Numerical Analysis of Cracking in Concrete Pavements Subjected To Wheel Load and Thermal Curling PDF

Numerical Analysis of Cracking in Concrete Pavements
Subjected to Wheel Load and Thermal Curling
A dissertation submitted to the
Graduate School
of the University of Cincinnati
in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
in Civil Engineering
in the School of Advanced Structures

of the College of Engineering and Applied Sciences
by
Temesgen W. Aure
M.Sc., Addis Ababa University (2008)

B.Sc., Bahir Dar University (2003)
September 2013
Committee Chair: Anastasios M. Ioannides, Ph.D.

Committee Members: Anant R. Kukreti, Ph.D.
Richard A. Miller, Ph.D.
Jeffery R. Roesler, Ph.D.
Abstract
The main goal of this research is to implement recent advances in nonlinear fracture mechanics,
most notably the introduction of the cohesive zone concept, in investigating the post-cracking
behavior of concrete pavements, subjected to wheel load and curling. The cohesive zone is
assumed to lie along a specified direction known a priori, and cohesive elements are inserted
along this path. The study follows a step-by-step approach, beginning with fracture analysis of
simply supported beams. In this initial step, experimental and numerical studies available in
literature are reproduced and excellent agreement is observed. Furthermore, important fracture
parameters and numerical challenges are identified, pertinent to the cohesive zone concept. It is
observed that post-crack responses of the beams are sensitive to choices regarding the solution
type, the concrete softening curve, and the uncracked region mesh sizes.
Single slabs-on-grade under wheel loads located at the slab edge or interior are considered next.
In this phase, it is observed that the increased size of the problem inhibits generating as refined
mesh as for the beams, and consequently obtaining a convergent solution poses a significant
challenge. This is resolved by using so-called viscous regularization, in which a small viscosity
term is introduced. Accordingly, a small deviation of traction stresses beyond the pre-defined
material softening curve can be tolerated. Once again, the simulation in this phase is verified by
reproducing experimental and numerical results available in literature. The effects of concrete
softening curve, cohesive zone mesh, solution method, fracture energy, and tensile strength on
the fracture process are investigated. It is observed that the fracture energy is the major
parameter that influences the responses.

ii
In a third phase of the study, a single slab-on-grade is subjected to wheel load and curling,
individually or in combination. In both cases, it is observed that the diurnal temperature cycle
and the shape of its profile through the slab thickness plays a significant role on the post-crack
responses of the slab. When the slab top is warmer, unstable cracks form; in contrast, a warmer
bottom results in stable cracks, thereby increasing the resistance of the slab and avoiding sudden
failure.
The final two phases of the research are devoted to the study of jointed concrete pavements, also
subjected to wheel load and temperature variations: the fourth phase encompasses aggregate
interlock joints and the fifth phase pertains to dowel bar joints. Linear and nonlinear aggregate
interlock mechanisms are simulated and their repercussions on the fracture responses of the slabs
are examined. Similarly, the effects of numerical idealization techniques for the dowel-slab
interaction, joint size and dowel looseness on the fracture process are examined.
It is concluded that the cohesive zone approach is a very promising tool in the ongoing
exploration of fracture behavior in concrete pavements. The techniques can be extended to
general loading situations that involve fatigue and crack branching. The results from this study
will contribute to the development of a more mechanistic failure analysis of concrete pavements.
iii
iv
To my parents: Ajebush Amante and Wondimu Aure
v
Acknowledgements
I would like to express my sincere gratitude to my academic advisor Dr. Anastasios M. Ioannides
for his continuous guidance during the course of this study. I also appreciate his financial support
for the first three years of this study. The discussions we had on both technical and non-technical
matters have broadened my outlook.
I would also like to thank my committee members Dr. Anant R. Kukreti, Dr. Richard A. Miller
and Dr. Jeffery R. Roesler for their valuable comments and suggestions. I am indebted, in
particular, to Dr. Kukreti for his encouragement and financial support through graduate
assistantship during the final stage of this research.
I am grateful to my friends Dawit Alemayehu, Demelash Hailemariam, Aschalew Kassa, Assefa
Yeshitla, Gashaw Assefa, Meron Tekle, Ebssa Duguma and Asnake Eticha for their
encouragement and support during my stay at the University of Cincinnati. I am also indebted to
the inspiration I obtained from my office mates Raj Dehal and Akshat Saxena.
My special gratitude goes to the Cincinnati Mahidere Selam Ethiopian Orthodox Tewahedo
Church board members and its congregation for providing me financial and moral supports
during the difficult times of this study. My blessing particularly goes to Yeneta (=Reverend)
Yemaneberhan Haile, Mr. Asrat Getnet and his family, Mr. Mesfin Lemma and his family, Mr.
Teshome Tefera and his family, Mr. Yohannes Elias, Mrs. Alemnesh, Deacon Lakew Dagne,
Mrs. Beletu, Mrs. Wudnesh Kassa, Ms. Alemshet Mekonnen, Mr. Gezahegn Tamene, and Mr.
vi
Ashenafi. I would like to extend my gratitude to Ato Berhanu Abebe for his encouragement and
support during the last year of this study.
I would also like to thank my wife Meskerem for her love, patience, and encouragement, all of
which ultimately resulted in the completion of this study. Finally, I would not have achieved this
academic level without the commitment of my parents Ajebush Amante and Wondimu Aure,
who instilled interest and curiosity in education since my early childhood.
vii
Table of Contents
Abstract .......................................................................................................................................... ii
Acknowledgements ...................................................................................................................... vi
Table of Contents ....................................................................................................................... viii
List of Tables .............................................................................................................................. xiv
List of Figures ............................................................................................................................. xvi
List of Symbols and Abbreviations ......................................................................................... xxii
Chapter 1 Research Overview ............................................................................................... 1
1.1 Introduction ..................................................................................................................... 1
1.2 Review of Concrete Fracture .......................................................................................... 3
1.3 Research Objective and Significance.............................................................................. 6
1.4 Research Scope ............................................................................................................... 7
1.5 Thesis Organization ........................................................................................................ 7
1.6 References ....................................................................................................................... 9
Chapter 2 Simulation of Crack Propagation in Concrete Beams ..................................... 19
2.1 Introduction ................................................................................................................... 20
2.2 Methodology ................................................................................................................. 23
2.3 Analysis of Linear Elastic Response............................................................................. 23
2.4 Simulation of Crack Propagation Using Built-in Cohesive Elements .......................... 26
2.4.1 Idealization of Cohesive Zone .................................................................................. 27
2.4.2 Sensitivity Study for Proposed Discretization .......................................................... 33
viii
2.5 Comparison with Previous Numerical Studies ............................................................. 38
2.5.1 GTSTRUDL/CRACKIT by Sengupta (1998) .......................................................... 38
2.5.2 ABAQUS® - JOINTC by Ioannides et al. (2005) ..................................................... 39
2.5.3 ABAQUS® - 2D UEL by Roesler et al. (2007a) ....................................................... 39
2.5.4 ABAQUS® - 2D COH2D4 by Gaedicke and Roesler (2009) ................................... 40
2.6 Comparisons with Experimental Results ...................................................................... 41
2.6.1 Experimental Results by Liu (1994) ......................................................................... 41
2.6.2 Experimental Results by Roesler et al. (2007a) ........................................................ 42
2.7 Conclusions ................................................................................................................... 42
2.8 References ..................................................................................................................... 44
Chapter 3 Numerical Analysis of Fracture Process in Pavement Slabs .......................... 67
3.1 Introduction ................................................................................................................... 68
3.2 Literature Review.......................................................................................................... 69
3.3 Finite Element Discretization ....................................................................................... 71
3.4 Linear Elastic Analysis ................................................................................................. 72
3.5 Sensitivity Studies ......................................................................................................... 73
3.5.1 Effect of Concrete Softening Curve .......................................................................... 73
3.5.2 Effect of Viscous Regularization .............................................................................. 78
3.5.3 Effect of Subgrade Idealization ................................................................................ 80
3.5.4 Effect of Cohesive Zone Width ................................................................................ 81
3.5.5 Effect of Cohesive Zone Mesh ................................................................................. 82
3.5.6 Effect of Solution Technique .................................................................................... 82

ix
3.5.7 Effect of Loading Mode ............................................................................................ 83
3.5.8 Effect of Tensile Strength and Fracture Energy ....................................................... 84
3.6 Crack Propagation under Interior Loading ................................................................... 86
3.7 Crack Propagation in a Slab Supported by Base Layer ................................................ 87
3.8 Comparison with Other Numerical and Experimental Studies ..................................... 88
3.8.1 Comparison with Ioannides et al. (2006) .................................................................. 88
3.8.2 Comparison with Gaedicke and Roesler (2009) ....................................................... 89
3.9 Conclusions ................................................................................................................... 91
3.10 References ..................................................................................................................... 92
Chapter 4 Curling Effects on Concrete Pavement Fracture ........................................... 119
4.1 Introduction ................................................................................................................. 120
4.2 Finite Element Discretization ..................................................................................... 121
4.2.1 General Considerations ........................................................................................... 122
4.2.2 Fracture Process Zone (FPZ) .................................................................................. 123
4.3 Pre-Crack Analysis of Slab under Thermal Curling Alone ........................................ 124
4.3.1 Comparison with 3D FE Analysis .......................................................................... 124
4.3.2 Comparison with 2D FE Analysis and Westergaard (1927) ................................... 126
4.4 Pre-Crack Analysis of Nonlinear Curling-and-Wheel Load ....................................... 129
4.5 Fracture Response of Pavement Slab Subjected to Thermal Curling Alone .............. 131
4.5.1 Effect of Notch Depth ............................................................................................. 133
4.5.2 Effect of Concrete Age ........................................................................................... 134
4.5.3 Effect of Slab Self-Weight ...................................................................................... 135

x
4.5.4 Effect of Nonlinearity of Temperature Distribution ............................................... 136
4.5.5 Effect of Slab Size .................................................................................................. 137
4.6 Fracture Response of Slab Subjected to Wheel Load and Thermal Curling .............. 138
4.6.1 Slab under Constant Thermal Gradient and Increasing Wheel Load ..................... 139
4.6.2 Slab under Constant Wheel Load and Increasing Thermal Gradient ..................... 146
4.7 Summary and Conclusions ......................................................................................... 149
4.8 References ................................................................................................................... 151
Chapter 5 Crack Propagation in Pavement Slabs with Aggregate Interlock................ 193
5.1 Introduction ................................................................................................................. 194
5.2 Mechanics of Aggregate Interlock: A Review............................................................ 196
5.3 Pre-Crack Aggregate Interlock Load Transfer Analysis ............................................ 200
5.3.1 Finite Element Discretization ................................................................................. 200
5.3.2 Validation of Linear Aggregate Interlock Discretization ....................................... 206
5.3.3 Validation of Nonlinear Aggregate Interlock Discretization .................................. 209
5.4 Post-Crack Analysis of Slabs with Aggregate Interlock under Wheel Load Alone ... 211
5.4.1 Linear Aggregate Interlock ..................................................................................... 212
5.4.2 Nonlinear Aggregate Interlock ............................................................................... 214
5.5 Pre- and Post-Crack Analysis of Slabs with Aggregate Interlock under Constant
Temperature Distribution and Increasing Wheel Load ........................................................... 218
5.5.1 Linear Aggregate Interlock ..................................................................................... 219
5.5.2 Nonlinear Aggregate Interlock ............................................................................... 221

xi
5.7 References ................................................................................................................... 224
Chapter 6 Numerical Analysis of Cracking in Doweled Concrete Pavements .............. 251
6.1 Introduction ................................................................................................................. 252
6.2 Finite Element Idealization of Doweled Pavement System ........................................ 253
6.2.1 Discretization of Slab and Subgrade ....................................................................... 253
6.2.2 Discretization of Dowel Bars .................................................................................. 255
6.2.3 Discretization of the Fracture Process .................................................................... 256
6.3 Validation of Finite Element Discretization: Pre-Crack Analysis .............................. 257
6.4 Crack Propagation in Slabs with Doweled Joints ....................................................... 259
6.4.1 Effect of Initial Joint Opening ................................................................................ 259
6.4.2 Effect of Dowel–Concrete Interaction .................................................................... 260
6.4.3 Effect of Dowel Bar Yielding ................................................................................. 263
6.4.4 Effect of Temperature Differential ......................................................................... 264
6.6 References ................................................................................................................... 266
Chapter 7 Summary, Conclusions and Recommendations for Future Work ............... 283
7.1 Research Summary ..................................................................................................... 283
7.2 Concluding Remarks ................................................................................................... 288
7.3 Recommendations for Future Research ...................................................................... 294
7.4 References ................................................................................................................... 297
xii
This page is intentionally left blank
xiii
List of Tables
Table 2.1 Geometry and Material Properties of Beams Studied .................................................. 49
Table 2.2 Deflection at Midspan of Beam A for Different Types of Elements ............................ 49
Table 2.3 Comparison of JOINTC and COH2D4 for Different Mesh Sizes (Beam A) ............... 50
Table 3.1 Geometry, Material Properties and Discretization of Slabs Considered .................... 101
Table 3.2 Comparison of Different Solid Elements Available in ABAQUS ............................... 102
Table 4.1 Comparison with Experimental and Numerical Results ............................................. 158
Table 4.2 Coefficients A, B, and C for Daily Cycle in June (Choubane and Tia, 1992) ............ 159
Table 4.3 Comparison of 3D Linear and Nonlinear Curling ...................................................... 160
Table 4.4 Comparisons for Linear Curling ................................................................................. 161
Table 4.5 Comparison with 2D FE Results for Quadratic Temperature Profile ......................... 162
Table 4.6 Comparison of 3D FE Combination and Superposition Analysis .............................. 163
Table 4.7 Comparison Among Different FE Programs .............................................................. 163
Table 4.8 Material Properties for Concrete at Different Ages (Jenq et al., 1993) ...................... 164
Table 4.9 Geometry, Material Properties and Discretization of Slab Considered ...................... 165
Table 4.10 Coefficients of Cubic Polynomial Used to Estimate Temperature Distribution ...... 166
Table 5.1 Walraven’s Equations for Contact Areas per Unit Crack Area .................................. 229
Table 5.2 Walraven’s Equations for Contact Areas per Unit Crack Area .................................. 230
xiv
Table 5.3 Joint Stiffness Characteristics Used for Linear Aggregate Interlock ......................... 232
Table 6.1 Geometry, Material Properties and Discretization of Pavement System.................... 271
xv
List of Figures
Figure 1.1 Fracture Process Zone in Concrete .............................................................................. 17
Figure 2.1 Comparison of Theoretical and Finite Element Results (Beam A) ............................. 51
Figure 2.2 Bilinear Softening Curve ............................................................................................. 52
Figure 2.3 Effect of Analysis technique (Beam A)....................................................................... 53
Figure 2.4 Effect of Analysis technique (Beam B) ....................................................................... 54
Figure 2.5 Effect of Mesh Fineness (Beam A) ............................................................................. 55
Figure 2.6 Effect of Mesh Fineness (Beam B) ............................................................................. 56
Figure 2.7 Effect of Notch Depth (Beam B) ................................................................................ 57
Figure 2.8 Effect of Cohesive Zone Width (Beam B) .................................................................. 58
Figure 2.9 Effect of Softening Curve (Beam B) ........................................................................... 59
Figure 2.10 Comparison of 2D CRACKIT with 3D Cohesive Element (Beam B) ....................... 60
Figure 2.11 Comparison between JOINTC and Cohesive Element (Beam A) ............................. 61
Figure 2.12 Comparison with Numerical Results by Roesler et al. (2007a) (Beam C) ................ 62
Figure 2.13 Comparison with Numerical Results by Gaedicke and Roesler (2009) (Beam C) ... 63
Figure 2.14 Comparison with Experimental Results by Liu (1994) (Beam B) ............................ 64
Figure 2.15 Comparison with Experimental Results by Roesler et al. (2007) (Beam C) ............. 65
Figure 3.1 Finite Element Discretization of Slab (SL1) ............................................................. 103
xvi
Figure 3.2 Finite Element Mesh Pattern Using C3D8R Elements (SL1) ................................... 103
Figure 3.3 Different Types of Concrete Softening Curves ......................................................... 104
Figure 3.4 Effect of Concrete Softening Curve and Analysis Technique ................................... 105
Figure 3.5 Effect of Viscous Regularization on Slab Response ................................................. 106
Figure 3.6 Effect of Subgrade Idealization on Slab Response.................................................... 107
Figure 3.7 Effect of Cohesive Zone Width ................................................................................. 108
Figure 3.8 Effect of Cohesive Zone Mesh .................................................................................. 109
Figure 3.9 Effect of Loading Mode ............................................................................................ 110
Figure 3.10 Effect of Tensile Strength ........................................................................................ 111
Figure 3.11 Effect of Fracture Energy ........................................................................................ 112
Figure 3.12 Variation of Normalized Peak Load, LLD and CMOD with Brittleness Number... 113
Figure 3.13 Slab Responses under Different Loading Scenaria Applied at Interior and Edge .. 114
Figure 3.14 Crack Growth Along Slab Length and Width at Points Shown in Fig. 3.13 ........... 115
Figure 3.15 Effect of Base Stiffness on Slab Response .............................................................. 116
Figure 3.16 Comparison with Numerical Results by Ioannides et al. (2006) ............................. 117
Figure 3.17 Comparison with Numerical and Experimental Results by Gaedicke and Roesler
(2009) .......................................................................................................................................... 118
Figure 4.1 Comparison Among Different 3D FE Simulations (Tension is Positive) ................. 167
xvii
Figure 4.2 Variation of Daytime Temperature (9 am - 8 pm) .................................................... 168
Figure 4.3 Variation of Longitudinal Bending Stress Caused by Daytime Temperature ........... 169
Figure 4.4 Variation of Nighttime Temperature (10 pm - 6 am) ................................................ 170
Figure 4.5 Variation of Longitudinal Bending Stress Caused by Nighttime Temperature (3D
Nonlinear Curling) ...................................................................................................................... 171
Figure 4.6 Comparison Among Three FE Programs .................................................................. 172
Figure 4.7 Effect of Notch Depth on Cracking Due to Curling .................................................. 173
Figure 4.8 Comparison with Jenq et al. (1993) ........................................................................... 174
Figure 4.9 Effect of Concrete Age on Fracture Due to Curling.................................................. 175
Figure 4.10 Variation of Maximum Dimensionless Temperature Differential with Brittleness
Number ....................................................................................................................................... 176
Figure 4.11 Effect of Concrete Unit Weight on Fracture Due to Curling .................................. 177
Figure 4.12 Effect of Temperature Profile on Fracture Due to Curling ..................................... 178
Figure 4.13 Effect of Slab Size on Fracture Due to Curling ....................................................... 179
Figure 4.14 Finite Element Idealization of Slab with Properties Shown in Table 4.9 ............... 180
Figure 4.15 Load-Plus-Curling Response Curves under Diurnal Thermal Distributions........... 181
Figure 4.16 Crack Growth along Slab Width for Load Levels Shown in Fig. 4.15 ................... 182
Figure 4.17 Crack Growth along Slab Width for Load Levels Shown in Fig. 4.15 ................... 183
Figure 4.18 Comparison of Linear and Cubic Temperature Profiles.......................................... 184

xviii
Figure 4.19 Effect of Slab Length and Width ............................................................................. 185
Figure 4.20 Effect of Slab Self-Weight ...................................................................................... 186
Figure 4.21 Variation of Dimensionless Peak Load versus Dimensionless Self-Weight ........... 187
Figure 4.22 Effect of Tensile Strength ........................................................................................ 188
Figure 4.23 Effect of Total Fracture Energy ............................................................................... 189
Figure 4.24 Fracture Responses of Slab under Constant Wheel Load and Increasing Daytime
Temperature Differential ............................................................................................................ 190
Figure 4.25 Effect of Notch Depth on Peak Linear Temperature Differential (Curling Only) .. 191
Figure 4.26 Comparison of Two Loading Scenaria .................................................................... 192
Figure 5.1 Typical Normal and Shear Stress Variations at Different Crack Openings .............. 233
Figure 5.2 Comparison Among Three Linear Aggregate Interlock Idealizations and Previous
Studies ......................................................................................................................................... 234
Figure 5.3 Verification of Relationship between Load Transfer Efficiencies ............................ 235
Figure 5.4 Top View, Boundary Conditions and Applied Load ................................................. 236
Figure 5.5 Comparison with Experimental and FE Results by Other Researchers (Nonlinear
Aggregate Interlock) ................................................................................................................... 237
Figure 5.6 Planar View, Location of Applied Load and Assumed Fracture Planes ................... 238
Figure 5.7 Effect of Joint Stiffness on Slab System Response ................................................... 239
Figure 5.8 Pre-Crack and Post-Crack Load Transfer Responses ................................................ 240
xix
Figure 5.9 Variation of Load Transfer Efficiency with Total Applied Load.............................. 241
Figure 5.10 Effect of Initial Joint Opening on Load-Displacement Response ........................... 242
Figure 5.11 Effect of Initial Joint Opening on Load Transfer at Peak Load .............................. 243
Figure 5.12 Variation of Load Transfer Efficiency with Total Applied Load............................ 244
Figure 5.13 Effect of Maximum Aggregate Size (Nonlinear Aggregate Interlock) ................... 245
Figure 5.14 Effect of Daytime Temperature (Linear Aggregate Interlock) ................................ 246
Figure 5.15 Effect of Nighttime Temperature (Linear Aggregate Interlock) ............................. 247
Figure 5.16 Effect of Daytime Temperature ............................................................................... 248
Figure 5.17 Effect of Nighttime Temperature ............................................................................ 249
Figure 5.18 Effect of Temperature on Joint Efficiency .............................................................. 250
Figure 6.1 Finite Element Mesh.................................................................................................. 272
Figure 6.2 Slab Geometry and Loading Layout.......................................................................... 273
Figure 6.3 Finite Element Mesh for Slab LSM-2 Tested by Hammons (1997) ......................... 274
Figure 6.4 Comparison with Other Experimental and Numerical Studies ................................. 275
Figure 6.5 Effect of Initial Joint Opening (D/skl ~ 30) .............................................................. 276
Figure 6.6 Effect of Initial Joint Opening on Load Transfer Response ...................................... 277
Figure 6.7 Effect of Dowel Slip and Embedment (ωi = 0.2 in.).................................................. 278
Figure 6.8 Effect of Dowel Bar Yielding.................................................................................... 279
xx
Figure 6.9 Cracked Concrete Pavement Slab under Daytime Temperature and Wheel Load .... 280
Figure 6.10 Effect of Temperature.............................................................................................. 281
xxi
List of Symbols and Abbreviations
2D two-dimensional
3D three-dimensional
a notch depth
AASHTO American association of state highway and transportation officials
Ax , Ay most probable projected contact areas in x and y-directions
AGG joint stiffness per unit length
b beam width
B brittleness number
CMOD crack mouth opening displacement
CTODc elastic critical crack tip opening displacement
D dimensionless damage variable, particle size
Dmax maximum aggregate grain size
Dv viscous stiffness degradation parameter
Dγ dimensionless unit weight
E Young’s modulus
EB Euler beam
Eff Joint effectiveness
FE finite element
FPZ fracture process zone
FCM fictitious crack model
fn, ft normal and tangential crack interface stress functions

xxii
ft ' concrete tensile strength
f c' concrete compressive strength
G shear modulus
Gf initial fracture energy
GF total fracture energy
h beam or slab thickness
I moment of inertia
ITI initial time increment
k modulus of subgrade reaction
Knn, Kss, Ktt nominal stiffness in normal and two shear directions
KIC stress intensity factor
L beam length
l radius of relative stiffness
lch characteristic length of material
lmax maximum arc-length increment
lmin minimum arc-length increment
lperiod total arc-length scale factor
LEFM linear elastic fracture mechanics
LLD load line displacement
LTEO load transfer efficiency with respect to crack mouth opening displacement
LTEδ load transfer efficiency with respect to deflection
xxiii
LTEσ load transfer efficiency with respect to stress
MEPDG mechanistic-empirical pavement design guide
MnTI minimum time increment
MxTI maximum time increment
NLFM nonlinear fracture mechanics
NC number of nodal columns
NR number of nodal rows
P applied load
PCC Portland cement concrete
p pressure, percent of aggregate passing sieve size D
pk aggregate volume per unit volume of concrete
PT total transferred load
Pmax peak load
q aggregate interlock factor
S beam span
SEM size effect method
T0 initial width of cohesive zone
TB Timoshenko beam
TE theory of elasticity
TLE transferred load efficiency
TPFM two parameter fracture model
TPS time period of step
xxiv
t traction stress
tn, ts, tt nominal stress along normal and two shear directions
t′i traction stress corresponding to w without cracking
t′s traction stress for separation w along the softening curve
v deflection
vmax maximum deflection
UEL user element
w crack opening
w1 crack opening corresponding to initial fracture energy
wn,ws,wt elastic cohesive displacements along normal and two shear directions
wcr cohesive zone separation at crack initiation
we effective traction displacement
wf crack opening at zero traction
wk cohesive zone separation at kink point
α coefficient of thermal expansion, concrete softening parameter
β dimensionless peak load
γ concrete unit weight
Δ vertical displacement at peak load
δL vertical deflection of loaded slab
Δlin initial increment in arc-length
δn, δs, δt normal and two shear interface displacements
ΔT temperature differential
xxv
δU vertical deflection of unloaded slab
εn, εs, εt nominal strain in along normal and two shear directions
ζ dimensionless load line displacement
η dimensionless crack mouth opening displacement
κ joint stiffness for corner node
λ joint length
μ Poisson’s ratio, coefficient of friction
σmax maximum bending stress at the bottom of loaded area
σn, σt normal and tangential interface stresses
σpu aggregate-paste matrix yielding strength
σxx longitudinal horizontal stress through slab thickness
ψ kink point stress ratio
ωi initial crack opening displacement
м viscosity parameter
xxvi
xxvii
Chapter 1 Research Overview
1.1 Introduction
Distresses in concrete pavement slabs are caused primarily by traffic loads and thermal curling.
The first analytical solution for pavement system responses under such actions was provided by
Westergaard (1926, 1927), and was limited by several simplifying assumptions (Ioannides et al.,
1985). Due to the complex nature of the problem, many subsequent studies have been devoted to
the elimination of these assumptions through the use of computerized numerical techniques,
particularly the finite element (FE) method (Huang and Wang, 1973; Tabatabaie and Barenberg,
1978; Ioannides and Korovesis, 1990; Khazanovich, 1994). As a result of this effort, several
stand-alone finite element programs for concrete pavement analysis have been developed over
the last 40 years, including among others, KENSLABS (Huang and Wang, 1973), ILLI-SLAB
(Tabatabaie and Barenberg, 1978), JSLAB (Tayabji and Colley, 1986), FEACONS (Tia et al.,
1987), EverFE (Davids et al., 1998) and Pave3D (Nishizawa et al., 2001).
Such analytical efforts have gradually precipitated a major shift in the pavement design paradigm
of the American Association of State Highway and Transportation Officials (AASHTO),
culminating in the abandonment of the heretofore statistical/empirical serviceability oriented
approach (AASHTO, 1993), in favor of a mechanistic-empirical pavement design guide
(MEPDG) aimed at distress prevention (AASHTO, 2008). According to the MEPDG, stresses
and strains are computed on the basis of engineering mechanics principles using a computer
program, and are provided as inputs to statistical/empirical algorithms, usually termed transfer
functions, in order to determine anticipated individual distress levels. For concrete pavements,
finite element program ISLAB2000 (Khazanovich et al., 2000), a commercial version of ILLI-
1
SLAB, is employed in the calculation of pavement responses. The accuracy of these responses is
not particularly critical since the statistical/empirical algorithms into which they are introduced
as inputs exhibit coefficients of determination ranging from less than 30% to about 60%
(AASHTO, 2008). The rationale of transfer functions for the purpose of converting responses to
distresses is founded on the so-called cumulative linear fatigue damage hypothesis (Miner,
1945), which originally pertained to cracking in metals but which has been extended without
justification to practically all pavement material distresses. Such phenomenological applications
may, therefore, be expected to lead to unrealistic pavement distress level predictions, and
statistical/empirical transfer functions remain one of the weakest links in the AASHTO (2008)
MEPDG, urgently in need of replacement by a more rational and mechanistic alternative. To
accomplish this objective, it is necessary to go beyond a linear elastic material characterization
toward a general nonlinear analysis implementing concepts of fracture mechanics, as is done in
other branches of engineering (Ioannides, 1997).
With the development of parallel processing, large memory computers and multi-purpose FE
commercial software, fracture mechanics promises to enhance our understanding of structural
behavior of concrete structures beyond the elastic limit, as cracking is initiated and propagates.
Application of fracture mechanics to pavement systems has recently gained momentum as
encouraging results are emerging (Ioannides et al., 2006; Song et al., 2006; Roesler et al., 2007;
Gaedicke and Roesler, 2009). This study contributes to such ongoing efforts by employing state-
of-the art tools of fracture mechanics in investigating concrete pavement slab cracking as
affected by several pertinent input parameters. Results presented elucidate the post-crack
2
response of single-slab and jointed concrete pavements subjected to wheel load and thermal
curling.
1.2 Review of Concrete Fracture
Concrete is a two phase heterogeneous material (cement paste and aggregate particles), whose
fracture involves a complex phenomenon comprising “micro-cracking, crack deflection [caused
by aggregate particles], aggregate bridging, crack-face friction, crack tip blunting by voids, and
crack branching”, usually termed as the fracture process zone (FPZ) (Shah et al., 1995). The size
of the FPZ in concrete is relatively larger than in other materials, both brittle and ductile. This
phenomenon was first noted by Kaplan (1961), whose experiments demonstrated the formation
of unstable cracks preceded by slow and stable crack growth, which indicates that transfer of
traction stresses occurs in the FPZ before a complete opening of two cracked faces (see Fig. 1.1).
Since linear elastic fracture mechanics (LEFM) assumes linear elastic material behavior and a
FPZ of negligible size, it cannot be applied to concrete (Kaplan, 1961; Shah and McGarry,
1971). Instead, nonlinear fracture mechanics (NLFM) has been pursued in concrete fracture
analysis. In general, two NLFM approaches are commonly used in the investigation of concrete
fracture (Shah et al., 1995). The first is called the “equivalent elastic crack” methodology, and
assumes that the energy release rate during cracking is equal to the critical energy release rate
consumed during the creation of the two crack surfaces. It effectively represents a modification
of LEFM inasmuch as it accounts for the FPZ and the specimen size effect by adjusting the stress
intensity factor, KIC. The two parameter fracture model (TPFM) by Jenq and Shah (1985) and the
size effect method (SEM) by Bažant and Kazemi (1990) are examples of this approach.
3
The second approach is the fictitious crack model (FCM), proposed by Hillerborg et al. (1976)
and representing an application to concrete of the work by Dugdale (1960) and by Barenblatt
(1962). The latter had assumed that the energy required to overcome the natural cohesive traction
in the material, so as to create a traction-free crack surface, is very large compared to the “energy
rate consumed during material fracturing in creating two surfaces” (Shah et al., 1995). Material
behavior prior to crack initiation is assumed to be linear elastic. Upon attaining the material
tensile strength ( f t ' ), traction stress in the FPZ gradually decreases to zero at a failure crack
width ( w f ), as depicted in Fig. 1.1.
One of the main advantages of the FCM is its suitability for numerical implementation. Different
researchers have implemented this approach in FE programs by converting traction stresses
along an assumed fracture path to equivalent nodal forces (e.g., Hillerborg et al., 1976;
Petersson, 1981; Gustafsson, 1985; Liu, 1994; Ioannides and Sengupta, 2003; Ioannides et al.,
2006). Others have created interface elements in accordance with the FCM approach in order to
study delamination and progressive damage in composite materials (Reedy et al., 1997; Chen et
al., 1999; Feih, 2005). Camanho and Dávila (2002) also created a user element (UEL) for
commercial FE package ABAQUS® in simulating mixed mode decohesion in composite
materials. This element was subsequently incorporated in ABAQUS® Version 6.5 “for modeling
deformation and damage in finite-thickness adhesive layers between bonded parts.” Similarly,
investigators at the University of Illinois, created UELs for ABAQUS® in order to simulate crack
propagation in concrete beam specimens (Roesler et al., 2007), in asphalt specimens (Song,
2006; Song et al., 2006), and in functionally graded materials (Evangelista et al., 2009; Park et
al., 2010). Responses obtained using these UELs are compared in the literature with
4
experimental measurements and satisfactory agreement is reported (Roesler et al., 2007).
Gaedicke and Roesler (2009) recently used cohesive elements that have been incorporated in
ABAQUS® in simulating crack propagation in simply supported beams, beams-on-grade, and
single slabs-on-grade subjected to wheel loads, and reported very encouraging results. Additional
progress is to be expected following the recent formulation of a more general mixed mode
cohesive zone idealization using potential energy fields (Park et al., 2009; Paulino et al., 2010).
A major challenge during numerical idealization of the FPZ is achieving a convergent solution,
since concrete fracture often involves stiffness degradation and snap-back behavior (Crisfield,
1991). The solution is sensitive to the type of softening curve adopted in describing the crack
opening progression once the material strength is attained, as well as the mesh size, penalty
stiffness and type of solver used (Chaboche et al., 2001; Camanho and Dávila, 2002; Gao and
Bower, 2004). The use of linear softening, moderately low penalty stiffness, and arc-length
solvers, along with a very fine mesh may be expected to contribute toward achieving convergent
solutions. Refining the mesh in large structural systems, such as concrete pavements, will, of
course, be computationally demanding. To overcome such numerical challenges, several
researchers have resorted to so-called viscous regularization, according to which a viscosity term
is introduced in the constitutive equations of the degrading material so as to control the rate of
viscous energy dissipation. In this manner, numerical instability is avoided, and the solution is
regularized, i.e., becomes convergent, even if somewhat less accurate (Chaboche et al., 2001;
Gao and Bower, 2004; Maimi et al., 2007; Lapczyk and Hurtado, 2007; Hamitouche et al.,
2008). ABAQUS® has implemented this technique in preventing convergence problems
associated with cohesive elements (ABAQUS, 2009).
5
Research in this thesis is intended as a contribution in this evolving field of study that aspires to
use FCM cohesive fracture simulation with viscous regularization to idealize cracking in single
and jointed concrete pavement slabs, subjected to wheel load and thermal curling.
1.3 Research Objective and Significance
The main objective of this dissertation is to implement the built-in cohesive elements that have
recently been added to the ABAQUS® element library in order to study crack propagation in
concrete pavement slabs subjected to temperature and wheel load. The study seeks to quantify
the effect of several variables involved in the fracture process, such as type of concrete softening
curve, analysis method, along with cohesive zone width and mesh size. Furthermore, the effects
of both linear and nonlinear temperature distributions, with and without a wheel load, on the
post-crack response of concrete pavement slabs are investigated.
The post-crack response of jointed concrete pavements with aggregate interlock and dowel bars
is also investigated. Both linear and nonlinear aggregate interlock formulations available in the
literature are reviewed and implemented in the FE simulation described. The effects of
parameters, such as initial joint width and aggregate size on the post-crack response of both the
loaded and unloaded slabs are examined. The influence of thermal curling on the post-crack
response and on joint efficiency is also investigated.
The study contributes to the ongoing development of rational failure criteria suited as substitutes
to the statistical algorithms derived from Miner’s hypothesis that are being used in the current
pavement design guides. By identifying important parameters pertinent to material
characterization and numerical idealization of plain concrete pavements, this research provides
6
guidelines for a fracture mechanics determination of distresses in pavement systems. Thus, not
only is the state-of-the art in pavement analysis and design enhanced, but also concrete fracture
in general is elucidated.
1.4 Research Scope
This study is limited to mode I fracture, which is examined using a discrete crack approach.
Accordingly, a crack plane is specified to develop along the anticipated principal stress axis,
where FCM cohesive elements are inserted. This stipulation is reasonable for concrete pavement
slabs loaded either at the edge or at their interior, and in both these cases the direction of the
maximum stress is readily apparent. Furthermore, the effect of shear stress can be reasonably
neglected in such structural systems. The intact or uncracked region of the pavement slab is
assumed to behave elastically. This study addresses the effect of some of the most important
features of a concrete pavement system subjected to static wheel load alone or in combination
with temperature curling, notably loading position, load transfer mechanism, and joint
characteristics.
1.5 Thesis Organization
The research described in this thesis has been conducted using multi-purpose commercial FE
program ABAQUS/STANDARD® version 6.7/9. The Ohio Supercomputer Center's IBM Cluster
1350, which “includes AMD Opteron multi-core technologies”, has been employed in executing
input files that were generated for the purposes of the study.
A systematic approach is followed throughout the study, by dividing it into five phases, each of
which is presented as an independent journal paper. Phase 1 is devoted to fracture simulation in

