Sunteți pe pagina 1din 11

Construction and Building Materials 208 (2019) 102–112

Contents lists available at ScienceDirect

Construction and Building Materials


journal homepage: www.elsevier.com/locate/conbuildmat

A calibrated mechanics-based model for top-down cracking of asphalt


pavements
Meng Ling a,⇑, Xue Luo b, Yu Chen c, Sheng Hu d, Robert L. Lytton e
a
Texas A&M Transportation Institute, Texas A&M University System, 3136 TAMU, DLEB 601C, College Station, TX 77843, United States
b
College of Civil Engineering and Architecture, Zhejiang University, 866 Yuhangtang Road, An-zhong Bldg, Hangzhou 310058, China
c
Texas A&M Transportation Institute, Texas A&M University System, 3136 TAMU, DLEB 508G, College Station, TX 77843, United States
d
Texas A&M Transportation Institute, Texas A&M University System, College Station, TX 77843, United States
e
Zachry Department of Civil Engineering, Texas A&M University, 3136 TAMU, DLEB 503A, College Station, TX 77843, United States

h i g h l i g h t s

 Top-down cracking growth in asphalt pavements was characterized using fracture mechanics and Paris’ law.
 Fracture and aging properties were recognized as two essential effects on top-down cracking propagation.
 Top-down cracking was identified as the precursor of fatigue cracking based on field performance.
 Transverse thermal stress was demonstrated to play a minor role in top-down cracking propagation.

a r t i c l e i n f o a b s t r a c t

Article history: A calibrated mechanistic-empirical model was developed to analyze top-down cracking (TDC) perfor-
Received 9 July 2018 mance with actual traffic load and thermal stress. The traffic-induced and thermal-induced J-integral
Received in revised form 12 February 2019 were employed in the Paris’ law with fracture parameters A0 and n0 estimated using asphalt mixture prop-
Accepted 14 February 2019
erties. Number of days for initiated TDC to reach the medium severity level was then calculated mecha-
nistically, which was related to field performance for calibration purposes. It was observed that when
TDC reached a critical length, the TDC length diminished and area of fatigue cracking increased. An asso-
Keywords:
ciated computer program was also developed, which normally completed a 30-year performance analysis
Asphalt pavement
Top-down cracking
within 1 min.
Oxidative aging Ó 2019 Elsevier Ltd. All rights reserved.
Load spectrum
Thermal fatigue
LTPP

1. Introduction instance, the AASHTO Mechanistic-Empirical (ME) Pavement


Design Guide Manual of Practice developed under NCHRP Project
Top-down longitudinal wheelpath cracking (TDC) is now identi- 01-37A provides a tentative methodology for the analysis and per-
fied worldwide as one of the dominant distress modes in asphalt formance prediction of TDC, which considered traffic-induced ten-
pavements. TDC permits the intrusion of water and foreign objects sile stress away from tires as the major driving force for TDC
into the asphalt surface layer, which deteriorates the surface struc- development [1]. In addition, recent work completed under NCHRP
tural integrity and increases the pavement maintenance cost. Project 01-42A developed a viscoelastic continuum damage crack
Recent studies have suggested some hypotheses in terms of TDC initiation model and an asphalt layer fracture mechanics crack
mechanisms (e.g. bending-induced surface tension and shear- propagation model [2]. However, additional research effort is
induced near-surface tension) and developed preliminary models needed to address the associated issues and to develop a calibrated
to predict TDC initiation and propagation characteristics. For ME model for incorporation into the AASHTO Pavement ME Design
procedures. Thus, to expand the function of the Pavement ME
⇑ Corresponding author. Design program, there is an urgent need to develop a model which
E-mail addresses: mengling@tamu.edu (M. Ling), xueluo@zju.edu.cn (X. Luo), yu. yields a more rational mechanistic TDC analysis and practical
chen@tamu.edu (Y. Chen), hu@tamu.edu (S. Hu), r-lytton@civil.tamu.edu design procedures of asphalt pavements and overlays.
(R.L. Lytton).

https://doi.org/10.1016/j.conbuildmat.2019.02.090
0950-0618/Ó 2019 Elsevier Ltd. All rights reserved.
M. Ling et al. / Construction and Building Materials 208 (2019) 102–112 103

