-vertical plane; and s and r are the modulus ratios.
k2 k3 k5 k6
M k1Pa oct ; M RH k4 Pa oct
V
R
Pa Pa Pa Pa
k8 k9
GVH k7 Pa oct ;
Pa Pa
where M R is the resilient modulus in the vertical direction; oct is the octahedral shear stress;
V
P a is the atmospheric pressure; M is the resilient modulus in the horizontal direction; GVH is
H
R
B-4
Mr 2555 CBR
0.64
where M r is the resilient modulus, and CBR is the California bearing ratio.
M r 1155 555 R
where M r is the resilient modulus, and R is the resistance R-value.
a
Mr 30000 i
0.14
where M r is the resilient modulus, and ai is the AASHTO layer coefficient.
0.64
75
M r 2555
1 0.728 wPI
where MR is the resilient modulus, wPI is the weighted plasticity index.
0.64
292
M r 2555 1.12
DCP
where M R is the resilient modulus, and DCP is the dynamic cone penetrometer index.
B-5
1 N
P
N
r N
where is the accumulated plastic strain; r is the resilient strain of granular material; N is
P
p
aN b
r
where p
is the accumulated plastic strain; r is the resilient strain of granular material; N is
the number of load cycles; and a and b are regression coefficients.
p 0e N
where pis the accumulated plastic strain; N is the number of load cycles; and 0 , , and
are regression coefficients.
p z p,z0 ekz
where p z is the plastic strain at depth z; p , z 0 is the vertical plastic strain at the top of
subgrade; z is the depth measured from the top of subgrade; and k is the model coefficient.
p s 0 e N v
r
where p is the accumulated plastic strain; r is the resilient strain of granular material; N is
the number of load cycles; 0 , , and are regression coefficients; s is a global calibration
coefficient, 1.673 for granular materials; r is the resilient strain imposed in the laboratory test;
and v is the average vertical resilient strain in the base layer of the flexible pavements.
p k6 oct k6 oct
log a0 a1
2a b b1
2b .log N
r Pa Pa
0
Pa Pa
where P is the accumulated plastic strain; r is the resilient strain of granular material; N is the
number of load cycles; is the bulk stress; P a is the atmospheric pressure; oct is the
octahedral shear stress; and a0 , a1 , a2 , b0 , b1 , and b2 are model coefficients.
B-6
R
p CN b
1 R
q
b d c'
q
f
where P is the accumulated plastic strain; N is the number of load cycles; C is the permanent
strain in the first loading cycle; b is a shear ratio parameter; R is the shear failure ratio
q 6sin c 6cos
1 3 , M , q0 , where c and are cohesion and friction angle;
q f q0 Mp 3 sin 3 sin
and d and c ' are material parameters.
D
p AN f
B C
max
d
where P is the accumulated plastic strain; N is the number of load cycles; d is the deviatoric
shear stress; f is shear stress; m ax is shear strength; and A , B , C , and D are regression
coefficients.
J
m
I1 K
n
p 0e N
2
2sin
3 3 sin
c 6cos
K
3 3 sin
where P is the accumulated plastic strain; N is the number of load cycles; and 0 , , and
are regression coefficients; J 2 is the second invariant of the deviatoric stress tensor; I 1 is the
first invariant of the stress tensor; 0 , , , m and n are model coefficients; and c and
are cohesion and friction angle, respectively.
log 0 0.80978 0.06626Wc 0.003077 106 Er
r
log 0.9190 0.03105Wc 0.001806 1.5 106 Er
log 1.78667 1.45062Wc 3.784 104 2 2.074 103Wc2 1.05 106 Er
where 0 , , and are Pavement ME Design model coefficients; r is the resilient strain of
granular material; N is the number of load cycles; W c is the water content; is the bulk
stress; and E r is the resilient modulus of granular material.
B-7
9
e 0.15 e 10 20
0
r 2
log 0.61119 0.017638Wc
1
4.89285
10 9
9
1 10
where 0 , , and are Pavement ME Design model coefficients; r is the resilient strain of
granular material; and W c is the water content.
c n tan
where is the shear stress; n
is the normal stress; c is the cohesion; and is the friction
angle.
B-8
c n tan
c 83.95 1.58 N 40 2.57 n 0.043 N 40 N 2 6.88 PLN GsN 0.14 0.81 tan
1.61 0.96 PI 0.88n 4.13 31.82Gsb
where N 40 is the percent of material passing 0.42 mm sieve size; n is the porosity; N 40 N is
normalized N 40 = N40 55.89 ; PLN is normalized plastic limit = PL 15.89 ; G sN is
normalized specific gravity of aggregate = G s 2.61 ; is matric suction; PI is plasticity
index; and Gsb is specific gravity of binder content.
B-9
Pi m ESAL f d
log m 1.07 0.34 D
where Pi is the pumping index; E SA L is the equivalent 80 kN single axle loads; f d is the
drainage adjustment factor; m is the model coefficient; and D is the slab thickness.
P 36.67 NPI
m
C 2 ni
Percent erosion damage 100
i 1 N i
where N is the allowable number of load repetitions based on a pressure of a PSI of 3.0; C1 is
the adjustment factor; P is the pressure on the foundation under the slab corner; m is the total
number of load groups; C 2 is the model coefficient; ni is the predicted number of repetitions
for the ith load group; and N i is the allowable number of repetitions for the ith load group.
f %Erosion f0e D N
B-10
v is time delay before the appearance of visible (measurable) damage; and is shape factor
related to the erosion rate.
FOUNDATION MODELS OF SUBGRADE
p(x, y) kw(x, y)
where p x , y is the distributed load applied in the x-y plane; w x , y is the displacement in
the vertical direction; and k is the foundation modulus.
x y
constant tension of a stretched elastic membrane of the top ends of the springs.
x y
flexural rigidity of the plate.
layer.
k G
(1 ) p 2 p kw G 2 w
c c
c 3k
4E f
k
3H
4
G HG f
9
where p is the distributed load; w is the displacement in the vertical direction; 2 is the
2
Laplace operator in x and y ( 2 2 2 );
2
c is the spring constant of upper spring layer;
x y
k is the spring constant of lower spring layer; G is the constant of the shear layer; E f is
Young’s modulus; G f is the shear modulus of the foundation material; and H is the thickness
of the foundation.
B-11
Susceptibility criterion.
Accuracy criterion.
Development criterion.
SUSCEPTIBILITY CRITERION
The susceptibility criterion refers to how the model responds to the changes in the
operational conditions, including moisture, heat, traffic stress, and load-induced/particle-induced
anisotropy. As listed in Tables 2 to 5 of the final report, the performance of flexible and rigid
pavement is closely related to the operational conditions of unbound layers and subgrade. For
example, as shown in Table 2, a flexible pavement is more susceptible to the load-related
cracking (alligator and longitudinal cracking) when the modulus of the base course decreases.
When the cross-anisotropy is considered, the fatigue life is normally shorter than when using an
isotropic modulus for the base course. In addition, the modulus of the base course significantly
reduces as the degree of moisture increases, which results in more severe load-related cracking.