7
simply supported beams, whose geometric and material properties are selected in conformity
with earlier studies conducted by independent researchers. The simulation is used to reproduce
numerical and experimental studies available in literature, thereby adding to the credibility of the
idealization.
In the second phase, the same investigation is extended to individual pavement slabs-on-grade.
The geometry and material properties of the slabs is again selected per previously reported
investigations conducted by other researchers. Multiple FE runs are carried out to assess the
effect of the width of the cohesive zone as well as of the mesh fineness used in this region, the
character of the loaded area, the analysis method, the concrete softening curve adopted, the slab
size and the fracture parameters. In concrete pavement slabs subjected to wheel loads alone, the
critical load position causing maximum bending stress is at the edge. In a concrete pavement slab
under both thermal and wheel loads, on the other hand, an interior position may sometimes be
critical, depending on the geometry of the slab. Consequently, to simulate crack propagation in a
slab loaded at the interior, cohesive elements are inserted along two assumed orthogonal fracture
planes, emanating from the slab center and extending parallel to its edges. The FE simulation of
the fracture process in phase 2 is also validated by comparisons with numerical and experimental
studies reported by other researchers.
In the third phase, fracture of an individual pavement slab-on-grade subjected to temperature
curling and wheel load is studied. Linear elastic analysis of the slab under temperature curling
alone is first conducted, in order to validate the FE idealization formulated for this purpose.
Subsequently, the effect of linear, quadratic, and cubic temperature distributions through the slab
thickness is investigated. Both day-and nighttime conditions are examined, focusing on the post-
8
crack response of the slab under curling alone or when subjected to combined curling-plus-wheel
load.
In phase 4, fracture behavior of concrete pavements equipped with aggregate interlock joints is
investigated. Linear and nonlinear aggregate interlock formulations available in literature are
examined. The effect of aggregate interlock joint characteristics on the post-crack response of the
concrete pavement slab under wheel load and curling is studied.
Post-crack responses of dowelled concrete pavements are investigated in phase 5. The effect of
joint-related parameters, such as dowel looseness and joint width on the post-crack behavior of
the slabs is considered.
1.6 References
AASHTO (2008). Mechanistic-Empirical Pavement Design Guide - A Manual of Practice,
Interim Ed. American Association of State Highway and Transportation Officials, Washington,
DC.
AASHTO (1993). AASHTO Guide for Design of Pavement Structures. American Association of
State Highway and Transportation Officials, Washington, DC.
ABAQUS (2009). ABAQUS Analysis User's Manual - Version 6.9-2. Dassault Systémes,
Providence, RI.
Barenblatt, G. I. (1962). "The mathematical theory of equilibrium cracks in brittle fracture." Adv.
Appl. Mech., 7(C), Elsevier, Amsterdam, The Netherlands, 55-129.
9
Bažant, Z. P., and Kazemi, M. T. (1990). "Determination of fracture energy, process zone length
and brittleness number from size effect, with application to rock and concrete." Int. J. Fracture,
44(2), Kluwer Academic Publishers, Dordrecht, The Netherlands, 111-131.
Camanho, P. P., and Dávila, C. G. (2002). "Mixed-mode decohesion finite elements for the
simulation of delamination in composite materials." Rep. No. NASA/TM-2002-211737, National
Aeronautics and Space Administration, Hampton, VA.
Chaboche, J. L., Feyel, F., and Monerie, Y. (2001). "Interface debonding models: a viscous
regularization with a limited rate dependency." Int. J. Solids Struct., 38(18), Elsevier,
Amsterdam, The Netherlands, 3127-3160.
Chen, J., Crisfield, M., Kinloch, A. J., Busso, E. P., Matthews, F. L., and Qiu, Y. (1999).
"Predicting progressive delamination of composite material specimens via interface elements."
Mech. Compos. Mater. Str., 6(4), Taylor and Francis Group, London, UK, 301-317.
Crisfield, M. A. (1991). Non-linear Finite Element Analysis of Solids and Structures, Vol. 1.
John Wiley and Sons, Inc., West Sussex, UK.
Davids, W. G., Turkiyyah, G. M., and Mahoney, J. P. (1998). "EverFE: Rigid pavement three-
dimensional finite element analysis tool." Transp. Res. Rec. 1629, Transportation Research
Board, National Research Council, Washington, DC, 41-49.
Dugdale, D. S. (1960). "Yielding of steel sheets containing slits." J. Mech. Phys. Solids, 8(2),
Pergamon Press Ltd., London, UK, 100-104.
10
Evangelista, F., Roesler, J., and Paulino, G. (2009). "Numerical simulations of fracture resistance
of functionally graded concrete materials." Transp. Res. Rec. 2113, Transportation Research
Feih, S. (2005). "Development of a user element in ABAQUS for modeling of cohesive laws in
composite structures." Rep. No. Risø-R-1501(EN), Risø National Laboratory, Roskilde,
Denmark.
Gaedicke, C., and Roesler, J. R. (2009). "Fracture-based method to determine the flexural load
capacity of concrete slabs." FAA COE Rep. No. 31, Department of Civil and Environmental
Engineering, University of Illinois at Urbana-Champaign, Urbana, IL.
Gao, Y. F., and Bower, A. F. (2004). "A simple technique for avoiding convergence problems in
finite element simulations of crack nucleation and growth on cohesive interfaces." Modelling
Simul. Mater. Sci. Eng., 12(3), Institute of Physics Publishing, London, UK, 453-463.
Gustafsson, P. J. (1985). "Fracture mechanics studies of non-yielding materials like concrete:
modelling of tensile fracture and applied strength analyses.” Rep. No. LUTVDG/(TVBM-1007)/1-
422/(1985), Division of Building Materials, Lund Institute of Technology, Lund, Sweden.
Hamitouche, L., Tarfaoui, M., and Vautrin, A. (2008). "An interface debonding law subject to
viscous regularization for avoiding instability: application to the delamination problems." Eng.
Fract. Mech., 75(10), Elsevier, Amsterdam, The Netherlands, 3084-3100.
11
Hillerborg, A., Modéer, M., and Petersson, P. E. (1976). "Analysis of crack formation and crack
growth in concrete by means of fracture mechanics and finite elements." Cement Concrete Res.,
6(6), Elsevier, Amsterdam, The Netherlands, 773-782.
Huang, Y. H., and Wang, S. T. (1973). "Finite element analysis of concrete slabs and its
implication for rigid pavement design." Hwy. Res. Rec. 466, National Research Council,
Washington, DC, 55-69.
Ioannides, A. M. (1997). "Fracture mechanics in pavement engineering: the specimen size
effect." Transp. Res. Rec. 1568, Transportation Research Board, National Research Council,
Ioannides, A. M., and Korovesis, G. T. (1990). "Aggregate interlock: a pure-shear load transfer
mechanism." Transp. Res. Rec. 1286, Transportation Research Board, National Research
Council, Washington, DC, 14-24.
Ioannides, A. M., Peng, J., and Swindler Jr, J. R. (2006). "ABAQUS model for PCC slab
cracking." Int. J. Pavement Eng., 7(4), Taylor and Francis Group, London, UK, 311-321.
Ioannides, A. M., and Sengupta, S. (2003). "Crack propagation in Portland cement concrete
beams: implications for pavement design." Transp. Res. Rec. 1853, Transportation Research
Ioannides, A. M., Thompson, M. R., and Barenberg, E. J. (1985). "Westergaard solutions
reconsidered." Transp. Res. Rec. 1043, Transportation Research Board, National Research
12
Jenq, Y., and Shah, S. P. (1985). "Two parameter fracture model for concrete." J. Eng. Mech.,
111(10), ASCE, New York, NY, 1227-1241.
Kaplan, M. E. (1961). "Crack propagation and the fracture of concrete." J. Am. Concrete I.,
58(5), ACI, Farmington Hills, MI, 591-610.
Khazanovich, L. (1994). "Structural analysis of multi-layered concrete pavement systems," Ph.D.
dissertation, University of Illinois at Urbana-Champaign, Urbana, IL.
Khazanovich, L., Yu, H. T., Rao, S., Galasova, K., Shats, E., and Jones, R. (2000). ISLAB2000-
Finite Element Analysis Program for Rigid and Composite Pavements- User’s Guide. ERES
Consultants, Urbana, IL.
Lapczyk, I., and Hurtado, J. A. (2007). "Progressive damage modeling in fiber-reinforced
materials." Compos. Part A, 38(11), Elsevier, Amsterdam, The Netherlands, 2333-2341.
Liu, P. (1994). "Time dependent fracture of concrete," Ph.D. dissertation, Ohio State University,
Columbus, OH.
Maimi, P., Camanho, P. P., Mayugo, J. A., and Dávila, C. G. (2007). "A continuum damage
model for composite laminates: Part II – Computational implementation and validation." Mech.
Mater., 39(10), Elsevier, Amsterdam, The Netherlands, 909-919.
Miner, M. A. (1945). "Cumulative damage in fatigue." J. Appl. Mech.-T. ASME, 12(3), ASME,
New York, NY, 159-164.
13
Nishizawa, T., Koyanagawa, M., Takeuchi, Y., and Kimura, M. (2001). "Study on mechanical
behavior of dowel bar in transverse joint of concrete pavement." Proc., Seventh International
Conference on Concrete Pavements, International Society for Concrete Pavements, Orlando, FL,
571-584.
Park, K., Paulino, G. H., and Roesler, J. R. (2009). "A unified potential-based cohesive model of
mixed-mode fracture." J. Mech. Phys. Solids, 57(6), Elsevier, Amsterdam, The Netherlands, 891-
908.
Park, K., Paulino, G. H., and Roesler, J. (2010). "Cohesive fracture model for functionally
graded fiber reinforced concrete." Cem. Concr. Res., 40(6), Elsevier, Amsterdam, The
Netherlands, 956-965.
Paulino, G. H., Park, K., Celes, W., and Espinha, R. (2010). "Adaptive dynamic cohesive
fracture simulation using nodal perturbation and edge‐swap operators." Int. J. Numer. Methods
Eng., 84(11), John Wiley and Sons, Ltd., Hoboken, NJ, 1303-1343.
Petersson, P. E. (1981). "Crack growth and development of fracture zones in plain concrete and
similar materials." Rep. No. LUTVDG/(TVBM-1006)/1-174/(1981), Division of Building
Materials, Lund Institute of Technology, Lund, Sweden.
Reedy Jr., E. D., Mello, F. J., and Guess, T. R. (1997). "Modeling the initiation and growth of
delaminations in composite structures." J. Composite Mater., 31(8), Technomic Publishing Co.,
Lancaster, PA, 812-831.
14
Roesler, J., Paulino, G. H., Park, K., and Gaedicke, C. (2007). "Concrete fracture prediction
using bilinear softening." Cement Concrete Comp., 9(4), Elsevier, Amsterdam, The Netherlands,
300-312.
Shah, S. P., and McGarry, F. J. (1971). "Griffith fracture criterion and concrete." J. Eng. Mech.,
97(EM6), ASCE, New York, NY, 1663-1676.
Shah, S. P., Swartz, S. E., and Ouyang, C. (1995). Fracture Mechanics of Concrete: Applications
of Fracture Mechanics to Concrete, Rock and Other Quasi-brittle Materials. John Wiley and
Sons, Inc., New York, NY.
Song, S. H. (2006). "Fracture of asphalt concrete: a cohesive zone modeling approach
considering viscoelastic effects," Ph.D. dissertation, University of Illinois at Urbana-Champaign,
Urbana, IL.
Song, S. H., Paulino, G. H., and Buttlar, W. G. (2006). "Simulation of crack propagation in
asphalt concrete using an intrinsic cohesive zone model." J. Eng. Mech., 132(11), ASCE, Reston,
VA, 1215-1223.
Tabatabaie, A. M., and Barenberg, E. J. (1978). "Finite element analysis of jointed or cracked
concrete pavements." Transp. Res. Rec. 671, Transportation Research Board, National Research
Tayabji, S. D., and Colley, B. E. (1984). "Analysis of jointed concrete pavements: Interim
Report." Rep. No. FHWA/RD-86/041, Federal Highway Administration, Washington, DC.
15
Tia, M., Armaghani, J. M., Wu, C. L., Lei, S., and Toye, K. L. (1987). "FEACONS III computer
program for an analysis of jointed concrete pavements." Transp. Res. Rec. 1136, Transportation
Research Board, National Research Council, Washington, DC, 12-22.
Westergaard, H. M. (1927). "Analysis of stresses in concrete pavements due to variations of
temperature." Proc. Hwy. Res. Board 6, Highway Research Board, National Research Council,
Westergaard, H. M. (1926). "Stresses in concrete pavements computed by theoretical analysis."
Public Roads, 7(2), Federal Highway Administration, Washington, DC, 25-35.
16
External applied load
Traction (t)
ft '
wf Crack Opening (w)
Traction free zone Fracture process zone Intact zone
Figure 1.1 Fracture Process Zone in Concrete
17
18
Chapter 2 Simulation of Crack Propagation in Concrete Beams1
Abstract
This paper discusses the simulation of crack propagation in concrete beam specimens using
commercial finite element package ABAQUS® 6.7-1. Special-purpose cohesive elements are used
to simulate the fracture process in accordance with the fictitious crack model. Two- as well as
three-dimensional finite element discretizations are carried out. Parameters influencing the
responses, such as mesh fineness, cohesive zone width, type of softening curve and analysis
technique are studied. The responses are then compared with previous experimental and
numerical investigations conducted by various independent researchers. It is demonstrated that
cohesive elements can be used in idealizing crack propagation as required in pavement
engineering.
1
A shorter version of this chapter has appeared in Transp. Res. Rec. 2154, Transportation
Research Board, National Research Council, Washington, DC, 2010, pp. 12-21.
19
2.1 Introduction
The development of a mechanistic-empirical approach for the analysis and design of pavement
systems has received increased attention recently, reigniting the debate over the use of
statistical/empirical transfer functions, whose experimental verification is questionable, at best
(Ioannides, 1997a; Khazanovich and Tomkins, 2005). Following an exhaustive examination of
various fracture mechanics options suggested as potential replacements to Miner’s hypothesis
(Miner, 1945), Hillerborg’s fictitious crack model (FCM) (Hillerborg et al., 1976) was found to
be the most promising for studying crack propagation in Portland cement concrete (PCC)
pavements, and a step-by-step approach was outlined for its implementation (Ioannides, 1997a
and 1997b). Accordingly, Ioannides and Sengupta (2003) formulated a two-dimensional (2D)
numerical procedure to simulate crack propagation on the basis of the FCM for a simply
supported beam. The response of the beam over the elastic range was analyzed using commercial
finite element (FE) software GTSTRUDL (GTSTRUDL, 1993), while its fracture behavior was
studied using a specially coded FORTRAN program, called CRACKIT. To facilitate the
generalized application of the concepts implicit in the GTSTRUDL/CRACKIT combination,
Ioannides et al., (2005) subsequently implemented their approach by using the general purpose
FE package, ABAQUS® (ABAQUS, 2003). They reported at the time that the applicability of the
built-in fracture analysis capabilities of ABAQUS® was too limited for pavement engineering,
especially since the FCM was not used. Consequently, in their 2D study of simply supported
beams, the investigators employed a nonlinear spring element from the ABAQUS® library,
JOINTC, to idealize the fracture zone. An alternative approach was developed by Song (2006),
who coded a user-defined subroutine in ABAQUS®, thereby creating a user element (UEL); an
application of this UEL is described in the 2D study of crack propagation in simply supported
20
asphalt concrete beams by Song et al. (2006). A series of such 2D 4-noded UEL elements were
inserted at the center of the beam to simulate fracture.
A similar approach to that by Song et al. (2006) was followed by Roesler et al. (2007a) who
simulated crack propagation in simply supported concrete beam specimens. A UEL for
ABAQUS® created by Park (2005) was used to discretize the cohesive zone in accordance with
the FCM. The parameters that define the FCM bilinear softening curve were directly determined
from the three-point bending test (ASTM D790, 2010). The total fracture energy, GF, was
determined from the area of load versus crack mouth opening displacement (P-CMOD) curve
and the initial fracture energy, Gf, was computed per either the two parameter fracture model
(TPFM) (Jenq and Shah, 1985) or the size effect method (SEM) (Bažant and Kazemi, 1990). The
location of the kink point was assumed to be at 25% of the tensile strength of the specimen.
The same simulation concept found in Roesler et al. (2007a) was extended to idealize the
fracture process in simply supported beams with functionally graded concrete materials (Roesler
et al., 2007b). A specimen made partially of plain concrete and partially of fiber reinforced
concrete was tested so as to obtain fracture properties. The numerical simulation employed a
trilinear softening curve that was constructed by accounting for the bridging effect of the fiber
reinforced concrete on the fracture process zone on the bilinear curve used by Roesler et al.
(2007a). The parameters that define the trilinear curve were obtained from experiments
conducted during the study (ASTM D790, 2010). A UEL created for ABAQUS® implementing
the trilinear curve was used to idealize the cohesive zone.
21
The corresponding case involving a three-dimensional (3D) PCC pavement slab-on-grade was
first considered by Ioannides and Peng (2004), using once again JOINTC elements from the
ABAQUS® library. A more versatile 3D UEL implementing a cohesive zone idealization was
subsequently formulated by Song (2006), and was applied to a cylindrical asphalt concrete
specimen.
Such efforts received a boost with the release in early 2005 of ABAQUS® version 6.5, which for
the first time included “a family of cohesive elements for modeling deformation and damage in
finite-thickness adhesive layers between bonded parts. Cohesive elements are typically
connected to underlying elements with surface-based TIE CONSTRAINTS, so the mesh used for
the cohesive layer can be independent of the mesh used for the bonded components” (ABAQUS,
2004). Gaedicke and Roesler (2009) reported using these cohesive elements in their study of
crack propagation in concrete beams and slabs.
The investigation presented in this paper is a continuation of the step-by-step development and
application of fracture mechanics tools using the FCM in pavement engineering, initiated at the
University of Cincinnati in the late 1990s. The main objective is to implement the built-in
cohesive elements that have recently been added to the ABAQUS® library, and to compare the
performance of these elements to that reported in earlier investigations. It is hoped that in this
manner, a contribution will be made to the ongoing effort for more mechanistic-empirical
pavement design procedures that will utilize fracture mechanics concepts, thereby replacing the
purely statistical/empirical transfer functions and Miner’s hypothesis, which are currently in use.
Validating a FE simulation of crack propagation in a simply supported beam is considered to be
22
a necessary precursor to a more comprehensive analysis of slabs-on-grade, required for in situ
pavement systems.
2.2 Methodology
The present study focuses on the post-cracking response of simply supported PCC beams, using
commercial general purpose FE program ABAQUS®/STANDARD, version 6.7-1. The geometric
and material properties of the beams considered in this paper are shown in Table 2.1. To begin
with, however, a linear elastic analysis using 2D and 3D elements is described, and the results
are compared with available closed-form solutions. This initial step is considered essential in
ensuring the robustness of the proposed FE idealization. Upon the successful conclusion of the
linear elastic analysis, the simulation of crack propagation can be carried out, by implementing
the built-in cohesive elements of ABAQUS®, in accordance with the FCM for fracture. In all
analyses, elements CPS4 and C3D27 are used for 2D and 3D discretizations of the intact
material, respectively, while the cohesive fracture zone is simulated with COH2D4 (2D) and
COH3D8 (3D) cohesive elements. Program runs reported herein capture the effects of numerical
analysis technique, mesh fineness, cohesive zone width, and type of softening curve. The
resulting simulation is finally used to reproduce numerical and experimental studies conducted
by other independent researchers, thereby adding to its credibility.
2.3 Analysis of Linear Elastic Response
This section reports numerical testing to ascertain the robustness of the proposed FE idealization
through simulation of the linear elastic response of the simply supported beam designated as
Beam A in Table 2.1. This beam was originally used by Sengupta (1998) and was later adopted
23
by Ioannides et al. (2005), whose results are, therefore, available for comparison purposes, along
with published closed-form solutions. Both concentrated and uniformly distributed loads are
considered.
Beam A was assumed to be simply supported on rollers, requiring degree of freedom 2 (vertical
displacement) to be fixed at the two support nodes. Because of symmetry, half of the beam was
meshed uniformly with 64×48 CPS4, and 5×40×30 C3D27 elements, for the 2D and 3D
discretizations, respectively. All elements used were nearly square, which thereby eliminated
any significant aspect ratio effects. Once the mesh for the right-hand-side half of the beam was
defined, a mirror image was created along the plane of symmetry, and surface-based TIE
CONSTRAINTS were used to connect surfaces on either side of the symmetry plane.
The simulations first considered a concentrated load of 33.7 kips, applied at the midspan. For the
2D simulation, the load was applied on the top node at the midspan. To avoid localized effects
in the 3D idealization, a small loaded area of size 0.4 by 1.5 in. (two elements along the
symmetry line) was defined at the center of the beam, on which a pressure of 56.17 ksi was
applied. The same beam was also subjected to a uniformly distributed load of magnitude 10
kips/in. This load was applied as a pressure of magnitude 6.667 ksi (10 kips/in. divided by1.5 in.)
over the top surface elements.
Simulation results were compared with closed-form solutions. For a beam subjected to a
concentrated load (P), the maximum midspan deflection (vmax) is given by Timoshenko and
Goodier (1970) as:
24
PL3 PL  3 3 3  P
vmax        0.21 (2.1)
48EI 4cb  4G 10 E 4 E  Eb
where c is half of the beam thickness, h; b, L, I are beam’s width, length and moment of inertia;
 is the Poisson’s ratio, E is the Young’s modulus, and G is the shear modulus of the material,
which can be computed from:
E
G (2.2)
2(1  μ)
The results are shown in Table 2.2, where it is observed that the FE discretizations show a
discrepancy from the theoretical solution of 2.3% and 0.3%, for 2D and 3D, respectively. For the
beam subjected to uniformly distributed load ( qo ), results were compared with values computed
according to the following theories, as summarized by Shames and Dym (1985):
Euler-Bernoulli’s Theory:
qo L4  x  4 3  x  2 5 
v  x         (2.3)
24EI  L  2  L  16 
Theory of Elasticity:
qo L4  x 4 3  x 2 5  h 2  12 3  1 x  2 
v  x             μ     (2.4)
24EI  L  2  L  16  L   5 2  4 L  
Timoshenko Beam Theory:
25
qo L4   x 4 3  x 2 5   h 2 2(1  μ)  x  2 1  
v  x                 
24EI   L  2  L  16   L  k  L  4   (2.5)
10(1  μ)
k
(12  11μ )
Note that x and v in the above equations represent any point of interest along the span of the
beam and the corresponding deflections, respectively. It was found that the FE idealization
exhibited near perfect agreement with the theory of elasticity accounting for Poisson’s ratio
effect. Moreover, the FE solutions provided excellent approximations (98.86% and 98.66% for
2D and 3D, respectively) to the theoretical solution for a beam with shear deformation
(Timoshenko theory), as shown in Fig. 2.1. The Euler beam theory, on the other hand, gave
lower results (77%) as compared to other theories as well as the FE results. It is, therefore,
concluded that the FE discretization implemented in this study is robust as far as the linear elastic
aspects of the analysis are concerned, and that it may be considered to be a good candidate for
investigating the more demanding fracture mechanics issue of crack propagation in simply
supported beams.
2.4 Simulation of Crack Propagation Using Built-in Cohesive Elements
This section outlines the FE formulation incorporating cohesive elements recently introduced in
ABAQUS®, and their implementation in accordance with the FCM for investigating the post-
fracture response of simply supported concrete beams. The sensitivity of the FE solution to a
variety of aspects of the numerical procedure employed is also examined.
26
2.4.1 Idealization of Cohesive Zone
Among the three classes of cohesive elements that are available in ABAQUS®, those
implementing the so-called “traction-separation” formulation are the most suitable for use in
crack propagation studies that use the FCM. Accordingly, the load-displacement responses can
be subdivided into three stages: pre-crack; initiation of crack; and post-peak (or softening)
behavior.
a. Pre-Crack Behavior
During the pre-crack stage, the material is considered to experience a very small but finite
separation, and the cohesive element response is governed by the following elastic strain-
displacement relation (ABAQUS, 2007):
ε n   wn 
  1  
ε s    ws  (2.6)
 ε  T0 w 
 t  t
where:  and w are the nominal strain and the elastic separation vectors, respectively, in the
normal (n) and two shear directions (s and t); and T0 is initial width of the cohesive zone. The
elastic traction stress components can then be computed from Eq. (2.7):
t n   K nn K ns K nt   n 
    
t s    K ss K st   s  (2.7)

t   symm K tt   
t   t 
where K is a nominal stiffness (also referred to as penalty stiffness), illustrated in Fig. 2.2, and t
27
is the nominal stress, in the normal and two shear directions, respectively. If the shear and
normal components are uncoupled, Eq. (2.7) will reduce to:
t n   K nn 0 0   n 
    
t s    0 K ss 0   s  (2.8)

t   0 K tt   
t  0  t 
For a 2D idealization, only the first two rows and columns of Eq. (2.7) are used. Selection of the
initial width of the cohesive zone and of the penalty stiffness, K, relies largely on prior
experience with the software, yet it can influence the solution convergence significantly.
ABAQUS (2007) recommends computing the penalty stiffness from: Knn = E/T0, where E is the
Young’s modulus of the intact (or uncracked) material. Similarly, it may be assumed that Kss =
Ktt = G/T0, where G is the corresponding shear modulus of the intact material.
b. Initiation of Crack and Post-Peak Behavior
Crack initiation refers to the beginning of the degradation of the material. In PCC crack
propagation studies, it is often assumed that the crack initiates when the stress reaches the tensile
strength, f t ' , of the material (Liu, 1994; Ioannides and Sengupta, 2003; Roesler et al., 2007a).
Once the crack initiates, material damage evolves according to a predefined softening law. Park
et al. (2008) provided a comprehensive list of softening options proposed for concrete. At first,
Hillerborg et al. (1976) used a linear softening curve. Nowadays, it is more common to employ a
bilinear curve characterized by three points, as shown in Fig. 2.2: (a) the crack initiation point,
defined by traction stress f t ' and separation wcr ; (b) the kink point, at which the separation-
28
traction stress pair is ( wk , ψ f t ' ); and (c) the critical (or maximum) separation, w f , for which the
traction stress is zero.
The crack evolution for any kind of softening is facilitated by using a dimensionless damage
variable defined as (ABAQUS, 2007; Gaedicke and Roesler, 2009):
t's
D  1 for wcr  w  w f (2.9)
t'i
where: t´s is the traction stress for separation w, along the softening curve; tí is the traction
stress that would have corresponded to w had the pre-crack stiffness endured, as explained in
Fig. 2.2.
The traction stress for a given separation before the kink point can easily be derived as:
  w - wcr  
t's  ft ' 1   -1    for wcr  w  wk (2.10)
  wk - wcr  
Similarly, for the second softening segment, t´s can be obtained as:
 wf  w 
t's   ft '   for wk  w  w f (2.11)
 w f  wk
 
The elastic traction stress that would have been obtained had the material not become damaged,
tí, can easily be obtained from:
tí=Knnw (2.12)
Substituting Eqs. (2.12) and (2.10) into Eq. (2.9) and simplifying, one can obtain:
29
D
 w  wcr  wk  wcr  for w  w  w
(2.13)
w  wk  wcr 
cr k
A similar approach can be followed to obtain an expression for D pertaining to the remaining
segment wk  w  w f . ABAQUS® requires D to be specified as a function of the effective
separation, we , defined as the separation with respect to the crack initiation point, wcr .
The above equations are useful for a bilinear softening curve, whose kink point location is
known. Instead of assuming the location of the kink point , Roesler et al. (2007a) used the initial
fracture energy, Gf, a parameter defined as the area under the first two limbs of softening curve
shown in Fig. 2.2. This is used to determine the separation at which the second limb of the
softening curve intersects the separation line, w1 . From Fig. 2.2, therefore, w1 can be found as:
2G f
w1  (2.14)
ft'
The traction t´s, is now easily determined from:
 w w 
ts'  f t '  1  for wcr  w  wk (2.15)
 w1  wcr 
The equation for the traction stresses for separation values between wk and w f remains the same
as Eq. (2.11). The kink point separation, wk , can be calculated using Eq. (2.14) for t´s= ψ f t ' as
follows:
wk  w1   w1  wcr  (2.16)
30
The separation at the zero traction stress, w f , depends on the difference between the total and
initial fracture energy, and is given as (Roesler et al., 2007a):
2
wf  G  1   G f  (2.17)
 ft'  F
ABAQUS® incorporates only linear and exponential softening curves. The bilinear or any other
kind of curve, however, may be specified by the user in tabular form, as illustrated below.
c. Numerical Example
Typical calculation procedures that are used to generate the necessary input values of material
properties for cohesive elements in accordance with a bilinear curve are demonstrated below for
Beam A and Beam C.
i. Beam A
The material properties for this beam are shown in Table 2.1 as: E = 4000 ksi; f t ' = 0.46327 ksi.
The bilinear curve parameters are (Ioannides et al., 2006): ψ f t ' = 0.15442 ksi (thus ψ = 1/3), wk =
0.000745 in., w f = 0.0033525 in. The initial width of the cohesive zone is assumed to be T0 =
0.001 in.
The initial stiffness is: Knn = E/T0 = 4×106 ksi/in. The critical separation, wcr , can be computed
from Eq. (2.12) as wcr = f t ' /Knn = 1.158×10-7 in. Considering a point on the bilinear softening
curve at w = 0.0005 in., for example, Eq. (2.10) yields:
31
  1   0.0005  1.158(10) 7  
t's  0.46327 1    1  7  
 0.256006 ksi (2.18a)
  3   0.000745  1.158(10)  
The elastic traction stress that would have been obtained had there been no damage is: tí = Knn w
= 4×106×0.0005=2000 ksi. The damage variable D is now computed by using Eq. (2.9) as:
0.256006
D  1  0.9998719969 (2.18b)
2000
The effective displacement, we, which will be defined along with D, is: we = w-wcr = 0.0005-
1.158×10-7 = 4.998842 ×10-4 in. Next, the damage variable and the effective displacement are
specified in the input file as (D, we). At w = 0.0005 in., it is (0.9998719969, 0.0004998842). A
similar approach can be used for Beam B, which has the same softening curve and material
properties, except for E, as that of Beam A.
ii. Beam C
The material properties of this beam are also shown in Table 2.1 as: E = 32000 MPa; f t ' = 4.15
MPa; Gf = 56.57 N/m; GF = 167 N/m. T0 is assumed to be 1 mm; the stress ratio at the kink point,
ψ, is taken to be 0.25 (Roesler et al., 2007a). Using Eq. (2.14) and (2.17) one can obtain:
2  56.57(10) 3
w1   0.02726265 mm
4.15 (2.18c)
2
wf  167  (1  0.25)  56.57  103  0.24014 mm
0.25  4.15
The penality stiffness is: Knn = E/T0 = 32000 MPa/mm. The critical separation is: wcr = f t ' /Knn =
1.296875×10-4 mm. The kink point separation, wk can be obtained from Eq. (2.16) as: wk =
0.02726265-0.25(0.02726265-1.296875×10-4) = 0.0204794 mm. Considering a separation, w, of

32
0.01 mm (which is less than wk), for example, Eq.(2.15) gives:
 0.02726265  0.01 
t's  4.15    2.64033 MPa
 0.0272626500  0.0001296875 
t'i  32000  0.01  320MPa (2.18d)
2.64033
D  1  0.991748966
320
The effective displacement is: we = w-wcr = 0.0098703125 mm. The softening behavior is
provided in the form of a table in the input file; for example, (D, we) is (0.991748966,
0.0098703125) at w = 0.01 mm. It is noted that in this section, SI units have been retained per
Roesler et al. (2007a) as an illustration of such use in this field.
2.4.2 Sensitivity Study for Proposed Discretization
Previous investigators have examined in detail the effect of several variables influencing
numerical solutions analogous to that proposed in the present paper. Thus, Song et al. (2006)
studied the effect of total fracture energy, GF (defined as the area under the bilinear curve shown
in Fig. 2.2), of tensile strength, f t ' , and of cohesive zone width, T0, on the fracture of asphalt
concrete beams. For his part, Park (2005) examined the sensitivity of the solution to the initial
fracture energy, Gf, as well as to the location of the kink point, for PCC specimens. The latter
was also investigated by Gaedicke and Roesler (2009), who employed built-in cohesive
elements.
In this section, the sensitivity of the proposed FE discretization to the analysis technique, mesh
size, notch depth, cohesive zone width and type of softening curve is investigated. The beams are
idealized using 3D elements, C3D27 for the intact material and COH3D8 for the cohesive zone,
respectively.
33
a. Effect of Analysis Technique
For the purposes of this investigation, there are currently two analysis options in ABAQUS®: the
general (or default) procedure, which uses a Newton-Raphson method; and the modified Riks
approach (ABAQUS, 2007). The latter is particularly suited for potentially unstable problems
that occasionally exhibit negative stiffness values, or present convergence difficulties, e.g.,
buckling and snap-back of the load-displacement curve. Since cohesive elements are inserted in
materials experiencing softening resulting from progressive damage, their application may also
be fraught with such numerical problems. To overcome such challenges when using the Newton-
Raphson approach, “viscous regularization” (ABAQUS, 2007) may be applied, but a preferable
alternative in certain cases is to employ the modified Riks procedure. For example, Song et al.
(2005) encountered divergence problems when using the Newton-Raphson method, whereas the
modified Riks approach produced convergence. For their part, Yang and Proverbs (2004)
studied the efficacy of various solution strategies for fracture, and concluded that arc-length
solvers (such as that used in the modified Riks method) would capture the softening behavior in
snap-back type of load-displacement relations.
To study the effect of the two analysis options currently available in ABAQUS®, Beam A and B
in Table 2.1 are considered in an unnotched configuration. The cohesive zone width is set to
0.001 in. The penalty stiffnesses, Knn , Kss , Ktt , are computed as noted earlier. Linear softening
is used for simplicity.
For the Newton-Raphson method (Crisfield, 1991), the loading parameters that need to be
specified are: initial time increment (ITI), time period of the step (TPS), minimum time
increment (MnTI) and maximum time increment (MxTI). These are set to 6×10-3, 1.0, 1×10-9,
34
3×10-2, respectively, on the basis of previous work by Ioannides et al. (2005). The maximum
number of time increments is set to 200.
The modified Riks approach (Riks, 1979) requires four parameters to be defined: the initial
increment in the arc-length along the static equilibrium path, Δlin; the total arc-length scale
factor, lperiod; and the minimum and maximum arc-length increments, lmin, and lmax , respectively.
A convenient way to assign values to these parameters is to retain those specified above: 6×10-3,
1.0, 1×10-9, 3×10-2, respectively. Moreover, the terminal increment is similarly set to 200.
The results obtained in this manner are plotted in Fig. 2.3 and Fig. 2.4 for Beams A and B,
respectively, as pairs of load versus crack mouth opening displacement (P-CMOD) and load
versus load line displacement (P-LLD) curves. It is clear in Fig. 2.3 that for Beam A, the trends
captured by the two methods are significantly different from one another; those according to the
modified Riks approach are considered to be more realistic because they reflect the snap-back
behavior expected in the softening stage. For Beam B, however, both techniques give identical
results when the P-LLD curve is examined, and only a minor difference is observed in the in the
two P- CMOD curves, as shown in Fig. 2.4. Evidently, the difference between the results of the
two techniques is more pronounced in deeper beams (lower span-to-thickness ratio), and in such
cases resorting to the modified Riks procedure may be advantageous when built-in cohesive
elements are employed.
b. Effect of Mesh Fineness
A set of runs involving three types of mesh were considered for Beams A and B in Table 2.1;
these discretizations involved multiple mesh configurations for the intact region of the beams,
35
and a constant mesh for the cohesive zone. Linear softening was used for simplicity. The
coarsest mesh consisted of 8×6×1 and 24×6×2 C3D27 elements in the length, depth and width
directions, whereas the finest mesh had 32×24×4 and 60×15×8 C3D27 elements; the median
mesh had 16×12×2 and 36×9×4 C3D27 subdivisions for Beams A and B, respectively. The
cohesive zone mesh was maintained as 10 times finer than the median mesh in the depth and
width directions, and had one element in the length direction. The results are shown in Fig. 2.5
and Fig. 2.6 for Beams A and B, respectively. It is observed that the effect of mesh fineness is
surprisingly insignificant for the meshes adopted, especially regarding the CMOD response. In
as much as there are differences between the coarsest and the finest meshes for the LLD
response, it is observed that a finer mesh generally results in a slightly lower load before the
peak, and a slightly higher load after the peak, but such differences do not exceed 5%.
c. Effect of Notch Depth
The effect of notch depth, a, on the P-CMOD response was studied by using the 60×15×8
element discretization of Beam B. The cohesive zone mesh used previously was retained here. It
is observed from Fig. 2.7 that as the notch depth increases, the CMOD value increases but the
peak load decreases For the 6.67% notch depth-to-beam thickness ratio, for example, the peak
load is 12.5% lower than for the unnotched beam, whereas the CMOD value is 28.9% higher.
d. Effect of Cohesive Zone Width
To investigate the influence of the cohesive zone width, three levels were considered: T0 = 1.0,
0.01, and 0.001 in. Three-dimensional FE discretization of Beam B with linear softening was
carried out. Figure 2.8 shows that as the cohesive zone width increased, the elastic deformation
36
of the cohesive zone increased, increasingly contributing to the elastic strain energy until damage
began. Once damage started, the responses were not sensitive to the cohesive zone width. An
increase in cohesive zone width decreased the peak load that could be supported by the beam.
The 0.01 and 0.001 in. width, however, yielded almost the same result, and this led to the
decision to use 0.001 in. in all subsequent sections.
e. Effect of Softening Curve
In the cases considered above, linear softening was used for its simplicity. A comparison
between linear and bilinear curves employed in conjunction with Beam B indicated that linear
softening would over-predict the peak load by about 11%, as shown in Fig. 2.9. After the peak,
however, the bilinear curve eventually gave a higher load than the linear one, and there was a
crossover on the curves. The areas under P-CMOD curves for both softening assumptions were
approximately the same, reflecting the equality of the areas under the two softening curves.
On the basis of the preceding investigations into the sensitivity of the proposed fracture
formulation to the various discretization parameters examined, it can be concluded that careful
attention should be given to the selection of the type of solver, of the width of the cohesive zone
and of the type of softening curve in idealizing crack propagation in simply supported beams
when cohesive elements are employed. Equipped with enhanced understanding of the effect of
each parameter involved in cohesive zone FE analysis, one may proceed with the application of
the approach to reproducing numerical and experimental results obtained by other researchers, as
described in the following sections.
37
2.5 Comparison with Previous Numerical Studies
In this section, comparison is made of simulation results obtained from this study with those
from numerical studies conducted by other independent researchers. The purpose is to validate
the proposed FE procedure using corroborating evidence confirming its credibility.
2.5.1 GTSTRUDL/CRACKIT by Sengupta (1998)
In the earliest University of Cincinnati effort to simulate crack propagation in simply supported
concrete beams, Sengupta (1998) developed a combination approach that employed a
commercial software FE package, GTSTRUDL (GTSTRUDL, 1993), for the elastic response and
the associated flexibility matrix, in tandem with a specially coded FORTRAN computer program
named CRACKIT for the ensuing fracture behavior, in accordance with a bilinear softening law
and the FCM. To illustrate his approach, Sengupta (1998) reproduced the response of a beam
that had first been studied by Liu (1994), and discretized it with 4-node plane stress elements of
size 0.2×0.5 in. Liu’s beam is Beam B in Table 2.1, and is considered in the present study with a
3D idealization using C3D27 elements of size 0.2×0.2×0.2 in. Results obtained using the 2D
GTSTRUDL/CRACKIT combination approach by Sengupta (1998) and the 3D COH3D8
procedure employed for the cohesive fracture zone in this study are shown in Fig. 2.10. The P-
LLD curves appear to agree better than the P-CMOD curves. The difference between the P-LLD
curves can be explained by the difference in the intact region elements used (2D 4-node element
versus 3D 27-node element), whereas the P-CMOD discrepancy is mainly due to the implicit
assumption in CRACKIT that the notch remains undeformed until the crack begins to propagate.
38
2.5.2 ABAQUS® - JOINTC by Ioannides et al. (2005)
In order to simulate crack propagation in concrete beams using ABAQUS®, Ioannides et al.
(2005) used CPS4 elements for the intact region and a nonlinear spring element, JOINTC, for the
fracture zone, prescribing the same bilinear FCM curve as Sengupta (1998). Beam A in Table
2.1 was considered and discretized with a coarse FE mesh consisting of elements of size 1×1 in.
The notch depth-to-beam thickness ratio was 1/3. Identical mesh pattern and element type were
used in the present 2D study, in which each intact zone element was a CPS4 and the cohesive
zone was discretized by using COH2D4 instead of JOINTC elements.
The results are shown in Fig. 2.11. The cohesive fracture simulation gave higher load for a given
CMOD as compared to the JOINTC fracture idealization in the post-peak stage. In general,
however, it can be said that the two approaches are in good agreement for the particular mesh
considered. Nonetheless, the wavy curves in Fig. 2.11 suggest some convergence difficulties,
which may easily be overcome when the mesh is refined. To complicate matters, however, mesh
refinement is also found to increase the discrepancy between JOINTC and the cohesive elements,
as shown in Table 2.3. Further research is required to identify the cause of this phenomenon; the
preliminary postulate is that it is related to differences in the assumptions of the two elements
regarding pre-crack behavior.
2.5.3 ABAQUS® - 2D UEL by Roesler et al. (2007a)
Roesler et al. (2007a) employed a UEL for the cohesive zone, and a 4-node plane stress element
for the intact material, in a mesh that was significantly finer near the cohesive zone than further
away. The geometry and material properties of their beam are shown in Table 2.1, where it is
39
designated as Beam C. The following parameters were also adopted: T0 = 0.04 in.; E = 4.6 Mpsi;
Gf = 0.323 lb/in.; GF = 0.954 lb/in.; ψ = 0.25.
To reproduce the results of the 2D analysis presented by Roesler et al. (2007a), Beam C was
meshed uniformly in this study with 0.2×0.2 in. elements in both directions for the intact
material, whereas the mesh of the cohesive zone was made 5 times as fine. The penalty
stiffnesses for the given E and T0 values were computed as Knn = 118 Mpsi/in. and Kss = 51
Mpsi/in.
The P-CMOD curve obtained in the present study along with that presented by Roesler et al.
(2007a) is shown in Fig. 2.12. As it can be seen, good agreement is obtained between the two
numerical simulations. The small difference in the elastic region may be due to the respective
approaches followed in establishing the penalty stiffnesses.
2.5.4 ABAQUS® - 2D COH2D4 by Gaedicke and Roesler (2009)
Gaedicke and Roesler (2009) were the first to use the built-in 2D ABAQUS® cohesive element
COH2D4 to idealize the fracture process of simply supported beam for a variety of kink point
locations. Their mesh pertained to Beam C and was similar to that used by Roesler et al.
(2007a). Their P-CMOD curve for ψ = 0.25 is plotted in Fig. 2.13, along with the corresponding
results from the present study. The peak load predictions differ by about 7%; Gaedicke and
Roesler (2009) had reported that their discretization “under-predicted the peak load by 12% and
7% with respect to the average and minimum experimental peak load, respectively.” This may
again be attributed to penalty stiffness differences.
40
From the comparisons with previously reported results, it may be concluded that the proposed
use of built-in ABAQUS® cohesive elements is effective in simulating PCC fracture in simply
supported beams. Comparison with experimental measurements reported by other researchers
appears in the following section.
2.6 Comparisons with Experimental Results
In this section, FE simulations conducted by means of the proposed procedure that implements
cohesive elements in ABAQUS® are compared with experimental measurements reported by
various independent researchers. The simulations employ 3D discretizations with bilinear
softening curves in all cases.
2.6.1 Experimental Results by Liu (1994)
Liu (1994) tested notched beam specimens under center-point loading. The pertinent geometry
and average material properties reported are shown in Table 2.1, under the Beam B designation.
A comparison of the P-CMOD and P-LLD curves is shown in Fig. 2.14. Good agreement is
observed between the numerical solution in the present study and Liu’s experimental results,
especially for the P-CMOD curve. The small difference in the elastic portion of the P-LLD was
explained by Liu (1994) as “the result of support settlement, which can cause the measured load-
point deflection larger than the actual one. As a result, the predicted curves are stiffer in
comparison with the measured ones. A more sophisticated testing set-up is needed to overcome
this problem.”
41
2.6.2 Experimental Results by Roesler et al. (2007a)
Beam C of Table 2.1 is considered here. Good agreement is obtained between the experimental
results reported by Roesler et al. (2007a) and the FE simulation conducted in the present study,
as shown in Fig. 2.15. In the elastic range, the present idealization gave a smaller load for any
given CMOD as compared to the experimental results. This can be attributed to the use of low
penalty stiffness. The numerical procedure reproduced the peak load very well. The post-peak
behavior is accurately reproduced up to CMOD of 0.0063 in. If one keeps in mind the variability
in the experimental results from replicate specimens reported by Roesler et al. (2007a), it can be
concluded that the numerical simulation has reasonably captured the fracture process.
The results shown in Fig. 2.14 and 2.15 indicate the potential use of cohesive elements available
in ABAQUS® to simulate crack propagation in concrete on the basis of the FCM. The findings
from this study affirm the potential of the proposed numerical procedure when this is extended to
PCC slabs-on-grade in the near future.
2.7 Conclusions
This study focused on the use of 2D and 3D cohesive elements that have recently become
available in the commercial FE package ABAQUS® in studying crack propagation in simply
supported concrete beams. The input parameters required for traction-separation cohesive
elements include the tensile strength, the fracture energy, and the softening curve type. These
elements were inserted in the anticipated cohesive fracture zone, and their top and bottom faces
were tied to the beam elements on either side of the crack plane. Analyses conducted examined
the effect of solution technique, mesh size, and width of cohesive zone.
42
In a comparison of the proposed approach with other numerical studies, good agreement was
found with Sengupta’s GTSTRUDL/CRACKIT combination. The small discrepancy observed can
be ascribed to assumptions implicit in CRACKIT. The proposed FE formulation also gave good
agreement with the results from the UEL created by Park (2005). Results from the present study
were also compared with experimental data reported by different researchers and good
agreement was again found.
The main advantages of cohesive elements over other numerical simulations presented can be
summarized as: (a) capability to be applied in 3D FE analysis; (b) feasibility of implementing
different kind of softening curve assumptions; and (c) possibility of accommodating a variety of
different failure criteria. Cohesive elements, however, are computationally demanding and tax an
analyst’s discretization skills. For example, exploring material damage introduces nonlinearity to
the system, which may result in convergence problems especially if the Newton-Raphson
solution algorithm is used. To avoid this problem, the modified Riks method (or the Newton-
Raphson method with viscous regularization) can be used.
From the findings in this study, it is anticipated that cohesive elements implementing traction-
separation will be used in realistic problems involving PCC pavement slabs resting on layered
foundations. It is hoped that the use of fracture mechanics concepts will eventually lead to the
definition of more reliable and realistic failure criteria, which will address the current
weaknesses of statistical/empirical transfer functions commonly employed in pavement design
guides.
43
2.8 References
ABAQUS (2007). Analysis User’s Manual - Version 6.7-1. Dassault Systèmes, Providence, RI.
ABAQUS (2003). Analysis User's Manual - Version 6.4-1. Hibbitt, Karlsson and Sorensen, Inc.,
Pawtucket, RI.
ABAQUS (2004). Analysis User's Manual - Version 6.5. Hibbitt, Karlsson and Sorensen, Inc.,
Pawtucket, RI.
ASTM D790 (2010). Standard Test Methods for Flexural Properties of Unreinforced and
Reinforced Plastics and Electrical Insulating Materials. ASTM International, West
Conshohocken, PA.
44
GTSTRUDL (1993). Finite Element Computer Software System for Structural Analysis and
Design - User’s Manual. Georgia Tech Research Corporation, Georgia Institute of Technology,
Atlanta, GA.
Ioannides, A. M., and Peng, J. (2004). "Finite element simulation of crack growth in concrete
slabs: implications for pavement design." Proc., Fifth International Workshop on Fundamental
Modeling of Concrete Pavements, International Society for Concrete Pavements, Istanbul,
Turkey, 1-11.
beams: implications for pavement design." Transp. Res. Rec. 1853, Transportation Research
Ioannides, A. M. (1997a). "Pavement fatigue concepts: a historical review." Proc., Sixth
International Purdue Conference on Concrete Pavement Design and Materials for High
Performance, Purdue University, Indianapolis, IN, 147-159.
45
Ioannides, A. M. (1997b). "Fracture mechanics in pavement engineering: the specimen size
Ioannides, A. M., Peng, J., and Swindler Jr, J. R. (2005). "Simulation of concrete fracture using
ABAQUS." Proc., Eighth International Conference on Concrete Pavements, International
Society for Concrete Pavements, Copper Mountain, CO, 1138-1154.
111(10), ASCE, New York, NY, 1227-1241.
Khazanovich, L., and Tomkins, D. (2005). "Singularities in concrete pavement analysis."
Workshop on Fracture mechanics for concrete pavements: Theory to practice, International
Columbus, OH.
Park, K. (2005). "Concrete fracture mechanics and size effect using specialized cohesive zone
model," M.S. thesis, University of Illinois at Urbana-Champaign, Urbana, IL.
46
Park, K., Paulino, G. H., and Roesler, J. R. (2008). "Determination of the kink point in the
bilinear softening model for concrete." Eng. Fract. Mech., 75(13), Elsevier, Amsterdam, The
Riks, E. (1979). "An incremental approach to the solution of snapping and buckling problems."
Int. J. Solids Struct., 15(7), Pergamon Press, Ltd., Oxford, UK, 529-551.
Roesler, J., Paulino, G. H., Park, K., and Gaedicke, C. (2007a). "Concrete fracture prediction
300-312.
Roesler, J., Paulino, G., Gaedicke, C., Bordelon, A., and Park, K. (2007b). "Fracture behavior of
functionally graded concrete materials for rigid pavements." Transp. Res. Rec. 2037,
Transportation Research Board, National Research Council, Washington, DC, 40-49.
Sengupta, S. (1998). "Finite element simulation of crack growth in concrete beams: implication
for concrete pavement design," M.S. thesis, University of Cincinnati, Cincinnati, OH.
Shames, I. H., and Dym, C. L. (1985). Energy and Finite Element Methods in Structural
Mechanics. Hemisphere Publishing, New York, NY.
Urbana, IL.
47
Song, S. H., Paulino, G. H., and Buttlar, W. G. (2005). "Cohesive zone simulation of Mode I and
Mixed-Mode crack propagation in asphalt concrete." GeoFrontires 2005, Advances in Pavement
Engineering, ASCE, Austin, TX, 1-10.
Song, S. H., Paulino, G. H., and Buttlar, W. G. (2006). "Simulation of crack propagation in
VA, 1215-1223.
Timoshenko, S. P., and Goodier, J. N. (1970). Theory of Elasticity. McGraw-Hill, New York,
NY.
Yang, Z. J., and Proverbs, D. A. (2004). "A comparative study of numerical solutions to non-
linear discrete crack modelling of concrete beams involving sharp snap-back." Eng. Fract.
Mech., 20(1), Elsevier, Amsterdam, The Netherlands, 81-105.
48
Table 2.1 Geometry and Material Properties of Beams Studied
Young's Tensile Fracture