It should be noted that this study aimed at pavement material fatigue cracking appears. This phenomenon was widely identified
and structural design to resist load-related TDC, and the assess- in the LTPP database and is illustrated in the next section, which
ment of construction quality such as segregation or debonding is was also in accordance with the observations of Svasdisant et al.
out of the scope of this study. Two load mechanisms (tire- [9]. The TDC ME model was then calibrated using the LTPP data
pavement contact stresses and thermal stress) are usually recog- distributed in four climatic zones. Meanwhile, a practical TDC ME
nized as the loading effect to drive the surface-initiated crack to model should be compatible with the Pavement ME Design soft-
propagate downwards. Specifically, TDC length is cumulated with ware for further implementation. To facilitate the computation of
repeated vehicular loading and increase of load level, which is the monthly TDC growth, a windows-based standalone computer
accelerated by daily and seasonal temperature variations and program with a user interface was developed accordingly for rou-
oxidative aging of surface asphalt mixtures. Therefore, the propa- tine use, which makes the whole model convenient for TDC pave-
gation of TDC with time is primarily a fatigue damage type. ment design and maintenance strategies.
Besides, the contact stresses are actually non-uniformly dis- Fig. 1(a) illustrates the TDC ME model developed in this study
tributed within tire patch, which introduce a mixed tensile and and how it will be incorporated within the Pavement ME Design
shear stress state in and near the wheelpaths and are essential software. Fig. 1(b) demonstrates the analysis procedure of this
for accurate TDC analysis [3]. In one previous study, the non- study representing the major components contributed to the corre-
uniform three-dimensional (3D) contact stresses consisting of ver- sponding sub-models. After all of the sub-models were constructed,
tical, transverse and longitudinal stresses for a typical 315/80R22.5 the TDC ME model was assembled to calculate the ‘‘pre-fatigue life”
tire were measured using stress-in-motion systems by De Beer mechanistically and calibrated to field performance.
et al. [4]. The 3D stress patterns were further simplified and imple-
mented into finite element models (FEM) to compute the traffic 2. Fracture mechanics modeling of TDC propagation
load-induced J-integral at various crack depths with different
material properties and pavement structures [5]. The determined Paris’ law is widely applied as the state of practice to characterize
J-integral at the tip of TDC is critical to calculate the cumulative the steady fatigue damage stage in asphalt pavements [7,10–14]. In
crack growth with pavement depth using the Paris’ law and frac- this study, the generalized J-integral based Paris’ law was utilized to
ture mechanics. In addition, thermal fatigue induced by tempera- compute the load induced and thermally induced TDC growth with
ture variations is another potential influencing factor to TDC time, which is presented in Eq. (1). Note that for the TDC initiation
propagation. However, transverse thermal stress should be focused phase, damage density was adopted rather than the length of crack
on rather than longitudinal thermal stress due to the direction of due to the nature of coexistence of multiple micro-cracks, and the
TDC. In fact, the transverse thermal stress was examined to be J-integral in the crack initiation phase was much smaller compared
smaller than the longitudinal thermal stress due to pavement to that in the crack propagation phase [8].
boundary conditions [6]. Using a similar fatigue analysis method
like that for the traffic load, TDC growth induced by the thermal dc 0
¼ A0 ð J Þ n ð1Þ
fatigue can also be computed [7]. On the other hand, the increase dN
of aging level especially for the asphalt surface layer results in a
where c is the single macro-crack depth induced by traffic load or
brittle material that is imperative in the TDC ME analysis. Oxida-
thermal stress; N is the number of traffic load repetitions or temper-
tive aging is accelerated by high air void content and high air tem-
ature variations; A0 and n0 are the asphalt surface layer fracture
perature, which is mainly responsible for reducing the crack
properties, both of which change with aging level, moisture content
resistance and stress relaxation capability of pavement surface
and asphalt mixture composition; and J is the J-integral at the tip of
materials as an environmental factor. It is expected that after
TDC under traffic load or thermal stress.
long-term field aging, the pavement surface mixture becomes stif-
fer than the layers below and the modulus gradient is developed, The fracture parameters (A0 and n0 ) and J-integral can be deter-
making it easy for a crack to develop at the pavement surface. mined experimentally through various types of mechanical tests
A TDC initiation model has been developed recently based on such as repeated tension/compression test and monotonic load
the viscoelastic micro-fracture mechanics [8]. Required inputs col- test, and the fracture parameters A0 and n0 were found to be mate-
lected from the long-term pavement performance (LTPP) database rial dependent. For example, laboratory fatigue tests and analysis
were used to calculate the TDC initiation energy parameter and for viscoelastic materials indicated that n0 is inversely proportional
TDC initiation time for pavement sections in different climatic to the m-value by Schapery and Lytton et al. [15,16], which is the
zones. Tensile stress was attributed as the major load factor to slope of log creep compliance versus log loading time. Further-
the initiation and coalescence of multiple micro-cracks moving more, a strong linear relationship was exhibited between log A0
upwards to the pavement surface from the near surface. It was and n0 , once the parameter n0 is obtained, parameter A0 can be pre-
shown that this model was capable of predicting TDC initiation dicted using the regression equation such as the ones developed in
performance and being conducted as a practical TDC initiation [11,12]. As a result, two input levels for fracture parameters A0 and
design tool. Based on the findings of the previous study, the TDC n0 were achieved in this study, level one was based on testing
initiation phase is mainly dominated by the combined effects of results for associated asphalt mixture, and level two was based
traffic and low temperature, number of high temperature days, on the regression analysis from the fundamental asphalt material
asphalt mixture relaxation capability and pavement structure. In properties as an alternative approach.
this study, a calibrated mechanics-based TDC evolution model On the other hand, the traffic load induced J-integral took into
was developed to compute the ‘‘pre-fatigue life” and characterize account the effects of load spectrum, temperature and modulus
the TDC propagation feature. The ‘‘pre-fatigue life” was defined gradient with asphalt surface layer, moduli of base and subgrade
herein as the period for a single initiated TDC to propagate from layers, and pavement structural geometry [5]. As seen from
low severity level to medium severity level. When a critical TDC Fig. 2, the load-induced J-integral with asphalt layer depth reached
length was reached, multiple longitudinal wheelpath cracks were a peak value near the surface then decreased, which indicates that
identified at the pavement surface and then became intercon- TDC could initiate near the surface then propagate upwards. The
nected, which were rated as fatigue cracking. As such, it was ter- thermally induced J-integral accounted for temperature variation,
med as ‘‘pre-fatigue life” as it is the number of days when a temperature and modulus gradient, coefficients of thermal con-
single longitudinal crack at low severity grows until the associated traction and moduli of asphalt surface layer and base course, and
104 M. Ling et al. / Construction and Building Materials 208 (2019) 102–112