Based on the results in Tables 2 to 5, each unbound layer/subgrade model should be evaluated
under these operational conditions. The degree of susceptibility of the model is divided into the
following three levels:
High level.
Medium level.
Low level.
ACCURACY CRITERION
The accuracy criterion refers to how close the predictions made by an unbound
layer/subgrade model are to the actual behaviors of these underlying materials. More
specifically, the model should be verified by comparing to the laboratory measurements on
unbound layer and subgrade materials. In addition, the model needs to be compared with the
performance prediction that is made by its counterpart in Pavement ME Design through a
sensitivity analysis. The degree of accuracy is described by three levels:
C-1
High level.
Medium level.
Low level.
DEVELOPMENT CRITERION
The development criterion refers to the efforts required to develop, validate, and α-test
the unbound layer/subgrade model for the enhancements of Pavement ME Design. It is used to
ensure that essential development issues can be identified and solved (e.g., whether the data
elements that are needed for the model are available and/or whether the test methods and
equipment that are needed to provide inputs for the model are available). Furthermore, the model
can be validated by making predictions of the observed performance of pavements in the Long-
Term Pavement Performance (LTPP) database and/or from state departments of transportation.
This criterion will serve as the basis of the development and implementation of enhancements for
Pavement ME Design in Phase II of this project. The ease of development of an unbound
layer/subgrade model is rated as three levels:
Simple level.
Involved level.
Complex level.
C-2
Table C.1. Scoring Results of Modulus Models of Unbound Layers and Subgrade.
Evaluation Criteria
Total
Model OC1
DA2 ED3 Score
M4 H5 A6 T7
Nonlinear Stress-dependent Model (51)
3 3 3 2 3 2 16
M R k1 3k 2
Nonlinear Stress-dependent Model (52)
3 3 3 2 3 2 16
M R k1 k2
Nonlinear Stress-dependent Model (53)
MR k2 k3 k1 d k1 d 3 3 3 2 3 2 16
MR k2 k4 d k1 k1 d
Nonlinear Stress-dependent Model (54)
a ' b ' d 3 3 3 2 3 2 16
MR
d
Nonlinear Stress-dependent Model (55)
3 3 3 2 3 2 16
M R k1 k 2 dk3
Nonlinear Stress-dependent Model (56)
k k
I
2 3
3 3 3 1 2 2 14
M R k1Pa 1 oct
Pa Pa
Nonlinear Stress-dependent Model (57)
k k
I
2 3
3 3 3 1 2 2 14
M R k1Pa 1 oct 1
Pa Pa
Nonlinear Stress-dependent Model (58)
I1 2 J 3 3 3 1 2 2 14
M R M Pa R 2
Pa Pa
Moisture-sensitive Model (19)
MR ba
log a 2 3 3 1 2 2 13
M Ropt b
1 exp ln km S Sopt
a
Moisture-sensitive and Stress-dependent Model (59)
MR k2 k3 k1 d ks ua uw 2 3 3 2 2 2 14
1
Operational condition
2
Degree of accuracy
3
Ease of development
4
Moisture
5
Heat
6
Anisotropy
7
Traffic
C-3
MR k2 k4 d k1 ks ua uw
M R k P net
'
1 a 1 o 1
Pa Pa Pa
Moisture-sensitive and Stress-dependent Model (8)
k2 k3
3k6 oct 1 3 3 1 2 3 13
kus pa a w
M R k1 pa b k7
pa p a
C-4
C-5
Table C.2. Scoring Results of Permanent Deformation Models of Unbound Layers and
Subgrade.
Evaluation Criteria
Total
Model OC
DA ED Score
M H A T
Non-stress-dependent Mechanistic-empirical Model (70)
1 N
P
3 3 3 1 2 1 13
N
r N
Non-stress-dependent Mechanistic-empirical Model (71)
p 3 3 3 1 2 2 14
aN b
r
Non-stress-dependent Mechanistic-empirical Model (72)
3 3 3 1 2 1 13
p 0e N
max
d
J
m 1 3 3 1 1 3 12
I1 K
n
p 0e N
2
C-6
e 0.15 e 10 20
0
r 2
2 3 3 1 3 1 13
log 0.61119 0.017638Wc
1
4.89285
109
1 10
9
C-7
Table C.3. Scoring Results of Shear Strength Models of Unbound Layers and Subgrade.
Evaluation Criteria
Total
Model OC
DA ED Score
M H A T
Non-moisture-sensitive Model (78)
3 3 3 1 3 1 14
c n tan
C-8
m
3 3 3 1 3 2 15
Cn
Percent erosion damage 100 2 i
i 1 N i
C-9
C-10
Researchers have recently developed a new constitutive model (66) for unbound base
courses considering both nonlinear cross-anisotropic behavior and moisture-sensitive
characteristics, and incorporating the proposed constitutive model into the finite element model
of the base layer to quantify the influence of moisture content on the pavement performance.
More specifically, the saturation factor and the matric suction of the unsaturated unbound
aggregates are applied to the proposed constitutive model to reflect the moisture dependence.
Additionally, a new user-defined material subroutine (UMAT) is developed to characterize the
moisture-sensitive and stress-dependent nonlinear cross-anisotropic behavior of base materials in
the software ABAQUS. The formulation of the model is given as follows:
k2 k3
I 3 fhm oct
M k1Pa 1
V
R (D.1)
Pa Pa
M RH G
s V
; r VHV (D.2)
MR MR
where M V
R
is the resilient modulus in the vertical direction; I1 is the first invariant of stress
tensor; oct is the octahedral shear stress; Pa is the atmospheric pressure; is the volumetric
water content; is the saturation factor, 1 f 1 ; hm is the matric suction; k1 , k2 , and k3 are
f
regression coefficients; M RH is the resilient modulus in the horizontal direction; GVH is the shear
modulus in the horizontal–vertical plane; and s r are the modulus ratios.
In order to verify the accuracy of the modulus model in Equation D.1, researchers used
the repeated load triaxial test (Figure D.1) measurements on three selected materials at three
different moisture contents. The matric suction value in Equation 4.1 is obtained from the filter
paper test, required as Level 1 input. For Level 2, the suction will be calculated from the relation
between the Thornthwaite moisture index (TMI) and the equilibrium suction value (details are
presented in Appendix I, “Subgrade Subroutine for Flexible and Rigid Pavements”). Figure D.2
presents the comparison between the predicted moduli using Equation 4.1 and the measured
moduli from the triaxial tests. The model prediction provides a good agreement with the test
measurements. This indicates that the constitutive model proposed in Equation D.1 can reflect
the moisture-sensitive and stress-dependent behavior of unbound aggregates. After verification,
Equations D.1 and D.2 are coded into a UMAT to develop a moisture-sensitive and stress-
dependent nonlinear program that incorporates cross-anisotropy.