Thickness Width Length Span
Modulus Strength Energy
Beam
h b L S E ft ' GF
(in.) (in.) (in.) (in.) (ksi) (ksi) (lb/in.)
A (Sengupta, 1998) 6 1.5 16 16 4000 0.463 0.431
B (Liu, 1994) 3 1 12 12 5405 0.463 0.431
C (Roesler et al.,
9.8 3.2 43.3 39.4 4641 0.602 0.954
2007a)
Table 2.2 Deflection at Midspan of Beam A for Different Types of Elements
Element Type No. of Elements vmax×10-2 (in.) Percentage (%)
Theory NA 3.528 100.0
B22 160 3.673 104.1
CPS4 64×48 3.609 102.3
C3D27 5×40×30 3.538 100.3
49
Table 2.3 Comparison of JOINTC and COH2D4 for Different Mesh Sizes (Beam A)
Mesh Response JOINTC (a) COH2D4 (b) Col. b ÷ Col. a (%)
Pmax (kips) 0.5658 0.5656 99.97
8×6 CMOD/2 (in.) 0.0007 0.0006 92.12
LLD (in.) 0.0016 0.0016 100.00
Pmax (kips) 0.4954 0.5914 119.38
16×12 CMOD/2 (in.) 0.0006 0.0009 147.12
LLD (in.) 0.0016 0.0022 132.88
Pmax (kips) 0.4117 0.5933 144.11
64×48 CMOD/2 (in.) 0.0007 0.0008 125.63
LLD (in.) 0.0018 0.0024 134.87
Note: Applied vertical displacement = 0.03 in., a/h = 33.33%, bilinear softening curve, Pmax =
maximum load resisted by the beam
50
0.12
FE (C3D27)
0.10 TB
TE
FE (CPS4)
0.08 EB
Deflection (in.)
0.06
0.04
0.02
0.00
0 2 4 6 8
Distance Along Beam Span (in.)
Figure 2.1 Comparison of Theoretical and Finite Element Results (Beam A)
Note: FE = finite element, TB = Timoshenko beam, TE = theory of elasticity, EB = Euler’s beam
51
(wi, tí)
tí
Tractions (t)
f´t (wi, t´s)
Knn (wk, ψf´t )
wcr wi w1 wf
Crack Opening (w)
Figure 2.2 Bilinear Softening Curve
52
1.8
LLD Riks
LLD NR
1.6
CMOD Riks
CMOD NR
1.4
Applied Load (kips)
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 0.001 0.002 0.003 0.004 0.005
CMOD/2 or LLD (in.)
Figure 2.3 Effect of Analysis technique (Beam A)
53
0.45
CMOD NR
0.40 CMOD Riks
LLD NR
LLD Riks
0.35
0.30
Applied Load (kips)
0.25
0.20
0.15
0.10
0.05
0.00
0 0.001 0.002 0.003 0.004 0.005
CMOD/2 or LLD (in.)
Figure 2.4 Effect of Analysis technique (Beam B)
Note: NR = Newton-Raphson
54
1.80
CMOD 8x6x1
LLD 8x6x1
CMOD 16x12x2
1.60 LLD 16x12x2
CMOD 32x24x4
LLD 32x24x4
1.40
1.20
Applied Load (kips)
1.00
0.80
0.60
0.40
0.20
0.00
0.000 0.001 0.002 0.003 0.004 0.005
CMOD/2 or LLD (in.)
Figure 2.5 Effect of Mesh Fineness (Beam A)
55
0.50
CMOD 24x6x2
LLD 24x6x2
CMOD 36x9x4
LLD 36x9x4
0.40 CMOD 60x15x8
LLD 60x15x8
Applied Load (kips)
0.30
0.20
0.10
0.00
0 0.001 0.002 0.003 0.004 0.005
CMOD/2 or LLD (in.)
Figure 2.6 Effect of Mesh Fineness (Beam B)
56
0.45
a/h=0
0.40
a/h=1/15
0.35 a/h=1/5
0.30
Applied Load (kips)
0.25
0.20
0.15
0.10
0.05
0.00
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030
CMOD/2 (in.)
Figure 2.7 Effect of Notch Depth (Beam B)
Note: a is notch depth, and h is beam thickness
57
0.45
T0=0.001 in.
T0=0.010 in.
0.40
T0=1.000 in.
0.35
0.30
Applied Load (kips)
0.25
0.20
0.15
0.10
0.05
0.00
0.000 0.001 0.002 0.003 0.004 0.005
CMOD/2 (in.)
Figure 2.8 Effect of Cohesive Zone Width (Beam B)
Note: T0 is cohesive zone width
58
0.20
LLD Linear
CMOD Bilinear
0.18
LLD Bilinear
CMOD Linear
0.16
0.14
0.12
Applied Load (kips)
0.10
0.08
0.06
0.04
0.02
0.00
0 0.001 0.002 0.003 0.004 0.005
CMOD or LLD (in.)
Figure 2.9 Effect of Softening Curve (Beam B)
59
0.18
0.16
0.14
0.12
Applied Load (kips)
0.10
0.08
0.06 LLD 2D CRACKIT

CMOD 2D CRACKIT
LLD 3D COH3D8
0.04
CMOD 3D COH3D8
0.02
0.00
0 0.0005 0.001 0.0015 0.002 0.0025 0.003 0.0035
CMOD/2 or LLD (in.)
Figure 2.10 Comparison of 2D CRACKIT with 3D Cohesive Element (Beam B)
60
0.7
LLD 2D JOINTC
CMOD 2D JOINTC
LLD 2D COH3D8
0.6 CMOD 2D COH3D8
0.5
Applied Load (kips)
0.4
0.3
0.2
0.1
0
0 0.0005 0.001 0.0015 0.002 0.0025 0.003 0.0035 0.004
CMOD/2 or LLD (in.)
Figure 2.11 Comparison between JOINTC and Cohesive Element (Beam A)
61
1.6
Roesler et al. (2007a) 2D UEL
1.4 This study 2D COH2D4
1.2
Applied Load (kips)
1.0
0.8
0.6
0.4
0.2
0.0
0.000 0.005 0.010 0.015 0.020
CMOD (in.)
Figure 2.12 Comparison with Numerical Results by Roesler et al. (2007a) (Beam C)
62
1.6
Gaedicke and Roesler (2009) 2D COH2D4
This study 2D COH2D4
1.4
1.2
Applied Load (kips)
1.0
0.8
0.6
0.4
0.2
0.0
0.000 0.005 0.010 0.015 0.020
CMOD (in.)
Figure 2.13 Comparison with Numerical Results by Gaedicke and Roesler (2009) (Beam C)
63
0.18
LLD Experimental
CMOD 3D COH3D8
0.16
LLD 3D COH3D8
0.14 CMOD Experimental

Applied Load (kips)
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0 0.001 0.002 0.003 0.004 0.005
CMOD or LLD (in.)
Figure 2.14 Comparison with Experimental Results by Liu (1994) (Beam B)
64
1.6
This study 3D COH3D8
Roesler et al. (2007a) Experimental

1.4
1.2
Applied Load (kips)
1.0
0.8
0.6
0.4
0.2
0.0
0.000 0.005 0.010 0.015 0.020
CMOD (in.)
Figure 2.15 Comparison with Experimental Results by Roesler et al. (2007) (Beam C)
65
66
Chapter 3 Numerical Analysis of Fracture Process in Pavement Slabs2
Abstract
This paper presents a numerical analysis of the fracture behavior of pavement slabs, using
special-purpose cohesive elements. Hilleborg’s fictitious crack model is employed in sensitivity
studies exploring the effect of a number of parameters on edge loading responses. Moreover, the
case of interior loading is investigated, anticipating a future thermal stress analysis. Results are
compared with previous experimental as well as numerical investigations conducted by other
independent researchers. It is shown that cohesive elements are suitable for studying crack
propagation as required in pavement engineering. It is envisaged that the approach presented in
this study can be extended to more realistic in situ pavement systems, thereby addressing the
limitations of current mechanistic-empirical pavement design procedures.
2
A shorter version of this chapter has appeared in Canadian Journal of Civil Engineering, Vol.
39, No. 5, 2012, pp. 506-514.
67
3.1 Introduction
Portland cement concrete (PCC) slabs-on-grade are subjected to traffic loads and environmental
stressors, mainly temperature and moisture, resulting in the formation of discrete cracks
(Westergaard, 1927). Analytical as well as numerical idealizations of the slab behavior up to its
elastic limit under those actions have been well documented (Huang and Wang, 1973).
Additionally, considerable research has been devoted to post-elastic characteristics, following
crack initiation (Ang et al., 1963; Ramsamooj, 1993; Ioannides and Peng, 2004). Naturally, the
performance of the system is greatly affected by the processes of cracks formation and
propagation. The present study is intended as yet another small contribution in this evolving field
of research, which aspires to eventually replace the purely statistical/empirical transfer functions
and Miner’s cumulative linear fatigue hypothesis (Miner, 1945), which are currently in use in
mechanistic-empirical pavement design guides (AASHTO, 2008). More specifically, two
problems frequently encountered in dealing with post-crack slab behavior are addressed in this
paper. The first is that the complexity of the phenomenon defies closed-form theoretical
treatment, especially for quasi-brittle materials like concrete, while the second is that its
numerical simulation poses enormous computational challenges of its own.
For this study, post-crack analysis of slab-on-grade pavements is accomplished using a
commercial finite element package ABAQUS®, in which a nonlinear fracture mechanics (NLFM)
application of the fictitious crack model (FCM), first proposed by Hillerborg et al. (1976), is
implemented. A crucial aspect of the simulations presented is the use of cohesive elements for
capturing deformation and damage, leading to a detailed discussion of how related computational
issues may be resolved.
68
3.2 Literature Review
The majority of analytical approaches developed to date for crack simulation have been limited
to linear elastic fracture mechanics (LEFM) concepts (Ramsamooj, 1993, 1994; Shah and
Ouyang, 1994; Roesler and Khazanovich, 1997; Gotlif et al., 2006). Such approaches neglect
the quasi-brittle nature of concrete and the creation of a large fracture process zone (FPZ)
(Kaplan, 1961; Shah and McGarry, 1971). The relatively fewer attempts to employ concepts of
NLFM to characterize concrete fracture have been summarized by Shah et al. (1995). These aim
primarily at capturing the physics of the problem, while simultaneously remaining rather
simplistic, lest their numerical implementation become prohibitively demanding. The FCM, first
proposed by Hillerborg et al. (1976), is a case in point, since it has been found to simulate well
the cracking process, while lending itself well for numerical algorithms. The FCM takes into
account the interdependence of the traction pressure transferred at the crack tip and the
corresponding material delamination occurring in the FPZ. Energy expended in creating new
crack surfaces is neglected, since it is assumed to be very small compared to that needed to cause
crack propagation.
The FCM has been implemented in finite element (FE) investigations of the crack process,
beginning with two-dimensional problems (Hillerborg et al., 1976; Bažant and Oh, 1983; Liu,
1994). In order to extend the approach to increasingly more complex practical situations, a step-
by-step approach has been pursued by researchers at the University of Cincinnati since the late
1990s, following an extensive review of fracture mechanics applications in pavement
engineering (Ioannides, 1997a, 1997b). At first, a standalone computer program was coded and
applied to simply supported beams (Ioannides and Sengupta, 2003). Subsequently, appropriate
69
elements in the commercially available FE program ABAQUS® Version 5.6 were sought
(Ioannides et al., 2005). At the time, nonlinear spring element, JOINTC, was found to be suitable
for this purpose, and this was adopted for concrete beams and slabs-on-grade subjected to
mechanical loads (Ioannides and Peng, 2004; Ioannides et al., 2006). Similarly, investigators at
the University of Illinois, created user elements (UEL) relying on FCM, and implemented them
locally in commercial software to simulate crack propagation in concrete beam specimens (Park,
2005; Roesler et al., 2007), in asphalt specimens (Song, 2006; Song et al., 2006a, 2006b), and in
functionally graded concrete materials (Evangelista et al., 2009; Park et al., 2010). Although
these elements were two-dimensional, responses obtained were compared with experimental
measurements and adequately good agreement was reported. Additional theoretical progress is
to be expected following the recent formulation of a more general cohesive zone concept using
potential energy fields (Park et al., 2009).
Numerical implementation of the FCM has been boosted significantly by the 2005 release of
ABAQUS® Version 6.5, which for the first time included “a family of cohesive elements for
modeling deformation and damage in finite-thickness adhesive layers between bonded parts.”
Both aforementioned University research groups quickly embarked on related but independent
simulations of crack propagation in beams and slabs-on-grade subjected to mechanical loads,
with some very encouraging results (Gaedicke and Roesler, 2009; Aure and Ioannides, 2010).
The present paper is the most recent product of this continuing effort, which may eventually
enable engineers to develop rational failure criteria that can replace the statistical/empirical
algorithms in current pavement design practice (AASHTO, 2008).
70
3.3 Finite Element Discretization
The present study utilizes general purpose FE program ABAQUS/STANDARD® version 6.9-2 for
the analysis of pavement slabs-on-grade, whose geometry plus material properties are shown in
Table 3.1. Unless and otherwise stated, the slab designated as SL1 in Table 3.1 is considered in
most sections; this slab was previously analyzed by Ioannides et al. (2006), using JOINTC
elements for the FPZ idealization.
To minimize resources expended, bulk material and cohesive FPZ are discretized independently:
a coarse mesh of C3D27 elements is used for the intact material, pursuant to a conclusion from a
linear elastic analysis of the slab to be presented in the next section, and a relatively fine mesh of
0.001×0.12×0.12 COH3D8 elements for the FPZ idealization, as indicated in Fig. 3.1. The two
mesh sections are connected using surface-based TIE CONSTRAINTS, which enforce identical
boundary conditions on nodes that lie on the interface.
Cohesive elements, COH3D8, are inserted along the anticipated fracture plane, in accordance
with a discrete crack approach. This is deemed a reasonable representation for pavement slabs
loaded either at the edge or at the interior, since the fracture plane is anticipated in the direction
of the maximum stress. Moreover, Mode I or opening mode is assumed, since the contribution of
the shear stresses is expected to be negligible.
Two Winkler-type subgrade idealizations are considered: SPRING1 elements that can only
support compression; and FOUNDATION elements that can resist both tension and compression.
Moreover, two loading scenaria are considered during the fracture analysis of the slabs: a unit
displacement applied at two nodes on either side of the symmetry line at (x, y, z) = (3, 6, 6 in.)
71
from the edge, as shown in Fig. 3.1, and a unit displacement applied over the nodes that lie
within a 12 by 12 in. loading area, so that a rigid plate type of loading can be realized.
The responses monitored are as follows: load line displacement (LLD), i.e., vertical displacement
at the center of the loaded area to the right of the symmetry (also, fracture) plane; maximum
bending stress at the bottom of the loaded area (σmax); crack mouth opening displacement
(CMOD), i.e., the horizontal displacement at the bottom of the FPZ directly under the applied
load; and total applied load (P), i.e., the sum of reaction forces at all loaded nodes.
3.4 Linear Elastic Analysis
In order to select an appropriate element for the discretization of its intact (uncracked) mass, a
linear elastic analysis of slab SL1 is first conducted. Three discretization trials employing
C3D27, C3D20R and C3D8R elements, which are described as 27-node full integration, 20-node
reduced integration and 8-node reduced integration solid elements, respectively, are considered.
The mesh is uniform in all three cases with 40×20×3 elements in the length, width and depth
directions, respectively, resulting in an element size of 6×6×2 in. The subgrade is simulated by
using the tension capable linear elastic FOUNDATION option available in ABAQUS®.
Pressure is incrementally applied at the slab edge over an area of 12×12 in., The maximum
bending stress (σmax) and load line displacement (LLD) values attained at the specified maximum
pressure of 100 psi are given in Table 3.2, along with the analytical results obtained from
Westergaard’s edge load solution (Westergaard, 1948). It is clearly observed that linear solid
elements, C3D8R, result in a maximum stress in the slab that is only about 65% of that predicted
by Westergaard and reproduced more precisely by the higher order elements.
72
To improve the performance of the C3D8R elements, the alternating mesh pattern shown in Fig.
3.2 is considered; this is similar to the meshes used by Gaedicke and Roesler (2009). At the
symmetry line of the slab, the element size is 0.1875×0.1875×0.1875 in. whereas at the far ends
of the slab, the element size is 12×12×3 in. The total number of C3D8R elements employed is
174,240, resulting in a total of 202,718 nodes. This is about 8 and 15 times the number of nodes
generated with uniformly meshed C3D27 and C3D20R elements, respectively. The maximum
bending stress and deflection values for the alternating C3D8R simulation are also shown in
Table 3.2. The maximum stress now approaches 95% of the Westergaard value, but the cost of
such an improvement in terms of computational resources expended is deemed prohibitive for
general application. Therefore, simpler and coarser meshes of higher order elements are
recommended, in order to ensure savings in computational resources without compromising the
accuracy of the responses. Additionally, C3D8R elements exhibit spurious modes in the vicinity
of the loaded area, particularly when a concentrated load is applied. Therefore, unless otherwise
stated, the full integration C3D27 element with a mesh size of 6×6×2 in. is adopted in all
subsequent sections in discretizing the intact material.
3.5 Sensitivity Studies
3.5.1 Effect of Concrete Softening Curve
The FCM assumes that the traction stress is purely a material property, independent of specimen
geometry and size. The softening curve that relates the traction stress to the opening
displacement is defined in terms of the total fracture energy, GF, the concrete tensile strength, f t ' ,
and the shape of the curve. Several researchers have argued that the shape of the softening curve
affects structural response significantly, particularly local failure behavior (Petersson, 1981;
73
Gustafsson, 1985; Roelfstra and Wittmann, 1986; Alvaredo and Torrent, 1987). Consequently,
numerous different curve shapes have been proposed (Shah et al., 1995; Park et al., 2008).
Hillerborg et al. (1976) started with a simple linear softening curve, as shown in Fig. 3.3. Since
the total fracture energy is equal to the area under the curve, the final (zero traction)
displacement, w f , is:
GF
wf  2 (3.1)
ft'
Subsequently, Petersson (1981) and Gustafsson (1985) employed a bilinear softening curve
characterized by an additional kink point, as shown in Fig. 3.3. If the coordinates of the kink
 G 1 
point, (wk, ψ f t ' ), are set at  0.8 F' , f t '  , the final displacement will be:
 ft 3 
GF
w f  3.6 (3.2)
ft'
The location of the kink point, leading to Eq. (3.2), is contingent on two empirical assertions,
represented by the coefficients 0.8 and 1/3, above, and has, therefore, been subject to debate.
Bažant (2002) essentially eliminated the first of these through the introduction of initial fracture
energy, Gf, a specimen size dependent parameter, whose determination had been outlined by
Bažant and Kazemi (1990). With regard to the bilinear softening curve, Gf is the area under the
first branch, thereby defining the (extrapolated) horizontal intercept of this branch as:
2G f
w1  (3.3)
ft'
74
Bažant (2002) retained the second coefficient, , as an empirical assertion, but estimated that its
value ranged from about 1/6 to 1/3. Accordingly, the final displacement can be written as:
2
wf  G  1   G f  (3.4)
 ft'  F
Eager to determine  experimentally, Park et al. (2008) suggested using yet another parameter
obtained from testing, this time as described by Jenq and Shah (1985), namely the elastic critical
crack tip opening displacement (CTODc). Assuming that this is equal to the displacement at the
kink point, the stress ratio, , can then be established as:
CTODc f t '
  1 (3.5)
2G f
Additional softening curves have been proposed by various investigators, including exponential
and power forms. For example, an exponential traction stress versus displacement is used by
Gaedicke and Roesler (2009):
  
w  wcr
w f  wcr

  1  exp 
ts'  f t ' 1     (3.6)
  1  exp 
  
In this expression, wcr is the cohesive zone separation at the crack initiation point. The softening
parameter α may be determined by equating the integrand of Eq. (3.6) to the total fracture
energy. In this study, numerical computing package MATLAB (Moler, 2004) was used for this
purpose. Similarly, a power law softening shape is used by Song (2006):
75

 w 
ts'  f t '  1   (3.7)
 wf
 
Integration can be used again to compute the exponent parameter α, for a given value of wf.
In ABAQUS®, the softening curves are sub-divided into two regions for the purpose of
integration: the elastic or pre-crack response and the damage or post-crack behavior. The pre-
crack response is characterized by:
ε n   wn 
  1  
ε s    ws  (3.8)
 ε  T0 w 
 t  t
where:  is a nominal strain and w is the elastic separation vectors, in the normal and two shear
directions, respectively; and T0 is initial width of the cohesive zone. The elastic traction stress
components can then be computed from:
t n   K nn K ns K nt   n 
    
t s    K ss K st   s  (3.9)

t   symm K tt   
t   t 
where K is a nominal stiffness (also referred to as penalty stiffness) and t is the nominal traction
stress vector, in the normal and two shear directions, respectively. If uncoupled traction is
assumed (as done in the present study), the off-diagonal terms in Eq. (3.9) are zero. The values
of the penalty stiffness can be approximated by using the elastic moduli, E and G of the intact
material, and the so-called characteristic width of the cohesive zone, T0, i.e., Knn = E/T0, and Kss
= Ktt = G/T0 (Daudeville et al., 1995).
76
Following the onset of the crack, and for as long as the strength of the cohesive zone exceeds
that of the intact material, damage evolves according to the stiffness degradation variable, D,
which is defined as:
t' s
D  1 for wcr  w  w f (3.10)
t'i
where: t´s is the traction stress for separation w, along the softening curve; tí is the traction stress
that would have corresponded to w had the pre-crack stiffness endured (ABAQUS, 2009).
In the present study, the repercussions of softening curve selection are examined using as test
slab the one designated as SL1 in Table 3.1. Four softening curves are considered, namely,
linear, bilinear, exponential and power. All correspond to the same total fracture energy given in
Table 3.1, as shown Fig. 3.3. For bilinear softening, the location of the kink point is situated per
Petersson (1981) and Gustafsson (1985). Setting the final displacement for the exponential and
power softening curves to the same value as obtained from Eq. (3.2) results in parameter α,
determined by integration of Eqs. (3.6) and (3.7), as 3.06 and 2.60, respectively. Results of the
FE runs with each of the four softening curves are shown in Fig. 3.4. Linear softening is
observed to give a slightly higher peak load than the rest. The difference between the linear and
the bilinear curves is less than 2%, which is even lower than that reported for simply supported
beams (Aure and Ioannides, 2010). In fact, all differences are negligible from a practical point of
view. Recalling that previous studies (Roesler et al, 2007; Gaedicke and Roesler, 2009; Aure
and Ioannides, 2010) indicated that bilinear softening predicts results that are in good agreement
with experimental measurements, this relatively simple representation was adopted for the
77
present study. Convergence concerns (Gaedicke and Roesler, 2009) are addressed through
viscous regularization, below. The analysis is carried out using both Newton-Raphson and
modified Riks, which are to be discussed in a subsequent section below.
3.5.2 Effect of Viscous Regularization
A major challenge posed by materials exhibiting progressive damage and stiffness degradation is
achieving a convergent solution, since this is sensitive to the type of softening curve, mesh size,
penalty stiffness and type of solver used (Chaboche et al., 2001; Camanho and Dávila, 2002; Gao
and Bower, 2004). The use of linear softening, moderately low penalty stiffness, and arc-length
solvers, along with a very fine mesh may be expected to result in convergent solutions, but
refining the mesh in large structural systems, like concrete pavements, can be computationally
demanding. To overcome this problem, several researchers have resorted to viscous
regularization. The idea was first proposed by Needleman (1988) for simulating plastic behavior,
and was subsequently widely accepted (Chaboche et al., 2001; Gao and Bower, 2004; Maimi et
al., 2007; Lapczyk and Hurtado, 2007; Hamitouche et al., 2008). By adjusting the value of a
viscous term introduced in the constitutive equations of the degrading material, the rate of
viscous energy dissipation is controlled, numerical instability is avoided, and the solution is
regularized.
In order to address convergence problems with cohesive elements, ABAQUS® incorporates this
concept in the form of a viscous stiffness degradation parameter, Dv, defined by:
d 1
 Dv    D  Dv  (3.11)
dt м
78
where м is “the viscosity parameter representing the relaxation time of the viscous system,” D is
“the degradation variable evaluated in the inviscid backbone model”, and t denotes time
(ABAQUS, 2009). The traction stress is then computed using Dv as:
t's  1  Dv  t'i (3.12)
The viscous system approaches the inviscid ideal as (t/м) tends to infinity, making the viscous
energy dissipated negligble. Viscous energy during unstable crack growth depends on the
specimen geometry and material properties (Gao and Bower, 2004).
Accordingly, a sensitivity study was conducted in order to select the appropriate size of м for the
problem size considered in the present study. Ideally, м should be as small as possible since it
represents an artificial departure from the actual material constitutive relationship, but this
requirement is tempered by the fact that larger м values are computationally less demanding.
Setting м to zero, results in abrupt solution termination due to lack of convergence. Different
values of м ranging from 10-9 to 10-3 were, therefore, considered, and the results are presented in
Fig. 3.5. Four stages are discernible in the load-displacement curve: linear, nonlinear, softening,
rebound. The linear stage corresponds to the elastic behavior of the system, prior to crack
initiation. Departure from linearily heralds the onset of cracking, but its location is sensitive to
the value of м used. The second stage is nonlinear, suggesting that stable cracks are forming but
the maximum load capacity of the slab is not yet attained. In the third stage, softening occurs
evincing the formation of unstable cracks as the material continues to detriorate. Finally, the
fourth stage is reached, during which the subgrade springs assume the responsibility of
sustaining the load. Figure 3.5 also indicates that the results for м = 10-6 and м = 10-9 are almost
79
identical, i.e., м values lower than 10-6 do not affect the response; therefore, м = 10-6 was
adopted for the present study. Beyond this value, the solution becomes very sensitive to м and
diverges from the inviscid (actual) behavior, presumably because in its search for equilibrium,
the solver employs grossly fictitious traction levels from Eq. (3.12) that exceed the actual
material tensile strength, and increase the viscous energy dissipated.
The effect of viscous parameter is more pronounced when Newton-Raphson solution method is
used compared to modified Riks method. For м values greater than 10-6, the modified Riks
method predicts a lower peak load than the Newton-Raphson (e.g., 32 versus 69 kips for м = 10-
3
). For м values lower than 10-6, however, the two solvers predicted identical peak loads, as
shown in Fig. 3.5. The differences between the two solvers will be discussed in a subsequent
section below.
3.5.3 Effect of Subgrade Idealization
The Winkler subgrade in concrete pavement systems is generally idealized using spring elements
that can support both tension and compression or compression alone. The former approach is
commonly preferred for its numerical simplicity in spite of the fact that it hardly represents in
situ concrete pavement behavior. This section investigates the effect of both simulation
approaches on the post-crack load-displacement responses. The tension and compression
supporting subgrade is idealized using the built-in elastic FOUNDATION, providing the
subgrade reaction as an input, while SPRING1 elements are employed in discretizing a subgrade
that supports compression alone. The same modulus of subgrade reaction value is used in both
cases by converting it to equivalent forces on the spring element depending on its contributing
area. To compare the two simulations, a FE run of the SL1 slab is made and the load-
80
displacement curves are shown in Fig. 3.6. It can be clearly observed from the FOUNDATION
simulation result that although there is a slight change in the slope of the curve evincing the
initiation of cracking in the slab, there is no softening region, i.e., there is no indication of a
complete slab failure. The absence of softening can be explained by the fact that the
FOUNDATION elements pull down any slab elements attempting to lift up along the unloaded
edge of the slab. For the SPRING1 discretization, on the other hand, the crack initiation point and
the peak load point are clearly exhibited. It should be noted that the SPRING1 simulation
requires larger amounts of computational resources since the solver utilizes smaller time
increments to capture the solution path in the softening region. Generally, it can be concluded
that the load-displacement response as well as the amount of computational resources expended
are very much influenced by the subgrade idealization, and, therefore, the analyst must carefully
weigh both options against prevailing in situ conditions before deciding to adopt either of the
idealization schemes.
3.5.4 Effect of Cohesive Zone Width
The characteristic width of the cohesive zone, T0, is used to determine the penalty stiffness
components in the normal and shear directions for the cohesive element, as discussed above. In
beam fracture analysis, a cohesive width of 0.001 in. was found to be a reasonable value to give
a convergent solution (Aure and Ioannides, 2010).
The same approach may be extended to slabs. Four different T0 -values were tried: 0.001, 0.01,
0.1, and 1.0 in. Judging from the load-displacement curves shown in Fig. 3.7, changes in the
cohesive zone width do not change the response substantially, although an increase in T0 widens
the crack mouth opening displacement slightly. This is due to the fact that increasing cohesive
81
width decreases the penalty stiffness and consequently increases the horizontal elastic
displacement of the cohesive zone. As for beams, the 0.001 in. width is found to be
computationally attractive without compromising the stability and accuracy of the solution.
Therefore, it was adopted in this study.
3.5.5 Effect of Cohesive Zone Mesh
Three cohesive zone meshes, with aspect ratio of 3, 2, and 1, respectively, are considered. These
give rise to COH3D8 element sizes of 0.75×0.25 in., 0.40×0.20 in., and 0.12×0.12 in, along the
width and depth directions, respectively. For these cases, 1×160×24, 1×300×30, and 1×1000×50
cohesive elements, respectively, correspond to the 40×20×3 slab elements. Refining the cohesive
zone mesh is found to decrease the number of increments needed to complete the applied
displacement. Nonetheless, the load-displacement curves shown in Fig. 3.8 for all three cases are
observed to be essentially identical, and, therefore, the least demanding 0.12×0.12 in. cohesive
zone mesh is adopted in this study.
3.5.6 Effect of Solution Technique
In materials that exhibit snap-back type of load-displacement curves, it has been reported that the
Newton-Raphson method fails to capture the softening region (Riks, 1979; Crisfield, 1991;
Verhoosel et al., 2009). To remedy this situation, Yang and Proverbs (2004) conducted a study
of several numerical methods and recommended the use of arc-length solvers, like the modified
Riks, for problems involving material softening. In implementing the modified Riks algorithm,
ABAQUS (2009) suggests that it can also be used for problems involving snap-back and snap-
through load-displacement history. This suggestion was verified by Aure and Ioannides (2010),
82
who demonstrated the failure of Newton-Raphson to capture the softening region appropriately
during the fracture of simply supported beams. The usefulness of modified Riks was especially
noticeable for beams with smaller span-to-thickness ratios.
A similar comparison of the two methods is made in the present study for slab-on-grade fracture.
Responses are compared in Fig. 3.4, where it is clearly seen that the modified Riks method
captures the post-peak softening region, whereas the Newton-Raphson shows a vertical drop in
the load at constant displacement. Despite this numerical disadvantage, however, careful
adjustment of the modified Riks arc-length increment is required if the reloading curve is to be
reproduced, as well. This requires increased effort, which accrues increased computational
demand compared to Newton-Raphson. The inaccuracy observed by using the latter may in fact
be tolerable in certain cases, owing to its numerical simplicity and shorter computational time,
especially if only the peak load supported by the slab is of primary interest. In this study,
Newton-Raphson was, therefore, retained only in the study of viscous regularization and of
interior loading, for reasons explained elsewhere herein.
3.5.7 Effect of Loading Mode
In the preceding sections, the displacement is applied at the edge over two nodes located at (x, y,
z) = (3, 6, 6) in., per Gaedicke and Roesler (2009). This approach was compared to the case of a
unit displacement applied at all nodes that lie within a 12 by 12 in. area, per Ioannides et al.
(2006). The two scenaria correspond roughly to the application of concentrated and rigid plate
loads, respectively. It is found that the cohesive elements start to be damaged earlier under the
concentrated load scenario, resulting in a significantly lower peak load: 18 compared to 43 kips,
respectively, as shown in Fig. 3.9. This is probably attributable to stress redistribution occurring
83
when a rigid plate load is applied, toward the corner nodes of the loaded area and away from the
assumed fracture plane. The actual loading condition on in situ slabs lies between these two
extreme scenaria, probably being closer to the concentrated load case. The latter is, therefore,
retained in this study for edge loading; for interior loading the rigid plate load is deemed more
appropriate, as discussed in the pertinent section, below.
3.5.8 Effect of Tensile Strength and Fracture Energy
The softening curves depicted in Fig. 3.3 make it apparent that the two main material parameters
influencing the fracture process are tensile strength, f t ' , and total fracture energy. To quantify
the influence of these parameters on the load versus displacement response of the slab, one of
them may be kept constant while changing the other. To illustrate this more simply, the linear
softening curve and the Newton-Raphson solution method may be employed, as detailed next.
Let the tensile strength increase in 10% increments over the value indicated in Fig. 3.3, to 1.1 f t ' ,
1.2 f t ' , 1.3 f t ' , and 1.4 f t ' . The load versus displacement responses obtained are shown in Fig.
3.10: whereas the load at which the crack initiates increases slightly, the peak load and the post-
peak softening behavior remain unaffected. In all cases, the material exhibits brittle behavior,
with localized softening immediately following crack initiation. It is evident that although the
tensile strength has a limited influence along the small softening segment of the load-
displacement curve, it does not seem to play a major post-cracking role. Consequently, the peak
load and the corresponding crack mouth opening and vertical displacements remain almost the
same for all cases considered.
84
Similarly, five different values of total fracture energy, i.e., 0.9 GF, 1.1 GF, 1.2 GF, 1.3 GF and
1.4 GF may be considered, keeping the tensile strength constant. This simulation results in Fig.
3.11, which indicates that increasing the fracture energy gives rise to higher peak loads. The
dimensionless ratios β = [P/P1GF] of the peak load, ζ = [LLD/LLD1GF] of the vertical
displacement at the peak load, and η = [CMOD/CMOD1GF] of the crack mouth opening
displacement at the peak load for each case considered, to the corresponding peak load, LLD and
CMOD for the baseline fracture energy (1.0 GF), are plotted against a dimensionless parameter
called the brittleness number, B, defined by Bache and Vinding (1990) as:
ft 2 h h
B  (3.13)
EGF lch
where lch is the characteristic length of the material, first introduced by Hillerborg (1985), E is
Young’s modulus, and h is the slab thickness. The definition in Eq. (3.13) suggests that as the
fracture energy decreases, B increases, i.e., the material becomes more brittle. Figure 3.12
confirms β is inversely proportional to B. It may be postulated that a unique relationship exists
that would allow one to determine the peak load for a particular value of B, given the
corresponding peak load for a different brittleness number, thereby providing a linkage between
laboratory testing and field applications. A similar trend is observed in the plot of B versus ζ and
η corresponding to the peak load. Such observations are very encouraging in the quest for
mechanistic post elastic failure criteria in pavement engineering, and underscore the desirability
for treating fracture energy as equally important as the tensile strength.
85
3.6 Crack Propagation under Interior Loading
For PCC pavement slabs subjected to mechanical loads alone, edge loading causes maximum
bending stress. If both thermal and mechanical loads act together, however, critical stress may
sometimes occur at the interior of the slab. Anticipating a future study of thermal fracture,
cohesive elements are used in this section to idealize interior loading cracks and a comparison is
made with edge loading. The crack is assumed to propagate in two orthogonal directions from
the center of the slab, per Meda et al. (2004). Therefore, cohesive elements are inserted centrally
along the x- and y-axes, and are tied to the intact slab elements. The cohesive element properties
are identical in both directions. In order to ensure a direct comparison between interior and edge
loading, the rigid plate load scenario is used for both, in addition to the concentrated load (point
displacement) case.
As expected, Fig. 3.13 shows that the maximum load supported by the slab is higher under
interior than under edge loading: 72 versus 43 kips for the rigid plate load; 46 versus 18 kips for
the concentrated loads. The trace in the vicinity of the peak load on the load-displacement curve
indicates that the maximum stress is attained not at the center but at the corner of the loaded area,
closest to the fracture plane. The first peak shows the progression of damage of the cohesive
zone toward the center, whereas the second peak shows the complete failure of the cohesive zone
as the crack proceeds along the prescribed fracture plane. It is also found that at any given
applied load, the crack opens up more along the width (shorter dimension) than along the length.
Crack growth patterns along the bottom of the FPZ in the length and width directions at load
levels shown in Fig. 3.13 are plotted in Fig. 3.14. It is observed that at load level point 3, nodes
directly under the loaded area have exceeded the final (zero traction) displacement, i.e., the
86
displacement at which traction stress is zero, in the length direction, while the crack remains
stable in the width direction. At load level point 7, the displacement at the bottom of the slab in
both width and length directions has exceeded the zero traction value, indicating that the load is
now supported by the subgrade.
3.7 Crack Propagation in a Slab Supported by Base Layer
In general, concrete pavement structures incorporate a second man-made layer, commonly called
the base layer, on which the concrete slab is poured and compacted, in order to enhance
constructability. This section investigates the effect of the stiffness of the base layer on the post-
crack response of the slab by considering a base layer of the same plan dimensions as SL1, and 8
in. in thickness. To simplify the analysis, it is grossly assumed that the base layer remains intact,
behaves elastically and remains in full contact with the subgrade during the loading period.
Moreover, any influence of the base layer on the traction-separation relationship of the concrete
slab is neglected. An untreated aggregate base (UAB) and a cement treated base (CTB) with
Young’s modulus values of 25 and 1500 ksi, respectively, are assumed. The Poisson ratios of the
two materials are taken to be 0.35 and 0.2, respectively.
The slab mesh is kept identical to the previous cases, i.e., 40×20×3 C3D27 elements, whereas the
base is discretized with relatively coarser mesh size, 20×10×2 C3D27 elements. The interface
between the two layers is idealized using contact interactions, specifically INTERACTIONS,
HARD CONTACT, in which unlimited normal compressive pressure is transmitted from the slab
to the base layer. The two layers are assumed to be unbonded, i.e, no frictional stresses arise at
the interface. This is realized by assuming a zero coefficient of friction between slab and base.
87
Since the subgrade is assumed to remain in full contact with the base layer during the loading
period, the tension supporting elastic FOUNDATION is employed to simulate the subgrade.
The analysis results for the two base types are presented in Fig.3.15. As may be anticipated, the
presence of the UAB increased the peak load only very slightly (by about 5%), as compared to
the more prominent contribution by the CTB, which quadrupled this load. It can also be seen
from the CTB curve that the difference between the load at which the crack initiates, which is
made manifest by the change in slope of the curves, and the peak load attained is much larger
when the CTB replaces the UAB (35 compared to 4 kips, respectively). This indicates the
development of stable cracks in the system incorporating a CTB. The results reconfirm the fact
that the presence of a stiff base enhances the strength and fracture resistance of the slab, provided
drainage issues that may be associated with the use of a CTB are addressed.
3.8 Comparison with Other Numerical and Experimental Studies
This section compares the FE idealization employed in the present study with numerical analyses
provided by Ioannides et al. (2006) and by Gaedicke and Roesler (2009), as well as the
experimental results reported by Gaedicke and Roesler (2009).
3.8.1 Comparison with Ioannides et al. (2006)
Ioannides et al. (2006) used nonlinear spring element JOINTC, available in ABAQUS® to
simulate the fracture process in concrete pavement slabs implementing the FCM. The geometry
and material properties of the slab was identical to SL1 shown in Table 3.1. The slab was
discretized with a relatively coarser mesh of 40×10×3 C3D27R element as compared to the
40×20×3 C3D27 element outlined above. An initial notch depth of one-tenth and one-third of the
88
slab width and thickness, respectively, was provided to initiate the crack. Displacement control
was used with a maximum displacement of 1 in. over an area of 12×12 in., thereby simulating a
rigid plate loading scenario. The subgrade was idealized using FOUNDATION, which supports
tension as well as compression as described above.
In the present study, an identical FE simulation is adopted except that cohesive elements are used
instead of JOINTC elements for the idealization of the FPZ, and the C3D27 element is employed
instead of the C3D27R for the idealization of the intact material. The load-displacement results,
shown as P-CMOD to be consistent with Ioannides et al. (2006), are presented in Fig. 3.16. It is
observed that the simulation with the cohesive elements gives a lower load for a given CMOD,
which is a similar observation to that made in an earlier study for simply supported beams (Aure
and Ioannides, 2010). The main reason for these discrepancies may be the development of high
tensile stresses at the nodes of the elements shared by JOINTC and C3D27 elements along the
fracture plane. Ioannides et al. (2006) speculated that the occurrences of such tensile stresses
that are greater than the material failure stress are related to the “stress gradient criteria of
strength for quasi-brittle materials” (Kharlab and Minin, 1989). In the cohesive fracture
reproduction, on the other hand, the stresses along the fracture plane do not exceed the tensile
strength, since the latter is defined as the failure criterion. In either case, post-peak softening is
not exhibited on the load-displacement curves on account of the tension supporting foundation,
as explained earlier.
3.8.2 Comparison with Gaedicke and Roesler (2009)
Gaedicke and Roesler (2009) also used built-in traction-separation cohesive elements to simulate
cracking in PCC slabs-on-grade in ABAQUS®. Two slab geometries had been considered: 2.48
89
in. and 5.9 in. thick, by 78.74 ×78.74 in. in plan. For the present comparison, the first of these is
reproduced, assuming a one-third notch. The geometric and material properties of this slab are
described in Table 3.1, under slab label SL2.
A total of 283,904 elements had been employed out of which 21,504 were COH3D8 and the
remaining were C3D8R, the latter being used for the intact material. In view of the
computational intensity of this discretization, a coarser but uniform mesh is employed in its
reproduction during the present study, using a higher order element, C3D27, as delineated in
Table 3.1. A total of 12,160 elements out of which 2,560 were COH3D8 and 9,600 were C3D27
were employed. For simplicity, linear softening is retained herein, along with a penalty stiffness
computed from E/T0, where E is the slab Young’s modulus and T0 is the cohesive zone width, set
to 0.039 in. The foundation is simulated using tensionless SPRING1 elements, and horizontal
springs with a stiffness of 1/10th of that of the vertical springs are used to provide lateral
restraint, both per Gaedicke and Roesler (2009).
The simulation result is shown in Fig. 3.17. It is observed that the load-displacement curves from
the two studies are almost identical up to 4 kips. Beyond that point, cracking is initiated, and the
two simulations diverge until the peak load, which is also almost the same. The post-peak
unloading, during which the cohesive elements become completely damaged, occurs at a much
lower load in the present study. This is probably related to the much coarser mesh used in the
present study.
The corresponding experimental result reported by Gaedicke and Roesler (2009) is also plotted
along with the two numerical simulations in Fig. 3.17. Good agreement is obtained up to the
90
peak load. The post-peak behavior shows differences primarily due to the linear elastic
idealization of the subgrade. The experimental result shows a pronounced post-peak vertical drop
in the load, mainly due to local plastic yielding of the in situ soil. Considering the overriding
practical importance of the peak load, rather than of the softening region, the results obtained
from the simulation approach followed in the present study are deemed to be adequately accurate
as well as computationally efficient.
3.9 Conclusions
The use of cohesive elements in the simulation of PCC pavement slab fracture has been
investigated by studying the main parameters that affect pre- and post-peak responses.
Discretization with cohesive elements is computationally demanding as it involves material
softening, which influences the stability of the solution algorithm. Consequently, appropriate
solution control strategies (such as viscous regularization) need to be employed.
The effects of softening curve, cohesive zone width and mesh design, analysis technique, loading
mode, tensile strength and fracture energy has been investigated. It is found that the type of
softening curve, cohesive zone width and mesh design do not influence the response
significantly. They do, however, play a significant role in the convergence of the solution. The
influence of loading placement (at the interior or at the edge of the slab) has been examined, on
account of its importance for in situ pavements under simultaneous thermal and mechanical
loads. It is found that the fracture process is more sensitive to the fracture energy than to the
tensile strength.
91
It can be concluded from this study that the application of cohesive elements in fracture analysis
of PCC pavement slabs is promising, as demonstrated by the comparison with experimental
results. It is anticipated that the approach can be extended to pavements subjected to thermal
stresses and incorporating load transfer. This effort will contribute to the ongoing development
of rational failure criteria that can substitute the statistical/empirical algorithms along with
Miner’s hypothesis used in current mechanistic-empirical pavement design procedures.
3.10 References
Interim Ed. American Association of State Highway and Transportation Officials, Washington,
DC.
Providence, RI.
Alvaredo, A. M., and Torrent, R. J. (1987). "The effect of the shape of the strain-softening on the
bearing capacity of concrete beams diagram." Mater. Struct., 20(6), Springer, Berlin, Germany,
448-454.
Ang, D. D., Folias, E. S., and Williams, M. L. (1963). "The bending stress in a cracked plate on
an elastic foundation." J. Appl. Mech., 30(2), ASME, New York, NY, 245-251.
Aure, T. W., and Ioannides, A. M. (2010). "Simulation of crack propagation in concrete beams
using cohesive elements in ABAQUS." Transp. Res. Rec. 2154, Transportation Research Board,
National Research Council, Washington, DC, 12-21.