AASHTOWare Model Proposed Top-Down Cracking Model

Traffic Material Traffic Material


Properties Tire contact Pressure Properties

Climate INPUT
Pavement Climate Pavement
EICM Structure Structure
Temperature Model

Pavement Response Model


Pavement Response ( , ) Model: J-Integral
Multi-layer elastic system Crack Depth, Crack Width
MODELS Artificial Neural Network

Pavement Distress Models


Pavement Distress Models Transverse, Longitudinal,
Vertical, Thermal Stresses

Pavement Performance OUTPUT Pavement Performance Predictions


Predictions Top-Down Cracking Crack Depth
Extent and Severity

(a) Incorporation of TDC ME Model with Pavement ME Design Software

(b) Analysis Procedure of TDC ME Model


Fig. 1. Compatibility of TDC model and flowchart of TDC ME analysis.

pavement layer thicknesses [7]. In general, the J-integral was sen- factorial sets of J-integral to expedite the J-integral prediction for
sitive to pavement layers properties, temperature, load level, loca- both traffic load and thermal load, respectively. After training
tion of crack tip and pavement structure. and validating the ANN models, the normalized input and output
In order to calculate the crack length increment efficiently on variables, weight information, and neuron information for each
the pavement structural level, there is also a need to compute variable were retrieved from each of the ANN models, which were
the J-integral under traffic load spectra and temperature variations used to compute the J-integral in the computer program coded in
precisely. To make the J-integral readily applicable, FEM and artifi- C# developed and calculate the monthly crack growth and result-
cial neural network (ANN) models were constructed to obtain full ing ‘‘pre-fatigue life”.
M. Ling et al. / Construction and Building Materials 208 (2019) 102–112 105

J-integral (Pa*m) average daily truck traffic (AADTT) was utilized with an average
0 0.5 1 1.5 number of axles for each vehicle class and a normalized vehicle
0 distribution defined in the Pavement ME Design [18] was used to
Crack Depth/Asphalt Layer Thickness

compute the number of axle loads for each vehicle class and axle
0.2
type. A total of 6 ANN models were proposed for the following tire
Critical lengths and tire types of single axle load conditions: 64 mm (dual
J-integral
tire), 127 mm (dual tire), 229 mm (dual tire), 64 mm (single tire),
0.4 305 mm (single tire) and 406 mm (single tire). The six loading con-
ditions and associated load levels are illustrated in Fig. 3(b). After
the J-integral for the six conditions was calculated, linear interpo-
0.6
lation was applied to estimate the J-integral at other load levels
(tire lengths) for which the J-integral was not computed directly
0.8 from the FEM and ANN models. The interpolation process was
automatically completed in the C# based computer program. The
Critical predicted J-integral was then used in the Paris’ law to compute
J-integral
1 the monthly crack growth with load spectra, as shown in Eq. (2).
Fig. 2. Load-induced J-integral with pavement depth. X
384
n0
DC ¼ DNi A0 ðJid Þ ð2Þ
i¼1
A load spectrum model was developed to characterize the
where i is the load level i in a load spectrum and d is the location of
annual traffic load distribution as the level one input, since the
the crack tip below the pavement surface. There are 38 load levels
actual traffic load included various load levels, axle numbers, and
in each of the four axle types (single, tandem, tridem, and quadrem
tire numbers [8]. In the FEM, the tire-pavement contact area was
axles) in the LTPP database. DNi and J id are the monthly number of
assumed to be rectangular with a tire pressure of 689.5 kPa, and
loading cycles at load level i of a specific axle type and J-integral at
effective tire width remained nearly constant based on the mea-
the crack tip located at pavement depth d below the surface at that
surement conducted in [17]. Therefore, the greater contact area
load level and axle type, respectively.
due to a higher load level was represented by the increase of tire
length. The axle load distribution was defined as the number of
axles in each load interval of a specific axle spectrum. One example 2.1. Asphalt mixture fracture and aging sub-model
of a cumulative single axle load distribution with tire length for a
load spectrum is shown in Fig. 3(a). For the level two input, annual Oxidative aging of asphalt mixtures is one of the most crucial
environmental factors that affect the material and fracture proper-
ties of an asphalt mixture. The fracture parameters (A0 and n0 ) and
1 J-integral have been shown to be aging dependent. Particularly,
when the aging level increases, the m-value and E1 of the relax-
ation modulus in Eq. (3) decreases and increases, respectively. As
0.8
Cumulative Axle Load

a result, the asphalt material becomes less viscoelastic and gradu-


ally loses the relaxation capability, which results in the increase of
Distribution

0.6 the parameter n0 , as presented in Eq. (4). Eq. (4) was developed
based on the regression analysis with the test results of previous
0.4 studies. It was found that the decrease of air void content (i.e. less
initial damage density), increase of asphalt content (i.e. softer and
0.2 more resilient mixture) and decrease of aggregate gradation shape
parameter (i.e., more dense graded) defined in Eq. (5) improve the
fracture resistance of asphalt mixture.
0
The parameter n0 was determined to be a critical fracture index.
0 10 20 30 40 50 60
Generally, more fracture susceptibility is expressed as a larger
Tire Length (cm)
value of n0 . Besides, based on the limited laboratory testing results,
(a) Cumulative Axle Load Distribution with Tire Length the estimation of the decrease of m-value with aging time was
taken into account in this study, as shown in Eq. (6). Meanwhile,
10000 as the modulus increased with aging, J-integral also became aging
dependent, which was characterized in the FEM.
8000
Eðt Þ ¼ E1 t m ð3Þ
Axle Load (kg)