D-1
1000 A @ Optimum Moisture
Predicted Resilient Modulus
Content (OMC)
A @ 1.5% Above OMC
800
A @ 1.5% Below OMC
600
(MPa)
B @ OMC
400 B @ 1.5% Above OMC
B @ 1.5% Below OMC
200
C @ OMC
0 C @ 1.5% Above OMC
0 200 400 600 800 1000
C @ 1.5% Below OMC
Measured Resilient Modulus (MPa)
Figure D.2. Comparison of Predicted and Measured Resilient Moduli for Unbound Base
Materials. (A, B, C stand for 3 types of unbound aggregates)
Using this moisture-sensitive and stress-dependent nonlinear cross-anisotropic program,
the numerical study is conducted on a typical flexible pavement structure to examine its
capability to reflect the influence of unbound base on the pavement performance. The pavement
structure, finite element model, and modeling parameters are given in Figure D.3.
D-2
D-3
450
Tensile Strain at the bottom
of Asphalt Layer (µε)
400
350
300
250
200
Dry Opt Moist
800
Average compressive
700
strain in Base (µε)
600
500
400
300
200
Dry Opt Moist
(b) Compressive Strain in Unbound Base to Predict Rutting
Figure D.4. Demonstration of Effect of Moisture on Pavement Performance.
D-4
400
Tensile Strain at the bottom
of Asphalt Layer (µε)
360
320
280
240
200
Nonlinear Anisotropic Nonlinear Isotropic
Model Model
600
Average compressive strain
500
in Base (µε)
400
300
200
Nonlinear Anisotropic Nonlinear Isotropic Model
Model
Table D.1. Ranges of Input Parameters Used in ANN Models for Plastic and Non-plastic
Base Materials.
Plastic Soil Non-plastic Soil
Input parameters Range Input parameters Range
Percent passing No. 3/8" sieve 42–83 Percent passing No. 3/8" sieve 34–80
Percent passing No. 200 sieve 7.5–33.3 Percent passing No. 200 sieve 4–44.7
PL 12–48 Scale parameter, ϴ 13.7–51
PI 1–23 Shape parameter, Ѱ 0.165–0.47
OMC 5–20 OMC 4–20
MDD 103–151 MDD 98–153
Test MC 4.5–20 Test MC 3.4–19.8
D-5
Table D.2. Ranges of Input Parameters Used in ANN Models for Plastic and Non-plastic
Subgrade Materials.
Plastic Soil Non-plastic Soil
Input parameters Range Input parameters Range
Percent passing No. 3/8" sieve 50–100 Percent passing No. 3/8" sieve 53–100
Percent passing No. 200 sieve 1.8–98.4 Percent passing No. 200 sieve 1.7–96.7
PL 8–44 Scale parameter, ϴ 18.8–98.7
PI 1–75 Shape parameter, Ѱ 0.004–0.55
OMC 7–32 OMC 7–26
MDD 86–139 MDD 94–139
Test MC 6.4–35 Test MC 5.4–25.3
D-6
( zx ) 2 f
, ( zx ) 2 f s f (E.1)
v2
where sf is the shear strength of the base course.
Figure E.1. Illustration of In-situ Shear Stress in the Base Course on the PCC-Base
Interface Using a Mohr-Coulomb Failure Envelope.
Figure E.1 shows a schematic figure of the developed shear stress in the Portland Cement
Concrete (PCC)-base interface. The parameters v1 and v2 are shear stress on the interface in the
slab and base layer, respectively, when full shear is transferred. The parameter ( zx )2 f is the
shear stress in the base course on the interface for in situ conditions, which is limited by the
shear strength, s f , of the base course on the failure plane. When ( zx )2 f is greater than v2 , the
interface is considered as fully bonded. Depending on the ratio of the ( zx )2 f and v2 , the partial
bonding condition is defined in the PCC-base interface. The following section derives the
expressions of v2 and ( zx )2 f .
FORMULATION OF V2
The shear stress acting on the two faces of PCC-base interface can be expressed using the
theorem of elastic beam shear stress on the transformed section. It is assumed that full shear is
transferred through the interface of the transformed section:
VQ1
v1 ; (E.2)
Ib
VQ2
v2 (E.3)
E
Ib( b )
Es
E-1
The parameter Q is the first moment of area from the neutral axis of the transformed
section; herein Q 1 A 1 d 1 and Q 2 A 2 d 2 .
Substituting Eq. E.2 into Eq. E.3 yields:
Q2
v2 v1 * (E.4)
Q1(Eb / Es )
where v1 is determined using the Boussinesq point load solution (97) Figure E.2 illustrates the
shear stress acting on the PCC-base interface.
P 3 hs a 2 (1 2 )
x 2 (E.5)
2 ( hs a )
2 5/2
( hs a ) [( hs a ) hs ]
2 2 1/ 2 2 2 1/ 2
3P hs 3
z (E.6)
2 (hs 2 a 2 )5/2
3P ahs 2
v1 zx 2 2 5/ 2
(E.7)
2 ( hs a )
where P is the surface point load.
hs is the thickness of the slab.
E-2
hb
ahs 2
h ( h z)
3P b s
2
v2 (E.8)
2 ( hs 2 a 2 ) 5/ 2 h ( z hs )
s
2
FORMULATION OF (ΤZX)2ΘF
The expression of in-situ shear stress in the base course on PCC-base interface is derived
using the Mohr-Coulomb failure envelope, as shown in Figure E.3. The failure envelope is
defined by the shear strength parameters (i.e., cohesion, c, and friction angle, ).
Figure E.3. Maximum Shear Strength of Base Course in Mohr Coulomb Failure Envelope.
Herein, the state of plane stress on the slab-base interface is defined by x , z , and
zx , which is rotated by an angle of 2 from principal plane of stress. The angle of rotation,
2 , is expressed in Eq. E.10.
E-3
ah 2
3 2 s 2 2
tan 2
zx
(hs a )
z x 1 3(hs hs a2 )
3
(1 2 )
[ 2 2 2 ]
2 2 (hs a ) h 2 a2 1/2 h
s s
ahs 2
3 2 2 2
2 tan (
1 (hs a ) ) (E.10)
1 3(hs hs a 2 )
3
(1 2 )
[ ]
2 (hs 2 a 2 )2 h 2 a 2 1/2 h
s s
As illustrated by Figure E.3, the shear strength of the base course is calculated by Eqs.
E.11 and E.12:
s f r cos (E.11)
s f c f t a n c [ r s i n ] t a n (E.12)
where f
is the normal stress on the failure plane.
Therefore:
r (c tan)cos (E.13)
z x P 3( hs 3 hs a 2 ) (1 2 )
Herein, [ 2 2 ] (E.14)
2 4 ( hs a ) 2 5/ 2
( hs a ) [( hs 2 a 2 )1/ 2 hs ]
2 1/ 2
ahs 2
3 2 2 2
( zx )2 f (c tan )cos sin[tan (
1 (hs a ) )] (E.15)
1 3(hs3 hs a2 ) (1 2 )
[ ]
2 (hs 2 a2 )2 h 2 a2 1/2 h
s s
The expression of interface shear bonding, , is derived from Eqs. E.1, E.8, and E.15:
E-4
ah 2
3 2 s 2 2
(c tan ) cos sin[tan 1 ( ( hs a ) )]
1 3( hs hs a 2 )
3
(1 2 )
[ ]
2 ( hs 2 a 2 ) 2 h 2 a 2 1/2 h
s s
(E.16)
hb
3P ahs 2 hb ( hs 2 z )
2 ( hs 2 a 2 )5/2 h ( z hs )
s
2
The ranges of the parameters in ANN model are given in Table E.1.