92
Bache, H. H., and Vinding, I. (1990). "Fracture mechanics in design of concrete pavements."
Proc., Second International Workshop on the Design and Evaluation of Concrete Pavements,
CROW/PIARC, Sigüenza, Spain, 139-165.
Bažant, Z. P. (2002). "Concrete fracture models: testing and practice." Eng. Fract. Mech., 69(2),
Elsevier, Amsterdam, The Netherlands, 165-205.
Bažant, Z. P., and Oh, B. H. (1983). "Crack band theory for fracture of concrete." Mater. Struct.,
16(3), Springer, Paris, France,155-177.
Camanho, P. P., and Dávila, C. G. (2002). "Mixed-mode decohesion finite elements for the
simulation of delamination in composite materials." Rep. No. NASA/TM-2002-211737, National
Aeronautics and Space Administration, Hampton, VA.
Chaboche, J. L., Feyel, F., and Monerie, Y. (2001). "Interface debonding models: a viscous
regularization with a limited rate dependency." Int. J. Solids Struct., 38(18), Elsevier,
93
Daudeville, L., Allix, O., and Ladevze, P. (1995). "Delamination analysis by damage mechanics:
some applications." Compos. Eng., 5(1), Elsevier, Amsterdam, The Netherlands, 17-24.
Evangelista, F., Roesler, J., and Paulino, G. (2009). "Numerical simulations of fracture resistance
of functionally graded concrete materials." Transp. Res. Rec. 2113, Transportation Research
Gao, Y. F., and Bower, A. F. (2004). "A simple technique for avoiding convergence problems in
finite element simulations of crack nucleation and growth on cohesive interfaces." Modelling
Simul. Mater. Sci. Eng, 12(3), Institute of Physics Publishing, London, UK, 453-463.
Gotlif, A., Mallela, J., and Khazanovich, L. (2006). "Finite element study of partial-depth cracks
in restrained PCC slabs." Int. J. Pavement Eng., 7(4), Taylor and Francis Group, London, UK,
323-329.
Hamitouche, L., Tarfaoui, M., and Vautrin, A. (2008). "An interface debonding law subject to
viscous regularization for avoiding instability: application to the delamination problems." Eng.
Fract. Mech., 75(10), Elsevier, Amsterdam, The Netherlands, 3084-3100.
94
Hillerborg, A. (1985). "The theoretical basis of a method to determine the fracture energy GF of
concrete." Mater. Struct., 18(4), Springer, Berlin, Germany, 291-296.
Ioannides, A. M. (1997a). "Fracture mechanics in pavement engineering: the specimen size
beams: Implications for pavement design." Transp. Res. Rec. 1853, Transportation Research
slabs: Implications for pavement design." Proc., Fifth International Workshop on Fundamental
Turkey, 1-11.
95
Ioannides, A. M. (1997b). "Pavement fatigue concepts: a historical review." Proc., Sixth
International Purdue Conference on Concrete Pavement Design and Materials for High
Performance, Purdue University, Indianapolis, IN, 147-159.
Ioannides, A. M., Peng, J., and Swindler Jr, J. R. (2005). "Simulation of concrete fracture using
ABAQUS." Proc., Eighth International Conference on Concrete Pavements, International
111(10), ASCE, New York, NY, 1227-1241.
Kaplan, M. E. (1961). "Crack propagation and fracture of concrete." J. Am. Concrete I., 58(5),
ACI, Farmington Hills, MI, 591-610.
Kharlab, V. D., and Minin, V. A. (1989). "Strength criteria accounting for influence of stress
state gradient." Issledovaniya Po Mekhanike Stroil'Nykh Konstruktsii i Materialov (in Russian),
53-57.
Lapczyk, I., and Hurtado, J. A. (2007). "Progressive damage modeling in fiber-reinforced
materials." Compos. Part A, 38(11), Elsevier, Amsterdam, The Netherlands, 2333-2341.
Columbus, OH.
96
Maimi, P., Camanho, P. P., Mayugo, J. A., and Davila, C. G. (2007). "A continuum damage
model for composite laminates: Part II – Computational implementation and validation." Mech.
Mater., 39(10), Elsevier, Amsterdam, The Netherlands, 909-919.
Meda, A., Plizzari, G. A., and Riva, P. (2004). "Fracture behavior of SFRC slabs on grade."
Mater. Struct., 37(6), Springer, Berlin, Germany, 405-411.
Moler, C. B. (2004). Numerical Computing with MATLAB. Society for Industrial and Applied
Mathematics, Philadelphia, PA.
Needleman, A. (1988). "Material rate dependence and mesh sensitivity in localization problems."
Comput. Method Appl. M., 67(1), Elsevier, Amsterdam, The Netherlands, 69-85.
Park, K. (2005). "Concrete fracture mechanics and size effect using specialized cohesive zone
model," M.S. thesis, University of Illinois at Urbana-Champaign, Urbana, IL.
Park, K., Paulino, G. H., and Roesler, J. R. (2008). "Determination of the kink point in the
bilinear softening model for concrete." Eng. Fract. Mech., 75(13), Elsevier, Amsterdam, The
908.
97
Park, K., Paulino, G. H., and Roesler, J. (2010). "Cohesive fracture model for functionally
graded fiber reinforced concrete." Cem. Concr. Res., 40(6), Elsevier, Amsterdam, The
Ramsamooj, D. V. (1994). "Prediction of fatigue life of plain concrete beams from fracture
tests." Journal of Testing and Evaluation, 22(3), ASTM, West Conshohocken, PA, 183-194.
Ramsamooj, D. V. (1993). "Fracture of highway and airport pavements." Eng. Fract. Mech.,
44(4), Pergamon Press, Ltd., Oxford, UK, 609-626.
Riks, E. (1979). "An incremental approach to the solution of snapping and buckling problems."
Int. J. Solids Struct., 15(7), Pergamon Press, Ltd., Oxford, UK, 529-551.
Roelfstra, R. E., and Wittmann, F. H. (1986). "A numerical method to link strain softening with
fracture in concrete." Fracture Toughness and Fracture Energy in Concrete, Elsevier,
Roesler, J. R., and Khazanovich, L. (1997). "Finite-element analysis of Portland cement concrete
pavements with cracks." Transp. Res. Rec. 1568, Transportation Research Board, National
Research Council, Washington, DC, 1-9.
98
Roesler, J., Paulino, G. H., Park, K., and Gaedicke, C. (2007). "Concrete fracture prediction
300-312.
Shah, S. P., and McGarry, F. J. (1971). "Griffith fracture criterion and concrete." J. Eng. Mech.,
97(EM6), ASCE, New York, NY, 1663-1676.
Shah, S. P., and Ouyang, C. (1994). "Fracture mechanics for failure of concrete." Annu. Rev.
Mater. Sci., 24(1), Annual Reviews Inc., Palo Alto, CA, 293-320.
Shah, S. P., Swartz, S. E., and Ouyang, C. (1995). Fracture Mechanics of Concrete: Applications
of Fracture Mechanics to Concrete, Rock and Other Quasi-brittle Materials. John Wiley and
Sons, Inc., New York, NY.
Urbana, IL.
Song, S. H., Paulino, G. H., and Buttlar, W. G. (2006a). "Simulation of crack propagation in
VA, 1215-1223.
Song, S. H., Paulino, G. H., and Buttlar, W. G. (2006b). "A bilinear cohesive zone model
tailored for fracture of asphalt concrete considering viscoelastic bulk material." Eng. Fract.,
99
Verhoosel, C. V., Remmers, J. J. C., and Gutiérrez, M. A. (2009). "A dissipation-based arc-
length method for robust simulation of brittle and ductile failure." Int. J. Numer. Meth. Eng.,
77(9), John Wiley and Sons, Ltd., Hoboken, NJ, 1290-1321.
Westergaard, H. M. (1948). "New formulas for stresses in concrete pavements of airfields."
Trans., 113, ASCE, New York, NY, 425-439.
temperature." Proc. Hwy. Res. Board, 6, National Research Council, Washington, DC, 201-215.
Yang, Z. J., and Proverbs, D. A. (2004). "A comparative study of numerical solutions to non-
linear discrete crack modelling of concrete beams involving sharp snap-back." Eng. Fract.
100
Table 3.1 Geometry, Material Properties and Discretization of Slabs Considered
Slab Geometry
Slab SL1 SL2
Length (in.) 240 78.74
Width (in.) 120 78.74
Thickness (in.) 6 2.48
Material Properties
Young's Modulus (ksi) 4000 4641.28
Tensile Strength (ksi) 0.463 0.602
Fracture Energy (lb/in.) 0.431 0.679
Poisson's Ratio 0.15 0.15
Subgrade Characteristics
Modulus of Subgrade Reaction (psi/in.) 200 574.7
Numerical Discretization
Slab Element Size (in.) 6×6×2 2×2×0.4
Cohesive Zone Element Size (in.) 0.12×0.12 0.12×0.12
Concrete Softening Curve Bilinear Linear
Solver Type Riks or Newton-Raphson Riks
Subgrade Idealization SPRING1 SPRING1
Notes: 1 lb = 4.444 N; 1 in. = 25.4 mm; 1 ksi = 6.89 MPa
101
Table 3.2 Comparison of Different Solid Elements Available in ABAQUS
Finite Element Analytical Percentage

Element
σmax (psi) LLD (mils) σmax (psi) LLD (mils) σmax LLD
C3D8R 467.25 43.98 768.37 39.62 60.81 111.00
C3D20R 804.87 42.67 768.37 39.62 104.75 107.70
C3D27 785.87 42.66 768.37 39.62 102.28 107.67
Alternating C3D8R 743.56 43.16 768.37 39.62 96.77 108.93
102
Intact material (C3D27)
Cohesive zone (COH3D8)
Applied Load (P)
Subgrade (SPRING1)
Figure 3.1 Finite Element Discretization of Slab (SL1)
Figure 3.2 Finite Element Mesh Pattern Using C3D8R Elements (SL1)
103
0.45
f't = 0.46327 ksi
Exponential
Linear
0.36 α = 2.60
Power
α = 3.06 Bilinear
0.27
Traction Stress (ksi)
(wk, ψ f't) = (0.000745 in., 0.1544233 ksi)
0.18
wf = 0.0018618 in.
wf = 0.0033525 in.
0.09
w1
0.00
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030
Crack Mouth Opening Displacement (in.)
Figure 3.3 Different Types of Concrete Softening Curves
104
20.0
16.0
Applied Load (kips)
12.0
8.0 Bilinear Riks

Exponential Riks
Power Riks
Linear Riks
Bilinear NR
4.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Load Line Displacement (in.)
Figure 3.4 Effect of Concrete Softening Curve and Analysis Technique
Note: NR = Newton-Raphson
105
80.0
70.0
60.0
50.0
Applied Load (kips)
м = 1E-03 Riks
40.0
м = 1E-04 Riks
м = 1E-04 NR
30.0
м = 1E-06 NR
20.0 м = 1E-07 NR
м = 1E-09 NR
10.0 м = 1E-03 NR
0.0
0.00 0.08 0.16 0.24 0.32 0.40 0.48 0.56
Figure 3.5 Effect of Viscous Regularization on Slab Response
106
50.0
SPRING1
FOUNDATION
40.0
Applied Load (kips)
30.0
20.0
10.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Figure 3.6 Effect of Subgrade Idealization on Slab Response
107
20.0
16.0
Applied Load (kips)
12.0
T0 = 0.001 in.
T0 = 0.01 in.
T0 = 0.1 in.
T0 = 1.0 in.
8.0
4.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Figure 3.7 Effect of Cohesive Zone Width
108
20.0
16.0
Applied Load (kips)
12.0
8.0
Cohesive Element Size: 0.75 x 0.25 in.

Cohesive Element Size: 0.12 x 0.12 in.
4.0 Cohesive Element Size: 0.40 x 0.20 in.
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Figure 3.8 Effect of Cohesive Zone Mesh
109
50.0
Point Displacement
Uniform Displacement
40.0
Applied Load (kips)
30.0
20.0
10.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Figure 3.9 Effect of Loading Mode
110
20.0
1.0 f't
1.1 f't
16.0 1.2 f't
1.3 f't
1.4 f't
Applied Load (kips)
12.0
8.0
4.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Figure 3.10 Effect of Tensile Strength
Note: Baseline f t ' = 0.463 ksi
111
24.0
1.0 GF
1.1 GF
20.0
1.2 GF
1.3 GF
0.9 GF
16.0
1.4 GF
Applied Load (kips)
12.0
8.0
4.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Figure 3.11 Effect of Fracture Energy
Note: Baseline GF = 0.431 lb/in.
112
1.20
β=P/P1GF
ζ=LLD/LLD1GF
η=CMOD/CMOD1GF
1.14 Best fit (P/P1GF)
β = -0.766B + 1.5735
β, η or ζ
1.08 R² = 0.9929
1.02
0.96
0.90
0.45 0.55 0.65 0.75 0.85
Brittleness Number (B)
Figure 3.12 Variation of Normalized Peak Load, LLD and CMOD with Brittleness Number
113
100.0
Interior Rigid plate
Edge Rigid plate
Edge Point displacement 6
Interior Point displacement 8
80.0 5
7
3
2 4
60.0
Applied Load (kips)
40.0
20.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15
Figure 3.13 Slab Responses under Different Loading Scenaria Applied at Interior and Edge
114
0.015
Point 1 Point 2
Point 3 Point 4
Point 5 Point 6
Point 8 Critical displacement
0.010
CMOD/2 (in.)
0.005
wf/2
0.000
0 20 40 60 80 100 120
Half of Slab Length (in.)
0.015
Point 1 Point 2
Point 3 Point 4
Point 5 Point 6
Point 7 Critical displacement
Point 8
0.010
CMOD/2 (in.)
0.005
wf/2
0.000
0 10 20 30 40 50 60
Half of Slab Width (in.)
Figure 3.14 Crack Growth Along Slab Length and Width at Points Shown in Fig. 3.13
115
80.0
UAB E=25 ksi No base
CTB E=1500 ksi
70.0
60.0
50.0
Applied Load (kips)
40.0
30.0
20.0
10.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15 0.18 0.21
Figure 3.15 Effect of Base Stiffness on Slab Response
116
140.0
JOINTC Ioannides et al (2006)

COH3D8 This study
120.0
100.0
Applied Load (kips)
80.0
60.0
40.0
20.0
0.0
0.000 0.003 0.006 0.009 0.012
Crack Mouth Opening Displacement/2 (in.)
Figure 3.16 Comparison with Numerical Results by Ioannides et al. (2006)
117
12.0
This study
Gaedicke and Roesler (2009) - Experiment
10.5
Gaedicke and Roesler (2009) - Finite Element
9.0
7.5
Applied Load (kips)
6.0
4.5
3.0
1.5
0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Figure 3.17 Comparison with Numerical and Experimental Results by Gaedicke and
Roesler (2009)
118
Chapter 4 Curling Effects on Concrete Pavement Fracture
Abstract
This paper discusses finite element analysis of the behavior of pavement slabs-on-grade
undergoing cracking due to temperature and wheel loads. Traction-separation cohesive elements
are used to simulate the fracture process according to the fictitious crack model. To verify the
proposed discretization, pre-crack analysis of the slab subjected to curling alone, as well as to
curling combined with a wheel load, is carried out first, and the results are compared with
previous finite element and analytical solutions available in the literature. The three-dimensional
analysis presented herein produces lower curling stresses compared to earlier two-dimensional
calculations, but excellent agreement is confirmed with previous three-dimensional simulations.
The investigation is subsequently extended to the post-cracking stage, for which the effects of
linear and nonlinear temperature distributions, notch depth, slab size, slab self-weight, and
concrete age are investigated. All these factors are found to influence cracking due to curling,
except for slab self-weight, which is observed to be less significant for this loading stage than
before cracking. Findings presented elucidate concrete pavement fracture under wheel load and
curling. The procedures adopted may be extended to more complex in situ pavement systems
incorporating load transfer, thereby addressing the limitations in current pavement design
procedures that exclusively rely on statistical algorithms for the prediction of pavement
distresses.
119
4.1 Introduction
The first analytical attempt to address the problem of concrete pavement curling was due to
Westergaard (1927), who provided the well-known equations for the critical responses at the slab
edge and interior. In the derivation of these equations, Westergaard employed grossly restrictive
assumptions, including infinite slab self-weight, linear temperature distribution through the slab
thickness, and applicability of the principle of superposition when combining temperature-
related and load-induced stresses (Ioannides et al., 1999). Recognizing the importance of thermal
stresses, subsequent studies have endeavored to eliminate Westergaard’s assumptions,
particularly following the development of computer programs implementing two-dimensional
(2D) finite element (FE) analysis (Huang and Wang, 1974; Tia et al., 1987; Korovesis, 1990;
Khazanovich, 1994). Application of such curling investigations, however, has pertained solely to
the pre-crack aspects of slab behavior, in view of the limitations of 2D analysis.
More recently, three-dimensional (3D) FE analysis has attracted the attention of pavement
engineers, who have used either commercially available packages, such as ABAQUS ®(Kuo,
1994; Masad et al., 1996; Pane et al., 1998), ANSYS (Chen et al., 2002; Mahboub et al., 2004;
Dere et al., 2006; Siddique et al., 2006), and LS-DYNA (Shoukry et al., 2007), or stand-alone 3D
FE codes developed specifically for pavements, such as EverFE (Davids et al., 1998) and
Pave3D (Nishizawa et al., 2001; Shimomura et al., 2008). The most promising application of 3D
FE analysis is probably in the realm of post-fracture pavement behavior, and this has already
prompted a number of independent, yet complementary investigations of pavement slabs
subjected to wheel load acting without curling (Barros and Figueiras, 2001; Meda et al., 2004;
Ioannides et al., 2006; Gaedicke and Roesler, 2009; Aure and Ioannides, 2012). Such efforts,
120
however, have yet to be extended to the consideration of thermal curling, either by itself or in
combination with the wheel load. The only research effort in this direction has been that of
Channakeshava et al. (1993), who considered a dowel-jointed concrete pavement under wheel
load and linear temperature gradient, and who employed a smeared crack approach and a plastic
material constitutive relation. The latter two features are not pursued in this thesis, and this
precludes further consideration of this pioneering publication.
The main objective of this paper is to investigate the post-cracking behavior of a concrete
pavement slab subjected to both thermal curling and static wheel load. The concepts of nonlinear
fracture mechanics (NLFM) are employed, as incorporated in traction-separation cohesive
elements, recently implemented in a general purpose FE package, ABAQUS® (ABAQUS, 2009).
The presentation below is organized as follows. The proposed FE discretization is validated first
by application to the elastic (pre-crack) stage of the pavement slab under curling alone, as well as
curling combined with wheel load. The simulation is subsequently extended to the corresponding
post-cracking behavior.
4.2 Finite Element Discretization
The present study utilizes general purpose FE program ABAQUS/STANDARD® version 6.9-2 for
the analysis of pavement slabs-on-grade subjected to wheel load and thermal curling. Unless and
otherwise stated, the slab considered is 240 in. long, 120 in. wide and 6 in. thick. Typical
concrete material properties are: Young’s modulus, E, of 4,000 ksi; Poisson’s ratio, μ, of 0.2;
concrete unit weight, γ, of 150 pcf; and coefficient of thermal expansion, α, of 5×10-6/oF. The
modulus of subgrade reaction, k, was taken as 200 pci; this slab was previously analyzed by the
authors (Aure and Ioannides, 2012) considering wheel load alone.