6000  m
1 1
n0 ¼ 9:00498  40:8788  þ 2:8713 
4000 E1 m
Single Tire AV%
þ 1:0627  w þ 13:4772  ð4Þ
2000 AV% þ lnAC%
Dual Tire

0 f ðxÞ ¼ hxw ð5Þ


0 10 20 30 40 50
Tire Length (cm) m ¼ m0  0:009  year ð6Þ
(b) Axle Load Level with Tire Length
where E1 and m are the relaxation modulus parameters; h and w are
Fig. 3. Characterization of axle load in TDC ME model. cumulative aggregate gradation parameters; %AV and %AC are
106 M. Ling et al. / Construction and Building Materials 208 (2019) 102–112

percent air void content and percent asphalt content by volume, 2.1.2. Kinetics-based aging prediction for asphalt pavements
respectively; m0 is the unaged relaxation modulus parameter. A kinetics-based aging model developed previously was
selected to predict the aging effect on the back-calculated moduli
of asphalt pavements at different field aging times based on falling
2.1.1. Laboratory evaluation and modeling of asphalt field cores
weight deflectometer (FWD) data collected from the LTPP database
Two aging features (non-uniform aging and long-term aging)
[21]. The model contained kinetic parameters for asphalt mixtures
occur in asphalt pavements simultaneously which were charac-
(e.g. aging activation energy). For each LTPP section, the rheological
terized in this study. First, direct tension tests were performed
properties and aging properties of the asphalt mixtures were
at 10, 20 and 30 °C on field core specimens with different field
determined to characterize the temperature-dependency and pre-
aging times, and strains at top, center and bottom of the field
dicted long-term field aging of asphalt pavements, respectively.
cores and monotonic tensile load with time were recorded. The
The prediction equation for the moduli in four climatic zones
complex modulus gradient of the field core specimen was inves-
including Wet-freeze (WF), wet no-freeze (WNF), dry-freeze (DF),
tigated with an iteration process using elastic bending theory,
and dry no-freeze (DNF) is presented as follows:
elastic-viscoelastic correspondence principle and pseudo strain
 
concept. The detailed test and analysis procedures can be found jE jt ¼ jE ji þ ðjE j0  jE ji Þ 1  ekf t þ kc t ð10Þ
at Ling et al. [19]. The modulus gradient of an asphalt layer at a
specific loading frequency and temperature is shown in Eqs. (7) E
RT af
and (8). in which kf ¼ Af e field ð11Þ
 n
dz E
RT ac
EðzÞ ¼ Ed þ ðE0  Ed Þ ð7Þ kc ¼ Ac e field ð12Þ
d
where Ei is the initial modulus which can be predicted using the
E0 ANN model developed by Bari and Witczak [22], jE j0 is the inter-
k¼ ð8Þ cept of the constant-rate line of the modulus; kf is the fast-rate
Ed
reaction constant; kc is the constant-rate reaction constant; t is
where E (z) is the dynamic modulus in pavement depth z at a speci- the aging time, day; Af is the fast-rate pre-exponential factor; Eaf
fic loading frequency and temperature; Ed and E0 are the dynamic is the fast-rate aging activation energy;Ac is the constant-rate pre-
moduli at the bottom and top at the same loading and temperature exponential factor; Eac is the constant-rate aging activation energy;
condition, respectively; d is the thickness of the field core specimen; R is the universal gas constant (8.314 J/molk); and T field is the field
n is the model parameter, which presents the shape of the stiffness aging absolute temperature, Kelvin. The values in Table 1 were
gradient; and k is the ratio of the modulus at the top to the modulus obtained from the back-calculated field modulus LTPP data, which
at the bottom. were categorized according to the climate zones.
The modulus gradient parameters n and k were determined as
complex numbers and they are actual material properties of field- 2.2. TDC performance sub-model
aged asphalt mixtures [19]. The two parameters were then used
in the FEM and ANN models to characterize the non-uniform TDC performance data and asphalt surface layer material prop-
aging feature of asphalt surface layer [5]. Based on the test erties of LTPP pavement sections were acquired from the LTPP
results, the parameter n did not show consistent aging depen- database. Specifically, the pavement structure (thicknesses of
dence and the parameter k increased with aging time t (month) pavement layers), asphalt mixture material properties (unaged
in Eq. (9). dynamic modulus master curve, air void content, binder content,
aggregate gradation), annual traffic load (load spectra and AADTT),
k ¼ 1 þ 0:397t 0:2933 ð9Þ temperature history (annual numbers of days above 32 °C and
After the complex moduli at different temperatures and aging below 0 °C, and hourly air temperature) and pavement perfor-
levels were determined, the modified Christensen-Anderson- mance (longitudinal wheelpath cracking length with time) were
Marasteanu (CAM) model was used to construct the dynamic mod- the required inputs to develop and calibrate the mechanics-based
ulus master curves for the same asphalt mixtures at different ages. TDC performance prediction model. It should be noticed that only
As aging level increased, the magnitude of dynamic modulus the unaged or slightly aged material properties of asphalt mixtures
increased and the shape of the master curve became flatter and were collected, since the aging processes are different for various
less time-dependent. Hence, the glassy modulus and rheological asphalt mixture types and climatic zones.
index in the modified CAM model were identified as a vertical shift A sigmoidal-shaped characteristic curve in Eq. (13) was applied
factor and rotation factor to characterize the higher upper asymp- to fit the observed total lengths of longitudinal wheelpath cracking
totes and smaller slopes of the master curves for the asphalt mix- with time.
tures at longer aging times, respectively. After applying the vertical q b