E-5
F-1
F-2
Figure F.1. Effect of Various Input Variables on Relationship between Degree of Bonding
and Modified k-value.
F-3
The stress calculation for a single tire is based on one point load applied on the surface of
a half space. According to Figure G.1, the analytical solutions (96) for normal stresses and shear
stresses at any point caused by the point load P are expressed as:
1 3 2
1 2 (G.1)
2
1 3 2
1 2 (G.2)
2
3
(G.3)
2
1 3 2 (G.4)
1 2
2
3 (G.5)
2
3 (G.6)
2
where is the point load; is the distance between an arbitrary point (x, y, z) and the x-axis;
is the distance between an arbitrary point (x, y, z) and the y-axis; is the vertical depth of the
point P from the surface; is the distance between an arbitrary point (x, y, z) and the point P;
and is Poisson’s ratio.
G-1
The normal stresses immediately beneath the point load P is considered for stress
calculations. Accordingly, both of and are equal to zero and the value of is equal to . The
analytical Eqs. G.1–G.6 with input values of , , and yield:
1 1 2 1
0 1 2 1 (G.7)
2 2 4
1 1 2 1
0 1 2 1 (G.8)
2 2 4
3
(G.9)
2
0 (G.10)
The average stresses in the base course immediately beneath the point load P are required
to reflect the stress effects on the development of permanent deformation in the base course and
thus on faulting. Thus, the average term of the vertical depth ( ) in the analytical Eqs. G.7–G.9
of normal stresses is determined by integration from the bottom of the concrete slab to the
bottom of the base course and expressed as:
1 1 1 1
(G.11)
(G.12)
where is the depth of mean vertical stress; is the thickness of slab; and is the total
thickness of slab and base course. In this way, Eqs. G.7–G.9 are expressed as:
G-2
2 1
(G.13)
4
3
(G.14)
2
The stress calculation for dual tires depends on two point loads applied on the surface of
a concrete slab as shown in Figure G.2. The point loads and are the same for dual tires.
The final normal stresses are the sum of the normal stress produced by the point loads
and . The normal stresses produced by the point load are determined by Eqs. G.18–G.20
and substituting 0, 0, and ( ) and expressed as:
2 1
(G.15)
4
3
(G.16)
2
0 (G.17)
where , , and are normal stresses caused by the point load . The normal stresses
caused by the point load are determined by Eqs. G.1–G.3 and substituting 0, , and
√ ( ) and generated as:
1
2 1 (G.18)
2
G-3
1 3 2
1 2 (G.19)
2
3
(G.20)
2
0 (G.21)
3 (G.22)
2
where is the distance between the two point load and also the tire spacing, which is typically
12 in.
Given the normal stresses from single and dual tires, the first invariant of stress tensor
and the second invariant of the deviatoric stress tensor in the center of the base course at the axle
load level can be determined by:
(G.23)
1
6 (G.24)
G-4
The vehicle classes are classified by FHWA into 13 classes relying upon whether they
convey travelers or products as presented in Table H.1. Load-related distresses are mostly caused
by heavy load groups including the vehicle classes 4 to 13 rather than light axle load groups of
vehicle classes 1 to 3. Therefore, the heavy axle groups are taken into consideration in this study
for the following determination of the traffic input in the load-related faulting model.
4 Buses
Table H.2 presents the load intervals for each axle type including single, tandem, tridem,
and quadrem. The various stress levels in the second faulting model can be obtained from the
H-1
axle load intervals for axle types so that the stress terms ( and ) are determined according to
the different axle loads. The same axle loads for each axle type should be grouped together to
provide continuous and non-repeating load levels. In other words, the axle load distributions for
each axle type should be firstly concatenated together and then grouped into the unique load
level. This procedure is schematically depicted in the Figure H.1. For example, the load level of
12,000 lb is contained in each load distribution table for four axle types (single, tandem, tridem,
and quad). When concatenating the four load distribution tables, the load level of 12,000 lb is
repeated four times so that the next step is to group the corresponding distribution (%) together
to provide a non-repeating load level. Thus the continuous and non-repeating load levels were
determined for determination of stress term ( and ).
H-2
According to the number of tires, the load distributions are categorized into eight groups
as shown in Figure H.2. The single axles of vehicle classes 4 to 7 have single tires and the others
have dual tires. The rest of the axles including tandem, tridem, and quadrem of vehicle classes 4
to 5 have single tire and others have dual tires. Therefore, the number of tires (single and dual
tire) form a matrix of vehicle classes and axle types into two major categories as shown in the
gray and white areas of Figure H.2.
H-3
The AADTT is the annual average 24-hour volume of traffic passing through a specified
section of highway and can be obtained from the LTPP database. The truck traffic in the AADTT
includes the heavy traffic of vehicle classes 4 to 13. The AADTT was adopted to be converted to
the number of axle loads if the WIM data are unavailable in some LTPP sections. In order to
convert the AADTT to annual number of axle loads for each vehicle class and axle type, two
truck-traffic adjustment factors are needed including normalized vehicle class distribution and
number of axles per truck.
The vehicle class distribution represents the percent of the AADTT for each vehicle for
the base year and is normalized for truck vehicle classes 4 to 13. The summation of the
normalized distribution factors of truck classes must be 100. Table H.3 presents the default value
of distribution factors obtained from the literature (30), and these distribution factors are
assumed to be constant every year. The annual number of truck for each vehicle class with a base
year is determined by:
365 (H.1)
where k is a specific vehicle class (class 4 to 13); is annual number of trucks for a vehicle
class, k; is normalized vehicle class distribution percentage for a truck class, k.
H-4
The number of axle types is the average number of individual axles within each vehicle
class and axle type including single, tandem, tridem, and quadrem. The default values of the
average number of axles for each truck class is predicted depending upon the LTPP traffic data,
as shown in Table H.4. The number of axle loads for each axle type with each truck class can be
determined by:
(H.2)
where is a specific axle type (single, tandem, tridem, or quad); is the annual number of
axle loads for an axle type in a vehicle class; is the average number of axles by axle type
for each truck class. In this way, the number of axles for each axle type and truck class are
determined by the AADTT and default axle load spectra, so the total number of axles can be
obtained.