121
The fracture plane is anticipated perpendicular to the length direction of the slab, in accordance
with a discrete crack approach. This is deemed a reasonable representation for pavement slabs
loaded either at the edge or at the interior, since the fracture plane is anticipated in the direction
of the maximum stress. Moreover, Mode I or opening mode is assumed, since the contribution of
the shear stresses is expected to be negligible. Further discussions on the FE discretization of the
entire pavement system is given in the following sections.
4.2.1 General Considerations
Among available options for discretizing a slab-on-grade system, previous researchers had used
either the 20-node quadratic element (Pane et al., 1998; Chen et al., 2002; Mahboub et al., 2004;
Siddique et al., 2006) or the reduced integration 27-node element (Kuo, 1994; Hammons, 1997;
Ioannides et al., 2006). The main disadvantage of the 20-node element is the lack of mid-face
nodes, which are helpful in specifying the temperature variation through the slab thickness. On
the other hand, the reduced integration 27-node element is often found to produce spurious
deformations at unrestrained nodes, particularly when the slab loses contact with the subgrade
due to curling (ABAQUS, 2009). Consequently, the 27-node full integration element, available in
ABAQUS® under the name C3D27, is adopted for the analyses presented below. A uniform
mesh pattern is used for the pavement slab, with 6×6×2 in. elements (length×width×thickness,
respectively).
The subgrade is idealized using SPRING1 elements supporting only compression. The force-
displacement relationship for each element of this type is hand-calculated with reference to the
tributary areas surrounding its node, and is provided in the input file as a table. For numerical
stability, lateral restraint is provided using additional SPRING1 elements, whose horizontal
122
stiffness is set to 1/10th of the value of the vertical springs, per Gaedicke and Roesler (2009).
This may be justified as simulating friction between the slab and the subgrade.
Temperature variation through the slab thickness is idealized by assigning nodal values. This is
accomplished using the keyword TEMPERATURE. It is assumed that the temperature remains
identical at all nodes lying on the same horizontal plane of the slab. When needed, a wheel load
is applied as a point displacement at two nodes on either side of the fracture plane at (x, y, z) =
(3, 6, 6) in. from the symmetry line at the edge, as explained further below.
4.2.2 Fracture Process Zone (FPZ)
To minimize resources expended, the fracture process zone (FPZ) and the bulk uncracked
material are discretized independently. The FPZ is idealized using traction-separation cohesive
elements (COH3D8) of finer mesh size of 0.12×0.12×0.001 in. This is connected to the much
coarser mesh of C3D27 elements representing the intact mass using surface-based TIE
CONSTRAINTS, which enforce identical boundary conditions at the nodes that lie on the
interface.
The cohesive elements are inserted along the anticipated fracture plane of the slab, consistent
with a discrete crack approach; this is deemed to be a reasonable assumption for pavement slabs
loaded at the edge, and it is justified by the commonly observed in situ crack patterns. Moreover,
Mode I fracture is assumed, since the contribution of shear stresses is expected to be negligible.
The fracture process can be distinguished into pre- and post-crack initiation phases, depending
on the stress level developing in the FPZ, as detailed in two earlier publications pertaining to
cracking in a simply supported beam, as well as an isolated slab-on-grade under static load (Aure
123
and Ioannides, 2010; Aure and Ioannides, 2012). The penalty stiffness and viscous regularization
techniques used in those earlier studies are retained here, as well. Moreover, linear and bilinear
concrete softening is retained for the curling-only and combined curling-with-wheel load cases,
respectively, chiefly on the grounds of expediency. A complete discussion of the softening
options available is provided by Aure and Ioannides (2012).
The responses monitored are as follows: longitudinal horizontal stress through the slab thickness
(σxx); load line displacement (LLD), i.e., the vertical displacement at the center of the loaded area
to the right of the fracture plane; crack mouth opening displacement (CMOD), i.e., the horizontal
displacement at the top or bottom of the FPZ; and the total applied load (P), i.e., the sum of the
conjugate “reaction” forces at the two loaded nodes, RF3.
4.3 Pre-Crack Analysis of Slab under Thermal Curling Alone
As a prelude to the study of fracture behavior, pre-crack elastic analysis of the slab subjected to a
temperature distribution alone is carried out first, and the results are compared with available
analytical, experimental and FE results found in the literature. This initial step is considered
essential for assessing the robustness of the proposed FE discretization of the intact slab mass
and the subgrade.
4.3.1 Comparison with 3D FE Analysis
In the aforementioned 3D FE study by Pane et al. (1998), a 30-ft long, 24-ft wide, and 9-in. thick
slab was subjected to a temperature distribution described by a cubic polynomial, per Thompson
et al. (1987) and Mohamed and Hansen (1996). The temperature profile at 3 pm is re-examined
in the present study, retaining the following cubic variation through the slab thickness:
124
T ( z)  1.4762 - 0.5291z  0.2685z 2 - 0.0514 z 3 (4.1)
Here, z = 0 denotes the mid-depth of the slab and z = h/2 its bottom, where h is the thickness of
the slab. Typical material properties are retained from Pane et al. (1998): Young’s modulus, E, of
4,000 ksi; Poisson’s ratio, μ, of 0.15; concrete unit weight, γ, of 150 pcf; and coefficient of
thermal expansion, α, of 5×10-6/oF. The modulus of subgrade reaction, k, was taken as 100 pci.
The variation of the longitudinal stress σxx (positive if tensile) through the thickness of the slab
ocurring at the slab center is shown in Fig. 4.1. It can be verified that the 3D FE results obtained
in this study agree well with those reported by Pane et al. (1998), despite differences in the type
and number of elements used.
As an additional comparison, the same slab is analyzed in this study using 3D FE software
EverFE v. 2.24 (Davids et al., 1998). The latter can only accommodate a trilinear temperature
distribution, and this is easily extracted from Eq. (4.1) by dividing the slab thickness into three
equal segements. Surprisingly, the simulation results plotted in Fig. 4.1 disclose that EverFE
overestimates the stress at the bottom of the slab (z = +4.5 in.) by about 15%. This may be due to
the linearization of the temperature profile, differences in element types used as well as
numerical solution strategies implemented in the two softwares.
The 3D FE simulation scheme adopted in this study is also applied in reproducing
experimentally measured curling stresses reported by Teller and Sutherland (1935) on a concrete
pavement slab 20-ft long, by 10-ft wide, and 6-in. thick. Temperature differential values (ΔT)
along with corresponding curling stresses had been provided (measured between April 18 and
May 19, 1934). Kuo (1994) adopted a linear temperature distribution in his own 3D FE
125
reproduction of the measurements, and assumed Young’s modulus of 5000 ksi, Poisson’s ratio of
0.15, concrete unit weight of 0.083 lb/in.3, and modulus of subgrade reaction of 200 pci. Table
4.1 presents a three-way comparison of the center bending stresses from these two previous
investigations, as well as from the present study. It is observed that the 3D FE idealizations
agree with one another, yet overestimate the measured curling stresses, typically by about 10%.
This discrepancy could be the result of the relatively coarse meshes adopted, as well as the fact
that “the observations extended over a considerable period of time… [and] the temperature
differentials [and therefore the corresponding stresses] were typical of the highest average
values”, per Teller and Sutherland (1935). Nonlinearity in the in situ temperature distribution, as
well as the existence of any so-called built-in upward curling arising during the construction
process could also contribute to the observed differences (Eisenmann and Leykauf, 1990).
4.3.2 Comparison with 2D FE Analysis and Westergaard (1927)
Choubane and Tia (1992) monitored pavement temperatures in six slab-on-grade pavements
using thermocouples embedded at the center of one of the slabs at 1.0, 2.5, 4.5, 6.5, and 8.0 in.
below its top surface, and reported that the variation of the temperature profile was nonlinear, for
which the following simple quadratic form was suggested:
T= A + Bz + Cz2 (4.2)
where: A, B, and C are coefficients that can be determined from the measured data; T is the
temperature in degrees Fahrenheit; z is the slab depth, with z = 0 at the top and z = h at the
bottom; and h is the slab thickness. Note that Choubane and Tia (1992) adopted the quadratic
assumption merely for the sake of expediency, even though several researchers caution that the
126
actual temperature variation may require a higher order polynomial (Teller and Sutherland, 1935;
Thompson et al., 1987).
Each slab was 20-ft long, 12-ft wide and 9-in. thick. Choubane and Tia (1992) conducted a 2D
FE analysis of one of the slabs using the program FEACONS-IV and compared the results with
the theoretical solution by Westergaard (19927), which they misguidedly attributed to Bradbury
(1938). The following material properties were assumed: Young’s modulus of 4,500 ksi;
Poisson’s ratio of 0.2; coefficient of thermal expansion of 6×10-6 /°F; and unit weight of 150 pcf.
The modulus of subgrade reaction was taken to be 300 pci.
The same slab was re-examined using the 3D FE simulation of the present study, retaining the
nonlinear temperature profiles reported for the month of June. The corresponding coefficients for
Eq. (4.2) are shown in Table 4.2.
Figures 4.2 and 4.3 show the temperature variation, in °F, and the computed longitudinal
bending stress, σxx, respectively, at the slab center through the slab thickness for the quadratic
distribution considered in this study between 9 am and 8 pm (called daytime). Similarly, Fig. 4.4
and 4.5 show the temperature profile and the corresponding longitudinal bending stress, σxx,
respectively, for the nighttime (10 pm to 6 am). As expected, the stress profiles reflect the
quadratic shape of the temperature distribution; any kinks observed in Fig. 4.3 and 4.5 are
attributable to the relatively coarse mesh employed. Almost without exception, the critical
bending stress occurs at the slab’s extreme fibers, where it may trigger either bottom-up or top-
down cracking. Nonetheless, at 9:00 am, maximum tension occurs at about 3 in. from the
bottom, which suggests that thermal cracks may in fact initiate within the slab mass. The overall
maximum tensile stress is observed at 3:00 pm, which may be the time of traffic congestion, as
127
well. Such an unfavorable combination of traffic and thermal loads will certainly aggravate the
formation of pavement cracks.
The bending stresses at the top (σt) and at the bottom (σb) of the slab computed by 3D nonlinear
curling analysis are compared with those from an additional 3D run of the corresponding linear
distribution using the temperature differential (ΔT) between the top and bottom of the slab given
in Table 4.2. Results are shown in Table 4.3. It is observed that the linear profile overestimates
the bottom-fiber stresses and underestimates the top-fiber stresses, except at 3:00 pm and 8:00
pm. The off-trend characteristic exhibited at these two times can be explained by inspecting
more closely the temperature profiles shown in Fig. 4.2 and 4.4. During the daytime, it can be
seen that the temperature profile is convex to the right, except at 3:00 pm, when it exhibits an
opposite curvature. The latter gives rise to lower temperatures according to the linear profile as
compared to the nonlinear one. At 8:00 pm, the temperature is almost identical at the top and
bottom. Consequently, tensile stresses of almost the same magnitude are obtained at the top and
at the bottom for the nonlinear temperature profile. At some other points, however, compressive
stresses are observed. This observation has significant consequences for the slab post-crack
response, which will be shown later to be influenced by the overall stress distribution through the
thickness of the slab.
The 3D FE results obtained in this study for the linear distribution are also compared in Table 4.4
to those reported by Choubane and Tia (1992), from their 2D FE analyses. For the sake of
completeness, the results from the analytical solution by Westergaard (1927) are also presented.
It is observed that the present 3D FE idealization generally gives results that are lower than those
reported from the 2D analysis, discrepancies ranging between 2 to 25%. A similar trend is also
128
observed when comparison is made with the analytical result. These discrepancies can be partly
explained by the relatively coarse 3D mesh size adopted in this study. Another cause may be the
theoretical differences between 2D plate theory and 3D continuum FE formulations, as inferred
from the closeness of the 2D FE result to that of Westergaard. It is recalled that Westergaard’s
“special theory” is an effort to recapture some of the lost strain energy in the plate-on-dense
liquid idealization (Ioannides et al., 1985)
For the nonlinear temperature distribution, the stresses obtained at the top and bottom using the
present 3D FE analysis are compared in Table 4.5 to those reported by Choubane and Tia (1992)
as resulting from a combination of 2D FE analysis and their proposed analytical solution for
nonlinear curling. As for the linear distribution, the 3D FE output records smaller bending
stresses than the 2D FE result (in the range of 4 to 40%), except at 8:00 pm, presumably for the
same reasons explained above.
4.4 Pre-Crack Analysis of Nonlinear Curling-and-Wheel Load
A pre-crack 3D FE discretization similar to that described in the previous section is also
employed here for the case of a nonlinear temperature distribution combined with a wheel load.
The 3D FE analysis of the slab under both actions is achieved by splitting the solution process
into two steps, namely temperature and wheel loading, per Mahboub et al. (2004). In the
temperature step, the responses due to curling and self-weight are computed, whereas in the
second step the effect of the wheel load on the already temperature-stressed slab is assessed. A
general nonlinear Newton-Raphson solution technique is employed (Aure and Ioannides, 2012)
to capture the geometric nonlinearities caused by curling and the tensionless subgrade.
129
The proposed ABAQUS® 3D FE simulation is first applied to a 12-ft wide, 15-ft long, and 9-in.
thick concrete pavement slab resting on a 200 pci subgrade, previously investigated by
Khazanovich (1994) using the 2D FE code ILSL2. Typical material properties are retained:
Young’s modulus of 4,000 ksi; Poisson’s ratio of 0.15; and coefficient of thermal expansion of
5×10-6/°F. The slab is loaded by a wheel load of P = 10,000 lbs, at a pressure p = 100 psi
applied at its corner in addition to its self-weight of 150 pcf. The temperature distribution
reported by Richardson and Armaghani (1987) in the month of August at Gainesville, FL, is
adopted. A quadratic temperature profile per Eq. (4.2), with coefficients for the daily cycle given
in Table 4.6, is assumed.
Simulation runs are made for the temperature profiles given in Table 4.6, in which the principal
bending stresses at the top and bottom of the slab are recorded for two scenaria. The first set of
results is termed the combination approach and corresponds to the proposed FE solution, in
which the application of the temperature distribution is followed by the wheel load applied on
the deformed shape of the slab. The second scenario is termed the superposition approach,
according to which individual responses from each of the two actions are simply summed up. For
this purpose, the 3D FE run of the slab under the wheel load alone is conducted and the
maximum principal bending stress at the top of the slab is found to be about 0.245 ksi, and
occurs at 28.69 in. from the corner of the slab. Table 4.6 indicates that the two approaches
generally yield similar results, except at noon and 4 pm, when the top is decidedly warmer than
the bottom. At these times, superposition leads to gross overestimates of the maximum bending
stress: 175% and 155% of the corresponding combination stresses, respectively. The discrepancy
is attributed to the difference in boundary conditions in the two scenaria, exacerbated by the
130
application of a relatively light load on a relatively stiff subgrade. The adequacy of
superposition, therefore, cannot be taken for granted, confirming previous observations in the
published literature (Ioannides and Salsilli-Mura, 1989). The overall largest diurnal maximum
principal bending stress is observed at midnight, presumably because the critical stress due to
both corner loading and nighttime temperature distribution occurs at the top.
Results from the proposed ABAQUS® simulation are compared in Fig. 4.6 to those reported by
Khazanovich (1994) on the basis of 2D FE results using program ILSL2, and to those from the
3D FE program EverFE v. 2.24 executed during this study. Similar trends are exhibited by all
three software. It is observed, however, that EverFE tracks the ABAQUS® results more closely
than ILSL2, especially at the slab top. Comparing stress values at any specified time may point
to discrepancies sometimes exceeding 50% (see Table 4.7), and may lead to misleading
conclusions as to the repercussions of using each particular analytical tool. The sensitivity of the
results to mesh refinement in each case should be carefully assessed in the context of the
respective theory employed, i.e., continuum or plate.
The discrepancies between 2D and 3D FE results certainly warrant further exploration,
preferably employing carefully measured field responses. For the purposes of this study,
however, the 3D FE formulation proposed is judged to be consistent with reliable state-of-the-art
efforts, and is, therefore, suitable for the investigations presented in the remainder of this paper.
4.5 Fracture Response of Pavement Slab Subjected to Thermal Curling Alone
Cracking in concrete pavement slabs may occur under thermal curling alone especially at an
early age, when the slab exhibits lower material strength. Using 2D FE methods, Jenq et al.
131
(1993) studied the effect of temperature on the formation of cracks in concrete pavement slabs at
the age of 16 hours, 24 hours, and 36 hours. They investigated the effect of notch depth, a,
ranging from 0 to 6 in.; and slab thickness, h, ranging from 6 to 12 in. The notch is formed at the
top of the slab, just as saw-cutting is conducted in the field; consequently, a nighttime
temperature differential is considered, which causes maximum curling stresses at the top, and
encourages a crack beneath the notch to propagate to the slab bottom. They reported that the
notch depth and saw-cut timing influenced significantly the critical temperature resisted by the
slab. For a 6-in. thick unnotched slab of age 16 hours, for example, a nighttime (cooler top)
temperature differential of about 30°F would cause random thermal cracks, and, therefore, a cut
should be made earlier. The 2D FPZ idealization adopted by Jenq et al. (1993) entailed node-by-
node crack progression. The applied temperature resisted may, therefore, have been
overestimated, whereas the corresponding crack mouth opening displacement may have been
underestimated.
In this study, the material properties provided by Jenq et al. (1993) as shown in Table 4.8, are
retained in an investigation of the fracture behavior of a concrete pavement slab, 240-in. long,
120-in. wide, and 6-in. thick, under thermal curling alone. The subgrade reaction and the
concrete unit weight are assumed to be 200 pci and 150 pcf, respectively. A linear concrete
softening curve is used in this section for simplicity (Aure and Ioannides, 2012). Consequently,
the FPZ damage initiates when the tensile strength is attained; thereupon the cohesive stress that
resists cracking decreases linearly until all the fracture energy is consumed. In order to capture
accurately the post-peak characteristics of the crack mouth opening versus applied temperature
differential and profile responses, the modified Riks solver is employed (Aure and Ioannides,
132
2010). The effects of notch depth, concrete age, self-weight, nonlinearity of temperature profile,
and slab size on the fracture process are discussed below.
4.5.1 Effect of Notch Depth
Joints are cut in concrete pavement slabs in order to avoid random cracks that may result from
thermal stresses at an early age. The depth and width of the cut and its timing are very important
in controlling the extent of the cracks (Jenq et al., 1993; Soares, 1997). In this study, the effect of
notch depth, a, is investigated by adopting the material properties of concrete at the age of 16
hours given in Table 4.8. Four different notch depths of 0, 1/2, 2, 3 in., are considered,
corresponding (for the 6-in. slab adopted) to notch depth-to-slab thickness ratios (a/h) of 0, 1/12,
1/3, and 1/2, respectively. Cracks are assumed to propagate from top to bottom as a nighttime
(cooler top) linear temperature differential is applied, toward a specified maximum temperature
difference of 54°F.
Figure 4.7 shows the variation of applied temperature differential (ΔT) versus CMOD. As might
be expected, as the notch depth increases, the critical temperature differential supported by the
slab decreases. Stable cracks are formed until the critical temperature differential is attained
(e.g., 33°F for the unnotched case); this is followed by softening, signaling the complete damage
of the cohesive elements. Once the cohesive elements along the assumed fracture plane are
damaged completely, the slab is divided into two essentially independent segments and the
location of the maximum thermal stress moves toward their respective centers. If the temperature
differential exceeds the values indicated in Fig. 4.7, unstable cracks will form and the slab would
fail at the age of 16 hours.
133
The results from this study are also compared with those from Jenq et al. (1993) in Fig. 4.8, in
which a similar trend is exhibited. The discrepancies between the two responses may be
attributed to differences in the respective simulations, including in the idealization of the FPZ,
the linear versus quadratic temperature profiles employed, the inclusion of a base layer and the
adoption of an elastic foundation by Jenq et al. (1993), as well as mesh fineness. The effect of
the temperature profile is pursued further in subsequent sections.
4.5.2 Effect of Concrete Age
The effect of concrete age on the fracture process is examined at 16 hours, 24 hours, and 36
hours, through increases in the concrete Young’s modulus, tensile strength and fracture energy,
as measured in the laboratory by Jenq et al. (1993). A constant notch depth of 0.5-in. at the top
of the slab is retained here along with the nighttime (cooler top) linear temperature differential
specified above. Results in Fig. 4.9 indicate that, as expected, the critical temperature differential
resisted by the slab increases significantly with age, reflecting concrete maturation.
Concomitantly, the corresponding CMOD decreases slightly with age. It is also noted that in the
post-peak or softening portion of the response curves, a much more pronounced reduction in the
temperature differential resisted is experienced as concrete maturity increases. This is
attributable to the increased fracture energy with age (see Table 4.8), which results in slower
damage in the FPZ, exhibited as a smaller (fracture energy-to-temperature differential loss) ratio.
The effect of age on the critical linear temperature differential resisted by the slab may be
examined using dimensionless parameters αΔT, and B, in which: α is the coefficient of thermal
expansion, a concrete material characteristic assumed to remain constant over time; ΔT is the
134
critical temperature differential between the bottom and top of the slab; and B is the brittleness
number, defined by Bache and Vinding (1990) as:
ft 2 h h
B  (4.3)
EGF lch
where lch is a characteristic length of the material; f t ' is the tensile strength; E is Young’s
modulus; GF is the fracture energy; and h is the slab thickness. The result is plotted in Fig. 4.10,
where it is observed that as the brittleness number increases, the maximum linear temperature
differential resisted by the slab increases almost linearly. The corresponding results from Jenq et
al. (1993) are also plotted in Fig. 4.10. Differences between the two simulations outlined above
notwithstanding, a similar trend are observed in their results, inviting further exploration of
dimensional analysis for interpretation of post-crack responses in pavement slabs.
4.5.3 Effect of Slab Self-Weight
Concrete self-weight can exacerbate curling stresses and the formation of discrete cracks. For the
sake of comparison, consider the same 6-in. concrete slab as in previous sections, with four
different assumed unit weights of 37.5, 75, 150, and 300 pcf, at 16 hours of age. The slab has a
0.5-in. top notch, and is subjected to the nighttime (cooler top) linear temperature differential
specified above. As the slab curls upward, the portion of the slab that loses contact with the
subgrade is acted upon by its self-weight, in a cantilever action with the support in the FPZ,
contributing to curling stresses. Consequently, lower curling stresses develop in slabs of a
smaller self-weight, and such lighter slabs fail at a higher critical temperature differential, as
observed in the results presented in Fig. 4.11. Nonetheless, this effect is not significant for
typical concrete unit weight values (e.g., between 50% and 200% of the commonly assumed self-
135
weight of 150 pcf). Similar observations are anticipated for a daytime (warmer top) temperature
differential, as well, since the latter has an analogous effect, except for causing bottom-up
cracking. For their part, Jenq et al. (1993) did not study the effect of concrete self-weight.
4.5.4 Effect of Nonlinearity of Temperature Distribution
In the pre-fracture analysis of curling stresses presented above, it was found that the bending
stress distribution through the slab thickness depends on the profile of the temperature
distribution. In some instances, it was observed that a linear temperature variation results in
higher top or bottom bending stresses than the corresponding nonlinear distributions considered.
To study this effect in the aftermath of crack initiation, the 6-in. pavement slab at 16 hours of age
with 0.5-in. notch at the top is considered again. Three nighttime (cooler top) temperature
profiles are investigated: a linear; a quadratic with downward curvature; and a quadratic with
upward curvature, as shown in Fig. 4.12 (inset). Note that in all three cases, the applied
temperature differential between the bottom and top remains at 54°F. The responses shown in
Fig. 4.12 reflect the pattern displayed by the temperature distributions, inasmuch as the linear
one lies in-between its two quadratic counterparts. The results also indicate the sensitivity of slab
response to the entire temperature profile and not merely to the temperature differential, retained
as constant in these analyses. Moreover, recalling the results in Figs. 4.2 through 4.5, it is
asserted that pavement slab response here reflects the entire distribution of curling stresses, and
not merely the extreme-fiber values. Slab failure results from the dissipation of fracture energy
through the entire slab thickness. Such dissipation depends on the overall prevailing stress state
and not merely on the extreme-fiber stresses. The presumed conservative nature of the
commonly used linear temperature distribution in concrete pavement analysis that only considers
136
extreme-fiber stresses is, therefore, called into question. This is especially true at an early age,
when temperature variation is the sole cause of cracking; the degree to which the same
observation also applies once the concrete hardens and the pavement is opened to traffic will be
taken up again in Section 4.6, below.
4.5.5 Effect of Slab Size
In his pioneering paper on curling, Westergaard (1927) demonstrated that pre-crack curling
stress increases almost linearly as slab size increases, until a dimensionless slab size, L/l or W/l,
of about 8 is reached, beyond which stresses remain nearly constant. Here, L denotes the slab
length, W the slab width and l the radius of relative stiffness of the slab-subgrade system. In this
study, the effect of slab size on the post-crack curling response is investigated by considering
once again the 16-hour old 6-in. slab, with a 0.5-in. top notch and nighttime (cooler top) linear
temperature variation scenario. The following sizes are considered: L = 120, 180, 240, 360, and
480 in., keeping W = 120 in.; W = 120, 150, 180, and 240 in., keeping L = 240 in.
The dimensionless critical temperature differential, αΔT, for each case is plotted in Fig. 4.13, as a
function of the dimensionless length and width, L/l and W/l. It is observed that as slab length
increases, the critical temperature differential drops drastically, until L/l reaches 10. Beyond L/ l
= 10, however, the response remains almost unaffected, with only a slight increase of the critical
temperature differential after about L/l =13, hearkening back to Westergaard’s own observations.
On the other hand, for the particular cases considered in the present study, increasing the slab
width leads to critical temperature differentials that decrease slightly in a linear manner. Note
that the FPZ is assumed to extend along the shorter (width) dimension of the slab.
137
For their part, Jenq et al. (1993) simulated the same slab supported on 9-in. base of modulus 50
ksi resting on elastic solid foundation of modulus 30 ksi, underlain by a rigid base. Their FE
analysis employed plane strain conditions, i.e., their implicit slab width is infinite. They retained
slab length of 24 ft and considered thicknesses of 6, 9, and 12 in., reporting the attained ΔT value
in each case. The complete data were reported in Liaw (1992), and are also plotted in Fig. 4.13
(assuming k = 500 pci), in which they appear to confirm the trends obtained in this study.
From the preceding investigations of slab fracture due to curling alone, the following conclusions
can be reached:
a. Thermal cracking is significantly affected by the age of the concrete, which dictates the time
and size of joint sawing. The dimensionless brittleness number, B, can be used to
characterize the effect of concrete age on the fracture process;
b. Surprisingly, the slab self-weight is observed to be unimportant with regard to the fracture
process under curling alone, except possibly for very light-weight materials;
c. The complete temperature profile through the slab thickness exerts a significant influence on
the fracture process under curling alone, and evinces the importance of the entire stress
distribution rather than of merely the extreme-fiber stresses. This result calls for the accurate
determination of pavement temperature profiles;
d. The fracture process due to curling is particularly sensitive to slab sizes below L/l of 10, at
least for the cases considered in the present study.
4.6 Fracture Response of Slab Subjected to Wheel Load and Thermal Curling
In this section, the post-crack behavior of a pavement slab subjected to both thermal curling and
a wheel load is investigated for the first time in the published literature. Geometry and material
138
properties for the slab are selected to be consistent with earlier analyses conducted by Ioannides
et al. (2006) and Aure and Ioannides (2012); these are provided in Table 4.9. The FE
discretizations for both the fracture process zone and the bulk (intact) material remain identical
with those in Section 4.3, above.
When examining post-crack behavior, a wide range of options is available with regard to the
combined curl-plus-load condition. Two extreme situations are easily envisaged: (a) a constant
thermal gradient is applied in combination with an increasing applied wheel load; (b) a constant
applied load exists in combination with an increasing thermal gradient. The actual field situation
is probably somewhere between these two extremes, with the proportion of stress contributed by
each of the two factors being in a state of constant flux. Only the two extreme combinations
among the host of possibilites will be pursued below.
4.6.1 Slab under Constant Thermal Gradient and Increasing Wheel Load
In this situation, the two-step pre-crack FE procedure adopted by Mahboub et al. (2004) is
retained. In the first step, the constant temperature distribution is applied along with the slab
self-weight, whereas in the second step, the increasing wheel load is additionally applied. In the
discretization proposed herein, measured nonlinear temperature profiles are considered. The
temperature distributions reported by Thompson et al. (1987) and by Choubane and Tia (1992)
cannot be used, since in both those cases the slabs were thicker. Consequently, the measured
temperature profiles for a 6-in. slab reported by Teller and Sutherland (1935) were adopted.
Diurnal and monthly variations are available. In this study, the profiles for July 12 and 13, 1932
are selected, and these are discretized using a cubic polynomial, so that temperature values may
be assigned at each node in the FE mesh. The equation of the cubic polynomial is of the form:
139
T  Az 3  Bz 2  Cz  D (4.4)
where z = 6 in. at the bottom of the 6-in. thick slab and z = 0 at its top. Coefficients, A, B, C, and
D of the polynomial, are given in Table 4.10. For comparison purposes, temperature differentials
corresponding to each hour are also listed. The wheel load is applied as an increasing point
displacement placed symmetrically at (x, y, z) = (3, 6, 6) in., per Fig. 4.14. Post-crack responses
are tracked, as noted in Section 4.2.2, above. These load-plus-curl results are compared to the
corresponding load-only outputs from Chapter 3 (Aure and Ioannides, 2012) in order to assess
the influence of introducing thermal regime considerations. More specifically, the sensitivity of
post-fracture response to the diurnal cycle, to the nonlinearity of the temperature profile, and to
the geometric and material properties of the system is investigated in the following sections.
a. Effect of Daytime and Nighttime Variations on Fracture
The FE simulation is performed for the constant temperature profiles described in Table 4.10
combined with an increasing applied load. Selected results are presented in Fig. 4.15, along with
previously published data by Aure and Ioannides (2012) for the no-curling (i.e., wheel load only)
case. During the daytime, at 9:30 am, the peak load resisted by the slab is approximately 50% of
the value obtained under wheel load alone (8 kips versus 18 kips). This is explained by the fact
that a significant portion of the bending strength of the slab is consumed by thermal stresses,
prior to the application of the wheel load. The reverse is observed at 7:00 pm, when the peak
sustainable load reaches more than 200% of the curl-free condition (50 kips versus 18 kips).
During the daytime the slab curls down and acts as a plate supported solely along its boundaries,
since only its periphery remains in contact with the subgrade. Unstable cracks form, followed by
a pronounced softening effect. During the nighttime, on the other hand, as the slab curls up, a
140
very substantial portion of its bottom surface remains supported by the subgrade. This condition,
combined with curling stresses that counteract the loading stresses, result in stable crack
formation, followed by only a relatively mild capacity reduction during softening. It is
interesting to note in Fig. 4.15 that during the night, the bulk of the wheel load is applied after
crack initiation (manifested in an abrupt slope change at load level 1), whereas during the day,
the peak load is reached shortly after the onset of cracking, at load level 1. This also evinces the
greater stability of fracture during the nighttime as compared to the daytime.
Figures 4.16 and 4.17 show the variation of crack mouth opening displacement across the bottom
of the width of the slab along the assumed fracture plane. The curves correspond to specific load
levels of interest as already indicated in Fig. 4.15, for the temperature profiles reported at 9:30
am and 7:00 pm. Also plotted is the failure displacement, w f , i.e., the CMOD at which the
traction stress reaches zero. At 9:30 am, Fig. 4.16 shows that about 40% of the slab (45 to 55 in.)
width has experienced bottom cracking at load levels 1 and 2. The failure displacement has
exceeded over more than half the cracked portion (35 to 45 in.). By load level 3, the slab has
completely failed, since the cracks along the entire width of the slab exceeded the failure
displacement; evidently, any additional load beyond load level 3 is carried by the subgrade.
For the 7:00 pm temperature profile, Fig. 4.17 indicates that the failure displacement is not
exceeded at all at load level 1, suggesting that the observed cracks are stable and cohesive
elements are still offering support. As the load increases, unstable cracks begin to form. At load
levels 2 and 3, for example, more than 55% and 90%, respectively, of the width of the slab
exhibits crack mouth opening displacements greater than the failure displacement. By load levels
4 and 5, the entire slab width has experienced crack mouth openings greater than the failure
141
displacement, indicating that cohesive elements no longer offer resistance to the applied load.
Comparing crack growth at the two peak load levels (load level 2 at 9:30 am and load level 3 at
7:00 pm), it is apparent that crack mouth opening displacement is much larger at 7:00 pm than at
9:30 am. For example, at 30 in. from the origin, the CMOD is 0.035 in at 7:00 pm (Fig. 4.17,
load level 3) compared to 0.006 in. at 9:30 am (Fig. 4.16, load level 2). Therefore, it may be
concluded that although the slab exhibits a substantially higher strength during the nighttime, its
concomitant larger crack mouth openings may lead to other unforeseen problems, such as
pumping. It is envisaged that this type of analysis will become increasingly meaningful as the
profession advances toward design procedures that rely on distress determination from
mechanistic analysis rather than from statistical algorithms.
b. Comparison of Linear and Nonlinear Temperature Distributions
Analysis of a slab subjected to thermal curling alone presented earlier in this Chapter has
confirmed fracture responses are sensitive to the shape of the temperature profile through the
slab thickness. It has also been observed that these responses depend on the entire distribution
and not merely on the extreme-fiber values. This effect is re-examined in this section by
considering the linear and cubic temperature profiles (per Table 4.10) at 9:30 am and 7:00 pm, in
combination with an increasing wheel load.
The load-displacement responses from the pertinent FE runs are plotted in Fig. 4.18. For both
cases considered, results show that the peak loads when linear temperature distribution is
assumed are faintly greater than the corresponding values when cubic temperature profile is
adopted. At 9:30 am, for example, the peak load corresponding to the linear temperature
distribution is 15% greater than that of the cubic profile (8.8 kips versus 7.7 kips). This may
142
appear at first to be an unexpected result: the FE output indicates that prior to the load
application, the curling-only extreme-fiber bending stress due to the linear temperature profile is
higher than for the cubic distribution, by 13%. The latter observation is also consonant with
some earlier conclusions (Section 4.5.4) that a linear distribution may lead to higher extreme-
fiber stresses, which may facilitate crack initiation. The peak load, however, depends not only on
those stresses, but also on the overall stress gradient through the slab thickness. This
phenomenon was first noted by Kharlab and Minin (1989), who proposed using a “stress
gradient criterion of strength” rather than relying on the maximum stress alone. The FE results
obtained in this study confirm the necessity of accurate temperature profile determination, as
well as the need for a 3D FE analysis if fracture under curling is to be tracked.
c. Effect of Slab Length and Width
As discussed earlier in Section 4.5.5 for cracking due to curling alone, the critical temperature
differential resisted by the slab is not sensitive to its length for L/l value when this is greater than
10. In this section, it is desired to examine whether the same conclusion holds true when the slab
is subjected to thermal curling and wheel load. To investigate this, a constant daytime (warmer
top) linear temperature differential of 24°F is applied followed by an increasing wheel load. The
slab sizes used earlier are retained.
The peak load, Pmax, i.e., the load beyond which softening occurs, may be rendered
dimensionless and plotted against the dimensionless slab length, L, or slab width, W; this is
accomplished by retaining the ratios (L/l) and (W/l), per Westergaard (1927), and introducing
(Pmax / f t ' h2), in accord with Bache and Vinding (1990). The trend revealed in Fig. 4.19 is
similar to that from the case of cracking due to curling only: slab size beyond (L/l) of 10 does not
143
affect the peak load supported by the slab. The width effect is similar, albeit much less
pronounced, as might be expected. A more comprehensive dimensionless representation of
phenomena like this would entail dimensionless parameters for the geometry of the applied load
(tire print radius, a) and for the applied temperature distribution, e.g., (a/l) and (αΔT), per
Ioannides and Salsilli-Murua (1989), as well as for the slab’s self-weight, Dγ. The latter was first
suggested in the former Soviet Union by Korenev and Chernigovskaya (1962), and subsequently
applied in the United States by Lee and Darter (1994); its influence is discussed further below.
d. Effect of Slab Self-Weight
In order to investigate the effect of the self-weight of the slab, γ, the typical value of 150 pcf
assumed in the previous sections is incremented through multiplication by a factor of 0.5, 2, 5,
and 10, specified for the 240- by 120-in. slab. A constant daytime (warmer top) linear
temperature differential of 24°F is applied, followed by an increasing wheel load, as done for the
size effect study above. Results pertaining to the load-displacement behavior, shown in Fig. 4.20,
indicate that a change in the slab self-weight does not significantly affect the peak load resisted
by the slab. The vertical load line displacement at the peak load, Δ, however, increases with an
increase in self-weight. This displacement may be rendered dimensionless using the standard
(Δkl2/P) form (Ioannides et al., 1985), and plotted against the dimensionless self-weight, Dγ,
which is defined as:
 h2
D  (4.5)
kl 2
where k is the modulus of subgrade reaction, l is the radius of relative stiffness; P is Pmax, Δ is the
LLD corresponding to Pmax, and h is the slab thickness, respectively. The result shown in Fig.
144
4.21 confirms that the lighter the slab, the smaller the curling-plus-load deflection at the peak
load, presumably because the edge load location on the lighter slab lifts up more. As its self-
weight increases, more of the slab remains in contact with the subgrade and a more gradual
fracture process takes place. The softening segment of the load-displacement curves in Fig. 4.20
is seen to decrease.
e. Effect of Tensile Strength and Total Fracture Energy
During fracture analysis of a pavement slab subjected to wheel load alone (Aure and Ioannides,
2012), it was found that tensile strength influences the load at which cracking begins (manifested
in an abrupt slope change in the P-LLD curve), whereas fracture energy determines the peak load
at which the slab ruptures completely along the assumed fracture plane (evinced by softening). A
similar investigation is replicated in this section using the same 120 by 240-in. slab subjected
first to a constant linear daytime (warmer top) temperature differential of 24°F, followed by an
increasing wheel load. The tensile strength, ft ' , baseline value of 0.463 ksi (per Table 4.9) is
modified to 0.9 f t ' and 1.4 f t ' , keeping the total fracture energy constant. Similarly, in examining
the effect of total fracture energy (GF), the baseline value of 0.431 lb/in. (see Table 4.9) is
incremented to 0.9GF, 1.2GF and 1.4GF, keeping the tensile strength constant. The load-
displacement response for each case is shown in Fig. 4.22 for the tensile strength and Fig. 4.23
for the fracture energy factorials. Also shown in both figures are some of the corresponding
results for a slab under wheel load alone, presented in Chapter 3 (Aure and Ioannides, 2012).
Figure 4.22 shows that as tensile strength increases, the peak load also rises modestly, but the
change in the softening segment is barely significant, since the fracture energy is assumed to be
constant. It is observed that the presence of curling causes the peak load to be reached
145
immediately after crack initiation. On the contrary, when only wheel load is applied, there is a
localized softening after crack initiation, followed by a stable crack growth until peak load is
attained. This is caused by the pre-existing widespread curling stresses along the entire assumed
fracture plane when wheel load is applied, compared to the localized stresses that first develop
when the wheel load is applied by itself, and that continue to increase gradually along the
fracture plane.
Figure 4.23 shows that increasing total fracture energy results not only in a higher peak load, but
also in a diminished post-peak loss due to softening. Moreover, displacement experienced during
the softening stage also increases, and the softening segment becomes less precipitous, i.e., the
material exhibits increased ductility, indicating the development of a stable crack. In general, it
can be concluded that daytime curling combined with wheel load reduces the extent of stable
crack growth in the slab, a phenomenon that can be compensated by increasing total fracture
energy, for example, by using fibers.
4.6.2 Slab under Constant Wheel Load and Increasing Thermal Gradient
In this loading combination situation, a constant magnitude of wheel load is applied in the first
step, followed by an increasing daytime differential temperature toward a maximum value ΔT =
96°F. To investigate the effect of the magnitude of the applied load (P), FE runs for P values
ranging from 0.001 kips to 7 kips are made. The differential temperature (ΔT) versus load line
displacement (LLD) response for selected cases is shown in Fig. 4.24. As may be anticipated,
when the applied load increases, the differential temperature needed to cause the slab failure
decreases, since some of the strength of the material is consumed by the applied load prior to the
application of temperature. When 4 kips and 7 kips wheel loads are applied, for example, the
146
peak temperature differentials (i.e., the values corresponding to slab failure) are only about 22°F
and 12°F, respectively.
Examining the deformed shape of the slab under the 7-kip load alone, it is found that crack starts
propagating before temperature application begins. Therefore, as temperature differential
increases gradually, so does the load line displacement, evincing stable crack growth. In contrast,
under the 4-kip or 1-kip loads, crack initiation occurs at a specific temperature differential, and is
reflected in the flat segment of the response in Fig. 4.24.
It is also clearly seen from Fig. 4.24 that as the applied load decreases, LLD increases in the
negative direction (upward): the slab curls down significantly with a small load (i.e., the
midpoint of the edge where the load is applied moves upward with respect to the subgrade). This
is clearly evident, for example, for the curve with P = 0.1 kips. For P values less than about 0.3
kips, Fig. 4.24 shows that there is no softening at all (i.e., no flat segment) in the responses,
indicating no cracking. Therefore, the presence of a very small cut or notch is necessary to
initiate crack growth under daytime curling alone or curling combined with very small wheel
load.
To find out the notch size at which crack initiation under curling alone occurs, five different
notch depths of 0.001, 0.1, 0.5, 1 and 2 in. are introduced at the bottom of the slab,
corresponding to notch depth-to-slab thickness (a/h) ratios of 1/6000, 1/60, 1/12, 1/6 and 1/3.
Finite element runs are made for the slab subjected to daytime curling alone. The variation of the
peak differential temperature with notch size is plotted in Fig. 4.25. As expected, a peak
temperature differential is observed even for the smallest notch depth of 0.001 in. (a/h = 1/6000).
147
As notch depth increases, the peak differential temperature decreases almost linearly. Therefore,
for daytime temperature to initiate bottom-up cracking, a very small notch should exist at the
bottom of the slab. In contrast, as noted in Section 4.5.1 (see Fig. 4.7), under nighttime
temperature alone (warmer bottom), top-down cracking may initiate without any notch in the
slab. This may be related to the different effect slab self-weight has on the respective curled
boundary conditions, but this issue requires further exploration.
To compare the two extreme loading scenaria presented in this study, i.e., scenario (a) constant
daytime temperature differential combined with increasing wheel load, and scenario (b) constant
wheel load followed by increasing daytime temperature differential, the variation of the peak
temperature differential versus the applied load is plotted in Fig. 4.26. Note that responses for
scenario (a) are obtained from Section 4.6.1(a) corresponding to the peak loads obtained at
constant daytime temperature differentials for the time periods given in Table 4.10 (see also Fig.
4.18). The curve for scenario (b) is extracted from Fig. 4.24. A similar trend is observed in both
cases: an increase in the magnitude of one of the stressors decreases the magnitude of the other
stressor required to cause slab failure. Under pre-crack elastic analysis, the two scenaria may be
expected to yield the same curve. Yet, other things being equal, Westergaard’s curling analysis
(which assumed infinite slab weight) appears to conform better with scenario (b), which is a
more conservative approach than scenario (a). Scenario (a) is preferable to scenario (b), since it
allows a higher load a given temperature and a higher temperature at a given load.
Stresses at the bottom of the slab caused by a wheel load and daytime temperature are of the
same sense (both tensile), but the corresponding deformations are in the opposite direction,
changing the boundary condition of the slab as one of the stressors is applied compared to what
148
prevailed under the other. Moreover, as mentioned above, cracking under wheel load first
initiates locally beneath the load and then gradually propagates, following the assumed failure
path. Under thermal curling, on the other hand, cracking may begin almost at the same time at all
points along the assumed fracture plane, since temperature (therefore, curling stress) is assumed
to be constant for all points on the same plane. Wheel load alone results in a relatively gradual
failure of the slab, whereas the curling only causes a sudden collapse of the slab once the failure
strength is attained. Depending on the magnitude of the stressor applied first, the responses under
combined analysis can, therefore, exhibit stable or unstable cracking behaviors. The large flat
softening segment exhibited in Fig. 4.24 (especially at the lower applied loads), for example,
replicates the sudden failure of the slab under curling only. As the load increases, on the other
hand, for example at P = 7 kips, stable crack growth is observed without almost any softening,
reminiscent of load-only behavior. Observations in Fig. 4.24 and Fig. 4.26 evidently challenge
the validity of linear superposition if it were to be applied for post-crack responses due to
individual stressors.
4.7 Summary and Conclusions
The post-crack behavior of a concrete pavement slab subjected to two main actions, temperature
and wheel load, has been investigated, preceded by a pre-crack (elastic) analysis. The proposed
3D FE simulation gave thermal stresses smaller than those from 2D FE approaches and
analytical results. Nonetheless, excellent agreement is observed with other 3D FE results
available in the literature. Post-crack behavior of an early age concrete slab-on-grade subjected
to thermal curling alone has also been investigated. The age of concrete is the dominant factor
affecting material properties, and determines the timing and depth of joint sawing. It has also
149
been observed that slab self-weight is not significant with regard to the critical temperature
differential resisted by the slab.
The influence of a fixed diurnal temperature distribution through the slab thickness acting in
combination with increasing wheel load has been examined. It has been observed that daytime
conditions significantly reduce the peak load capacity of the slab. During the nighttime, cracks
are stable, mainly due to the counteraction of curling stresses with the loading stresses, as well as
the full contact between the slab and the subgrade. In both cracking due to curling alone and
fixed curling combined with increased wheel loading, responses are observed to depend not only
on the extreme-fiber stresses that may initiate the crack, but also on the stress distribution
through the slab thickness, reflecting the temperature profile.
From the parametric studies conducted, it can be concluded that the influence of slab size on the
fracture process is negligible for L/l ratios greater than 10. Moreover, the peak load is insensitive
to the self-weight of the slab. The deflection at the peak load increases as the slab self-weight
increases. An increase in the fracture energy is observed to increase the post-peak load and to
result in stable (more ductile) crack formation, whereas an increase in the tensile strength affects
the load at which cracks initiate and results in brittle failure.
The loading scenario involving a fixed wheel load followed by an increasing daytime
temperature has also been considered. It is observed that the stability of the crack and the
magnitude of the peak temperature differential largely depend on the magnitude of the wheel
load. A relatively high wheel load, encourages stable crack growth with almost no softening,
beginning at a lower temperature differential. In contrast, a rather small wheel load, allows the
150
temperature differential to increase significantly before crack initiation, but the peak is followed
by sudden slab failure with pronounced softening.
It can be concluded that the application of cohesive elements to the fracture analysis of concrete
pavement slabs is promising. It is anticipated that the approach can be extended to the
idealization of fracture in in situ jointed concrete pavements. This step-by-step effort can
contribute to the ongoing development of rational failure criteria for pavement structures.
Consequently, the exclusively statistical algorithms that are used to predict pavement distresses
on the basis of Miner’s hypothesis (Miner, 1945) in current pavement design procedures may
eventually be eliminated.
4.8 References
Providence, RI.
Aure, T. W., and Ioannides, A. M. (2012). "Numerical analysis of fracture process in pavement
slabs." Can. J. Civ. Eng., 39(5), NRC Research Press, Ottawa, ON, Canada, 506-514.
Bache, H. H., and Vinding, I. (1990). "Fracture mechanics in design of concrete pavements."
Proc., Second International Workshop on the Design and Evaluation of Concrete Pavements,
CROW/PIARC, Sigüenza, Spain, 139-165.

151
Barros, J. A., and Figueiras, J. A. (2001). "Model for the analysis of steel fiber reinforced
concrete slabs on grade." Comput. Struct., 79(1), Elsevier, Oxford, UK., 97-106.
Bradbury, R. D. (1938). Reinforced Concrete Pavements, Wire Reinforcement Institute,
Washington, DC.
Channakeshava, C., Barzegar, F., and Voyiadjis, G. Z. (1993). "Nonlinear FE analysis of plain
concrete pavements with doweled joints." J. Transp. Eng., 119(5), ASCE, New York, NY, 763-
781.
Chen, H. M., Dere, Y., Sotelino, E., and Archer, G. (2002). "Mid-panel cracking of Portland
cement concrete pavements in Indiana." Rep. No. FHWA/IN/JTRP-2001/14, Purdue University,
West Lafayette, IN.
Choubane, B., and Tia, M. (1992). "Nonlinear temperature gradient effect on maximum warping
stresses in rigid pavements." Transp. Res. Rec. 1370, Transportation Research Board, National
Davids, W. G., Turkiyyah, G. M., and Mahoney, J. P. (1998). "EverFE: Rigid pavement three-
dimensional finite element analysis tool." Transp. Res. Rec. 1629, Transportation Research
Dere, Y., Asgari, A., Sotelino, E. D., and Archer, G. C. (2006). "Failure prediction of skewed
jointed plain concrete pavements using 3D FE analysis." Eng. Failure Anal., 13(6), Elsevier,
152
Eisenmann, J., and Leykauf, G. (1990). "Effect of paving temperatures on pavement
performance." Proc., Second International Workshop on the Design and Evaluation of Concrete
Pavements, CROW/PIARC, Sigüenza, Spain, 419-433.
Hammons, M. I. (1997). "Development of an analysis system for discontinuities in rigid airfield
pavements." Rep. No. GL-97-3, US Army Corps of Engineers, Vicksburg, MS.
Huang, Y. H., and Wang, S. T. (1974). "Finite element analysis of rigid pavements with partial
subgrade contact." Transp. Res. Rec. 485, Transportation Research Board, National Research
Ioannides, A. M., and Salsilli-Murua, R. A. (1989). "Temperature curling in rigid pavements: an
application of dimensional analysis." Transp. Res. Rec. 1227, Transportation Research Board,
Ioannides, A. M., Thompson, M. R., and Barenberg, E. J. (1985). "Westergaard solutions
Ioannides, A. M., Craig, M. D., and Christopher, M. W. (1999). "Westergaard curling solution
153
Ioannides, A. M., Peng, J., and Swindler Jr, J. R. (2006). " ABAQUS model for PCC slab
Jenq, Y. S., Liaw, C. J., and Kim, S. C. (1993). "Effect of temperature on early crack formation
in Portland cement concrete pavements." Transp. Res. Rec. 1388, Transportation Research
Kharlab, V. D., and Minin, V. A. (1989). "Strength criteria accounting for influence of stress
state gradient." Issledovaniya po Mekhanike Stroil'nykh Konstruktsii i Materialov (in Russian),
Leningrad, 53-57.
Khazanovich, L. (1994). "Structural analysis of multi-layered concrete pavement systems," Ph.D.
Korenev, B. G., and Chernigovskaya, E. I. (1962). Analysis of Plates on Elastic Foundation.
Gosstroiizdat, Moscow, Russia.
Korovesis, G. T. (1990). "Analysis of slab-on-grade pavement systems subjected to wheel and
temperature loadings," Ph.D. dissertation, University of Illinois at Urbana-Champaign, Urbana,
IL.
Kuo, C. M. (1994). "Three-dimensional finite element analysis of concrete pavements," Ph.D.
154
Liaw, C. J. (1992). "Effects of temperature on the fracture behavior of early aged concrete
pavement and asphalt concrete overlay, " M.S. thesis, The Ohio State University, Columbus,
OH.
Lee, Y. H., and Darter, M. I. (1994). "Loading and curling stress models for concrete pavement
design." Transp. Res. Rec. 1449, Transportation Research Board, National Research Council,
Mahboub, K. C., Liu, Y., and Allen, D. L. (2004). "Evaluation of temperature responses in
concrete pavement." J. Transp. Eng., 130(3), ASCE, Reston, VA, 395-401.
Masad, E., Taha, R., and Muhunthan, B. (1996). "Finite element analysis of temperature effects
on plain jointed concrete pavements." J. Transp. Eng., 122(5), ASCE, Reston, VA, 388-398.
Meda, A., Plizzari, G. A., and Riva, P. (2004). "Fracture behavior of SFRC slabs on grade."
Mater. Struct., 37(6), Springer, Berlin, Germany, 405-411.
Mohamed, A. R., and Hansen, W. (1996). "Prediction of stresses in concrete pavements
subjected to nonlinear gradients." Cement Concrete Comp., 18(6), Elsevier, Amsterdam, The
Nishizawa, T., Koyanagawa, M., Takeuchi, Y., and Kimura, M. (2001). "Study on mechanical
behavior of dowel bar in transverse joint of concrete pavement." Proc., Seventh International
Conference on Concrete Pavements, International Society for Concrete Pavements, Orlando, FL,
571-584.
155
Pane, I., Hansen, W., and Mohamed, A. (1998). "Three-dimensional finite element study on
effects of nonlinear temperature gradients in concrete pavements." Transp. Res. Rec. 1629,
Richardson, J. M., and Armaghani, J. M. (1987). "Stress caused by temperature gradient in
Portland cement concrete pavements." Transp. Res. Rec. 1121, Transportation Research Board,
Shimomura, T., Nishizawa, T., and Ozeki, T. (2008). "Evaluation of thermal stress in airport
concrete pavement slab by 3D-FEM analysis." Proc., Ninth International Conference on
Concrete Pavements, International Society for Concrete Pavements, San Francisco, CA, 740-
754.
Shoukry, S. N., Fahmy, M., Prucz, J., and William, G. (2007). "Validation of 3D FE analysis of
rigid pavement dynamic response to moving traffic and nonlinear temperature gradient effects."
Int. J. Geomech., 7(1), ASCE, Reston, VA, 16-24.
Siddique, Z., Hossain, M., and Meggers, D. (2006). "Curling and curling stresses of new
concrete pavements." Airfield and Highway Pavements, The 2006 Airfield and Highway
Pavement Specialty Conference, ASCE, Atlanta, GA, 671-682.
Soares, J. B. (1997). "Concrete characterization through fracture mechanics and selected
pavement applications," Ph.D. dissertation, Texas A and M University, College Station, TX.
156
Teller, L. W., and Sutherland, E. C. (1935). "Observed effects of variation in temperature and
moisture on the size, shape, and stress resistance of concrete pavement slabs." Public Roads,
16(9), Bureau of Public Roads, Washington, DC, 169-197.
Thompson, M. R., Dempsey, B. J., Hill, H., and Vogel, J. (1987). "Characterizing temperature
effects for pavement analysis and design." Transp. Res. Rec. 1121, Transportation Research
Board, National Research Council, Washington, 11-19.
Tia, M., Armaghani, J. M., Wu, C. L., Lei, S., and Toye, K. L. (1987). "FEACONS III computer
program for an analysis of jointed concrete pavements." Transp. Res. Rec. 1136, Transportation
Research Board, National Research Council, Washington, DC, 12-22.
temperature." Proc. Hwy. Res. Board 6, National Research Council, Washington, DC, 201-215.
157
Table 4.1 Comparison with Experimental and Numerical Results
σxx (ksi) σxx (ksi) σxx (ksi)

ΔT
Teller and Sutherland (1935) Kuo (1994) Present study
(°F)
14 0.186 0.187 0.187
18 0.220 0.243 0.239
18 0.209 0.243 0.239
19 0.266 0.255 0.251
20 0.252 0.268 0.264
20 0.320 0.268 0.264
21 0.195 0.280 0.276
158
Table 4.2 Coefficients A, B, and C for Daily Cycle in June (Choubane and Tia, 1992)
Time A (°F) B (°F/in.) C (°F/in.2) ΔT (°F)
09:00 am 98.7382 -3.3178 0.2396 10.4529
11:00 am 112.3161 -4.9194 0.2633 22.9500
01:00 pm 122.8802 -5.9518 0.2816 30.7543
03:00 pm 109.1800 -0.3571 -0.1429 14.7851
05:00 pm 115.2114 -2.5486 0.0571 18.3086
08:00 pm 96.5053 1.7280 -0.1833 -0.7076
10:00 pm 91.1598 1.7961 -0.1359 -5.1564
12:00 pm 87.8053 1.6622 -0.0976 -7.0586
02:00 am 85.1843 1.6057 -0.0800 -7.9714
06:00 am 82.3145 1.2467 -0.0412 -7.8818
159
Table 4.3 Comparison of 3D Linear and Nonlinear Curling
Linear Nonlinear Percentage

Time (a) (b) (c) (d)
a/c b/d
σt σb σt σb
12:00 am 0.112 -0.113 0.151 -0.074 73.9 153.5
02:00 am 0.125 -0.126 0.157 -0.094 79.4 134.5
06:00 am 0.123 -0.125 0.140 -0.108 88.1 115.4
09:00 am -0.154 0.154 -0.251 0.057 61.4 269.2
11:00 am -0.266 0.266 -0.373 0.159 71.4 166.9
01:00 pm -0.314 0.314 -0.428 0.200 73.3 157.0
03:00 pm -0.197 0.198 -0.140 0.255 141.4 77.3
05:00 pm -0.229 0.229 -0.252 0.206 90.8 111.2
08:00 pm 0.012 -0.012 0.086 0.062 13.7 -20.1
10:00 pm 0.084 -0.085 0.139 -0.030 60.2 283.7
Note: σt is the top bending stress in ksi; σb is the bottom bending stress in ksi
160
Table 4.4 Comparisons for Linear Curling
Analytical 2D FE 3D FE Percentage
Time (a) (b) (c) (d) (e) (f)
f/b f/d
σt σb σt σb σt σb
12:00 am 0.118 -0.118 0.116 -0.116 0.112 -0.113 96.3 97.8
02:00 am 0.133 -0.133 0.129 -0.129 0.125 -0.126 95.0 97.8
06:00 am 0.132 -0.132 0.128 -0.128 0.123 -0.125 95.1 97.6
09:00 am -0.175 0.175 -0.164 0.164 -0.154 0.154 88.6 94.1
11:00 am -0.384 0.384 -0.310 0.310 -0.266 0.266 69.5 85.8
01:00 pm -0.514 0.514 -0.418 0.418 -0.314 0.314 61.2 75.1
03:00 pm -0.247 0.247 -0.228 0.228 -0.197 0.198 80.1 86.7
05:00 pm -0.306 0.306 -0.263 0.263 -0.229 0.229 75.0 87.1
08:00 pm 0.012 -0.012 0.012 -0.012 0.012 -0.012 105.5 103.7
10:00 pm 0.086 -0.086 0.086 -0.086 0.084 -0.085 -97.4 98.9
Note: σt is the top bending stress in ksi; σb is the bottom bending stress in ksi
Analytical: After Westergaard (1927)
2D FE: Choubane and Tia (1992) using 2D FE program FEACONS IV
3D FE: Present study employing ABAQUS®
161
Table 4.5 Comparison with 2D FE Results for Quadratic Temperature Profile
2D FE 3D FE
Percentage
(Choubane and Tia, 1992) (This Study)
Time
(a) (b) (c) (d)
c/a d/b
σt σb σt σb
12:00 am 0.152 -0.080 0.151 -0.074 99.6 92.4
02:00 am NA NA 0.157 -0.094 NA NA
06:00 am 0.143 -0.113 0.140 -0.108 97.9 95.8
09:00 am NA NA -0.251 0.057 NA NA
11:00 am -0.406 0.214 -0.373 0.159 91.8 74.5
01:00 pm -0.521 0.315 -0.428 0.200 82.2 63.5
03:00 pm -0.176 0.280 -0.140 0.255 79.3 91.2
05:00 pm -0.284 0.242 -0.252 0.206 88.8 85.1
08:00 pm 0.079 0.055 0.086 0.062 108.8 112.3
10:00 pm NA NA 0.139 -0.030 NA NA
Note: σt is the top bending stress in ksi; σb is the bottom bending stress in ksi; NA: not available
162
Table 4.6 Comparison of 3D FE Combination and Superposition Analysis
Coefficients Combination Superposition Percentage
Hour A B C (a) (b) (c) (d)

c/a d/b
(°F) (°F/in.) (°F/in.2) σt (ksi) σb (ksi) σt (ksi) σb (ksi)
Midnight 95 0.8889 -0.1482 0.2980 0.2053 0.2962 0.2033 99.39 99.05
4:00 am 90 1.0000 -0.0741 0.2784 0.1830 0.2764 0.1830 99.28 99.98
8:00 am 87 1.0000 -0.0247 0.2645 0.1716 0.2625 0.1694 99.24 98.72
Noon 95 -2.0000 0.2963 0.1399 0.0750 0.2448 0.1606 174.94 214.10
4:00 pm 107 -2.8889 0.1975 0.1589 0.0965 0.2448 0.1605 154.05 166.24
8:00 pm 102 0.1111 -0.1728 0.2950 0.2108 0.2950 0.2110 100.00 100.08
Table 4.7 Comparison Among Different FE Programs
ILSL2 ABAQUS® EverFE Percentage