shift and rotation, the higher upper asymptotes and smaller slopes lðtÞ ¼ l0 eðtto Þ ð13Þ
of the dynamic modulus master curves at longer aging times were where lðtÞ is the total longitudinal crack length at service time t
normalized to the ones at the reference aging time. Based on the after construction, which was defined as the total measured crack
experimental data analysis, power functions were utilized to fit
and predict glassy modulus and rheological index at longer aging
times. Thereafter, an aging shift function similar to the time-
Table 1
temperature shift function was developed, which was determined Default values of aging properties of field asphalt pavements.
to be a function of aging time, aging temperature, activation energy
and lab-field acceleration factor, all of which made it possible to Climate Zone Eaf (kJ/mol) Af Eac (kJ/mol) Ac

horizontally shift the modified master curves of the asphalt mix- WF 15.21 1.04E + 01 40.33 6.87E + 07
tures at longer aging times. This allowed the shifted aged master WNF 28.80 1.31E + 03 46.90 2.69E + 07
DF 20.19 9.81E + 01 37.07 2.51E + 05
curves to superimpose upon the unaged one and predict the mas-
DNF 15.85 9.85E + 00 30.79 4.60E + 04
ter curves at other aging times [20].
M. Ling et al. / Construction and Building Materials 208 (2019) 102–112 107

lengths in three severity levels without weighting factor, m; l0 is 150

LongitudinalWheelpath Crack Length


the maximum theoretical longitudinal crack length as 300 m (i.e.
150 m is the length of a typical LTPP section); q and b are the 120
scale parameter and shape parameter of the distress curve,
respectively; t 0 is the crack initiation time, day; and t is the pave-

(m/300m)
90
ment service time, day. Prior to this study, the TDC initiation
time t 0 was reasonably predicted based on micro-fracture
mechanics and field performance [8]. In addition to t0 , both q 60

and b had physical meaning. The scale parameter q was identi-


fied mathematically as the service time (number of days) for 30
TDC to reach 1/e (36.8%) of the maximum crack length (300 m)
after initiation. Therefore, it was directly related to the ‘‘pre- 0
fatigue life”, which was computed with the sub-models using 0 1000 2000 3000 4000 5000 6000
Eq. (1). Particularly, a larger measured q represented a longer Service Time (day)
TDC ‘‘pre-fatigue life”. The parameter b was used to characterize Measured Fitted
the shape of the distress curve, which was normally inversely
proportional to the scale factor, q. The effects of change ofq (a) LTPP Section 12-0108
and b on TDC performance are exhibited in Fig. 4(a) and (b), 150
respectively. Nonlinear regression was performed using the Excel

Longitudinal Weelpath Crack Lengh


solver function to obtain parameters q and b for the selected
LTPP pavement sections. Examples of regression and associated 120
pavement performance are presented in Fig. 5. Calibration was
then conducted to both q and b, which is discussed in the next

(m/300m)
90
section.

60
Longitudinal Wheelpath Crack Length (m)

200
30

150 0
0 1000 2000 3000 4000 5000
= 3500 Service Time (day)
100 Measured Fitted

(b) LTPP Section 4-0117


50
Fig. 5. Examples of longitudinal cracking characteristic curve.

2.2.1. Evaluation of longitudinal wheelpath cracking and fatigue


0 cracking
0 2000 4000 6000 8000 Three severity distress levels of longitudinal wheelpath crack-
=0.7 =0.9 =1.1 =1.5 ing and corresponding crack length were recorded and categorized
Service Time (day) based on the LTPP severity level criterion. From the observations of
TDC performance data collected, only about 15–20% of longitudinal
(a) Effect of parameter on TDC performance curve
wheelpath cracking was in the medium and high severity levels.
Meanwhile, there were not many distress points for the crack
250
Longitudinal Wheelpath Crack Length (m)

propagation histories. Most pavement sections only had about 2–


5 points, especially for those in the DF and WF zones. It appeared
200 that the TDC occurred more frequently in the non-freeze zones
and TDC became alligator cracking when it reached a crack length
of approximate 50–100 m. Such examples of observations are illus-
150
trated in Fig. 6(a) and (b), which indicate that as the length of TDC
=1.1
(longitudinal wheelpath cracking) diminished, alligator cracking
100 took its place and alligator cracking area increased. As a result, this
suggests that not all fatigue cracking was initiated from the bottom
of asphalt layer, which further demonstrated that the appearance
50 of TDC is the first stage of fatigue cracking. After the longitudinal
cracking length increased, several longitudinal cracks were con-
nected by short transverse cracks, which were then rated as alliga-
0
0 2000 4000 6000 8000 tor cracking in the LTPP database. For the characterization of TDC
propagation in this study, only the distress points prior to the
=2000 =3000 =4000 =5000
decline of the TDC length were used.
Service Time (day)

(b) Effect of parameter on TDC performance curve 2.2.2. Empirical relationship between depth and width of TDC
In order to correlate the calculated TDC depth with the measur-
Fig. 4. Effects of sigmodal-shaped curve parameters on TDC performance. able crack width, the field core data extracted from Kumara et al.
108 M. Ling et al. / Construction and Building Materials 208 (2019) 102–112

80 80 Table 2
Measured crack width to crack depth ratio [23].
Longitudinal Wheelpath Crack Length (m)

70 70
Crack Depth (cm) Crack Width (cm)
60 60 2.755 0.376

Alligator Crack Area (m2)