H-5
Vehicle Class Single Axle Tandem Axle Tridem Axle Quadrem Axle
4 1.62 0.39 0.00 0.00
5 2.00 0.00 0.00 0.00
6 1.02 0.99 0.00 0.00
7 1.00 0.26 0.83 0.00
8 2.38 0.67 0.00 0.00
9 1.13 1.93 0.00 0.00
10 1.19 1.09 0.89 0.00
11 4.29 0.26 0.06 0.00
12 3.52 1.14 0.06 0.00
13 2.15 2.13 0.35 0.00
In summary, the WIM data are used to determine traffic inputs of number of axles by
rearranging them into two main categories of single and dual tires by various axle load levels for
all axle types. When the WIM data are unavailable in some LTPP sections, the AADTT is
converted into number of axle loads within each vehicle class and axle type at various axle load
levels. In order to perform this conversion, the axle load distributions are required, which
indicate the percentage of the number of axles within axle load levels by axle types of vehicle
classes. The WIM data at the same section in recent years or in adjacent sections are used to
determine the axle load distribution. The total number of axles determined by the AADTT and
default axle load spectra and the axle load distribution determined by available WIM tables are
used for conversion from AADTT to WIM data.
H-6
The regression equations for equilibrium suction for vegetation canopy and bare soil are
given as:
To identify the presence of vegetation in the soil surface, a vegetation cover map of
continental United States is presented below.
I-1
The depth to constant suction is the depth below the surface of the natural soil at which
there is no more annual and seasonal variations of moisture or suction. It is also called the depth
of the moisture active zone. This depth has been observed in the field to vary between 9.4 and
21 feet, depending upon the depth of penetration of the roots of vegetation. The depth to constant
suction must be used when a water table is deeper than 21 feet. In the absence of vegetation, the
depth to constant suction is at the minimum. The subgrade subroutine sets the depth to constant
suction at 9.4 feet where there is no vegetation at the surface and sets it at:
The equilibrium suction map in Figure 4.14 of the final report shows colored regions of
ranges of suction expressed in cm. The conversion to the pF scale is as follows:
Table I.2 lists the conversion of typical ranges in Equilibrium suction map from cm scale
to pF scale.
I-2
The Fredlund and Xing equation is used to estimate the relationship between matric
suction and water content and expressed as:
1
(I.5)
ln
ln 1
1 (I.6)
1.45 10
ln 1
where S is the degree of saturation (unit is %); is soil matric suction (unit is psi); and , ,
and are soil fitting parameters. These four fitting parameters are the outputs of ANN models
and are predicted by the input variables into SWCC models for plastic and non-plastic materials.
Tables I.2 and I.3 present the associated inputs for SWCC model ( , , and ) of plastic
and non-plastic subgrade materials, respectively.
Table I.3. Input Variables to SWCC Model Parameters for Plastic Materials.
Table I.4. Input Variables to SWCC Model Parameters for Non-plastic Materials.
The resilient modulus of subgrade considering the moisture and suction dependency is
given:
3
(I.7)
I-3
where is the resilient modulus (unit is MPa); is first invariant of the stress tensor (kPa);
is volumetric water content; is matric suction; is octahedral shear stress; is
atmospheric pressure (101.325 kPa); is saturation factor (1 ); , and are
regression coefficients that can be predicted by ANN models for plastic and non-plastic
materials. Tables I.4 and I.5 present the input variables into ANN models ( , and ) of
plastic and non-plastic materials.
Loads.
Layer thickness and Poisson’s Ratio.
Layer Moduli.
Figure I.1 presents the flowchart of incorporating the subgrade models into Pavement ME
Design for flexible and rigid pavements.
I-4
Figure I.2. Flowchart of Incorporating the Subgrade Models into Pavement ME Design for
Flexible and Rigid Pavements.
I-5
The Fredlund and Xing equation is used to estimate the relationship between matric
suction and water content and expressed as:
1
(J.3)
ln
ln 1
1 (J.4)
1.45 10
ln 1
where S is the degree of saturation (unit is %); is soil matric suction (unit is psi); and , ,
, and are soil fitting parameters. These four fitting parameters are the outputs of ANN
J-1
models and are predicted by the input variables into SWCC models for plastic and non-plastic
materials. Tables J.2 and J.3 present the associated inputs for SWCC model ( , , , and )
of plastic and non-plastic base materials.
Table J.2. Input Variables to SWCC Model Parameters for Plastic Materials.
Table J.3. Input Variables to SWCC Model Parameters for Non-plastic Materials.
The resilient modulus of subgrade considering the moisture and suction dependency is
given:
3
(J.5)
where is the resilient modulus (unit is MPa); is first invariant of the stress tensor (kPa);
is volumetric water content; is matric suction; is octahedral shear stress; is
atmospheric pressure (101.325 kPa); is saturation factor (1 ); , and are
regression coefficients that can be predicted by ANN models for plastic and non-plastic
materials. Tables J.4 and J.5 present the input variables into ANN models ( , and ) of
plastic and non-plastic materials.
J-2
The new mechanistic-empirical permanent deformation model for unbound materials was
proposed, which can predict the permanent deformation behavior at different stress state using
the single stage test protocol. The formulation of the model is given as:
(J.6)
2
(J.7)
√3 3
6 ′
(J.8)
√3 3
J-3
Loads.
Layer thickness and Poisson’s Ratio.
Layer Moduli.
Figure J.1 presents the flowchart of incorporating the base course models into Pavement
ME design for flexible pavements.
Iterate
Modulus model coefficients, Calculate
levels of input I, II, III Equilibrium suction in base course
Pavement deformation model Variation of volumetric water content
coefficients, levels of input I, Suction at varying degrees of saturation
II, III Modulus of base course
Anisotropic, stress and suction
dependent
Pavement ME Design
Materials Permanent deformation properties
Figure J.1. Flowchart of Incorporating the Base Course Models into Pavement ME Design
for Flexible Pavements.
J-4
Models of equilibrium suction, SWCC, and resilient modulus of base course for rigid
pavements subroutines are the same as in unbound base course subroutine for flexible pavements
in Appendix J. There is no need for the redundant repetition.
The moisture-sensitive shear strength model for unbound materials is given as:
′ (K.1)
where is the shear strength; is the normal stress; ′ is the effective cohesion; ′ is the
effective friction angle; is the volumetric water content; is the saturation factor; and is
the matric suction. Table K.1 presents the inputs to the shear strength model.
Two faulting models were developed to estimate the development of faulting over time
and traffic repetitions, respectively. One is to predict the entire faulting development over time.
This model of the full faulting curve shows that there is an inflection point in the faulting curve.
Before reaching the inflection point, the accumulation of faulting is caused by the permanent
deformation of the supporting layers. The formulation of the first faulting prediction model is
expressed as:
ln (K.2)
K-1
where is the faulting depth; is the number of days after pavement construction date. is the
number of days when faulting initiates; is the number of days to failure due to erosion; and
and are model coefficients. Table K.2 presents the inputs to the first faulting model
parameters.
Table K.2. Input Variables into First Faulting Prediction Model Parameters.
After passing the inflection point, faulting accelerates due to the action of erosion. The
second model is to predict the faulting depth below the inflection point with traffic and expressed
as:
(K.3)
2 (K.4)
√3 3
6 (K.5)
√3 3
where is the total faulting depth; is the cumulative number of axles at axle load level ;
is the number of the axle when faulting occurs; is atmospheric pressure, 101.305 kPa; is
the second invariant of the deviatoric stress tensor at stress level ; is the first invariant of the
K-2
stress tensor at stress level ; and , , , and are model coefficients. Table K.3 presents
the inputs to the second faulting prediction model parameters.