Hour (a) (b) (c) (d) (e) (f)
c/a d/b e/c f/d
σt σb σt σb σt σb
Midnight 0.291 0.229 0.298 0.205 0.300 0.200 102.4 89.6 100.7 97.4
04:00 am 0.270 0.206 0.278 0.183 0.281 0.186 103.1 88.9 100.9 101.6
08:00 am 0.254 0.190 0.265 0.172 0.267 0.173 103.9 90.2 100.9 100.8
Noon 0.102 0.077 0.140 0.075 0.143 0.088 136.8 97.1 102.2 117.7
04:00 pm 0.237 0.129 0.159 0.096 0.164 0.158 67.1 75.1 103.2 163.7
08:00 pm 0.281 0.235 0.295 0.211 0.296 0.211 105.2 89.7 100.3 100.1
Note: σt is the top principal bending stress in ksi; σb is the bottom principal bending stress in ksi
163
Table 4.8 Material Properties for Concrete at Different Ages (Jenq et al., 1993)
Age 16 hrs 24 hrs 72 hrs
E (ksi) 1209 2130 3051
GF (kips/in.) 5.80×10-5 17.2×10-5 29.80×10-5
f t ' (ksi) 0.084 0.209 0.348
μ 0.2 0.2 0.2
α (×10-6/°F) 5.562 5.562 5.562
164
Table 4.9 Geometry, Material Properties and Discretization of Slab Considered
Slab Geometry
Length (in.) 240
Width (in.) 120
Thickness (in.) 6
Material Properties
Young's Modulus (ksi) 4000
Coefficient of Thermal Expansion (in./in./°F) 5×10-6
Tensile Strength (ksi) 0.463
Fracture Energy (lb/in.) 0.431
Poisson's Ratio 0.15
Modulus of Subgrade Reaction (psi/in.) 200
Discretization Characteristics
Slab Element Size (in.) 6×6×2
Cohesive Zone Element Size (in.) 0.12×0.12
Concrete Softening Curve Bilinear
Solver Type Riks or Newton-Raphson
Subgrade Idealization SPRING1
Note: 1 lb = 4.444 N; 1 in. = 25.4 mm; 1 ksi = 6.89 MPa
165
Table 4.10 Coefficients of Cubic Polynomial Used to Estimate Temperature Distribution
Temp. Differential Cubic: T(z) = Az3+Bz2+Cz+D

Time
ΔT (°F) A (°F/in.3) B (°F/in.2) C (°F/in.) D (°F) R2
07:00 am -3.4806 -0.0306 0.2750 0.0317 73.5550 0.9910
04:00 am -4.8438 -0.0389 0.3667 0.0079 75.0600 0.9865
01:00 am -3.7638 -0.0333 0.2643 0.2405 78.4860 0.9865
10:00 am -4.1064 -0.0083 0.0036 0.9619 82.1210 0.9678
07:00 pm -3.9780 0.0167 -0.3595 2.2190 87.0240 0.9969
09:30 am 21.8274 0.3333 -2.5536 -0.3155 99.8570 0.9595
11:30 am 18.6882 0.1389 -1.2381 -0.6865 101.6900 0.9911
04:00 am 16.6674 0.1111 -1.1429 0.0794 104.4800 0.9828
02:00 pm 14.9652 0.0556 -0.3929 -2.1389 107.8800 0.9953
Source of data: Teller and Sutherland (1935).
166
Longitudinal Stress (ksi)
-0.30 -0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10
-4.5
This study
Pane et al. (1998)

-3.0
EverFE
-1.5
Depth (in.)
0.0
1.5
3.0
4.5
Figure 4.1 Comparison Among Different 3D FE Simulations (Tension is Positive)
167
Computed Temperature (°F)
80 85 90 95 100 105 110 115 120 125
0
3
11:00 am
01:00 pm
Depth (in.)
4
03:00 pm
05:00 pm
5
08:00 pm
09:00 am
6
Figure 4.2 Variation of Daytime Temperature (9 am - 8 pm)
168
-0.50 -0.30 -0.10 0.10 0.30
0
2 09:00 am
11:00 am
3 01:00 pm
03:00 pm
05:00 pm
Depth (in.)
4
08:00 pm
Figure 4.3 Variation of Longitudinal Bending Stress Caused by Daytime Temperature
(3D Nonlinear Curling)
169
Computed Temperature (°F)
80 85 90 95 100
0
3
Depth (in.)
4
12:00 am
5 02:00 am
06:00 am
6
10:00 pm
7
Figure 4.4 Variation of Nighttime Temperature (10 pm - 6 am)
170
-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20
0
12:00 am
1 02:00 am
06:00 am
2
10:00 pm
4
Depth (in.)
Figure 4.5 Variation of Longitudinal Bending Stress Caused by Nighttime Temperature
(3D Nonlinear Curling)
171
0.35
0.30
Maximum Bending Stress (ksi)
0.25
0.20
0.15
0.10
ABAQUS
0.05 EverFE
ILSL2
0.00
0 10 20 30 40 50
Time (hr)
Figure 4.6 Comparison Among Three FE Programs
172
35.0
a/h=0
a/h=1/12
30.0
a/h=1/3
a/h=1/2
25.0
Applied Temperature Differential (°F)
20.0
15.0
10.0
5.0
0.0
0.0000 0.0010 0.0020 0.0030 0.0040 0.0050
Figure 4.7 Effect of Notch Depth on Cracking Due to Curling
173
35.0
Jenq et al. (1993) a/h=0
Jenq et al. (1993) a/h=1/12
30.0 This study a/h=0
This study a/h=1/12
25.0
Applied Temperature Differential (°F)
20.0
15.0
10.0
5.0
0.0
0.0000 0.0002 0.0004 0.0006 0.0008 0.0010 0.0012
Figure 4.8 Comparison with Jenq et al. (1993)
(16 hrs age concrete)
174
60.0
72 hours
24 hours
16 hours
50.0
Applied Temperature Differential (0F)
40.0
30.0
20.0
10.0
0.0
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030
Figure 4.9 Effect of Concrete Age on Fracture Due to Curling
(Notch size of 0.5 in.)
175
4.0E-04
This study
3.5E-04
Jenq et al. (1993)
3.0E-04
Critical Applied αΔT
2.5E-04
2.0E-04
1.5E-04
1.0E-04
5.0E-05
0.0E+00
0.50 0.56 0.62 0.68 0.74 0.80 0.86
Brittleness Number (B)
Figure 4.10 Variation of Maximum Dimensionless Temperature Differential with
Brittleness Number
(16-72 hrs, notch depth of 0.5 in.)
176
45.0
2γ
40.0
γ
0.5γ
35.0
0.25γ
30.0
25.0
20.0
15.0
10.0
5.0
0.0
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030
Figure 4.11 Effect of Concrete Unit Weight on Fracture Due to Curling
Note: γ = 150 pcf
177
40.0
Quadratic - Case 1
Quadratic - Case 2
35.0
Linear - Case 3
30.0 Case1:T ( z )  1.5 z 2

Case 2 :T ( z )  1.5 z 2  18 z
Case 3 :T ( z )  9 z
25.0
20.0
15.0
Case 2
10.0
Case 3
5.0 Case 1
0.0
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030
Figure 4.12 Effect of Temperature Profile on Fracture Due to Curling
178
4.0E-04
3.5E-04
Width effect
3.0E-04 Length effect
Length effect by Liaw (1992)
Critical Applied αΔT
2.5E-04
2.0E-04
1.5E-04
1.0E-04
5.0E-05
0.0E+00
5.0 10.0 15.0 20.0 25.0 30.0
L/l or W/l
Figure 4.13 Effect of Slab Size on Fracture Due to Curling
179
Intact (bulk) material (C3D27)
Cohesive zone (COH3D8)
Applied Load (P)
Subgrade (SPRING1)
Figure 4.14 Finite Element Idealization of Slab with Properties Shown in Table 4.9
180
60.0
5
3
50.0
4
40.0
2
Applied Load (kips)
07:00 pm
30.0
20.0 Load only
10.0
2 09:30 am
4
1
3
0.0
0.00 0.04 0.08 0.12 0.16 0.20 0.24 0.28
Figure 4.15 Load-Plus-Curling Response Curves under Diurnal Thermal Distributions
Note: Points 1, 2, 3, 4 and 5 represent load levels.
181
0.060
0.050 4
0.040
0.030
3
0.020
0.010 1
2 wf
0.000
0.0 20.0 40.0 60.0 80.0 100.0 120.0
Slab Width (in.)
Figure 4.16 Crack Growth along Slab Width for Load Levels Shown in Fig. 4.15
(9:30 am)
182
0.060
0.050
0.040 5
0.030
2
4
0.020
3
0.010 1
wf
0.000
0.0 20.0 40.0 60.0 80.0 100.0 120.0
Slab Width (in.)
Figure 4.17 Crack Growth along Slab Width for Load Levels Shown in Fig. 4.15
(7:00 pm)
183
50.0 Linear Cubic
40.0
07:00 pm
Applied Load (kips)
30.0
20.0
10.0
9:30 am
0.0
0.00 0.04 0.08 0.12 0.16 0.20 0.24
Figure 4.18 Comparison of Linear and Cubic Temperature Profiles
184
0.80
Width
Length
0.70
Pmax/f'th2
0.60
0.50
0.40
4.50 8.00 11.50 15.00 18.50
L/l or W/l
Figure 4.19 Effect of Slab Length and Width
185
25
0.5γ γ
20
2γ 10γ
5γ
Applied Load (kips)
15
10
0
-0.01 0.01 0.03 0.05 0.07 0.09
Figure 4.20 Effect of Slab Self-Weight
Note: γ = 150 pcf
186
0.80
0.70
Δkl2/P
0.60
0.50
0.40
0.45 5.45 10.45 15.45 20.45 25.45
Dγ ×105
Figure 4.21 Variation of Dimensionless Peak Load versus Dimensionless Self-Weight
187
21.00
0.9f't 1.3f't
No curling: 1.4f't
18.00 1.4f't 1.0f't
15.00
Applied Load (kips)
12.00 No curling: 1.0f't
9.00
6.00
3.00
0.00
0.00 0.02 0.04 0.06 0.08 0.10 0.12
Figure 4.22 Effect of Tensile Strength
Note: f t ' = 0.463 ksi
188
21.00 1.4GF
1.2GF
No curling: 1.4GF
0.9GF
18.00 1.0GF
15.00
Applied Load (kips)
12.00
No curling: 1.0GF
9.00
6.00
3.00
0.00
0.00 0.02 0.04 0.06 0.08 0.10 0.12
Figure 4.23 Effect of Total Fracture Energy
Note: GF = 0.431 lb/in.
189
80
P = 0.5 kips
Applied Differential Temperature (°F)
P = 0.1 kips
60
P = 7 kips
P = 0.01 kips
40
P = 1 kips
P = 4 kips
20
0
-0.05 -0.02 0.01 0.04 0.07 0.10 0.13 0.16
Load Line Displacement
Figure 4.24 Fracture Responses of Slab under Constant Wheel Load and Increasing
Daytime Temperature Differential
190
160
140 a/h = 1/6000
120
Peak Temperature Differential (°F)
100
80
60
40
20
0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
a/h
Figure 4.25 Effect of Notch Depth on Peak Linear Temperature Differential (Curling Only)
191
70
60
Scenario (b)
Applied or Peak Temperature Differential (°F)
Scenario (a)
50
40
30
20
10
0
0 2 4 6 8 10 12 14
Applied or Peak Load (kips)
Figure 4.26 Comparison of Two Loading Scenaria
192
Chapter 5 Crack Propagation in Pavement Slabs with Aggregate Interlock
Abstract
This paper discusses finite element analysis of crack propagation in pavement slabs with
aggregate interlock joints, using the finite element package ABAQUS®6.9-2. The fracture process
is idealized using cohesive elements, per the fictitious crack model. The joint mechanism is
simulated in accordance with linear and nonlinear approaches available in the literature. The
proposed discretization is first verified by comparing the pre-crack (elastic) responses with
analytical solutions and with experimental results published by independent researchers.
Parametric studies are conducted concerning the effects of joint stiffness, joint opening and
aggregate size. It is observed that when linear aggregate interlock is employed, load transfer
efficiency does not change significantly even after slabs undergo cracking. Under nonlinear
aggregate interlock discretization, on the other hand, load transfer efficiency increases
substantially and attains its maximum value at the peak loads resisted by the slabs and decreases
continuously thereafter. A daytime temperature profile is observed to reduce both the peak load
supported by the slab system and the load transfer efficiency of the joint, while a nighttime
temperature distribution results in modest increases in these metrics. It is concluded that the
proposed approach can be used as a basis for further exploration of crack propagation in concrete
pavement systems. The step-by-step methodology implemented in this study may contribute to
the ongoing development of rational failure criteria that can replace the statistical/empirical
algorithms currently used in pavement design procedures.
193
5.1 Introduction
In jointed concrete pavement slabs, load transfer may be accomplished by two primary
mechanisms: dowel bars and aggregate interlock. Dowel bars are often placed across a joint to
guarantee the longevity of the mechanism, which relies almost exclusively on each bar’s shear
resistance. Aggregate interlock, on the other hand, results from shear forces developing at the
rough interfaces constituting the joint due to the mechanical interaction of aggregate particles,
which may degrade with time. Critical responses in jointed concrete pavements are very sensitive
to the load transfer efficiency of the joint, i.e., its ability to transfer load applied on one slab to
the other. Therefore, it is essential to incorporate realistic joint characteristics in a fracture
analysis of such pavement systems.
Published literature indicates that the mechanics of aggregate interlock is a complex
phenomenon that depends on several parameters, including aggregate size and distribution,
concrete compressive strength, friction between the aggregate and the cement paste, crack (or
joint) opening, and interface sliding (Bažant and Gambarova, 1980; Walraven, 1981). Normal
and shear stresses developing along a cracked concrete interface have been observed to be
nonlinear functions of the corresponding displacements. A few researchers have accounted for
such nonlinearities for concrete pavements (Davids and Mahoney, 1999; Wattar, 2001), but most
have adopted a simplified approach by assuming a linear pure-shear interlocking mechanism.
The first use of the linear pure-shear assumption was made by Skarlatos (1950), who derived an
analytical relationship between responses in the loaded and unloaded slabs, respectively,
assuming that shear springs were distributed over the length of the pavement slab joint. This
approach is essentially identical to that subsequently implemented in two-dimensional (2D) finite
194
element (FE) programs, such as KENSLABS and ILLI-SLAB, commonly used in pavement
engineering (Huang and Wang, 1973; Tabatabaie and Barenberg, 1978). Results from ILLI-
SLAB for different slab geometries, load sizes and joint spring stiffness values were interpreted
using dimensional analysis by Ioannides and Korovesis (1990a).
A major limitation of the pure-shear approach is that it requires specifying a joint stiffness value,
which is not easy to determine with reference to the joint’s physical characteristics. Some
investigators have resorted to analysis of laboratory data conducted on aggregate interlock
jointed concrete pavements (Colley and Humphrey, 1967; Brink et al., 2005; Jensen and Hansen,
2006; Maitra et al., 2010), their conclusions, however, are only applicable to the particular set of
conditions they each considered and fail to address all the mechanistic phenomena observed.
Consequently, joint stiffness is commonly backcalculated from field observations (Ioannides and
Korovesis, 1990b).
The present study investigates the effect of both linear and nonlinear aggregate interlock on post-
crack responses of a typical concrete pavement under edge loading. The fracture process is
tracked using the fictitious crack model (FCM) first suggested for concrete by Hillerborg et al.
(1976), and employs cohesive elements, also implemented in earlier phases of this research effort
(Aure and Ioannides, 2010, 2012). A fixed fracture path is assumed along the center of both the
loaded and unloaded slabs, where principal stresses are expected, in accordance with a discrete
crack approach.
This study is organized as follows. In the first section, a review of aggregate interlock mechanics
is presented and a suitable nonlinear approach is identified. In the second section, pre-fracture
195
analysis of the slabs is conducted. Linear and nonlinear aggregate interlock idealizations are
verified by comparing with analytical, numerical and experimental results available in the
literature. In the third section, crack propagation in jointed slabs subjected to wheel loading is
examined. A similar approach is implemented in the fourth section for the case of a constant
temperature distribution, which is considered in addition to the increasing wheel load. The
effects of daytime and nighttime temperature variations on the post-crack responses of the slabs
are also investigated.
5.2 Mechanics of Aggregate Interlock: A Review
This section outlines a general formulation for load transfer due to aggregate interlock. A
cracked concrete surface contains irregularly-shaped, protruding aggregate particles. When their
surfaces slide with respect to one another, the aggregate particles interlock, thereby transferring
shear and normal stresses. These stresses depend on the size and distribution of the aggregate
particles, friction between the cement paste and the aggregate, compressive strength of concrete,
and the size of crack opening. According to Bažant and Gambarova (1980), the normal and shear
stresses at a cracked concrete interface, in a two-dimensional plane, are functions of the normal
and shear displacements of the interface, as follows:
 n  f n  n ,  t  (5.1a)
 t  f t  n ,  t  (5.1b)
where:  n is the normal stress;  t is the shear (or tangential) stress;  n and  t are the normal and
shear displacements, respectively, and f n and f t are pertinent (generally, nonlinear) functions to
be determined.
196
Differentiation of Eq. (5.1) results in:
 f n f n 
d n    n  t  d n   K nn

K nt  d n 
     (5.2)
 d t   ft ft   d t   Ktn Ktt   d t 
   t 
 n
f n f f f
in which: K nn  , K nt  n , Ktn  t , Ktt  t are crack stiffness coefficients that can be
 n  t  n  t
determined once the functions f n and f t are established. If the variation of functions f n and f t
is nonlinear with  n and  t , the crack stiffness coefficients may be sensitive to stress level, and
therefore, are subject to change as the load is applied. This behavior is identified as nonlinear
aggregate interlock mechanism. On the other hand, if crack stiffness coefficients remain constant
as the load is applied, linear aggregate interlock behavior results. When the two off-diagonal
terms of the crack stiffness matrix are identical (Ktn = Knt), the joint exhibits symmetrical
behavior, otherwise it is termed as unsymmetrical. Setting the two off-diagonal terms of the
crack stiffness matrix to zero results in uncoupled behavior, whereas if all the off-diagonal
coefficients are non-zero, the stiffness matrix defines coupled aggregate interlock.
Based on these definitions, therefore, three feasible combinations of aggregate interlock behavior
can be envisaged for each of the linear and nonlinear cases: linear-coupled-symmetrical, linear-
coupled-unsymmetrical and linear-uncoupled-symmetrical (pure-shear); nonlinear-coupled-
symmetrical, nonlinear-coupled-unsymmetrical and nonlinear-uncoupled-symmetrical (pure-
shear). In the present study, only linear-uncoupled-symmetrical and nonlinear-coupled-
unsymmetrical cases are considered.
197
According to Bažant and Gambarova (1980), interface stresses  n and  t must conform to the
following conditions: normal stress,  n , is always less than or equal to zero (i.e., always
compressive), whereas normal displacement,  n , is always greater than zero (i.e., crack only
opens); stress-displacement relations are continuous and smooth, ensuring that the differentiation
of functions f n and f t is feasible; normal stress,  n , is zero at zero shear displacement,  t ,
irrespective of the value of the normal displacement,  n , since there is no contact between the
two crack surfaces; at constant shear displacement,  t , both the normal,  n , and shear,  t , stresses
decrease as crack opening increases; at constant crack opening, both stresses increase as shear
displacement increases.
Following Bažant and Gambarova (1980), the majority of aggregate interlock mechanics studies
have been devoted to the determination of the functions f n,t  n ,  t  using either experimental
(Reinhardt and Walraven, 1982; Divakar et al., 1987) or theoretical (also called micro-
mechanical) approaches (Walraven, 1981; Divakar and Fafitis, 1992; Walraven, 1994). Notable
among these is the contribution by Walraven (1981), who formulated a theoretical constitutive
equation assuming concrete to be a two-phase material, and implemented “a statistical analysis
of the crack [or interface] structure and the associated contact areas between the crack faces [in
terms] of the displacements.” The stresses he computed from the most probable projected contact
areas of the aggregate particles are as follows:

 n   pu Ax   Ay  (5.3a)

 t   pu Ay   Ax  (5.3b)
198
where μ is the coefficient of friction between the aggregate and the cement paste, usually taken
as 0.4; Ax is the most probable projected contact area of a unit crack area in the x-direction; and
Ay is the most probable projected contact area of a unit crack area in the y-direction. The
aggregate-paste matrix yielding strength (σpu) depends on the uniaxial compressive strength of
concrete f c' and is given as:
 pu  56.7 fc' 0.56 psi (5.4)
Walraven (1981) provides formulae for determining the contact areas obtained using statistical
analysis of the size distribution of the aggregate particles and of the resulting deformation modes
when the aggregate bears against the cement paste. The areas are correlated to the maximum
aggregate size, crack opening, sliding displacement and aggregate particle size distribution.
Aggregate particle size distribution is described by the so-called Fuller curve (Fuller and
Thompson, 1907), given by:
D
p (5.5)
Dmax
where Dmax is the maximum aggregate particle size; D denotes a given particle size, and p gives
the percent of the aggregate that is finer than D. Walraven’s complete solutions for the contact
areas are given in Table 5.1.
After examining other micro-mechanics approaches, Davids (1998) implemented Walraven’s
constitutive relations in three-dimensional (3D) nonlinear FE program EverFE. He deemed the
approach to be relatively simple, suitable for numerical implementation, validated by other
independent researchers and of reasonable accuracy when estimating responses. Wattar (2001)
199
validated Walraven’s equations with experimental measurements, verifying their applicability to
the idealization of concrete pavement aggregate interlock joints. Walraven’s constitutive
relations were further extended to cyclic loads, experienced by concrete pavements (Walraven,
1994). This general (nonlinear) approach is, therefore, adopted in the present study for the
idealization of the nonlinear aggregate interlock mechanism in the broadest conceivable way.
Normal and shear stress variations, generated by numerical integration of Walraven’s
constitutive equations using numerical computing package MATLAB (Moler, 2004), are shown in
Fig. 5.1. These curves correspond to the typical properties of concrete for pavements given in
Table 5.2, and different crack openings (δn); they are used in the post-crack analysis of the
pavement slab in Section 5.4, below.
5.3 Pre-Crack Aggregate Interlock Load Transfer Analysis
5.3.1 Finite Element Discretization
Each of the two slabs, whose geometric and material properties are given in Table 5.2, is
discretized using a quadratic continuum element, C3D27, with a uniform mesh size of 6×6×2 in.
in the width, length, and depth directions, respectively. The subgrade is simulated using
SPRING1 elements that can only support compression. For computational stability and reflecting
the friction between slab and subgrade, horizontal restraint is provided by assigning the
SPRING1 elements with horizontal stiffness at 1/10th the value of their vertical stiffness, per
Gaedicke and Roesler (2009). Furthermore, the lateral thrust exerted by neighboring slab panels
is simulated by restraining the two slab sides opposite to the loaded edge in the direction
perpendicular to the joint plane.
200
There are three candidate elements for linear aggregate interlock simulation, namely, joint
elements (JOINTC), traction-separation cohesive elements (COH3D8), and PROJECTION
CARTESIAN type connector elements (CONN3D2). The nonlinear aggregate interlock
mechanism can only be simulated using CONN3D2. The application of each of these element
types in the discretization of both categories of aggregate interlock mechanism is described in
detail in the following sections.
a. Joint Elements for Linear Aggregate Interlock Idealization
Analysis of load transfer mechanism by aggregate interlock in concrete pavements dates back to
the late 1940s, culminating in the work of Skarlatos (1950), in which the behavior of the joint
was exclusively characterized as a pure-shear load transfer mechanism, i.e., all the crack stiffness
coefficients except Ktt were assumed to be zero. This approach was subsequently implemented in
2D FE programs, KENSLABS (Huang and Wang, 1973) and ILLI-SLAB (Tabatabaie and
Barenberg, 1978). Ioannides and Korovesis (1990a) employed dimensional analysis in
interpreting numerical results for different slab geometries and load sizes obtained from ILLI-
SLAB so as to extract the necessary joint stiffness terms. In this manner, they obtained a
relationship between a dimensionless joint stiffness parameter (AGG/kl) and the load transfer
efficiency with respect to deflection (LTEδ) and stress (LTEσ). Here, k is the subgrade modulus
and l is the radius of relative stiffness of the slab-subgrade system. Ioannides and Hammons
(1996) demonstrated that the joint stiffness parameter AGG, used in ILLI-SLAB to define
aggregate interlock factor, corresponds to Skarlatos’s symbol q.
Skarlatos’s aggregate interlock idealization was adopted in a 3D FE load transfer analysis by
Hammons (1997), who employed a general flexible joint element in ABAQUS® called JOINTC.
201
This is similar to conventional spring elements, but it also has the capability to simulate damping
that may be necessary in cyclic loading. The element can be ascribed linear or nonlinear
properties in three orthogonal directions, and, it can be used to idealize either linear-uncoupled-
symmetrical or nonlinear-uncoupled-symmetrical aggregate interlock behavior. In the present
study, therefore, this element is preferred over other spring elements available in ABAQUS®
(e.g., SPRING2 and SPRINGA), in anticipation of future research involving dynamic loading.
The use of Skarlatos’s pure-shear aggregate interlock idealization in 3D FE simulation as
employed by Hammons (1997) results in the interpenetration of the bulk elements lying along
the joint, as the two crack surfaces slide with respect to each other. To circumvent this problem,
normal stiffness (Knn) is provided in the present study along with the shear stiffness (Ktt). By first
assuming a particular value of Ktt, a trial-and-error approach is used to obtain Knn; accordingly,
the normal stiffness is set to 2.3 times the shear stiffness:
Knn  2.3Ktt (5.6)
This relationship may be justified by noting that the ratio of the Young’s modulus (E) to the
shear modulus (G) is 2 (1 + μ); for a concrete with Poisson’s ratio, μ, of 0.15, this ratio yields
2.3.
Following Hammons (1997), each JOINTC element is inserted between two nodes symmetrically
located on either side of a joint plane. For every assumed value of the dimensionless joint
stiffness AGG/kl, the corresponding JOINTC shear stiffnesses, (Ktt [FL-1]), of elements located at
the corner of the joint plane are determined per Hammons (1997), as follows:
202
AGG 
Ktt  (5.7)
4  N R  1 NC  1
where AGG [FL-2] is the joint stiffness per unit length, λ [L] is the joint length, NR is the number
of nodal rows, and NC is the number of nodal columns. The primary dimensions are abbreviated
here as [L] for length and [F] for force. Since the mesh adopted is uniform, the stiffnesses at the
edge and interior are two and four times those at the corner, respectively. Equation (5.6) is used
to obtain the corresponding normal stiffness, Knn. As noted earlier, the mesh was 6×6×2 in., in
the width, length and depth directions, respectively; from these, the number of nodal rows and
nodal columns can be determined. The JOINTC shear stiffnesses for elements located at the
corner of the joint corresponding to the assumed AGG/kl values are computed using Eq. (5.7)
and presented in Table 5.3.
b. Cohesive Elements for Linear Aggregate Interlock Idealization
Traction-separation cohesive elements (COH3D8), commonly employed for fracture analysis,
are also potential candidates for the discretization of linear aggregate interlock load transfer
mechanism. In such usage, a cohesive element performs the function of an interface element with
a linear constitutive relation. Its linear-coupled-symmetrical characteristics are similar to Eq.
(5.2), which when extended to 3D becomes (ABAQUS, 2009):
 n   K nn K ns K nt   n 
  1   
 s    K ss K st   s  (5.8)
  T0  Symm Ktt   
 t  t 
203
where: n, s, and t represent the normal and two shear directions, K is the stiffness along these
directions, T0 is the initial size of the cohesive element. The latter is assumed to be 0.001 in. as
suggested by a previous study (Aure and Ioannides, 2010).
Setting the off-diagonal terms to zero, the pure-shear aggregate interlock behavior becomes:
 n   K nn 0 0   n 
  1  0   
 s    K ss 0  s  (5.9)
  T0  0 Ktt   
 t 0  t 
The two shear stiffness coefficients, Kss and Ktt, can be related to the joint stiffness per unit
length, AGG, by using the following simple relation:
Kss h  Ktt h  AGG (5.10)
where h is the thickness of the slab. The factor of 2.3 suggested earlier may be used to relate the
normal and shear stiffnesses, as for JOINTC elements:
Knn  2.3 Kss  2.3 Ktt (5.11)
A large failure stress and negligible fracture energy is assumed for a COH3D8 element, so that it
remains elastic at all times. The top and bottom surfaces of the element are tied to the slab
elements using surface-based TIE CONSTRAINTS. This approach was found to work well in
earlier studies (Aure and Ioannides, 2010), and is especially useful in discretizing the joint with a
relatively finer mesh than the rest of the slab. It should be noted that linear-coupled-
unsymmetrical behavior cannot be accommodated by cohesive elements; the same is also true for
any kind of nonlinear behavior.
c. Connector Elements for General Aggregate Interlock Idealization
204
The main limitation of the JOINTC and COH3D8 elements is their inability to discretize the
more general nonlinear aggregate interlock load transfer mechanism represented by Eq. (5.2).
Consequently, another element type was sought in this study, and 3D connector element
(CONN3D2) was found to be the best candidates. This element can define a relative
displacement and rotation between two nodes in three orthogonal directions. ABAQUS® provides
varieties of connector types to accommodate all possible relative displacements and rotations
between the connected nodes, such as simple translational and rotational displacements. Among
a library of connector types, PROJECTION CARTESIAN was deemed to be appropriate for
idealizing both linear and nonlinear aggregate interlock behavior, provided adequate techniques
are devised to establish the required material properties, represented by the complete set of
matrix coefficients in Eq. (5.2).
Like JOINTC, a connector element is defined between two symmetrically located nodes along
the joint plane. When a connector element is used to discretize linear aggregate interlock
behavior, its stiffness coefficients (Knn and Ktt) are identical to those of a JOINTC element.
Nonlinear aggregate interlock behavior, on the other hand, requires the definition of the entire
force-displacement rule to be followed by ABAQUS® during the analysis.
In order to generate the force-displacement data required as inputs to the program from
Walraven’s constitutive equations, the following steps are followed: (a) Assume suitable ranges
for the normal and shear displacements. From Fig. 5.1, it is apparent that the interface stresses
are not influenced by normal and shear displacement values greater than 0.1 in. Therefore, the
upper limit in both cases may be set to 0.1 in. The lower limit can be zero for shear but needs to
be a small finite value for normal displacement, as indicated above (e.g. 0.0001 in.). To generate
205
a smooth curve, thereby avoiding any convergence problems during the FE analysis to follow,
each range should be subdivided into a sufficient number of intervals. The value of 100 was
found to ensure a convergent FE solution for both displacements. (b) Select an appropriate
numerical computing package to perform the integration for the most probable projected contact
areas, Ax and Ay , from Walaraven’s equations given in Table 5.1. The required variables are the
maximum aggregate size (Dmax) and the aggregate volume per unit volume of concrete (pk). The
latter is usually assumed to be 0.75 (Walraven, 1981). In this study MATLAB was selected for
computing the integrands “to within an error of 10-12 using recursive adaptive Lobatto
quadrature” (Moler, 2004). (c) Use Eqs. (5.3) and (5.4) to obtain the interface normal and shear
stresses (  n and  t ). (d) Compute the corresponding normal and shear forces by multiplying
these stresses by the contributing areas defining the connector element, i.e., the areas of those
elements located at the corner, edge, and interior of the joint plane, as done for the linear
aggregate interlock simulation. These forces are then provided as inputs to ABAQUS® manually
in tabular form, as functions of the displacements.
Examining the nonlinear aggregate interlock constitutive equations provided by Walraven, it is
observed that the crack stiffness coefficients given in Eq. (5.2) are unsymmetrical (i.e., Knt ≠ Ktn).
Consequently, unsymmetrical matrix storage is adopted in this study for nonlinear aggregate
interlock load transfer simulation; this is accomplished by using the keyword UNSYMM
(ABAQUS, 2009).
5.3.2 Validation of Linear Aggregate Interlock Discretization
206
The concrete pavement system described in Table 5.2 is used in this section to verify the
proposed linear aggregate interlock FE discretization. A pressure of 0.5 ksi is applied over a 2ϵ ×
2ϵ (12 by 12 in.) area, symmetrically located about the midpoint of the joint on the edge of the
loaded slab side. Here 2ϵ represents the side of the square area. The modulus of subgrade
reaction, l, is determined using the pertinent properties given in Table 5.2 to be 24.63 in. The
equivalent radius of the applied load, a, is 6.77 in. Therefore, a/l and ϵ/l becomes 0.275 and
0.244, respectively. Finite element runs are made for the eight assumed (AGG/kl) values given in
Table 5.3, using each of the three linear aggregate interlock discretization options outlined
above, involving JOINTC, COH3D8 and CONN3D2 elements. The following results are
monitored: load transfer efficiency with respect to deflection (LTEδ), calculated as the ratio of the
deflection at the center of the edge of the unloaded slab to the deflection of the loaded slab at that
point; transferred load efficiency (TLE), which is equal to the ratio of the sum of the vertical
reaction forces in the subgrade springs under the unloaded slab to the total applied load; and load
transfer efficiency with respect to bending stress (LTEσ), determined as the ratio of the bending
stress at the bottom of the unloaded slab at a point corresponding to the center of the edge of the
slab divided by the bottom fiber bending stress at that point on the loaded side of the joint.
The results are presented in Figs. 5.2 and 5.3, in which it is observed that the three linear
aggregate interlock idealizations exhibit excellent agreement with one another for all the
responses monitored. The CONN3D2 and JOINTC elements give exactly identical results with
each other, affirming the use of either of these elements in idealizing linear aggregate interlock.
The responses resulting from COH3D8 elements are observed to be slightly smaller than those
from the other two options. This is attributable to the surficial load transferring mechanism
207
pertaining to the COH3D8, in contrast to the nodal load transfer prevailing for the other two. For
LTEδ and TLE, almost identical results are obtained by the present study and those from
Ioannides et al. (1996) and Ioannides and Korovesis (1990a), respectively. For the LTEσ, on the
other hand, as joint stiffness increases, the present study points are lower than that by Ioannides
et al. (1996). This may be attributed to mesh fineness, plate versus continuum formulation and
point versus surficial load transfer mechanism. All three FE options in Fig. 5.3 result in data
points located slightly above the curve for the 2D FE results established by Ioannides and
Hammons (1996). The discrepancies may be caused by a combination of the theoretical approach
(continuum versus plate theory) as well as the difference in joint discretization.
It should also be noted that unlike JOINTC and CONN3D2 elements, meshing of traction-
separation COH3D8 elements can be refined as needed without any changes to the mesh of the
rest of the slab. Moreover, these elements might be useful in the future for simulation of linear
aggregate interlock with a coupled-symmetrical crack stiffness matrix, a capability not shared by
JOINTC. Therefore, in the subsequent post-crack analysis of the slabs, cohesive and connector
(COH3D8 and CONN3D2) elements are used to idealize linear and nonlinear aggregate interlock
mechanisms, respectively. This constitutes a departure from the conventional approach employed
for linear aggregate interlock idealization to date, which has favored the use of spring-type
(JOINTC) elements, per Hammons (1997).
A second observation is warranted at this point with regard to the method of load application. It
has been found that the uniform pressure approach described above gives rise to convergence
problems when retained in investigating post-fracture behavior. Consequently in what follows, it
is explained that a point load is applied symmetrically from the assumed fracture plane, instead.
208
5.3.3 Validation of Nonlinear Aggregate Interlock Discretization
The accuracy of the proposed nonlinear aggregate interlock simulation is verified by reproducing
the test data reported by Colley and Humphrey (1967). They examined the load transfer
efficiency of an aggregate interlock joint between two slab panels, for different concrete slab
thicknesses, joint openings, and base layers under static and cyclic loads. Davids and Mahoney
(1999) also reproduced this experimental study using EverFE, and their results are available for
comparison, as well. According to Colley and Humphrey (1967), the so-called “equivalent
subgrade reaction of the [sand-gravel] base and silt-clay soil” was taken as 145 pci. The size of
the joint opening was controlled by steel rods, anchored in the vertical faces of the two slabs
parallel to the joint line. The upward movement of the far ends of the two slabs during the
loading process was constrained by beams spanning the slab width. A static load of 9 kips was
applied over a 16-in. diameter steel bearing plate, placed at 1 in. from the joint, at mid-width on
the loaded slab.
The FE simulation conducted in the present study considers the two 7-in. thick, 48-in. wide and
108-in. long slab panels, resting on SPRING1 elements representing modulus of subgrade
reaction, k, of 145 pci (see Fig. 5.4). The concrete Young’s modulus is taken to be 4495 ksi, with
Poisson’s ratio of 0.2, per Davids and Mahoney (1999). Parameters needed for the nonlinear
aggregate interlock constitutive relations in Table 5.1 are selected per Walraven (1981) and
Davids and Mahoney (1999), as follows: maximum aggregate size of 1.5 in.; cement paste-
aggregate particle coefficient of friction of 0.4; concrete compressive strength of 5.5 ksi; and
aggregate volume per unit volume of concrete (pk) of 0.75. The required input forces for the
CONN3D2 elements are evaluated using Table 5.1, Eqs. (5.3) and (5.4). MATLAB is employed to
209
compute the most probable projected contact areas given in Table 5.1, using the steps given
earlier. The interface normal and shear stresses are then computed from Eqs. (5.3) and (5.4).
These stresses are multiplied by the contributing areas around the nodes that define the connector
element to obtain the corresponding normal and shear forces. Finally, the forces are provided as
input to ABAQUS® in a tabular form, depending on assumed interface normal and shear
displacements, as already discussed.
The two slabs are discretized using a uniform mesh with 3×3×1.75 in. elements, in the length,
width and thickness directions, respectively. The lateral supports provided by the anchored steel
rods during the experiment by Colley and Humphrey (1967) is simulated by restraining the nodal
displacements in both horizontal directions along the slab width at mid-depth. Moreover, the top
nodes along the far end edges are also restrained against vertical movement, so as to idealize the
effect of the transverse beams used in those locations. For mesh expediency, the 9-kip static load
is applied over a 12 by 12-in. area, resulting in a pressure of 0.0625 ksi.
Colley and Humphrey (1967) presented their results in terms of joint effectiveness (Eff), first
defined by Teller and Sutherland (1935) as:
2U 2 LTE
Eff  %   100  (5.12)
 L  U 1  LTE
where:  L is the vertical deflection of the loaded slab; U is the corresponding deflection for the
unloaded slab; and LTE is the load transfer efficiency with respect to deflection, given as:
U
LTE  %   100 (5.13)
L
210
The variation of LTEδ with respect to the assumed values of the initial joint opening, ωi, obtained
from the present FE study is plotted in Fig. 5.5, along with the experimental data from Colley
and Humphrey (1967) and the EverFE results obtained by Davids and Mahoney (1999),
respectively. The three studies generally exhibit good agreement, especially at smaller joint
openings. As the joint opening increases, EverFE estimates LTEδ values between those from the
experiment and from the present study. This is probably attributable to the fact that Davids and
Mahoney (1999) accounted for a 1-in. groove at the top and bottom of the joint during the
experiment, by inserting zero stiffness interface elements. To avoid the additional complexity it
would entail, this effect was not considered in the present study. In general, the disagreements
noted are insignificant for practical purposes. Therefore, the present nonlinear aggregate
interlock joint discretization can be deemed suitable for the post-crack analysis of jointed
pavement slabs in subsequent sections.
5.4 Post-Crack Analysis of Slabs with Aggregate Interlock under Wheel Load Alone
The geometric and material properties of the two identical slabs considered in Section 5.3.2 are
retained here, per Table 5.2 and Fig. 5.6. The fracture process zone (FPZ) is idealized using the
FCM per Hillerborg et al. (1976), realized by traction-separation cohesive elements, as in a
previous phase of this research (Aure and Ioannides, 2012). Cohesive elements are inserted along
the pre-defined fracture planes, where principal tensile stresses are expected, as shown in Fig.
5.6.
A bilinear concrete softening rule proposed by Petersson (1981) and used by Gustafsson (1985)
is employed. The pertinent fracture parameters defining the softening rule are given in Table 5.2.
211
These values are identical to those used in previous single-slab fracture analysis by Ioannides et
al. (2006) and by Aure and Ioannides (2012).
The load is applied as a unit displacement at each of two points, 3 in. away from the face of the
joint and 3 in. on either side of the anticipated fracture plane. To avoid convergence problems
that commonly arise during stiffness degradation of cohesive elements, viscous regularization is
employed with a viscosity, м, of 1×10-6. This value had been obtained from a sensitivity study
conducted for single-slab fracture analysis (Aure and Ioannides, 2012). A general Newton-
Raphson solution method is used, since it is less computationally demanding.
In addition to LTEδ and TLE defined earlier, the following responses are monitored: load line
displacement (LLD), i.e., the vertical displacement at either point of the applied load; the total
applied load (P), i.e., the sum of the conjugate “reaction” forces at the two loaded nodes, RF3;
and, load transfer efficiency with respect to crack mouth opening displacement (LTEO),
calculated as the ratio of the crack mouth opening displacement (CMOD) at the bottom edge of
the unloaded slab to that at the corresponding point on the loaded side of the slab joint. Analysis
results for both linear and nonlinear aggregate interlock simulations are presented in the
following sections.
5.4.1 Linear Aggregate Interlock
Finite element runs are made for the AGG/kl values assumed in Table 5.3, and the load-
displacement responses for some of the joint stiffnesses are shown in Fig. 5.7. Note that
COH3D8 elements are used to discretize the joint. It is observed that at smaller joint stiffness
values, e.g., AGG/kl = 0.01, the curves exhibit only one softening segment. This indicates that
212
the loaded slab has completely failed along the assumed fracture plane, while the unloaded slab
has remained intact. As the joint stiffness increases, a larger portion of the applied load begins to
be transferred and cracks begin to develop in the unloaded slab, as well. Consequently, two
softening regions are exhibited in the load-displacement curves for intermediate joint stiffness
values (e.g., AGG/kl = 0.1 and 1.0), indicating the failure of both slabs: the first peak load
indicates the failure of the loaded slab, whereas the second evinces the succumbing of the
unloaded slab. For AGG/kl = 10.0 and 100.0, only one softening event is again observed,
signifying that the two slabs are acting monolithically; the peak load represents the maximum
load that the two slabs can support together, before transverse cracking develops simultaneously
across both. In general, as the joint stiffness increases, the pavement system becomes more
competent, thereby causing the first peak load to increase continuously. The second peak load,
on the other hand, may develop at a relatively high applied load, but upon further joint stiffening,
its value quickly diminishes and it eventually coincides with the first peak load.
The effect of the dimensionless joint stiffness on several response parameters computed at the
first peak load is shown in Fig. 5.8. These parameters are: load transfer efficiency with respect
to deflection, LTEδ, and with respect to crack mouth opening displacement, LTEO; and
transferred load efficiency, TLE. Also shown are the elastic (pre-crack) curves reproduced from
the FE runs in Section 5.3.2. It is observed that after cracking, LTEδ remains almost identical to
the corresponding elastic curve, particularly for those cases exhibiting one softening segment.
This might be anticipated since joint characteristics are not altered by cracking, beyond any slab
size effects. The post-crack TLE, on the other hand, is slightly higher than the corresponding pre-
crack response particularly at higher joint stiffnesses, which may be due to differences in the
213
mode of load application in the two cases. These observations have the potential for some
interesting practical implications: knowledge of pre-crack joint responses, per Skarlatos (1950),
may hold clues for predicting aspects of post-cracking behavior, assuming that the dimensionless
joint stiffness does not depend on the magnitude of the applied load. The LTEO is observed to be
lower than the LTEδ for all joint stiffness values, but the two curves approach each other as joint
stiffness increases.
The variation of LTEδ with the total applied load for two typical joint stiffness values is further
examined in Fig. 5.9. It is observed that for AGG/kl =1, LTEδ remains constant (0.46) until the
total applied load equals 13.6 kips, which from Fig. 5.7 corresponds to the point at which the
load-displacement curve begins to change in slope. This point evinces the beginning of cracking
in the loaded slab. The LTEδ then slightly decreases until the total applied load becomes 41.2
kips (LTEδ = 0.39), which corresponds to the first peak load (failure of loaded slab). After this
load level, it increases until the total load equals 54.3 kips, which is the second peak load (failure
of unloaded slab) on the load-displacement curve shown in Fig. 5.7. A similar behavior is
exhibited for AGG/kl = 100, except that the offset in LTEδ becomes very small for the latter
between the beginning and end of cracking in the slabs.
5.4.2 Nonlinear Aggregate Interlock
In this section, the proposed nonlinear aggregate interlock joint discretization using CONN3D2
elements is employed to investigate post-crack responses of concrete pavement slabs. The
geometry of the two slabs is shown in Fig. 5.6, and the pertinent material properties as well as
the FE discretization employed are given in Table 5.2. The parameters required for Walraven’s
214
constitutive relations are also given in Table 5.2. The effects of initial joint opening and
maximum aggregate sizes on the responses are considered in subsequent sections.
a. Effect of Initial Joint Opening
Shrinkage and seasonal temperature variations may cause opening or closing of joint openings in
situ concrete pavement slabs. Application of mechanical loads may cause additional opening
changes, which may even be non-uniform over the joint depth, resulting in concomitant changes
in joint stiffness. Such fluctuations influence the transfer of deflections and of stresses to the
unloaded slab. The linear elastic aggregate interlock simulation presented previously cannot
accommodate this aspect of behavior, since that idealization simply assumes a constant joint
stiffness irrespective of the change in joint characteristics. The proposed nonlinear aggregate
idealization, on the other hand, accounts for the initial joint width, as well as for its subsequent
variation. Accordingly, the shear and normal forces occurring at the vertical faces of the joint are
allowed to vary in response to all possible normal and shear displacements.
To investigate the effect of initial joint opening (ωi), FE runs considering slabs with ωi values
ranging from 0.02 in. to 0.09 in. are conducted, and the load-displacement responses for some of
the joint openings are presented in Fig. 5.10. It is observed that as initial joint opening increases,
the slopes of the curves gradually decrease, denoting a reduction in the stiffness of the system.
This is due to the fact that in larger openings fewer aggregate particles can participate in
interlocking. As had been the case for linear aggregate interlock, two softening segments are
observed on these curves. The first (and lower) peak in the applied load indicates the failure of
the loaded slab, whereas the second (and higher) applied load peak denotes the failure of the
215
unloaded slab. With increasing joint opening, the second peak load becomes even higher, and
occurs much later than the first one, signaling that the decreased load transfer efficiency causes
the unloaded slab to fail at a higher level of the total applied load. This ensures the longevity of
the unloaded slab, whose diminished participation in resisting the applied load occurs at the
expense of the loaded slab. Also noteworthy in Fig. 5.10 is the nonlinearity of the curves,
especially discernible in the portions near the origin. Nonlinearity during the initial stages of
loading has never been noted before, and may be caused by the nonlinearity of the aggregate
interlock and/or cracking in the loaded slab.
Figure 5.11 presents plots of post-crack TLE, LTEδ and LTEO versus initial joint opening,
determined at the first peak load values. It is observed that all responses decrease linearly as
initial joint opening increases, although the trend exhibited by the TLE is much fainter, evincing
a smaller sensitivity to initial joint opening. Variations in the transferred load are muted by the
inertia of the pavement system, and reflect more global phenomena than localized cracking,
thereby thwarting uncontrollable behaviors.
The variation of LTEδ with respect to the applied load is plotted in Fig. 5.12 for initial joint
openings of 0.02 in. and 0.09 in. The results indicate that as the load is applied, LTEδ initially
increases up to a certain maximum value at the first peak load (representing the failure of the
loaded slab), and then gradually decreases as additional load is applied. This phenomenon is
caused by Walraven’s constitutive relations, which result in high interface stresses up to a certain
maximum value when the shear displacement increases and the normal (joint opening)
displacement decreases, as shown in Fig. 5.1. At early stages of load application, joint opening is
small and, therefore, aggregate particles interlocking capacity increases with applied load
216
thereby enhancing joint stiffness. As the joint opening and shear displacements continue to
increase, aggregate particles are no longer in contact, and therefore, the interface stresses
decrease, as depicted in Fig. 5.1. Consequently, beyond the second peak load (at which the
unloaded slab fails), the joint’s load transfer capability declines as the total applied load
increases. The effect of the size of initial joint opening on load transfer efficiency is also clearly
observed in Fig. 5.11 by comparing curves corresponding to the 0.02 in. and 0.09 in. openings.
b. Effect of Aggregate Size
One of the influential parameters in Walraven’s aggregate interlock constitutive equations is the
maximum diameter (Dmax) of the aggregate particles. Walraven (1981) conducted a sensitivity
study to investigate the effect of this parameter on interface stresses. He observed that the
interface shear stress is more influenced than the normal stress; the larger the maximum
aggregate size, the stiffer the shear stress versus shear displacement curve, especially at large
initial crack openings. This observation is important in aggregate interlock joints of concrete
pavements that rely on the interface shear stresses as load transfer mechanism.
The effect of maximum aggregate size on post-crack responses of pavement slabs is examined in
the present study by considering aggregate sizes of 0.65 in. and 1.3 in., while keeping other
parameters in Walraven’s constitutive relations unchanged (see Table 5.2). It may be argued that
the change in aggregate size may influence other material properties (e.g., compressive strength),
but this is considered beyond the scope of the proposed analysis. The simulation is conducted for
the two initial joint openings of 0.02 in. and 0.09 in., and the load-displacement responses are
presented in Fig. 5.13. It is observed that aggregate size matters little when the initial joint
217
opening is small. On the other hand, as initial joint opening increases, more interlocking can be
expected as Dmax increases, and therefore, larger aggregate particles result in stiffer joint
behavior, manifested in the reduced second peak load at which the unloaded slab failed. A
dimensionless parameter, ωi/Dmax, in conjunction with the peak load may be a useful parameter
to represent the effect of aggregate size on the slab response. It is concluded, therefore, that
nonlinear aggregate interlock simulation in tandem with fracture analysis is essential in assessing
the effect of joint opening and aggregate size.
5.5 Pre- and Post-Crack Analysis of Slabs with Aggregate Interlock under Constant
Temperature Distribution and Increasing Wheel Load
Diurnal variations of temperature cause stresses in concrete pavement slabs, especially when
combined with wheel loads, thereby reducing the system’s longevity. During the daytime, the
top surface of the slab is warmer than the bottom and the slab curls downward; the bottom of the
slab experiences tensile stresses, as well as widening of the joint opening. During the nighttime,
the opposite phenomenon occurs: the slab curls up, and tensile stresses arise at the top of the
slab, where joint opening broadens.
The effect of curling stresses on the pre- and post-crack responses of a single pavement slab has
been investigated in Chapter 4 of this study. It was observed that daytime temperature variation
through the slab thickness results in unstable crack formation and reduces significantly the peak
load resisted by the slab. In contrast, a nighttime temperature gradient spawns stable cracks and
a higher peak load. It was also observed that the fracture process depends not merely on the
extreme-fiber curling stresses, but on the entire stress distribution through the thickness, re-
emphasizing the importance of temperature profile estimation in concrete pavement structures.