3.747 0.655
50 50 3.931 0.687
4.408 0.562
40 40 4.518 0.532
4.775 0.992
30 30 5.767 0.787

20 20

10 10 used to determine the hourly temperature variation at the pave-


ment surface based on heat transfer fundamentals. FEM were
0 0
01-31-1993 10-28-1995 07-24-1998 04-19-2001 01-14-2004 10-10-2006 developed to obtain the far-field thermal stress and associated
Longitudinal Cracking Alligator Cracking
J-integral at the tip of the crack for numerous asphalt material
properties and pavement structure combinations [7]. The thermal
Service Date
stress was computed using the one dimensional viscoelastic
(a) LTPP Section 1-0101 constitutive equation [24], which was also one of the inputs in
70 70
the J-integral predicted by the ANN models. The constructed ANN
models had R2 values above 0.99 indicating a goodness of fit. In
Longitudinal Wheelpath Crack Length (m)

60 60
order to calculate the thermal stress, the relaxation modulus mas-
ter curve was converted from the dynamic modulus master curve
50 50 through the interconversion using the Prony series models. As a
Alligator Crack Area (m2)

result, a relationship between viscoelastic thermal stress and ther-


40 40 mally induced J-integral at the crack tip was established using the
ANN models with known material properties and pavement
30 30
structures.
Since the direction of thermal-induced TDC is parallel to the
direction of traffic flow, a maximum value of 69 KPa (10 psi) for
20 20
the transverse thermal stress was adopted based on the finding
by Zou and Roque [6]. The input variables of ANN models included
10 10 thickness of asphalt and base layers, viscoelastic thermal stress,
modulus gradient of asphalt layer, coefficients of thermal expan-
0 0 sion of the asphalt and base layers, modulus of base layer and
01-31-1993 07-24-1998 01-14-2004 07-06-2009 12-27-2014
TDC depth, and the output variable is the thermally induced J-
Longitudinal Cracking Alligator Cracking integral. It was observed that the thermal J-integral was normally
Service Date smaller than the traffic load induced J-integral. Therefore, the ther-
(b) LTPP Section 12-0107 mally induced TDC growth was expected to be negligible.

Fig. 6. Examples of longitudinal wheelpath cracking and alligator cracking in


asphalt pavements.
3. Computer program for TDC performance prediction

To facilitate the mechanistic computations of cumulative crack


[23] in Table 2 was used to develop the relationship between the
growth, two C# computer programs developed in Microsoft Visual
crack width and the crack depth. The best fitting function to the
Studio were used to calculate the crack growth due to traffic and
data in Table 2 is as follows:
thermal stress, respectively. An initial crack depth was assumed
dc ¼ 5:4  w0:555 ð14Þ to be 7.5 mm based on Lytton et al. [16]. It was noticed from the
results that the thermally related TDC propagation was not signifi-
where w is the width of the crack, cm; and dc is the depth of the cant compared to the traffic load related TDC growth, as only four of
crack, cm. Based on Eq. (14), the critical crack depth between low 40 LTPP sections exhibited limited thermal TDC ‘‘pre-fatigue life”,
and medium severity levels was predicted to be approximately and all of which were in the WF zone. Fig. 7 illustrates the interface
4 cm. of the standalone computer program, which includes pavement
structure, material properties, traffic information and climatic
2.3. Longitudinal thermal cracking sub-model information as the input components. The asphalt mixture modulus
was defined as the representative elastic modulus in [11], which
In addition to the traffic load, thermal stress is another factor was based on pulse time and average temperature. After running
that potentially influences the TDC evolution. Again, Eq. (1) was the program, the ‘‘pre-fatigue life” was computed accordingly,
utilized to compute the TDC growth due to thermal fatigue which accounted for all of the sub-models described above and
induced by temperature cycling. Similar to the traffic load induced was used for calibration purposes. In general, the program com-
J-integral, the computation and use of the thermally induced pleted the whole analysis over a period of 30 years within 1 min.
J-integral also followed the same calculation procedure as with
traffic loading. As the first step, the factorial design of J-integral 3.1. Input module
was developed from the FEM results, and then the ANN models
were constructed to predict the J-integral under various tempera- 3.1.1. Traffic load
ture conditions, material properties and pavement structures. A In the computer program, two traffic load levels were adopted,
one-dimensional pavement temperature prediction model was level one was the annual load spectrum model with three
M. Ling et al. / Construction and Building Materials 208 (2019) 102–112 109

Fig. 7. Interface of C# based computer program.

parameters (a,b,c) for single and dual tires loading and level two pavement structural geometry and traffic load spectra were ana-
was based on AADTT with a default distribution. The load spectrum lyzed based on the sub-models above. The parameters of q and b
is modeled as follows: were estimated after calibration, which are presented in the next
section. In addition, the associated TDC characteristic curve was
ðbcLij Þ
NðLi Þj ¼ aee ð15Þ automatically generated against the pavement service time.