Table K.3. Input Variables into Second Faulting Prediction Model Parameters.
K-3
The interface shear bonding, , is the ratio of in-situ shear stress in the base course on the
slab-base interface and the shear stress in the base layer on the interface when full shear is
transferred. The formulation of the degree of bonding is given:
(L.1)
where is the degree of bonding; is the shear stress in the base course on the slab-base
interface; and is the shear stress in the base layer on the interface when full shear is
transferred. Table L.1 presents the inputs to the degree of bonding model.
The equivalent thickness is calculated using the moment of inertia of the transformed
section. The transformed section that consists of concrete and base is converted into an
equivalent cross section composed of only concrete. The thickness of the equivalent section is
estimated by:
12
(L.2)
L-1
where is the slab-base equivalent thickness; and b is the width of slab. is the moment of
inertia of the transformed pavement section. Table L.2 presents the inputs to the equivalent
thickness model.
L-2
The equation of radius of relative stiffness based on the concept of equivalent thickness
and subgrade k-value is given:
(M.1)
12 1
where is the modulus of subgrade reaction; Es is the elastic modulus of the PCC; is the
Poisson’s ratio of the PCC; heq is the slab-base equivalent thickness; and le is the radius of
relative stiffness.
M-1
References
1 Shahji, S. 2006. Sensitivity Analysis of AASHTO’s 2002 Flexible and Rigid Pavement Design
Methods. Master of Sc. in the Dept. of Civil & Environmental Engineering—College of
Engineering and Computer Science—University of Central Florida. Orlando, FL.
2 Masad, S.A., and D.N. Little. 2004. Sensitivity Analysis of Flexible Pavement Response and
AASHTO 2002 Design Guide to Properties of Unbound Layers. Research Report ICAR 504-1.
International Center for Aggregates Research, Austin, TX.
3 Masad, S., D. Little, and E. Masad. 2006. “Analysis of Flexible Pavement Response and
Performance Using Isotropic and Anisotropic Material Properties.” Journal of Transportation
Engineering-ASCE, Vol. 132, No. 4, pp. 342–349.
5 Witczak, M.W., D. Andrei, and W.N. Houston. 2000. “Resilient Modulus as Function of Soil
Moisture—Summary of Predictive Models.” Development of the 2002 Guide for the
Development of New and Rehabilitated Pavement Structures. NCHRP 1-37A, Inter Team
Technical Report (Seasonal 1), Arizona State University, Tempe, AZ.
6 Butalia, T.S., J. Huang, D.G. Kim, and F. Croft. 2003. “Effect of Moisture Content and Pore
Water Pressure Buildup on Resilient Modulus of Cohesive Soils in Ohio.” ASTM Special
Technical Publication, No. 1437, pp. 70–84.
7 Wolfe, W., and T. Butalia. 2004. Continued Monitoring of SHRP Pavement Instrumentation
Including Soil Suction and Relationship with Resilient Modulus. Report No. FHWA/OH-
2004/007. U.S. Department of Transportation, Federal Highway Administration, Washington,
DC.
8 Gupta, S., A. Ranaivoson, T. Edil, C. Benson, and A. Sawangsuriya. 2007. Pavement Design
Using Unsaturated Soil Technology. Report No. MN/RC-2007-11. Final Research Report
submitted to Minnesota Department of Transportation, University of Minnesota, Minneapolis,
MN.
9 Cary, C.E., and C.E. Zapata. 2011. “Resilient Modulus for Unsaturated Unbound Materials.”
Road Materials and Pavement Design, Vol. 12, No. 3, pp. 615–638.
11 Adu-Osei, A., D.N. Little, and R.L. Lytton. 2001. “Cross-Anisotropic Characterization of
Unbound Granular Materials.” In Transportation Research Record: Journal of the
R-1
12 Sahin, H., F. Gu, Y. Tong, R. Luo, and R.L. Lytton. 2013. “Unsaturated Soil Mechanics in the
Design and Performance of Pavements.” Advances in Unsaturated Soils, pp. 87–99.
13 Masad, S.A., and D.N. Little. 2004. Sensitivity Analysis of Flexible Pavement Response and
AASHTO 2002 Design Guide to Properties of Unbound Layers. Research Report ICAR 504-1.
International Center for Aggregates Research, Austin, TX.
14 Brown, S.F. 1996. “Soil Mechanics in Pavement Engineering.” Geotechnique, Vol. 46, No. 3,
pp. 383–426.
15 Theyse, H. L., De Beer, M., and Rust, F. C. 1996. “Overview of South African Mechanistic
Pavement Design Method.” Transportation Research Record: Journal of the Transportation
Research Board, No. 1539, pp. 6-17.
16 Zhou, F., E.G. Fernando, and T. Scullion. 2010. Development, Calibration, and Validation of
Performance Prediction Models for the Texas M-E Flexible Pavement Design System. Research
Report FHWA/TX-10/0-5798-2. Texas Transportation Institute, College Station, TX.
17 Gabr, A.R., and D.A. Cameron. 2012. “Permanent Strain Modeling of Recycled Concrete
Aggregate for Unbound Pavement Construction.” Journal of Materials in Civil Engineering,
Vol. 25, No. 10, pp. 1394–1402.
20 Zhou, F., E.G. Fernando, and T. Scullion. 2010. Development, Calibration, and Validation of
Performance Prediction Models for the Texas M-E Flexible Pavement Design System. Research
Report FHWA/TX-10/0-5798-2. Texas Transportation Institute, College Station, TX.
21 Carpenter, S.H., and R.L. Lytton. 1977. Thermal Pavement Cracking in West Texas: Final
Report. Research Report 18-4F. Texas Transportation Institute, Texas A&M University, College
Station, TX.
22 Hansen, W., and E.A. Jensen. 2001. Transverse Crack Propagation of JPCP as Related to
PCC Toughness. Research Report: RC-1404. Michigan Department of Transportation, Lansing,
MI.
23 Jung, Y.S., and D.G. Zollinger. 2011. “New Laboratory-Based Mechanistic-Empirical Model
for Faulting in Jointed Concrete Pavement.” In Transportation Research Record: Journal of the
Transportation Research Board, No. 2226, pp. 60–70.
R-2
24 Jung, Y.S., D.G. Zollinger, and B.M. Ehsanul. 2012. “Improved Mechanistic-Empirical
Continuously Reinforced Concrete Pavement Design Approach with Modified Punchout Model.”
In Transportation Research Record: Journal of the Transportation Research Board, No. 2305,
pp. 32–42.
25 Vandenbossche, J.M., S. Nassiri, L.C. Ramirez, and J.A. Sherwood. 2012. “Evaluating the
Continuously Reinforced Concrete Pavement Performance Models of the Mechanistic-Empirical
Pavement Design Guide.” Road Materials and Pavement Design, Vol. 13, No. 2, pp. 235–248.