218
In the present study, the effect of curling stresses on the pre- and post-crack response of the two
slabs interconnected by aggregate interlock is investigated. The geometry of the setup is shown
in Fig. 5.6, and the pertinent properties are given in Table 5.2. Constant temperature profiles
measured by Teller and Sutherland (1935) at 7:00 pm and 9:30 am of June 12-13, 1932 are
considered; these are typical nighttime and daytime variations through a 6 in. slab. A cubic
polynomial was found in Chapter 4 to fit adequately these temperature variations; the following
best fit equations are used to assign temperature values at nodes through the slab thickness:
T  z   0.333z 3  3.446 z 2  5.042 z  78.036 at 9:30 am R 2  0.9595 (5.12a)
T  z   0.017 z 3  0.060 z 2  0.295z  90.995 at 7:00 pm R 2  0.9969 (5.12b)
Here z is measured upward from the bottom of the slab. It is assumed that the temperature
remains the same at all nodes lying on the same horizontal plane.
As described in Chapter 4, the analysis is carried out in two steps. In the first instance, the
constant temperature distributions given in Eq. (5.12) are applied; the application of the
increasing wheel load follows thereafter. The latter is represented by a unit displacement as
described in earlier sections. Both linear and nonlinear aggregate interlock idealizations are
considered. Note that the second loading scenario, in which a constant wheel load is applied first
followed by an increasing thermal curling, is not considered in this chapter.
5.5.1 Linear Aggregate Interlock
The resulting load-displacement curves for the specified nonlinear daytime temperature profile
(resulting in temperature differential, ΔT = 22°F) at three selected joint stiffness values are
shown in Fig. 5.14. The peak loads in this case are significantly lower than those for no-curling,
219
presented earlier in Fig. 5.7, which had ranged between 30 and 60 kips. This is attributed to the
curling stresses, which consume a substantial portion of the slab’s load-carrying capacity, even
prior to the application of the wheel load. At lower joint stiffness values, only one softening kink
is observed, indicating the failure of the loaded slab alone; this was also observed earlier in the
case of no curling. As the joint stiffness increases, the unloaded slab starts to share the applied
load and two softening regions may begin to be manifest. Finally, at very large joint stiffness
values, for example, AGG/kl=100, only one softening is observed once more, evincing that the
two slabs act monolithically. The translation of the load-displacement curves to the right with
decreasing joint stiffness is worth noting. This reflects the curl-only deflections at the onset of
loading, which are determined by the degree of monolithic or independent action of the two
slabs, depending on the joint stiffness.
A similar approach has been followed for the analysis of the slab under nonlinear nighttime
temperature distribution (resulting in ΔT = 4°F) and wheel loading. The corresponding load-
displacement curves obtained are shown in Fig. 5.15, in which the peak loads are significantly
higher than those for no-curling (see Fig. 5.7). Comparing the peak loads from these curves with
those in Fig. 5.14, the first peak nighttime load (at which softening begins) is found to be much
larger than for daytime temperatures for each joint stiffness value considered. At AGG/kl = 100,
for example, this load is six times higher than during the daytime (96 kips compared to 12 kips).
This change is attributable to compressive curling stresses arising during the nighttime at the
bottom of the slab, counterbalancing the tensile stresses due to the wheel load.
As a result of temperature variation through the thickness of the slab, the two joint faces, which
are initially vertical and parallel to one another (resembling the letter H), rotate with respect to
220
one another and assume inclined orientations, resembling letters V or Λ. Such movement
consists of displacements normal to the joint surfaces, which contribute to a change in the joint
stiffness. This is the essence of the coupling phenomenon between normal and shear interface
stresses, which the linear aggregate interlock idealization cannot simply accommodate. This
limitation is addressed by nonlinear aggregate interlock, as discussed below.
5.5.2 Nonlinear Aggregate Interlock
For the case of nonlinear aggregate interlock, load-displacement results for daytime temperature
are shown in Fig. 5.16 for three selected initial joint opening values. Two softening regions are
observed once again, an effect captured by the cascading failure of the loaded and unloaded
slabs, respectively. When compared to the responses from the cases without curling presented in
Fig. 5.10, the daytime peak loads are significantly lower (8 to 15 kips compared to 70 to 95
kips), as had been the case for linear aggregate interlock, as well.
During the daytime, the joint widens at the bottom and progressively closes at the top of the slab,
assuming a configuration resembling the letter Λ. When the wheel load is applied, the joint
opening continues to broaden at the bottom and remains closed at the top, thereby causing a
decrease in load transfer efficiency. This results in a reduction of the pavement system
competence, manifested in the reduced initial slope of the load-displacement curves as initial
joint opening increases.
Comparison of Fig. 5.17 with Fig. 5.10 indicates that peak loads increase very slightly during the
nighttime. A nighttime temperature variation results in progressive joint closing at the bottom of
the slab and widening at the top, in a configuration resembling the letter V. This may be
221
expected to increase joint efficiency, an effect that can be particularly significant at large
nighttime temperature differentials. This is verified by conducting additional nonlinear aggregate
interlock runs involving an initial joint opening of 0.07 in., positive (i.e., daytime) and negative
(i.e., nighttime) linear temperature differentials of 6, 12, and 18°F. Variations of transferred load
efficiency at the first peak load with respect to the temperature differentials are presented in Fig.
5.18. It is observed that during the nighttime, load transfer efficiency remains unaffected by an
increase in the temperature differential, while during the daytime it is inversely proportional to it.
Therefore, daytime temperature is deemed critical, not only because of the additional curling
stresses induced at the bottom of the slab, but also due to the loss of joint efficiency, which
hastens the failure of the loaded slab.
Simulations using different techniques for idealizing aggregate interlock load transfer in jointed
concrete pavement slabs are presented. Both pre- and post-crack analyses are conducted.
Previously published numerical and experimental studies are available only for pre-crack
conditions, and they are used to validate the proposed formulation. Post-crack responses are
used to investigate crack growth in both the loaded and the unloaded slabs, for linear as well as
nonlinear aggregate interlock idealizations. For linear aggregate interlock, it is found that at
smaller joint stiffness values, there is only one peak load, indicating the failure of only the loaded
slab. As the joint stiffness increases, however, softening begins to be manifest at two points on
the load-displacement curve, evincing the successive failure of both the loaded and unloaded
slabs. At a very large joint stiffness, the two slabs act as one unit, and only one softening region
is observed once again, at a relatively large peak load. The post-crack load transfer efficiencies
222
with respect to vertical deflections and the applied load are observed to be similar to those
reported in the literature for uncracked slabs. In general, the load transfer efficiency with respect
to crack mouth opening displacement (LTEO) is found to be lower than load transfer efficiency
with respect to deflection (LTEδ).
Unlike its linear precursor, nonlinear aggregate interlock idealization takes into account the
change in the joint opening during the loading process, and the concomitant change in the load
transfer efficiency of the joint. It is observed that the load transfer efficiency with nonlinear
aggregate interlock decreases linearly with increasing initial joint opening. It is also found that
LTEδ continuously increases until the peak load is attained and decreases thereafter.
The effects of both nighttime and daytime temperature are also investigated for the linear as well
as the nonlinear aggregate interlock idealizations. Daytime temperature variations result in a
lower peak load, and in decreased joint efficiency on account of a widened joint opening at the
bottom of the slab. Nighttime temperature curling, on the other hand, has the opposite effect and
may prevent a decrease of joint efficiency, if it does not actually enhance joint behavior.
When viewed in conjunction with earlier studies (Aure and Ioannides, 2010; Aure and Ioannides,
2012), this study affirms the conclusion that application of cohesive elements to the fracture
analysis of concrete pavement slabs is promising. The approach will be extended to more
complex in situ pavement systems that involve dowel-jointed slabs under combined wheel and
thermal loadings in the next Chapter. Experimental studies that can verify such numerical
approaches need to be conducted both in the field and in the laboratory, thereby contributing to
the ongoing development of rational failure criteria that can replace the statistical/empirical
223
algorithms along with Miner’s (Miner, 1945) hypothesis that are currently relied upon in
mechanistic-empirical pavement design procedures.
5.7 References
Providence, RI.
Bažant, Z. P., and Gambarova, P. (1980). "Rough cracks in reinforced concrete." Journal of the
Structural Division, 106(ST4), ASCE, New York, NY, 819-842.
Brink, A. C., Horak, E., and Visser, A. (2005). "Improvement of aggregate interlock equation
used in mechanistic design software." Int. J. Concrete Pavements, 1(1), Taylor and Francis
Group, London, UK, 1-22.
Colley, B. E., and Humphrey, H. A. (1967). "Aggregate interlock at joints in concrete
pavements." Bulletin 189, Highway Research Board, National Research Council, Washington,
DC, 1-18.
224
Davids, W. G. (1998). "Modeling of rigid pavements: joint shear transfer mechanisms and
solution strategies," Ph.D. dissertation, University of Washington, Seattle, WA.
Davids, W. G., and Mahoney, J. P. (1999). "Experimental verification of rigid pavement load
transfer modeling with EverFE." Transp. Res. Rec. 1684, Transportation Research Board,
National Research Council, Washington, D. C, 81-89.
Divakar, M. P., and Fafitis, A. (1992). "Micromechanics based constitutive model for interface
shear." J. Eng. Mech., 118(7), ASCE, New York, NY, 1317-1337.
Divakar, M. P., Fafitis, A., and Shah, S. P. (1987). "Constitutive model for shear transfer in
cracked concrete." J. Struct. Eng., 113(5), ASCE, New York, NY, 1046-1062.
Fuller, W.B. and Thompson, E. (1906). "The laws of proportioning concrete." Trans., ASCE,
LVII(2), New York, NY, 67-143.
225
Ioannides, A. M., Peng, J., and Swindler Jr, J. R. (2006). " ABAQUS model for PCC slab
Ioannides, A. M., and Hammons, M. I. (1996). "Westergaard-type solution for edge load transfer
problem." Transp. Res. Rec. 1525, Transportation Research Board, National Research Council,
Ioannides, A. M., Alexander, D. R., Hammons, M. I., and Davis, C. M. (1996). "Application of
artificial neural networks to concrete pavement joint evaluation." Transp. Res. Rec. 1540,
Ioannides, A. M., and Korovesis, G. T. (1990a). "Aggregate interlock: a pure-shear load transfer
mechanism." Transp. Res. Rec. 1286, Transportation Research Board, National Research
Ioannides, A.M., and Korovesis, G.T. (1990b). "Backcalculation of joint related parameters in
concrete pavements," Proc., Third International Conference on the Bearing Capacity of Roads
and Airfields, Norwegian University of Science and Technology, Trondheim, Norway, 549-558.
226
Jensen, E. A., and Hansen, W. (2006). "Nonlinear aggregate interlock model for concrete
pavements." Int. J. Pavement Eng., 7(4), Taylor and Francis Group, London, UK, 261-273.
Maitra, S. R., Reddy, K. S., and Ramachandra, L. S. (2010). "Load transfer characteristics of
aggregate interlocking in concrete pavement." J. Transp. Eng., 136(3), ASCE, Reston, VA, 190-
195.
Moler, C. B. (2004). Numerical Computing with MATLAB. Society for Industrial and Applied
Mathematics, Philadelphia, PA.
Reinhardt, H. W., and Walraven, J. C. (1982). "Cracks in concrete subjected to shear." J. Struct.
Div., 108(ST1), ASCE, New York, NY, 207-224.
Skarlatos, M. S. (1950). "Deflections and stresses in concrete pavements of airfields with
continuous elastic joints." Rep. No. AD628501, U. S. Army Corps of Engineers, Ohio River
Division Laboratories, Mariemont, OH.
227
16(9), Bureau of Public Roads, Washington, DC, 169-197.
Walraven, J. C. (1994). "Rough cracks subjected to earthquake loading." J. Struct. Eng., 120(5),
ASCE, New York, NY, 1510-1524.
Walraven, J. C. (1981). "Fundamental analysis of aggregate interlock." J. Struct. Div.,
107(ST11), ASCE, New York, NY, 2245-2270.
Wattar, S. W. (2001). "Aggregate interlock behavior of large crack width concrete joints in PCC
airport pavements," Ph.D. dissertation, University of Illinois at Urbana-Champaign, Urbana, IL.
228
Table 5.1 Walraven’s Equations for Contact Areas per Unit Crack Area
Case A :  t   n
Dmax 4 pk  D 
Ay   n2 t2 F    G1  n ,  t , D  dD
t   Dmax 
Dmax 4 pk  D 
Ax   n2 t2 F    G2  n ,  t , D  dD
t   Dmax 
Case B :  t   n
 n2  t2
4 pk  D  Dmax 4 p  D 
Ay   F   G3  n ,  t , D  dD   n2 t2 F   G1  n ,  t , D  dD
n k
2 n   Dmax  n   Dmax 
 n2  t2
4 pk  D  Dmax 4 p  D 
Ax   F   G4  n ,  t , D  dD   n2 t2 F   G1  n ,  t , D  dD
n k
2 n   Dmax  n   Dmax 
with
 t 


G1  n ,  t , D   D 3  D 2   n2   t2    umax   n  umax  umax
2


 n2   t2
 n 1 2


G2  n ,  t , D   D 3  t  D 2   n2   t2    umax   umax   t  
4
D   n  umax 
2
 n  t
2 2
  
1 2 1  u  1 2  2 n 
 n  D   n2  D 2 arcsin  n max   D arcsin  
4 4 1
 D  4  D 
 2  
229
Table 5.1 (Continued)
2
1 
G3  n ,  t , D   D  D   n 
3
2 
 1 2 1  2 
G4  n ,  t , D   D 3  D 2   n D   n2  D 2 arcsin  n  
8 4 4  D 
0.5 4 6 8 10
 D   D   D   D   D   D 
F   0.532    0.212    0.072    0.036    0.025  
 Dmax   Dmax   Dmax   Dmax   Dmax   Dmax 
1 1  2 D 
2
       
2 2
 n  n2   t2   n2  n2   t2 2
n  t
2
  n  t
2
 4 t   
2 2   2  
umax 
 n2   t2
pk  aggregate volume/unit volume of concrete
230
Table 5.2 Geometry, Material Properties and Discretization of Two Slabs Considered
Geometry of Each Slab

Length (in.) 240
Width (in.) 120
Thickness (in.) 6
Material Properties
Thermal Expansion Coefficient (in./in./°F) 5×10-6
Compressive Strength (ksi) 5
Unit Weight (pcf) 150
Walraven’s Parameters
Maximum Aggregate size (in.) 0.65
Coefficient of Friction 0.4
Aggregate Volume per Unit Volume of Concrete 0.75
Modulus of Subgrade Reaction (pis/in.) 200
Slab Element Size (in.) 6×6×2
Softening Bilinear
Solver Newton-Raphson
Subgrade Idealization SPRING1
Notes: 1 lb = 4.444 N; 1 in. = 25.4 mm; 1 ksi = 6.89 Mpa
231
Table 5.3 Joint Stiffness Characteristics Used for Linear Aggregate Interlock
Assumed JOINTC or CONN3D2 COH3D8

AGG
AGG Ktt (kips/in.) Ktt (kips/in.)
kl
(ksi)
By Eq. (5.7) By Eq. (5.10)
values
0.001 0.00493 0.00062 0.00082
0.010 0.04930 0.00616 0.00821
0.100 0.49300 0.06159 0.08212
1.000 4.93000 0.61587 0.82116
10.000 49.3000 6.15870 8.21160
100.000 493.000 61.5870 82.1160
1000.000 4930.00 615.870 821.160
10000.000 49300.0 6158.70 8211.60
Note: For JOINTC or CONN3D2: Kss= 0 and Knn is determined from Eq. (5.6);
Edge stiffness = 2 × corner stiffness; Interior stiffness = 4 × corner stiffness;
For COH3D8: Kss= Ktt and Knn is determined from Eq. (5.11);
Eh3
l4 is radius of relative stiffness and k is modulus of subgrade reaction.

12 1   2 k 
232
1.80 δn=0.005 in.
δn=0.01 in.
1.40
Shear Stress (ksi)
1.00 δn=0.03in.
0.60 δn=0.05 in.
δn=0.07 in.
0.20
δn=0.07 in.
-0.20
δn=0.05 in.
-0.60 δn=0.03 in.

Normal Stress (ksi)
-1.00
δn=0.01 in.
δn=0.005 in.
-1.40
-1.80
0 0.02 0.04 0.06 0.08
Shear Displacement (in.)
Figure 5.1 Typical Normal and Shear Stress Variations at Different Crack Openings
233
1.0
JOINTC
COH3D8
CONN3D2
Ioannides et al. (1996)
0.8 Ioannides and Korovesis (1990a)
LTEσ
LTEδ, LTEσ or TLE
0.6
LTEδ
Note: a/l = 0.275
TLE
0.4
0.2
0.0
1.0E-03 1.0E-02 1.0E-01 1.0E+00 1.0E+01 1.0E+02 1.0E+03 1.0E+04
AGG/kl
Figure 5.2 Comparison Among Three Linear Aggregate Interlock Idealizations and
Previous Studies
234
1.00
0.80
JOINTC
COH3D8
CONN3D2
Ioannides and Hammons (1996)
0.60
Note: a/l = 0.275
LTEδ
0.40
0.20
0.00
0.00 0.20 0.40 0.60 0.80 1.00
LTEσ
Figure 5.3 Verification of Relationship between Load Transfer Efficiencies
(Linear Aggregate Interlock)
235
Restrained in x and y
12 in.
48 in. Applied load = 9 kips 12 in.

y Restrained in x and y
x directions
216 in.
Figure 5.4 Top View, Boundary Conditions and Applied Load
(Colley and Humphrey, 1967)
236
100.0
80.0
LTEδ (%)
60.0
This study
Davids and Mahoney (1999)

40.0
Colley and Humphrey (1967)
20.0
0.0
0.00 0.02 0.04 0.06 0.08 0.10
ωi (in.)
Figure 5.5 Comparison with Experimental and FE Results by Other Researchers
(Nonlinear Aggregate Interlock)
237
Joint
120 in.
Expected Fracture Plane
Joint Line
+ + 3 in.
6 in.
Applied load
120 in.
Expected Fracture Plane
x
y
240 in.
Figure 5.6 Planar View, Location of Applied Load and Assumed Fracture Planes
238
120.0
AGG/kl=0.01
AGG/kl=0.10
100.0 AGG/kl=10.0
AGG/kl=100
AGG/kl=1.00
80.0
Total Applied Load (kips)
60.0
40.0
20.0
0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.60
Figure 5.7 Effect of Joint Stiffness on Slab System Response
239
1.0
Pre-crack
0.8 Post-crack
LTEδ
0.6
LTEδ, TLE or LTEO
LTEO
0.4
TLE
0.2
0.0
1.0E-03 1.0E-02 1.0E-01 1.0E+00 1.0E+01 1.0E+02 1.0E+03 1.0E+04
AGG/kl
Figure 5.8 Pre-Crack and Post-Crack Load Transfer Responses
240
1.2
AGG/kl = 100
1.0
0.8
AGG/kl = 1
LTEδ
0.6
Second peak load

0.4
First peak load

0.2
0.0
0 20 40 60 80 100
Figure 5.9 Variation of Load Transfer Efficiency with Total Applied Load
241
160.0
140.0
ωi = 0.05 in.
120.0
ωi = 0.07 in.
ωi = 0.03 in.
100.0 ωi = 0.09 in.
80.0
60.0
40.0
20.0
0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.60
Figure 5.10 Effect of Initial Joint Opening on Load-Displacement Response
(Nonlinear Aggregate Interlock, Dmax = 0.065 in.)
242
1.00
LTEδ
0.80
LTEO
0.60
TLE, LTEδ or LTEO
0.40
TLE
0.20
0.00
0.00 0.02 0.04 0.06 0.08 0.10
ωi (in.)
Figure 5.11 Effect of Initial Joint Opening on Load Transfer at Peak Load
243
1.2
1.0 First peak Second peak
0.8
ωi = 0.02 in.
LTEδ
0.6 First peak

Second peak
0.4
ωi = 0.09 in.
0.2
0.0
0 50 100 150 200
Figure 5.12 Variation of Load Transfer Efficiency with Total Applied Load
(Nonlinear Aggregate Interlock, Dmax = 0.065in.)
244
160.0
Dmax=0.65 in.
140.0 Dmax=1.30 in.
120.0
ωi = 0.02 in.
100.0
ωi = 0.09 in.
80.0
60.0
40.0
20.0
0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.60
Figure 5.13 Effect of Maximum Aggregate Size (Nonlinear Aggregate Interlock)
245
35.0
AGG/kl=0.0100
AGG/kl=1.0000
30.0
AGG/kl=100.00
25.0
20.0
15.0
10.0
5.0
0.0
-0.03 0.02 0.07 0.12
Figure 5.14 Effect of Daytime Temperature (Linear Aggregate Interlock)
246
200.0
AGG/kl=100.00
AGG/kl=1.000
180.0
AGG/kl=0.0100
160.0
140.0
120.0
100.0
80.0
60.0
40.0
20.0
0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.60
Figure 5.15 Effect of Nighttime Temperature (Linear Aggregate Interlock)
247
30.0
25.0
20.0
15.0 ωi = 0.02 in. ωi = 0.09 in.
10.0
ωi = 0.05 in.
5.0
0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Figure 5.16 Effect of Daytime Temperature
248
160.0
ωi = 0.05 in.
140.0
ωi = 0.07 in.
120.0
ωi = 0.09 in.
100.0 ωi = 0.03 in.
80.0
60.0
40.0
20.0
0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70
Figure 5.17 Effect of Nighttime Temperature
249
0.44
Daytime
Nighttime
0.42
TLE
0.40
0.38
0.36
5 10 15 20
ΔT (°F)
Figure 5.18 Effect of Temperature on Joint Efficiency
(Nonlinear Aggregate Interlock, Dmax = 0.065 in., ωi = 0.07 in.)
250
Chapter 6 Numerical Analysis of Cracking in Doweled Concrete Pavements
Abstract
This paper presents numerical analyses of crack propagation in concrete pavement slabs with
doweled joints using the finite element package ABAQUS® 6.9-2. Traction-separation cohesive
elements are employed in simulating the fracture process according to the fictitious crack model.
Both wheel load and thermal curling are considered. The effects of joint width, dowel-concrete
interaction idealization, and dowel slip on slab responses are investigated. It is observed that
joint width does not affect the maximum load supported, but it reduces the load transfer
efficiency of the pavement system. Dowel slip is found to decrease the peak load that is resisted
by the slab system. A daytime temperature variation significantly reduces the peak load,
whereas a nighttime temperature profile increases it. This step-by-step effort may contribute to
the ongoing development of rational failure criteria that can substitute the statistical/empirical
algorithms used in current pavement design procedures.
251
6.1 Introduction
Load transfer in jointed concrete pavements is accomplished by two primary mechanisms:
aggregate interlock and dowel bars. Aggregate interlock relies only on shear forces developing
on rough vertical joint interfaces due to mechanical interlock between aggregate particles.
Numerical idealization of aggregate interlock joints and its post-crack implications on pavement
system responses have been discussed extensively in Chapter 5.
Dowel bars are often placed across a joint to complement the aggregate interlock load transfer
mechanism, and to ensure its permanence. Friberg (1940) idealized dowel bars as beam elements
encased in an elastic medium, per an earlier suggestion by Timoshenko and Lessels (1925). This
approach has been incorporated in the two-dimensional (2D) finite element (FE) code ILLI-
SLAB, by considering both the bending and shear resistance of dowel bars (Tabatabaie and
Barenberg, 1978). Huang and Chou (1978) argued that the bending contribution of the dowel bar
over the very short span of the joint opening could be neglected, leaving shear resistance as the
sole load transfer mechanism of doweled joints. Korovesis (1990) adapted ILLI-SLAB to permit
activation of dowel bending, shear and torsional degrees of freedom. Ioannides and Korovesis
(1992) then used dimensional analysis to interpret results obtained from ILLI-SLAB and derived
an independent joint stiffness parameter (D/skl) corresponding to the dimensionless aggregate
interlock joint stiffness (AGG/kl). Here, D [FL-1] is composite joint shear stiffness, AGG [FL-2] is
aggregate interlock joint stiffness per unit length, k [FL-3] is the subgrade modulus, l [L] is the
radius of relative stiffness of the slab-subgrade system, and s [L] is dowel bar spacing. The
primary dimensions are abbreviated here as [L] for length and [F] for force.
252
A slightly different formulation for the interaction of the dowel beam element with the slab was
first presented by Nishizawa et al. (1989), who divided the dowel bar into three segments: two
bending components embedded in the slab and one shear-and-bending component across the
joint. This approach was further refined by Guo et al. (1995), who implemented it in an update of
2D FE code JSLAB (Tayabji and Colley, 1984). Subsequently, an analogous, if more elaborate,
concept was also employed in three-dimensional (3D) FE analysis (Davids and Mahoney, 1999;
Davids, 2000; Kim and Hjelmstad, 2003).
In the present study, 3D numerical pre- and post-crack analysis of a dowel-jointed pavement
system is conducted, using the general purpose finite element package ABAQUS® 6.9-2. The
slab system is subjected to both wheel and temperature loading. The fracture process of the slabs
is tracked using traction-separation cohesive elements inserted along the anticipated crack path in
both slabs. The effects on slab responses of dowel-concrete interaction idealization, joint opening
and dowel looseness are investigated.
6.2 Finite Element Idealization of Doweled Pavement System
6.2.1 Discretization of Slab and Subgrade
The geometry, material properties and FE discretization characteristics employed in the present
study are shown in Table 6.1. Properties, including tensile strength, f t ' , and fracture energy, GF,
are selected to be consistent with previous studies (Ioannides and Peng, 2004; Ioannides et al.,
2006; Aure and Ioannides, 2012). The FE mesh shown in Fig. 6.1 is generated using software
application ABAQUS®/CAE 6.9-2, a pre- and post-processing package. Reduced integration 20-
node quadratic elements, C3D20R, are employed for the bulk of the concrete slab material, since
253
these are the highest order continuum elements supported in ABAQUS®/CAE. The mesh is
considerably finer along the assumed fracture plane and the joint than elsewhere. Tetrahedron
elements, C3D10, are used in the transition zones to the coarser mesh regions, as shown in Fig.
6.1, since these are the highest order tetrahedron elements compatible with C3D20R.
The adoption of this graded mesh increases the amount of labor required in defining the stiffness
of individual spring elements for the supporting Winkler subgrade, which must not resist any
tension. To overcome this difficulty, membrane elements, M3D8R, of negligible Young’s
modulus (1 ksi) and zero thickness are defined beneath each slab, onto which FOUNDATION
elements that can support tension and compression are attached; the stiffness of the latter is
manually provided in the input file. This approach was first proposed by Kuo (1994) to idealize
slab-subgrade and slab-base interfaces. The membrane elements are discretized with the same
mesh size as the slabs; their horizontal interaction with the slabs is defined as FRICTION,
ROUGH (i.e., infinite coefficient of friction), whereas in the normal direction only compressive
forces are resisted. Consequently, the slab and the membrane can separate when a tensile force
acts between them.
The wheel load is applied as a unit displacement at nodes located at (x, y, z) = (114, 3, 6) in. and
(x, y, z) = (114, -3, 6) in., as shown in Fig. 6.2. This is intended to represent a 12 by 12 in. area at
mid-edge of the loaded slab. It is anticipated that the maximum stress will develop at the slab
bottom, along the joint, half-way between the points of displacement application. The two
vertical planes perpendicular to the joint are, therefore, assumed to be the fracture planes. The
results monitored are: the total applied load (P), i.e., the sum of the conjugate “reaction” forces at
254
the two loaded nodes, RF2; and the vertical displacement at the nodes where the displacement is
applied.
6.2.2 Discretization of Dowel Bars
The diameter, spacing, length, and material properties of the dowel bars employed in the present
study are shown in Table 6.1. These values are selected to represent prevailing practical
guidelines (Huang, 2004). The C3D20R element used for the intact slab material is also
employed in discretizing the dowel bars, thereby eliminating any loss of precision incurred when
dissimilar elements are connected. Two approaches are investigated when idealizing dowel-
concrete interaction. In the first approach, the dowel bars are placed in holes “drilled” in the
concrete slab using ABAQUS®/CAE instruction cut extrude, and SURFACE INTERACTION
options are used to describe their relative movement. If no relative slip is allowed, the TIE
CONSTRAINT option may be used, instead.
The second approach involves embedding the dowel bars in the slab without pre-drilling holes.
This method is often used to idealize reinforcement bars in concrete structures. In ABAQUS®, it
is achieved by the use of the EMBEDDED ELEMENT key word. The main advantage of this
approach over the former is that it simplifies the discretization of the slab near the dowels, which
has repercussions on convergence and computational resources required. The nodes of the
embedded element are constrained to the nodes on the host (concrete mass) element. The
program will “search for the geometric relationships between nodes on the embedded elements
and the host elements. If a node on an embedded element lies within a host element, the degrees
of freedom at the node will be eliminated by constraining them to the interpolated values of the
degrees of freedom of the host element” ABAQUS (2009). The main drawbacks of this approach
255
are that it does not simulate the in situ dowel-concrete interaction, which involves sliding on one
side of the joint, and that it does not account for any gaps present between the bar and the
concrete slab. The effects of such differences between the two approaches on the post-crack
response of the pavement system are examined in sections that follow.
As a baseline, linear elastic material properties are assumed for the dowel bars. It might be
argued, however, that the load transfer system could fail by yielding even before the slab fails,
i.e., before the peak load is attained. Therefore, yielding properties of the dowel bars are also
specified in Table 6.1. The resulting slab response is compared to the baseline observations later
in the study.
6.2.3 Discretization of the Fracture Process
Three-dimensional traction-separation cohesive elements, COH3D8, are used to discretize a very
narrow fracture process zone (FPZ), whose width is set to 0.001 in. as suggested by previous
experiences (Aure and Ioannides, 2010; Aure and Ioannides, 2012). The FPZ mesh involves a
uniform pattern of element size 0.2 by 0.2 in., which is much finer than the slab mesh size. The
cohesive elements are inserted along the assumed fracture plane, and surface-based TIE
CONSTRAINTS are used to connect them to the intact slab surfaces. This technique has been
used by the authors in earlier work (Aure and Ioannides, 2010; Aure and Ioannides, 2012) and
found to be efficient. A bilinear concrete softening curve is employed per Petersson (1981) and
 G 1 
Gustafsson (1985); the coordinates of the kink point are set at  0.8 F' , f t '  , as in a previous
 ft 3 
study (Aure and Ioannides, 2012). The pertinent fracture parameters are given in Table 6.1. Since
this is a problem involving stiffness degradation, viscous regularization is employed with the
256
general (Newton-Raphson) solver option in order to ensure a convergent solution; the so-called
viscosity is set to 1×10-6 (ABAQUS, 2009).
6.3 Validation of Finite Element Discretization: Pre-Crack Analysis
The pre-crack phase, during which all materials are linear and elastic, affords the opportunity to
verify the FE discretization, i.e., the adequacy of the mesh size, element types, and idealizations
of dowel-concrete interaction and subgrade. For this purpose, the experimental study conducted
by Hammons (1997) is reproduced. Among the six small-scale tests reported, LSM-2 is
considered. The system consists of two slabs, each of which is 36×48×2 in., connected by 0.25
in. diameter dowel bars that are 15.5 in. long and are spaced at 4 in. center-to-center. The joint
opening is 0.0625 in. The moduli, E, for the slab and for the dowel bar are taken as 4 Mpsi and
29 Mpsi, respectively; the corresponding values of Poisson’s ratio are 0.18 and 0.3 (per
Hammons, 1997). Additional details about the experimental setup are provided by Hammons
(1997). The rubber pad foundation used in the experiment was assigned a subgrade modulus, k,
of 0.33 ksi/in., backcalculated using a FE analysis by Hammons (1997).
The FE discretization of the test slab-dowel system is shown in Fig. 6.3. It is similar to the one
discussed in Section 6.2, above, which will be employed in the next section. The boundary
conditions are selected to represent the set-up in the experiment: translations in all directions at
nodes along the two outer 48-in. sides, as well as planar translations at nodes along the two outer
72-in. sides, are restrained. The load is applied as a pressure of 145 psi at a corner by the joint
over a square area of 16 in.2. The two dowel-concrete interaction idealizations discussed above
are implemented. In the first case, the dowel bars are assumed to be encased in the slabs (without
a hole), i.e., dowels experience no relative movement with respect to the slabs. Consequently, the
257
EMBEDDED ELEMENT approach is employed to discretize dowel-concrete interaction. In the
second case, which is closer to reality, holes are created in the unloaded slab and then dowel bars
are placed in therein. It is noteworthy that Hammons (1997) had applied grease to the dowel bars
and encased them in drinking straws on the unloaded side of the joint, a common practice in
order to ensure debonding during load transfer. While reproducing Hammons experimental
results, Davids and Mahoney (1999) estimated the thickness of the drinking straw to be 3.15
mills (0.08 mm). In the present study, this effect is accounted by providing a clearance of 3.15
mills between the dowel bars and the slab holes. The effect of grease is idealized, by providing a
coefficient of friction of 0.01.
Figure 6.4 shows the resulting deflection profile at a distance of 7 in. from the edge closest to the
loaded area; this is the location at which the displacement transducers had been placed during the
experiment (Hammons, 1997). The same experiment had been previously reproduced by Davids
and Mahoney (1999) in verifying the FE program EverFE, and their result is also shown. Good
agreement is observed between the present study and Davids and Mahoney (1999) for both
cases, no gap (embedded) and with gap. As compared to the experiment, it is observed that the
EMBEDDED ELEMENT approach results in a very stiff joint and results in a perfect load
transfer efficiency. The second approach, in which the gap between the dowel bars and the slab
holes is accounted, gives almost perfect deflection profile as compared to the experiment
especially around the joint. Relatively larger differences between the present study and the
experiment in the displacements as one moves away from the joint may be attributed to the effect
of the boundary condition of the slab, which is idealized in this study as perfectly rigid around
the peripheries of the slab. Obviously, perfect rigidity may not be attained in the experiment.
258
Other reasons may include numerical discretization errors, such as mesh fineness and contact
interaction between the slab and dowel bar. It worth noting that the idealization of dowel-
concrete interaction when a gap exists between the slab holes and the dowel bars is
computationally demanding, especially when such an approach is augmented with cracking in
the slabs.
From the results shown in this section, the numerical discretization adopted is deemed reliable
and applicable to the investigation of the more complex phase involving fracture of the slabs
subjected to both wheel load and thermal curling, as discussed in the following sections.
6.4 Crack Propagation in Slabs with Doweled Joints
Post-crack responses of the doweled pavement system described earlier are studied in this
section. Recall that geometry, material properties and FE discretization are shown in Table 6.1
and Figs. 6.1 and 6.2; slabs are connected by dowel bars that are nominally 14 in. long and of
diameter 0.75 in., spaced at 12 in. center-to-center, inserted in holes “drilled” into the concrete
slab. A parametric study is carried out to investigate the effect of joint opening, dowel
embedment, dowel-concrete interaction (dowel looseness), and yielding of the dowel bar, as well
as thermal curling.
6.4.1 Effect of Initial Joint Opening
The significance of joint opening, ωi, on the load-displacement response of the two-slab system
during the fracture process is investigated by considering three initial joint openings: 0.2 in.
(baseline), 0.4 in. and 0.8 in. The length of the dowel bars is adjusted slightly depending on the
joint width, so that 6.9 in. remain encased in the slab on each side of the joint; for the baseline
259
opening of 0.2. in., the length of the dowels is, therefore, 14 in. The dowel bar surfaces are
connected to the corresponding holes ‘drilled’ in the slab using the TIE CONSTRAINT option;
this eliminates dowel slip, but it is selected because of its implementation simplicity.
The load-displacement responses obtained in this manner are plotted in Fig. 6.5. The curves
exhibit only one softening event, evincing that the two slabs failed almost at the same time. The
more competent dowel system evidently results in monolithic action of the two slabs. It is
observed that an increase in joint opening decreases slightly the stiffness of the pavement
system, discernible as a decrease in the slope of the load-displacement curves. The reduction in
slope at about half the peak load evinces the onset of cracking in the loaded slab. The value of
the peak load resisted by the slabs remains largely unaffected by changes in joint opening. It
appears, therefore, that while an increase in initial joint opening may increase slightly the
displacement at which the peak load occurs, its influence on the peak load supported by the slabs
is rather insignificant. This observation is confirmed by Fig. 6.6, which shows the variation of
the load transfer efficiency with respect to deflection (LTEδ) as the initial joint opening increases.
It is evident that as the joint opening increases, load transfer efficiency decreases almost linearly.
6.4.2 Effect of Dowel–Concrete Interaction
The interaction between the concrete mass and the dowel bar is one of the major parameters
influencing load transfer efficiency in a doweled pavement system. Consequently, there have
been several attempts by different researchers to discretize this interaction realistically in 3D FE
analysis. In general, such efforts fall into two main categories: those employing spring elements
and those relying on contact elements.
260
An early attempt employing the former approach is described by Channakeshava et al. (1993).
The same methodology was subsequently implemented by Davids (2000) during the
development of 3D FE program EverFE. Accordingly, spring elements were used to represent
the relative movements of the dowel bar with respect to the concrete mass in which it is encased,
in the three orthogonal directions. Such movements are characterized using three distinct
stiffnesses, but since the normal and transversal components are physically indistinguishable,
only two spring stiffenesses need be specified. The first is the conventional modulus of dowel
support, K [FL-3] (Timoshenko and Lessels, 1925). The second stiffness, corresponding to the
slip of the dowel bar along its longitudinal direction, was dubbed the “dowel-slab restraint
modulus”. It has the same dimensions as K, but it is typically assumed to have only a fraction of
its magnitude. A similar approach was adopted by subsequent investigators (Bhattacharya, 2000;
Dere et al., 2006). The main limitations of this idealization technique are: (a) There are no sound
experimental or theoretical methods to determine the modulus of dowel support, K (Ioannides
and Korovesis, 1992), let alone its ratio to the dowel-slab restraint modulus; (b) This approach
ignores dowel looseness, and is incapable of simulating possible dowel-concrete separation; (c)
A gap alternative provided by Davids (2000), ignores geometric nonlinearities arising at the
dowel-concrete interface during the continual alternation between contact and separation.
Moreover, this alternative necessitates the elimination of dowel sliding, effectively requiring
full-bond between the dowel and the concrete upon gap closing.
The second approach employs contact elements, already implemented in FE programs selected
for simulation of dowel-concrete interaction in concrete pavement systems (William and
Shoukry, 2001; Riad et al., 2009; Maitra et al., 2009). Coloumb’s friction law is adopted to
261
idealize dowel slip, according to a specified coefficient of friction. A certain clearance is
provided between the dowel and the concrete to represent the gap; upon closure, the specified
modulus of dowel support is activated. Determination of these parameters is still fraught with
difficulties.
In the present study, the second approach is adopted, because it is deemed to be more
comprehensive and realistic, even though for numerical expediency, it is assumed that there is no
gap between the dowel bar and the slab. The bearing (or normal) resistance is idealized using the
HARD CONTACT option under SURFACE INTERACTION, which obviates the need to specify
K since the transfer of normal pressure between the two surfaces is controlled by their respective
stiffnesses and the FE mesh. Dowel slip is idealized using the Coulomb FRICTION option.
As noted in Section 6.2.2 above, fully bonded dowel bars can be simulated alternatively using
the EMBEDDED ELEMENT discretization, which eliminates the presence of holes in the
concrete mass and thereby simplifies the numerical discretization of the dowel-concrete region.
Four cases are considered involving dowels inserted in pre-drilled holes. In the first instance, no
relative movement is allowed between the dowel and slab (i.e., the two are assumed to be fully
bonded), whereas for the remaining two cases sliding between the dowel and slab is allowed, and
the coefficient of friction is set at 0.01 or 0.2. To compare with the fully bonded case, the
EMBEDDED ELEMENT discretization is considered as a fourth case. Only the loaded side
segment of the dowel bar is allowed to slip. In all the cases, the initial joint opening is kept at 0.2
in.
262
The load-displacement responses obtained are shown in Fig. 6.7. It is observed that there is a
significant reduction in the peak load (by about 20%) as the dowel slips. It is interesting to see
that there is also a noticeable decrease in stiffness of the system with slip, occurring after the
onset of cracking at about half the peak load. The load transfer efficiency with respect to
deflection (LTEδ) is also computed. It is found that for fully bonded, coefficient of friction of 0.2
and 0.01, LTEδ is 0.705, 0.700, and 0.696, respectively at an applied load of 35 kips. This
indicates a slight reduction of load transfer efficiency of the joint with increasing slip.
The load-displacement response corresponding to the EMBEDDED ELEMENT discretization
option is also shown in Fig. 6.7. The results indicate that this approach gives a slightly lower
peak load value as compared to the no slip case. The stiffness of the system, however, remains
the same in these two cases. It can be concluded that for dowel bars fully bonded to the concrete
slab, the EMBEDDED ELEMENT approach is not only attractive for its numerical simplicity but
also compares well with the more elaborate no-slip alternative.
6.4.3 Effect of Dowel Bar Yielding
As the load continues to increase past crack initiation, dowel bars may yield and precipitate load
transfer failure, even before ultimate structural collapse of the concrete. To investigate whether
this situation may happen for the joint configuration specified earlier (Table 6.1), dowel bars are
assigned the ABAQUS® nonlinear material constitutive relation of isotropic hardening suitable
for steel. Two yield stress levels of 36 ksi and 55 ksi are considered; these correspond to ultimate
stresses of 50 and 70 ksi, respectively. In both instances, ultimate plastic strain is assumed to be
10% of the elastic strain (Salmon et al., 2009).
263
The simulation result is presented as a load-displacement curve in Fig. 6.8. It is observed that
dowel bar yielding occurs much later than the peak load. Yielding is manifested in the slope
decrease after the end of softening, a phenomenon not observed in the linear, no-yield case (no-
slip in Fig. 6.7). It might be expected that premature load transfer failure due to dowel yielding
may occur only if dowels are of inadequate diameter, are too widely spaced, or are made of low
grade material.
6.4.4 Effect of Temperature Differential
The effect of temperature on the fracture of a single concrete pavement slab has been studied in
Chapter 4 of the present research. Two typical daytime and nighttime temperature cycles have
been considered, as originally recorded by Teller and Sutherland (1935) at 7:00 pm and 9:30 am
on July 12 and 13, 1932, respectively. The pavement system consisted of a 6-in. thick concrete
slab-on-grade, resting on silty clay subgrade. That case study is retained in this Chapter, which
also considers load transfer provided by steel dowels, as described in Table 6.1, above. Per
Chapter 4, cubic polynomials are used to assign temperature values, T(y), through the slab
thickness, as follows:
T  y   0.333 y3  3.446 y 2  5.042 y  78.036 at 9:30 am R 2  0.9595 (6.1a)
T  y   0.017 y3  0.060 y 2  0.295 y  90.995 at 7:00 pm R 2  0.9969 (6.1b)
in which y is the distance measured up from the bottom of the slab. It is assumed that the
temperature is constant at all points lying on the same horizontal plane.
The load is applied in two steps: first, temperature effects are considered alone, and subsequently
the wheel load is applied incrementally as a unit displacement, per Section 6.2.1, above. The
264
deformed shape of the cracked slab for the daytime temperature differentials obtained in this
manner is shown in Fig. 6.9; the corresponding nighttime (curled-up) depiction is very similar. In
this enlarged presentation, the transition elements described in Section 6.2.1 are clearly visible.
Figure 6.10 shows both daytime and nighttime load-displacement responses, along with the
corresponding response for the case without curling (no-slip case in Fig. 6.7). As might be
expected, the nighttime peak load (~95 kips) is significantly higher than the no-curling value
(~45 kips); the latter is itself more than double the corresponding daytime capacity (~18 kips).
These results reinforce the conventional understanding that load-plus-curling stresses are
cumulative during the daytime, both being tensile at the bottom of the slab, whereas they tend to
offset each other during the nighttime, as curling tension migrates to the top.
The effects of joint characteristics on the load carrying capacity and fracture process of a
doweled pavement system have been studied. An initial pre-crack analysis provided verification
for the robustness of the proposed FE idealization, which was confirmed by comparisons with
previous laboratory and numerical results. Fracture analysis of the slabs has been carried out to
investigate the maximum load the system can resist before failure. It is found that an increase in
initial joint opening decreases slightly the system stiffness (manifest in a slope reduction of the
load-displacement curve), but does not change significantly the peak load resisted. A major
parameter that influences slab response is identified to be the interaction between the dowel bar
and the slab, i.e., dowel slip; the peak load decreases by about 20% for the range of coefficient of
friction considered in this study (no slip, 0.01 and 0.2). If the dowel is fully bonded to the slab,
the computationally efficient EMBEDDED element approach for dowel placement is found to be
265
no less precise than the more demanding options. The influence of daytime and nighttime
temperature distributions on the peak load capacity of dowel jointed slabs is also investigated. It
is observed that daytime temperatures result in approximately half the peak load resisted during
the nighttime, due to their combination with load effects.
It can be concluded from this study that the application of cohesive elements in a fracture
analysis of doweled concrete pavement slabs is promising and may be extended to concrete
pavements subjected to repeated loading. This effort may contribute to the ongoing development
of rational failure criteria that can substitute the statistical/empirical algorithms employed along
with Miner’s hypothesis (Miner, 1945) in current mechanistic-empirical pavement design
procedures.
6.6 References
Providence, RI.
Bhattacharya, K. (2000). "Nonlinear response of transverse joints of airfield pavements." J.
Transp. Eng., 126(2), ASCE, Reston, VA, 763-781.