where N(Li)j is the number of axle loads at tire length Li (load level i)
of axle category j; a, b and c are magnitude parameter as the total 4. Calibration of TDC performance model
number of axles, width parameter and slope parameter of the
cumulative axle load distribution curve, respectively. In terms of the modeling of TDC propagation, there were two
‘‘pre-fatigue lives” consisting of the number of days for traffic load
solely and number of days for thermal stress solely to reach the
3.1.2. Temperature condition
critical crack depth of 4 cm as the average boundary line between
The climatic zone and temperature of the nearest weather sta-
low and medium severity within an asphalt surface layer. They
tion to the construction site and annual temperature information
were computed individually using the sub-models within the
including the number of days above 32 °C and below 0 °C were
software.
selected to determine the crack initiation and propagation charac-
The two calculated independent numbers of days for all the
teristics. The climatic histories with hourly temperature variations
LTPP sections selected were combined to predict the observed q
in recent consecutive 30 years and heat transfer parameters such as
and b using the extended Miner’s hypothesis, as shown in Eqs.
albedo and heat flux were collected to predict pavement tempera-
(16) and (17), which are regarded as the universal pavement crack-
ture internally, which were used to quantify the effect of tempera-
ing calibration equations when the effects of thermal stress and
ture on aging and viscoelastic properties of asphalt surface layer.
traffic load are obviously involved. The calibration equations can
be further expanded and applied to other distress types to include
3.1.3. Material properties and pavement structure
different load mechanisms such as the number of days due to ten-
For simplicity, a three-layer pavement system (asphalt, base
sile stress (Mode I) and shear stress (Modes II or III).
and subgrade layers) was selected in the performance analysis.
 
Relaxation modulus parameters (E1 and m-value) of the asphalt NTRi
surface layer, surface layer representative elastic modulus defined qi ¼ NTRi a0  a1 ð16Þ
NTHi
in [11], initial air void content, asphalt content, base and subgrade
layers moduli, thickness of asphalt layer and base layer were the  
1 N
inputs for the material properties and pavement structure charac- bi ¼ c0 þ c1 TRi ð17Þ
NTRi NTHi
teristics. Note that for other pavement structures with more than
three layers, Odemark’s method could be adopted to compute the where N TRi and N THi are the numbers of days for cracking with traffic
equivalent layer thickness, which was verified through the FEM in load and thermal stress to reach the critical depth, respectively; and
the reference [13]. a0 , a1 , c0 and c1 are calibration coefficients.
However, as stated in the previous section, only four pavement
3.2. Output module sections in the WF zone were determined to have the limited ‘‘pre-
fatigue lives” with the thermal effect, indicating that the thermal
The values of TDC initiation time t0 and ‘‘pre-fatigue life” for the stress was almost negligible to TDC performance. Therefore, two
TDC performance with a specific environment, material properties, simplified equations were proposed to estimate the values of q
110 M. Ling et al. / Construction and Building Materials 208 (2019) 102–112

and b for most pavement sections without substantial thermal bi ¼ c1 logðN TRi Þc2 ð19Þ
effect:
qi ¼ NTRi a1 þ a2 ð18Þ where a1 , a2 , c1 and c2 are calibration coefficients.
Table 3 summarizes all of the calibration coefficients for differ-
ent climatic zones. Note that due to the limited number of inde-
pendent LTPP pavement sections that have enough longitudinal
Table 3 wheelpath cracking data in the two freeze zones, the pavement
Summary of calibration coefficients for four climatic zones.
sections in the WF zone and DF zone were combined with the
Climatic Zone a1 a2 c1 c2 WNF zone and DNF zone as wet and dry zones, respectively. The
Wet 631. 04 2269.8 0.4201 1.097 computed ‘‘pre-fatigue lives” and corresponding values of q and
Dry 1617.6 1705.3 0.4201 1.097 b are depicted in Fig. 8. As seen, there was a fairly strong correla-
tion between ‘‘pre-fatigue life” and q, which was expected from
the fracture mechanics aspect. On the other hand, as the ‘‘pre-
20000
fatigue life” increased, b decreased, which indicates that for those
pavements with higher ‘‘pre-fatigue life”, the shapes of the distress
characteristic curves were normally flatter meaning a slower crack
15000 growth rate. As b increased gradually, the shape of the distress
(day)

R² = 0.841
curve changed from convex to linear and then to concave. It might
also possible to develop an empirical equation to estimate b as the
Calculated

10000 level two input without using the ‘‘pre-fatigue life”. For example,
as observed from the LTPP TDC performance data, as asphalt layer
thickness and modulus increased, b decreased, and as AADTT
5000 increased, b increased. However, some empirical equations have
been tried to correlate b with the aforementioned variables but
the R2 was always lower than that of Eq. (19).
0
0 5 10 15 20 25
Pre-fatigue life (month)
4.1. Parametric study of TDC designing program
(a) Prediction of in Wet Zone
A parametric study was conducted using the developed pro-
gram to assess the effects of different design variables on the
25000
TDC performance in the WNF zone. The selected influencing factors
included the traffic level (AADTT), E1 and m-value of the relaxation
20000 modulus master curve, asphalt layer thickness, and base layer
thickness. The levels of variables are listed in Table 4. Note that
(day)

R² = 0.896
the section to section variation of the m-value in Eq. (3) converted
15000
from dynamic modulus master curve was found to be insignificant
Calculated

compared to other variables. It is shown that the increase of AADTT


10000 and decrease of m-value reduced both of the initiation time and
‘‘pre-fatigue life” in Fig. 9(a) and (c). The increase of asphalt layer
5000 modulus shortened the crack initiation time but reduced the crack
propagation speed after reaching a specific crack length, as pre-
sented in Fig. 9(b). On the other hand, the increases of asphalt layer
0 thickness and base layer thickness in Fig. 9(d) and (e) were benefi-
0 2 4 6 8 10 12 14 16
cial to increase the ‘‘pre-fatigue life”. As expected, the effect of
Pre-fatigue life (month)
asphalt layer thickness was more notable to the TDC growth than
that of base layer thickness. The effects of volumetric properties
(b) Prediction of in Dry Zone
including air void content, binder content and aggregate gradation
3.5