26 Rao, C., and M.I. Darter. 2013. “Enhancements to Punchout Prediction Model in
Mechanistic-Empirical Pavement Design Guide Procedure.” In Transportation Research Record:
Journal of the Transportation Research Board, No. 2367, pp. 132–141.
27 Jeong, J.H., and D.G. Zollinger. 2001. “Characterization of Stiffness Parameters in Design of
Continuously Reinforced and Jointed Pavements.” In Transportation Research Record: Journal
of the Transportation Research Board, No. 1778, pp. 54–63.
28 Jung, Y.S., D.G. Zollinger, M. Won, and A.J. Wimsatt. 2009. Subbase and Subgrade
Performance Investigation for Concrete Pavement. Research Report 6037-1. Texas
Transportation Institute, Texas A&M University, College Station, TX.
29 Cleveland, G.S., J.W. Button, and R.L. Lytton. 2002. Geosynthetics in Flexible and Rigid
Pavement Overlay Systems to Reduce Reflection Cracking. Research Report 1777-1. Texas
Transportation Institute, The Texas A&M University System, College Station, TX.
30 Lytton, R. L., F.-L. Tsai, S. I. Lee, R. Luo, S. Hu, and F. Zhou. 2010. NCHRP Report 669:
Models for Predicting Reflection Cracking of Hot-Mix Asphalt Overlays. Transportation
Research Board of the National Academies, Washington, D.C.
31 Byrum, C.R., and R.W. Perera. 2005. “The Effect of Faulting on IRI Values for Jointed
Concrete Pavements.” Paper presented at the Proc. 8th International Conference on Concrete
Pavements, International Society for Concrete Pavements, Bridgeville, PA, and Purdue
University, West Lafayette, IN.
32 Bakhsh, K.N. 2014. Design Methodology for Subgrades and Bases under Concrete Roads and
Parking Lots. PhD diss., Texas A&M University, College Station, TX.
33 Ren, D.Y., L. Houben, and L. Rens. 2013. “Cracking Behavior of Continuously Reinforced
Concrete Pavements in Belgium Characterization of Current Design Concept.” In Transportation
Research Record: Journal of the Transportation Research Board ,No. 2367, pp. 97–106.
34 Jung, Y.S., D.G. Zollinger, and A.J. Wimsatt. 2010. “Test Method and Model Development
of Subbase Erosion for Concrete Pavement Design.” In Transportation Research Record:
Journal of the Transportation Research Board, Vol. 2154, pp. 22–31.
35 Jung, Y., D.G. Zollinger, B.H. Cho, M. Won, and A.J. Wimsatt. 2010. Subbase and Subgrade
Performance Investigation and Design Guidelines for Concrete Pavement. Research Report No.
FHWA/TX-12/0-6037-2. Texas Transportation Institute, The Texas A&M University System,
R-3
36 Bakhsh, K.N., and D. Zollinger. 2014. “Faulting Prediction Model for Design of Concrete
Pavement Structures.” Pavement Materials, Structures, and Performance, pp. 327–342. doi:
10.1061/9780784413418.033
38 Vaillancourt, M., L. Houy, D. Perraton, and D. Breysse. 2014. “Variability of Subgrade Soil
Rigidity and its Effects on the Roughness of Flexible Pavements: A Probabilistic Approach.”
Materials and Structures, pp. 1–10.
39 Yu, S.Y., and P. Dakoulas. 1993. “General Stress-Dependent Elastic-Moduli for Cross-
Anisotropic Soils.” Journal of Geotechnical Engineering-ASCE, Vol. 119, No. 10, pp. 1568–
1586. doi: 10.1061/(ASCE)0733-9410(1993)119:10(1568)
40 Oh, J.H., R.L. Lytton, and E.G. Fernando. 2006. “Modeling of Pavement Response Using
Nonlinear Cross-Anisotropy Approach.” Journal of Transportation Engineering-ASCE, Vol.
132, No. 6, pp. 458–468.
41 Khoury, N.N., and M.M. Zaman. 2004. “Correlation Between Resilient Modulus, Moisture
Variation, and Soil Suction for Subgrade Soils.” In Transportation Research Record: Journal of
the Transportation Research Board, No. 1874, pp. 99–107.
42 Yang, S.-R., W.-H. Huang, and Y.-T. Tai. 2005. “Variation of Resilient Modulus with Soil
Suction for Compacted Subgrade Soils.” In Transportation Research Record: Journal of the
Transportation Research Board, No. 1913, pp. 99–106.
44 Sawangsuriya, A., T.B. Edil, and C.H. Benson. 2009. “Effect of Suction on Resilient Modulus
of Compacted Fine-Grained Subgrade Soils.” In Transportation Research Record: Journal of the
Transportation Research Board, No. 2101, pp. 82–87. Doi 10.3141/2101-10
45 Khoury, C., G.A. Miller, and Y.N. Abousleiman. 2010. “Effect of Suction Hysteresis on
Resilient Modulus of Fine-grained Cohesionless Soil.” 8th Int. Conf. Bearing Capacity Roads,
Railways, and Airfields, Univ. of Illinois–Urbana Champaign, Champaign, IL, pp. 71–78.
46 Vanapalli, S.K., and Z. Han. 2013. “Prediction of the Resilient Modulus of Unsaturated Fine
Grained Soils.” Paper presented at the Proceedings of International Conference on
Advances in Civil Engineering, AETACE.
47 Li, Q., H.J. Lee, and S.Y. Lee. 2011. “Permanent Deformation Model Based on Shear
R-4
48 Oh, J., E.G. Fernando, and R.L. Lytton. 2007. “Evaluation of Damage Potential for
Pavements Due to Overweight Truck Traffic.” Journal of Transportation Engineering-ASCE,
Vol.133, No. 5, pp. 308–317.
50 Huang, Y.H. 1993. Pavement Analysis and Design: Prentice Hall, Englewood Cliffs, NJ.
51 Seed, H.B., F.G. Mitry, C.L. Monismith, and C.K. Chan. 1967. NCHRP Report 35:
Prediction of Flexible Pavement Deflections from Laboratory Repeated-Load Tests. HRB,
National Research Council, Washington, D.C.
52 Hicks, R.G., and C.L. Monismith. 1971. “Factors Influencing the Resilient Properties of
Granular Materials.” In Highway Research Record 345, pp. 15–31.
53 Thompson, M.R., and R.L. Robnett. 1979. “Resilient Properties of Subgrade Soils.”
Transportation Engineering Journal, ASCE, Vol. 105, No. TE1, pp. 71–89.
54 Drumm, E.C. 1990. “Estimation of Subgrade Resilient Modulus from Standard Tests.”
Journal of Geotechnical Engineering, Vol. 116, No 5, pp. 774–789.
56 Witczak, M.W. and J. Uzan. 1988. The University Airport Pavement Design System Report I
of V: Granular Material Characterization. Department of Civil Engineering, University of
Maryland, Silver Spring, MD.
57 Witczak, M.W. 2003. Harmonized Test Methods for Laboratory Determination of Resilient
Modulus for Flexible Pavement Design. Final Report of NCHRP Project 1-28A, Transportation
Research Board, Washington, D.C.