266
Channakeshava, C., Barzegar, F., and Voyiadjis, G. Z. (1993). "Nonlinear FE analysis of plain
concrete pavements with doweled joints." J. Transp. Eng., 119(5), ASCE, New York, NY, 763-
781.
Davids, W. G. (2000). "Effect of dowel looseness on response of jointed concrete pavements." J.
Transp. Eng., 126(1), ASCE, Reston, VA, 50-57.
Davids, W. G., and Mahoney, J. P. (1999). "Experimental verification of rigid pavement joint
load transfer modeling with EverFE." Transp. Res. Rec. 1684, Transportation Research Board,
Dere, Y., Asgari, A., Sotelino, E. D., and Archer, G. C. (2006). "Failure prediction of skewed
jointed plain concrete pavements using 3D FE analysis." Eng. Failure Anal., 13(6), Elsevier,
Friberg, B.F. (1940). "Design of dowels in transverse joints of concrete pavements." Trans.,
195(1), ASCE, New York, NY, 1076-1095.
Guo, H., Sherwood, J. A., and Snyder, M. B. (1995). "Component dowel bar model for load
transfer systems in PCC pavements." J. Transp. Eng., 121(3), ASCE, New York, NY, 289-298.
267
Huang, Y. H. (2004). Pavement Analysis and Design. 2nd Ed., Prentice-Hall, Upper Saddle
River, NJ.
Huang, Y. H. and Chou, Y. T. (1978). "Discussion on finite element analysis of jointed or
cracked pavements by Tabatabaie, A. M. and Barenberg, E. J." Transp. Res. Rec. 671,
slabs: implications for pavement design." Proc., Fifth International Workshop on Fundamental
Turkey, 1-11.
Ioannides, A. M., and Korovesis, G. T. (1992). "Analysis and design of doweled slab-on-grade
pavement systems." J. Transp. Eng., 118(6), ASCE, New York, NY, 745-768.
Kim, J., and Hjelmstad, K. D. (2003). "Three-dimensional finite element analysis of doweled
joints for airport pavements." Transp. Res. Rec. 1853, Transportation Research Board, National
268
Korovesis, G. T. (1990). "Analysis of slab-on-grade pavement system subjected to wheel and
temperature loadings," Ph.D. dissertation, University of Illinois at Urbana-Champaign, Urbana,
IL.
Kuo, C. M. (1994). "Three-dimensional finite element analysis of concrete pavements," Ph.D.
Maitra, S. R., Reddy, K. S., and Ramachandra, L. S. (2009). "Load transfer characteristics of
dowel bar system in jointed concrete pavement." J. Transp. Eng., 135(11), ASCE, Reston, VA,
813-821.
Nishizawa, T., Fukuda, T., and Matsuno, S. (1989). "A refined model of doweled joints for
concrete pavement using FEM analysis," Proc., Fourth International Conference on Concrete
Pavement Design and Rehabilitation, Purdue University, West Lafayette, IN, 735-745.
Riad, M. Y., Shoukry, S. N., William, G. W., and Fahmy, M. R. (2009). "Effect of skewed joints
on the performance of jointed concrete pavement through 3D dynamic finite element analysis."
Int. J. Pavement Eng., 10(4), Taylor and Francis Group, London, UK, 251-263.
269
Salmon, C. G., Johnson, J. E., and Malhas, F. A. (2009). Steel Structures: Design and Behavior
5th Ed., Prentice Hall, Upper Saddle River, NJ.
Tayabji, S. D., and Colley, B. E. (1984). "Analysis of jointed concrete pavements: Interim
Report." Rep. No. FHWA/RD-86/041, Federal Highway Administration, Washington, DC.
16(9), Bureau of Public Roads, Washington, DC,169-197.
Timoshenko, S., and Lessels, J. M. (1925). Applied Elasticity. Westinghouse Technical Night
School Press, East Pittsburgh, PA.
William, G. W., and Shoukry, S. N. (2001). "3D finite element analysis of temperature-induced
stresses in dowel jointed concrete pavements." Int. J. Geomech., 1(3), ASCE, Reston, VA, 291-
307.
270
Table 6.1 Geometry, Material Properties and Discretization of Pavement System
Slab Geometric Properties

Length (in.) 240
Width (in.) 120
Thickness (in.) 6
Dowel Bar Geometric Properties
Length (in.) 14
Diameter (in.) 0.75
Spacing (in.) 12
Joint Width (in.) 0.2
Slab Material Properties
Thermal Expansion Coefficient (in./in./°F) 5×10-6
Compressive Strength (ksi) 5
Unit Weight (pcf) 150
Dowel Bar Properties
Yield Stress (ksi) 36
Ultimate Stress (ksi) 50
Poisson's Ratio 0.3
Modulus of Subgrade Reaction (psi/in.) 200
Softening Curve Bilinear
Solver Newton-Raphson
Subgrade Idealization FOUNDATION
271
Cohesive zone (fracture plane)
Dowel bar
Cohesive zone mesh
Total number of nodes: 302464

Total number of elements: 85656
37000 elements of type C3D20R
388 elements of type C3D10
40960 elements of type COH3D8
7308 elements of type M3D8R
Figure 6.1 Finite Element Mesh
272
Expected fracture plane
120 in.
Location of applied unit displacement
6 in.
14 in.
C. L
in.
6 in.
+ +
20-0.75 in. dia. dowels @ 12 in.
120 in.
Expected fracture plane
z
x
240 in.
Figure 6.2 Slab Geometry and Loading Layout
273
48 in.
72 in.
p=145 psi
A=16 in. 2=4 in.
x 4 in.
2 in.
Slab mesh
Dowel bar mesh
Figure 6.3 Finite Element Mesh for Slab LSM-2 Tested by Hammons (1997)
274
Distance Along Loaded Edge (in.)
23 25 27 29 31 33 35 37 39 41
0.005
Hammons (1997) - Experiment
Davids and Mahoney (1999) - No gap
Davids and Mahoney (1999) - Gap
0.008
This study - No gap
This study - Gap
0.011 Joint
Vertical Displacement (in.)
0.014
0.017
0.020
0.023
Figure 6.4 Comparison with Other Experimental and Numerical Studies
275
50.0
Cracking plane
45.0 ωi=0.2 in. ωi=0.8 in.
ωi=0.4 in.
40.0
35.0
30.0
25.0
20.0
15.0
10.0
5.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15 0.18
Figure 6.5 Effect of Initial Joint Opening (D/skl ~ 30)
276
0.80
0.75
LTEδ
0.70
0.65
0.60
0.10 0.30 0.50 0.70 0.90
Initial Joint Opening (in.)
Figure 6.6 Effect of Initial Joint Opening on Load Transfer Response
277
50.0
Coefficient of friction=0.01
Coefficient of friction=0.20
Embedded dowel
No-slip
40.0
30.0
20.0
10.0
0.0
0.00 0.03 0.06 0.09 0.12 0.15 0.18
Figure 6.7 Effect of Dowel Slip and Embedment (ωi = 0.2 in.)
278
210.0
fy=36 ksi
180.0 fy=50 ksi
No yielding
150.0
120.0
90.0
60.0
30.0
0.0
0.00 0.20 0.40 0.60 0.80 1.00
Figure 6.8 Effect of Dowel Bar Yielding
279
Transition Elements
Applied Load
Figure 6.9 Cracked Concrete Pavement Slab under Daytime Temperature and Wheel Load
(ωi = 0.2 in.)
280
140.0
Daytime temperature
120.0 Nighttime temperature
No curling
100.0
80.0
60.0
40.0
20.0
0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
Figure 6.10 Effect of Temperature
281
282
Chapter 7 Summary, Conclusions and Recommendations for Future Work
7.1 Research Summary
The main objective of this research has been to implement nonlinear fracture mechanics
concepts, particularly the fictitious crack model (FCM), in the study of crack propagation in
concrete pavement slabs, using cohesive elements that have been recently incorporated in the
general purpose finite element (FE) package ABAQUS®. This has been accomplished through a
step-by-step procedure that investigates successively simply supported beams, individual slabs-
on-grade, as well as concrete pavement systems equipped with a load transfer mechanism and
subjected to load-plus-curling.
The approach was first verified by reproducing experimental and numerical data for simply
supported beam specimens reported by previous independent researchers. Next, the sensitivity of
post-crack responses was studied, as several pertinent parameters involved in the fracture process
were varied, including the width of the cohesive zone, mesh fineness of the bulk (intact) and
cohesive zone, concrete softening curve and solution techniques. Results were found to be more
sensitive to solution technique and concrete softening curve than to mesh fineness and width of
the cohesive zone.
A similar technique was then implemented for an individual slab-on-grade under a wheel load.
The main challenge during this phase was obtaining an efficient simulation, i.e., one that is
economical in its requirements for computing resources, while at the same time providing a
convergent solution without compromising the accuracy of the results. The challenge may be
largely attributed to the increase in the dimensions of the structural system (i.e., slab versus
283
beam), as well as the presence of a supporting subgrade. This problem was addressed by
designing an optimal mesh for both the cohesive zone and intact region, and by implementing a
numerical technique termed as viscous regularization, specifically developed to address
convergence issues caused by the damage of the cohesive zone. A sensitivity study was first
conducted to obtain the best value of the viscosity term. Simulation results in this phase have
been also verified by reproducing experimental and numerical data reported by other researchers.
Next, the effects of softening curve, cohesive zone width and mesh, analysis technique, loading
mode, tensile strength and fracture energy were investigated for the single slab-on-grade. It was
found that the type of softening curve and cohesive zone width and mesh do not influence post-
peak responses significantly, especially as compared to beams. They do, however, play a
significant role in the convergence of the solution. It was found that the fracture process is more
affected by the fracture energy than by the tensile strength. Newton-Raphson and modified Riks
solution methods have been compared, and it has been observed that the load-displacement
responses obtained by the latter exhibit more pronounced post-peak softening compared to the
former. Despite capturing accurately the post-peak responses, the modified Riks method was
found to be computationally more demanding and relatively inefficient in large problem sizes.
The influence of loading placement (at the interior or at the edge of the slab) was also examined,
on account of its importance for in situ pavements under both thermal and mechanical loads. In
this case, two softening regions were observed on the load-displacement curves, evincing
collapse of the slab first parallel to its width and subsequently parallel to its length.
The effect of curling with and without wheel load on the fracture behavior of a single slab-on-
grade was studied in the third phase of the present research. To begin with, the accuracy of the
284
numerical discretization was verified by conducting a pre-crack analysis of the slab under curling
alone and comparing responses obtained with analytical, as well as experimental, data available
in the literature. Subsequently, the post-cracking behavior of the concrete slab under thermal
curling alone was examined. The effects of parameters such as concrete age, notch depth, slab
size, temperature profile through slab thickness, and slab’s self-weight were investigated. It has
been observed that thermal cracks are affected by concrete age, which, therefore, dictates the
timing and the size of joint sawing. Cracking due to thermal curling was observed to be
insensitive to slab sizes greater than 10 times the radius of relative stiffness of the slab-
foundation system, an observation similar to Westergaard’s (Westergaard, 1927). Under curling
alone, the effect of slab’s self-weight on pavement slab cracking was found to be negligible. The
temperature profile through slab thickness was found to exert a significant influence on the
fracture process, and this evinces the importance of considering the entire temperature (and
stress) distribution, rather than relying merely on the maximum top and bottom temperature
differentials. This result calls for the accurate determination of pavement temperature profiles.
Fracture analysis of a single slab under both wheel load and thermal curling has been also carried
out by considering two loading scenaria: (a) constant temperature followed by increasing wheel
load; and (b) constant wheel load followed by increasing temperature. In the first loading case,
the effects of daytime and nighttime temperature profile have been investigated. It has been
observed that daytime curling forms unstable cracks and significantly reduces the slab’s
resistance to wheel load. Nighttime temperature, on the other hand, results in stable cracks and
increases resistance of the slab, but it results in a relatively large crack width at the bottom of the
285
slab. The effect of temperature profile through the slab thickness has once again been established
as the major parameter affecting the fracture process.
The loading scenario involving a fixed wheel load followed by an increasing daytime
temperature has also been considered. It is observed that the stability of the crack and the
magnitude of the peak temperature differential largely depend on the magnitude of the wheel
load. A relatively high wheel load, encourages stable crack growth with almost no softening,
beginning at a lower temperature differential. In contrast, a rather small wheel load, allows the
temperature differential to increase significantly before crack initiation, but the peak is followed
by a sudden slab failure with pronounced softening.
The fourth phase of the present research has focused on cracking in concrete pavement slabs
with aggregate interlock load transfer. Both linear and nonlinear aggregate interlock idealizations
have been considered. Post-crack responses were used in the study of crack growth in both the
loaded and the unloaded slabs, for linear as well as nonlinear aggregate interlock. For linear
aggregate interlock, it was found that at smaller joint stiffness values, there was only one peak
load, indicating the failure of the loaded slab alone. As the joint stiffness increased, however,
softening began to be manifest at two points on the load-displacement curve, evincing the
successive failure of both slabs. At very large joint stiffness values, the two slabs act
monolithically and only one softening region was noted once again, manifested at a relatively
large peak load. The load transfer efficiency with respect to vertical deflection and the
transferred load efficiency were observed to be similar to the corresponding values for uncracked
slabs.
286
Unlike linear aggregate interlock, the nonlinear aggregate interlock idealization takes into
account the change in joint opening occurring during the loading process, as well as the
concomitant alteration in the load transfer efficiency of the joint. It was observed that the load
transfer efficiency with nonlinear aggregate interlock decreases linearly with an increase in
initial joint opening.
The effects of both nighttime and daytime temperature distributions were also investigated for
linear as well as nonlinear aggregate interlock idealizations. Daytime temperature variations
result in a lower peak load, and in decreased joint efficiency, on account of a widened joint
opening at the bottom of the slab. Nighttime temperature curling, on the other hand, has the
opposite effect and may prevent a decrease in joint efficiency, if it does not actually enhance it.
The fifth and final phase of the present research was devoted to cracking in doweled concrete
pavements. From the post-cracking analysis of pavement slabs, it has been found that an increase
in initial joint opening decreases the system stiffness, but it does not change the peak load
resisted. A major parameter that influences slab response was identified to be the interaction
between the dowel bar and the slab, i.e., dowel looseness. If the dowel is fully bonded to the
slab, the EMBEDDED element approach for dowel placement has proven to be computationally
efficient. The influence of daytime and nighttime temperature distributions on the peak load
capacity of dowel jointed slabs was also investigated. It was observed that a daytime temperature
variation resulted in approximately a 50% reduction of the peak load resisted during the
nighttime.
287
7.2 Concluding Remarks
The present study demonstrates that the application of cohesive elements to fracture analysis of
plain concrete pavements holds significant potential as a means of exploring facets of
performance that have been dealt with only in a phenomenological way until now. It may even
be anticipated that with the advent of new techniques incorporating different fracture modes and
the emergence of the extrinsic cohesive element formulation from computational mechanics
(Park et al., 2009; Paulino et al., 2010), the development of rational failure criteria may be
forthcoming, so that reliance on the statistical/empirical algorithms used in current pavement
design guides may be drastically reduced, if not eliminated altogether. The main conclusions of
the study presented herein can be classified into two main categories, associated with numerical
simulation and structural system aspects, respectively. The former group includes those factors
that influence the accuracy of the results and the convergence of the solution, whereas the latter
class incorporates the effects of material characterization as well as of structural behavior of the
pavement slab, as summarized below.
a. Element choice
Quadratic elements (for example, C3D27) are generally found to produce better results than
linear elements (for example, C3D8R), even when comparing meshes with the same total number
of nodes. This is particularly important in fracture analysis of structural systems requiring
significant computational resources.
b. Subgrade characterization
The subgrade idealization exerts a significant influence on the convergence of the solution. The
preferred manner of simulating the subgrade in the present study involves using spring-type
288
elements, namely SPRING1. Each such spring is assigned properties reflecting that it supports
only compression; slab and subgrade separation can thereby be realized. On the other hand, when
the subgrade is represented using the FOUNDATION option which enforces permanent contact
between the slab and the soil, no softening in the load-displacement curves can be observed,
which may explain the relatively rapid convergence of the solution in these cases (i.e., fewer
time increments are needed). When a tensionless subgrade is employed, softening is clearly
exhibited on the load-displacement curve, but the solution requires a large number of increments
to convergence.
c. Inter-element interaction
This refers to the approaches adopted in simulating the interactions between independent
elements: for example, between cohesive elements and intact material (solid elements); between
dowel bars and the concrete slab; and between the slab and the underlying elements. Numerical
techniques were selected for this purpose so as to yield significant savings in computational
resources. This objective was tempered by concerns about undesirable side effects on the
convergence of the solution and the accuracy of the results. Surface-based TIE CONSTRAINTS,
HARD CONTACT INTERACTIONS, and EMBEDDED ELEMENT formulations were the
interaction idealization approaches employed in this study. Surface-based TIE CONSTRAINTS
are used to ensure independently discretized cohesive elements remain connected to the intact
material. This technique saves a significant amount of computational memory since the intact
material is discretized using a relatively coarser mesh. The HARD CONTACT INTERACTIONS
approach plays a major role in accelerating solution convergence, and it is mainly used to
simulate sliding friction between two individual components, such as between the dowel bar and
the slab, or between a base layer and the slab.

289
d. Solution technique
The load-displacement curves obtained during the post-peak responses of the simply supported
beam and of the pavement slab-on-grade most often exhibit snap-back instabilities. Moreover,
when the slab completely fails and the subgrade springs pick up the load, the load-displacement
curve exhibits a sudden change in slope making solution convergence very difficult. A Newton-
Raphson solution method fails to capture such behavior, since it employs a positive gradient in
the search for an equilibrium point in the solution space. A preferable solution approach is
available using arc-length type of solvers, as, for example, in the modified Riks method. The
main drawback of the latter, however, is that it requires smaller load increments in order to
ensure that the snap-back path of the load-displacement curve is successfully traversed;
consequently, the method consumes large amounts of computational resources. On the other
hand, the Newton-Raphson method is preferable over the Riks method if saving computational
resources is an overriding concern, especially when only the peak load needs to be determined.
e. Loading approach
In the present study, displacement and loading controls have been used, and the former was
found to be most appropriate in capturing post-peak responses of the slab. Loading control used
in conjunction with the Newton-Raphson solution approach fails to converge after the peak load
is attained. If loading controls as well as post-peak response are desired, the modified Riks can
be used, with the load applied over a point. When the load is applied as a pressure, the solution
fails to converge after the peak load even with the modified Riks method.
A uniform displacement was also stipulated over a certain area, as is the case when a rigid plate
is applied on the slab. This, however, resulted in stress concentrations at the corners of the
290
loaded area, creating convergence issues caused by localized failure. It was concluded, therefore,
that a single displacement at the centroid of the loaded area is the most suitable method of load
application for practical representation of the wheel load; at the same time the amount of
computer memory consumed was reduced.
f. Viscous regularization
To address the convergence challenge posed by an increase in the size of the structural system,
the present study adopted viscous regularization. This technique allows the solver to iterate using
values within a small but finite extent beyond the pre-defined concrete softening curve without
compromising the accuracy of the solution. The slab responses were observed to be very
sensitive to the viscosity value used. Therefore, a sensitivity study was conducted as the
structural size, element type, and mesh size employed changed. The viscous regularization
technique permits simulation of large and complex structural systems at a relatively low
computational resource expense.
g. Fracture parameters
Parameters considered include fracture energy, tensile strength and the concrete softening curve.
From the present study, it was concluded that fracture energy is the major parameter influencing
post-crack structural response. Tensile strength, on the other hand, was observed to have a
minimal effect on slab response, except for slabs subjected to temperature curling and wheel
load. As the tensile strength increases, load-displacement responses for slabs under wheel load
and temperature were observed to exhibit relatively pronounced softening, i.e., reduction in the
applied load from its peak level before reloading occurs, when compared to the corresponding
load-displacement responses for slabs under wheel load alone. Since fracture energy and tensile
291
strength are interrelated, it is necessary to consider their interaction, and the dimensionless
parameter called brittleness number, B, was selected for this purpose. The type of concrete
softening curve provided as input, e.g., linear, bilinear, etc., was found to have a minimal effect
on slab responses. It had, however, an effect on the convergence of the solution and the amount
of computational resources expended. In general, smoother softening curves were found to be
computationally more efficient.
h. Slab size
The effect of slab width and length on the post-crack response of the slab was found to be similar
to what has been observed in previous, more conventional (pre-crack) research after Westergaard
(1927). Slabs with dimensionless length (L/l) or width (W/l) greater than 10 have a negligible
effect on load-displacement responses. Due to the lack of experimental information concerning
the sensitivity of fracture energy and tensile strength to the slab thickness, as well as concerning
temperature variations, this conclusion was reached from analyses using only one slab thickness
(6 in.). Consequently, its generality has not been established. Given that fracture parameters
exhibit sensitivity to size, it is conceivable that Westergaard’s linear elastic observations may not
hold true for much thicker slabs. A more comprehensive future study would account for size
effects on material properties.
i. Curling effect
The temperature distribution along the thickness of the concrete slab was observed to be one of
the major factors influencing post-crack slab response. The peak load resisted by the slab was
most significantly altered not by the temperature differential between the top and bottom of the
slab, which is conventionally considered in pre-crack curling analysis, but by the temperature
292
profile through the slab thickness. This is precisely what necessitates the use of three-
dimensional finite element methods for post-crack analysis of concrete pavements.
Unstable cracks form when stresses due to the wheel load are of the same sense (compressive or
tensile) as those from thermal curling. For a slab loaded at its edge while subjected to
temperatures that decrease from top to bottom (daytime temperature profile), the load-
displacement response exhibits pronounced softening upon attaining the peak load. This is
attributed to the loss of contact between a large portion of the slab’s underside and the subgrade.
On the other hand, the same slab under temperatures increasing from top to bottom (nighttime
temperature profile) will remain in contact with the subgrade over a much larger area. This
explains the relatively smaller softening observed in the corresponding curves, as well as the
more stable crack growth. On the other hand, wheel loads cause nighttime cracks at the bottom
of the slab that were found to be wider than those developing during the daytime, and this may
lead to other undesirable effects, such as pumping. Temperature distribution was also observed
to alter joint opening, and thereby its load transfer efficiency.
j. Joint effect
The effect of aggregate interlock and dowelled joints on concrete slab response was examined.
The commonly adopted pure-shear aggregate interlock idealization oversimplifies the mechanics,
but it is used since it has huge computational advantage. To simulate joint behavior and its effect
on post-crack slab response more realistically, it is necessary to adopt nonlinear aggregate
interlock mechanics, so that the effects of concrete material properties, aggregate size and
distribution, as well as the change in joint width during the loading process, may be
accommodated. The effect of such joint characteristics on concrete pavement post-crack
293
response under static load and temperature curling was analyzed herein, possibly for the first
time. It was found that joint load transfer efficiency decreases linearly with increasing joint
width. The load-displacement curves exhibited one or two peak loads depending on the stiffness
of the joint. Dimensional analysis was employed for the limited sets of slab properties
considered in an effort to generalize observations made regarding the interaction among
individual design features affecting slab response.
7.3 Recommendations for Future Research
Following the development of new computational technologies, application of nonlinear fracture
mechanics to concrete structures has recently gained increased attention (Park et al., 2009). A
comparable implementation of such mechanistic approaches to pavement systems is still in its
infancy (Ioannides et al., 2006; Gaedicke and Roesler, 2009; Aure and Ioannides, 2010). In
contrast, the newly released (AASHTO, 2008) mechanistic-empirical pavement design guide
(MEPDG) employs computer programs but only to determine pavement responses, relying on
statistical algorithms for the prediction of pavement performance. Despite the considerable
mechanistic component of MEPDG, its empirical aspects are still thoroughly invested in the
statistical paradigm the new approach has inherited from its predecessors following the AASHO
Road Test (1958-1960), half a century ago. To overcome the lingering and significant
limitations, it is imperative that pavement analysis incorporate post-crack phenomena using self-
adaptive mesh generation strategies and crack branching techniques (Park et al., 2012). The
present study has endeavored in this direction through the assessment of the importance of
parameters pertaining to fracture analysis and post-crack behavior of concrete pavements. In
order to elucidate the complex interactions among these parameters, leading to the formulation of
294
more mechanistic pavement analysis and design procedures, the following are submitted as
potential areas for future investigations.
a. Fracture analysis of concrete pavements under mixed mode failure
The effects of shear stresses may be significant in thick concrete pavements, and in pavements
subjected to corner loads. Under such circumstances, the concrete may crack due to the
combined action of bending and shear, producing what is termed as mixed-mode failure.
b. Effects of cracking in non-surficial man-made layers
In general, pavement structures are constructed in such a way that the strength of man-made
layers decreases from top to bottom. Consequently, damage or cracking may occur in the base or
subbase before failure of the concrete slab, or simultaneously with it. Therefore, it may be
necessary to simulate the damage processes in non-surficial layers as well, in order to capture the
global response and behavior of the system.
c. Experimental verification of analytical studies into jointed concrete pavement cracking
Field and laboratory tests may be used to validate numerical studies, such as those conducted in
the present research on jointed concrete pavements. Early laboratory tests on jointed concrete
pavements had focused on pre-crack behavior (Colley and Humphrey, 1967; Hammons, 1997).
Consequently, the effects of joint characteristics on the fracture process of the slabs, and
conversely, the effect of slab cracking on load transfer efficiency have not been documented.
Such tests can also be used to identify variables that are not apparent in numerical analysis.
d. Fracture analysis of concrete pavements under fatigue loading
Pavement structures are subjected to cyclic environmental and traffic loads. Consequently,
failure in such structures is due to repeated action of such loads, but capturing this phenomenon
295
in a computer simulation poses significant numerical challenges. The quasi-brittle nature of
concrete and the difference between its compressive and tensile behaviors compound the
complexities commonly encountered in computational fracture analysis. The study of fatigue
cracking using cohesive zone simulation is barely at its earliest stage at this time (Park, 2009),
and much additional research is required before it can be applied to pavement systems.
e. Application of extrinsic numerical simulation in concrete pavements
In the present study, a fracture plane has been assumed along a fixed direction (usually termed as
discrete or intrinsic approach), along which cohesive elements are inserted. This assumption
neglects crack branching, a very common observation in concrete structures. Recent
developments in adaptive cohesive fracture simulation await application to concrete pavements
(Paulino et al., 2010; Park et al., 2012).
f. Application of dimensional analysis
Dimensional analysis can be invaluable in interpreting numerical data obtained from multiple
finite element runs for a factorial of geometric and material inputs. Its suitability in formulating
broad guidelines to be implemented by practicing engineers is particularly attractive, since
conclusions from theoretical developments must always be of consequence to design
applications.
g. Specimen size effect on fracture parameters
In contemporary practice, material properties for concrete pavement design are usually derived
from tests on small, simply supported beam specimens in a laboratory. It is almost universally
admitted that this approach may seriously misrepresent the strength characteristics of in situ
pavement slabs. Experimental and numerical studies are sorely needed in order to elucidate the
296
repercussions of the so-called size effect on concrete fracture parameters, as well as on the field
responses of concrete pavement systems.
h. Effects of shrinkage and creep
Early age concrete shrinkage may result in premature cracking shortly after a pour, especially in
longer jointed concrete pavements. The constitutive material relations adopted in the present
study do not account for the effects of shrinkage and creep. Further research is necessary to
account for the effects of such phenomena on the fracture process.
7.4 References
Interim Ed., American Association of State Highway and Transportation Officials, Washington,
DC.
Aure, T. W., and Ioannides, A. M. (2010). "Simulation of Crack Propagation in Concrete Beams
using Cohesive Elements in ABAQUS." Transp. Res. Rec. 2154, Transportation Research Board,
Colley, B. E., and Humphrey, H. A. (1967). "Aggregate interlock at joints in concrete
pavements." Bulletin 189, Highway Research Board, National Research Council, Washington,
DC, 14-24.
297
908.
Park, K. (2009). "Potential-based fracture mechanics using cohesive zone and virtual internal
bond modeling," Ph.D. dissertation, University of Illinois at Urbana-Champaign, Urbana, IL.
Park, K., Paulino, G. H., Celes, W., and Espinha, R. (2012). " Adaptive mesh refinement and
coarsening for cohesive zone modeling of dynamic fracture." Int. J. Numer. Methods Eng., 92(1),
John Wiley and Sons, Ltd., Hoboken, NJ, 1-35.
Paulino, G. H., Park, K., Celes, W., and Espinha, R. (2010). "Adaptive dynamic cohesive
fracture simulation using nodal perturbation and edge‐swap operators." Int. J. Numer. Methods
Eng., 84(11), John Wiley and Sons, Ltd., Hoboken, NJ, 1303-1343.
temperature." Proc. Hwy. Res. Board 6, Highway Research Board, National Research Council,
298
299

Numerical Analysis of Cracking in Concrete Pavements Subjected To Wheel Load and Thermal Curling PDF

Uploaded by

Copyright:

Available Formats

Numerical Analysis of Cracking in Concrete Pavements Subjected To Wheel Load and Thermal Curling PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Numerical Analysis of Cracking in Concrete Pavements Subjected To Wheel Load and Thermal Curling PDF

Uploaded by

Copyright:

Available Formats

Numerical Analysis of Cracking in Concrete Pavements

Subjected to Wheel Load and Thermal Curling

A dissertation submitted to the

of the University of Cincinnati

in partial fulfillment of the requirements for the degree of

in the School of Advanced Structures

M.Sc., Addis Ababa University (2008)

Committee Chair: Anastasios M. Ioannides, Ph.D.

parameter that influences the responses.

exploration of fracture behavior in concrete pavements. The techniques can be extended to

matters have broadened my outlook.

assistantship during the final stage of this research.

I am grateful to my friends Dawit Alemayehu, Demelash Hailemariam, Aschalew Kassa, Assefa

support during the last year of this study.

who instilled interest and curiosity in education since my early childhood.

Table of Contents ....................................................................................................................... viii

List of Tables .............................................................................................................................. xiv

List of Figures ............................................................................................................................. xvi

List of Symbols and Abbreviations ......................................................................................... xxii

Chapter 1 Research Overview ............................................................................................... 1

1.1 Introduction ..................................................................................................................... 1

1.2 Review of Concrete Fracture .......................................................................................... 3

1.3 Research Objective and Significance.............................................................................. 6

1.4 Research Scope ............................................................................................................... 7

1.5 Thesis Organization ........................................................................................................ 7

1.6 References ....................................................................................................................... 9

Chapter 2 Simulation of Crack Propagation in Concrete Beams ..................................... 19

2.1 Introduction ................................................................................................................... 20

2.2 Methodology ................................................................................................................. 23

2.3 Analysis of Linear Elastic Response............................................................................. 23

2.4 Simulation of Crack Propagation Using Built-in Cohesive Elements .......................... 26

2.4.1 Idealization of Cohesive Zone .................................................................................. 27

2.4.2 Sensitivity Study for Proposed Discretization .......................................................... 33

2.5.1 GTSTRUDL/CRACKIT by Sengupta (1998) .......................................................... 38

2.5.2 ABAQUS® - JOINTC by Ioannides et al. (2005) ..................................................... 39

2.5.3 ABAQUS® - 2D UEL by Roesler et al. (2007a) ....................................................... 39

2.5.4 ABAQUS® - 2D COH2D4 by Gaedicke and Roesler (2009) ................................... 40

2.6 Comparisons with Experimental Results ...................................................................... 41

2.6.1 Experimental Results by Liu (1994) ......................................................................... 41

2.6.2 Experimental Results by Roesler et al. (2007a) ........................................................ 42

2.7 Conclusions ................................................................................................................... 42

2.8 References ..................................................................................................................... 44

Chapter 3 Numerical Analysis of Fracture Process in Pavement Slabs .......................... 67

3.1 Introduction ................................................................................................................... 68

3.2 Literature Review.......................................................................................................... 69

3.3 Finite Element Discretization ....................................................................................... 71

3.4 Linear Elastic Analysis ................................................................................................. 72

3.5 Sensitivity Studies ......................................................................................................... 73

3.5.1 Effect of Concrete Softening Curve .......................................................................... 73

3.5.2 Effect of Viscous Regularization .............................................................................. 78

3.5.3 Effect of Subgrade Idealization ................................................................................ 80

3.5.4 Effect of Cohesive Zone Width ................................................................................ 81

3.5.5 Effect of Cohesive Zone Mesh ................................................................................. 82

3.5.6 Effect of Solution Technique .................................................................................... 82

3.5.8 Effect of Tensile Strength and Fracture Energy ....................................................... 84

3.6 Crack Propagation under Interior Loading ................................................................... 86

3.7 Crack Propagation in a Slab Supported by Base Layer ................................................ 87

3.8 Comparison with Other Numerical and Experimental Studies ..................................... 88

3.8.1 Comparison with Ioannides et al. (2006) .................................................................. 88