58 Lade, P.V., and R.D. Nelson. 1987. “Modeling the Elastic Behavior of Granular Materials.”
International Journal for Numerical and Analytical Methods in Geomechanics, Volume II, pp.
521–542.
59 Oloo, S.Y., and D.G. Fredlund. 1998. “The Application of Unsaturated Soil Mechanics
Theory to the Design of Pavements.” In Proc. 5th Intl. Conf. on the Bearing Capacity of Roads
and Airfields, Trondheim, Norway, pp. 1419–1428.
60 Lytton, R.L., J. Uzan, E.G. Fernando, R. Roque, D. Hiltunen, and S.M. Stoffels. 1993.
R-5
61 Lytton, R.L. 1995. “Foundations and Pavements on Unsaturated Soils.” Paper presented at the
Proceedings of the First International Conference on Unsaturated Soil, Paris.
63 Oh, J.H., and E.G. Fernando. 2008. Develop of Thickness Design Tables Based on the M-E
PDG. Research Report No. BDH10-1. Texas Transportation Institute, The Texas A&M
University System, College Station, TX.
64 Al-Qadi, I.L., H. Wang, and E. Tutumluer. 2010. “Dynamic Analysis of Thin Asphalt
Pavements by Using Cross-Anisotropic Stress-Dependent Properties for Granular Layer.” In
Transportation Research Record: Journal of the Transportation Research Board, No. 2154,
pp. 156–163.
65 Tutumluer, E., and M.R. Thompson. 1997. “Anisotropic Modelling of Granular Bases in
Flexible Pavements.” Transportation Research Record: Journal of the Transportation Research
Board, No. 1577, pp. 18–26.
66 Gu, F., X. Luo, Y. Zhang, H. Sahin, and R.L. Lytton. 2015. “Modeling of Moisture-Sensitive
and Stress-Dependent Nonlinear Cross-Anisotropic Behavior of Unbound Aggregates.”
67 Yau, A., and H.L. Von Quintus. 2002. Study of LTPP Laboratory Resilient Modulus Test
Data and Response Characteristics. Report No. FHWA-RD-02-051. Federal Highway
Administration Research, U.S. Department of Transportation.
68 ARA, Inc, 2004. Guide for Mechanistic-Empirical Design of New and Rehabilitated
Pavement Structures. Final Report of NCHRP Project 1-37A: Transportation Research Board,
Washington, D.C.
70 Kenis, W.J. 1977. Predictive Design Procedures, VESYS User’s Manual. Final Report, No.
FHWA-RD-77-154. Federal Highway Administration, McLean, VA.
72 Tseng, K.H., and R.L. Lytton. 1989. “Prediction of Permanent Deformation in Flexible
Pavement Materials.” In Implication of Aggregates in the Design, Construction, and
R-6
Performance of Flexible Pavement, ASTHMA STP 1016, ASTM, Philadelphia, PA., pp. 154–172.
73 Ayres, M.Jr. and M.W. Witczak. 1998. “AYMA: A Mechanical Probabilistic System to
Evaluate Flexible Pavement Performance.” Transportation Research Record 1629, pp. 137–148.
75 Chow, L.C., Mishra, D. and Turumluer, E. 2014. Aggregate base course material testing and
rutting model development. Final Report NCDOT Project 2013-18, FHWA/NC/2013-18.
76 Gu, F., Zhang, Y.Q., Droddy, C.V., Luo, R. and Lytton, R.L. 2015. “Development of a New
Mechanistic-Empirical Rutting Model for Unbound Granular Material.” Journal of
Materials in Civil Engineering, Vol. 28, No. (8).
77 Epps, J., S. Sebesta, B. Hewes, H. Sahin, R. Luo, J. Button, R.L. Lytton, C.A. Herrera, R.
Hatcher, and F. Gu. 2014. Development of a Specification for Flexible Base Construction.
Research Report No. FHWA/TX-13/0-6621. Texas A&M Transportation Institute, College
Station, TX.
78 Lambe, T.W., and R.V. Whitman. 1969. Soil Mechanics. Wiley, New York.
79 Abramento, M., and S. Carvalho. 1989. “Goetechnical Parameters for Study of Slope
Instabilization at Serra do Mar-Brazillan Southeast.” In Proceedings of the 12th International
Conference on Soil Mechanics and Foundation Engineering, Rio de Janeiro, Vol. 3, pp. 1599–
1602.
80 Fredlund, D.G., and H. Rahardjo. 1993. Soil Mechanics for Unsaturated Soils. John Wiley &
Sons, Inc., New York, NY.
82 Fredlund, D.G., A. Xing, M.D. Fredlund, and S.L. Barbour. 1996. “The Relationship of the
Unsaturated Soil Shear to the Soil-Water Characteristic Curve.” Canadian Geotechnical Journal,
Vol. 33, No. 3, pp. 440–448.
83 Rassam, D.W., and F.J. Cook. 2002. “Predicting the Shear Strength Envelope of Unsaturated
Soils.” Geotechnical Testing Journal, Vol. 28, pp. 215–220.
84 Titus-Glover, L., and E.G. Fernando. 1995. Evaluation of Pavement Base and Subgrade
Material Properties and Test Procedures. Research Report 1335-2. Texas Transportation
Institute, The Texas A&M University System, College Station, TX.
85 Rauhut, J.B., R.L. Lytton, and M.I. Darter. 1982. Pavement Damage Functions for Cost
R-7
Allocation, Vol. 1, Damage Functions and Load Equivalency Factors. Report No. FHWA/RD-
82/126. Federal Highway Administration, Washington, DC.
86 Markow, M.J., and B.D. Brademeyer. 1984. EAROMAR Version 2. Final Technical Report
FHWA/RD-82/086. Federal Highway Administration, U.S. Department of Transportation.
87 Larralde, J. 1984. Structural Analysis of Ridge Pavements with Pumping. PhD diss., Purdue
University, West Lafayette, IN.
88 Van Wijk, A.J. 1985. Rigid Pavement Pumping: (1) Subbase Erosion and (2) Economic
Modeling. Joint Highway Research Project File 5-10. School of Civil Engineering, Purdue
University, West Lafayette, IN.
90 Winkler, E. 1867. Die Lehre von der Elasticitaet und Festigkeit. Prag, Dominicus.
92 Hetenyi, M. 1950. “A General Solution for the Bending of Beams on an Elastic Foundation of
Arbitrary Continuity.” Journal of Applied Physics, Vol .21, pp. 55–58.
93 Pasternak, P.L. 1954. “On a New Method of Analysis of an Elastic Foundation by Means of
Two Foundation Constants.” Gosudarstrennoe Izdatel′stro Literaturi po Stroitel′stru i
Arkhitekture, Moscow, USSR (in Russian).
94 Kerr, A. 1965. “A Study of a New Foundation Model.” Acta Mechanica, Vol. 1, No. 2, pp.
135–147. doi: 10.1007/BF01174308
95 Timoshenko, S., and J. Goodier. “Theory of elasticity.” 1951. New York, Vol. 412, p. 108.
R-8