3 Table 4
Sensitivity analysis of critical variables for TDC propagation.
2.5
Variables AADTT E1 (MPa) m-value Total Base
Asphalt Thickness
2
Calculated

Thickness (inch)
(inch)
1.5
R² = 0.5527 AADTT 1000 1000 0.20 4 6
1 500 1000 0.20 4 6
E1 800 1500 0.20 4 6
0.5 800 1000 0.20 4 6
Asphalt Layer 800 1000 0.20 6 6
0
Thickness 800 1000 0.20 3 6
0 0.2 0.4 0.6 0.8 1 1.2
Base Layer 800 1000 0.20 4 10
log (Pre-fatigue life)
Thickness 800 1000 0.20 4 6
(c) Prediction of m-value 800 600 0.19 4 6
800 600 0.20 4 6
Fig. 8. Predictions of q and b using ‘‘pre-fatigue life”.
M. Ling et al. / Construction and Building Materials 208 (2019) 102–112 111
Longitudinal Wheelpath Crack Length (m)

Longitudinal Wheelpath Crack Length (m)


70 70

60
60

50
50
40
40
30
30
20
High Traffic 20
10 Low Traffic High m-value
10
0 Low m-value
0 1000 2000 3000 4000 5000 6000 7000 8000 0
0 1000 2000 3000 4000 5000 6000
Time (day) Time (day)
(a) Effect of AADTT (c) Effect of m-value
Longitudinal Wheelpath Crack Length (m)

80 80

Longitunal Wheelpath Cracking Length (m)


70 70

60 60

50 50

40 40

30 30

20 High AC Modulus 20
High AC Thickness
10 Low AC Modulus 10
Low AC Thickness
0 0
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Time (day) Time (day)

(b) Effect of Asphalt Layer Modulus (d) Effect of Asphalt Layer Thickness
Longitudinal Wheelpath Crack Length (m)

70

60

50

40

30

20
High Base Thickness
10
Low Base Thickness
0
0 1000 2000 3000 4000 5000 6000
Time (day)
(e) Effect of Base Layer Thickness

Fig. 9. Sensitivity analysis of input variables.

were directly represented with fracture parameter n0 and discussed Pavement ME Design for further implementation. Parametric study
in Subsection 2.1, and not repeated here. was also conducted using the computer-based TDC performance
prediction program, which showed the effect of each factor on
5. Conclusions and future work the TDC performance. The following conclusions are drawn based
on the finding of this study:
A calibrated mechanics-based TDC model was presented in this
study. The model consisted of five sub-models including viscoelas-  A ME model for TDC of asphalt pavements was developed to
tic aging properties of asphalt field cores, field aging prediction of predict crack initiation time and ‘‘pre-fatigue life”, which was
asphalt pavements, TDC initiation, TDC propagation and longitudi- calibrated to field performance;
nal thermal cracking. Based on the development and assembly of  Fracture properties were divided into two parts including the
sub-models and calibration using the LTPP data, a window-based simultaneous growth of micro-cracks that leads to the initiation
C# computer program was developed to predict the TDC field per- of a single crack and the growth of that single macro-crack into
formance. This model is a potential candidate to be used in the the pavement structure;
112 M. Ling et al. / Construction and Building Materials 208 (2019) 102–112

 Non-uniform aging and long-term aging of asphalt field mix- [6] J. Zou, R. Roque, Top-down cracking: enhanced performance model and
improved understanding of mechanisms, J. Assoc. Asphalt Paving Technol.
tures and asphalt pavements were evaluated in this study to
(2011) 80.
effectively simulate the field aging characteristics; [7] M. Ling, Y. Chen, S. Hu, X. Luo, R.L. Lytton, Enhanced model for thermally
 A connection between the appearance and growth of low- induced transverse cracking of asphalt pavements, Constr. Build. Mater.
severity TDC and increased growth of alligator cracking was (2019), https://doi.org/10.1016/j.conbuildmat.2019.01.179.
[8] M. Ling, X. Luo, Y. Chen, F. Gu, R.L. Lytton, Mechanistic-empirical models for
identified, indicating that TDC is the precursor of fatigue top-down cracking initiation of asphalt pavements, Int. J. Pavement Eng.
cracking; (2018), https://doi.org/10.1080/10298436.2018.1489134.
 Transverse thermal stress played a minor role in the propaga- [9] T. Svasdisant, M. Schorsch, G.Y. Baladi, S. Pinyosunun, Mechanistic analysis of
top-down cracks in asphalt pavements, Transp. Res. Rec. 1809 (2002) 126–136.
tion of TDC compared to the traffic load effect. [10] R.L. Lytton, F.L. Tsai, S.-I. Lee, R. Luo, S. Hu, F. Zhou, Models for Predicting
Reflection Cracking of Hot-Mix Asphalt Overlays (Report 669), National
The load spectrum model in this study used a standard tire, Cooperative Highway Research Program, Washington, D.C., 2010.
[11] X. Luo, Y. Zhang, R.L. Lytton, Implementation of pseudo J-integral based Paris’
which should also be constructed using other tire types in the law for fatigue cracking in asphalt mixtures and pavements, Mater. Struct. 49
future. Continued collection of the field data from the LTPP pro- (9) (2016) 3713–3732.
gram or other data sources is highly desirable. The calibration [12] M.M.J. Jacobs, P.C. Hopman, A.A.A. Molenaar, Application of fracture mechanics
in principles to analyze cracking in asphalt concrete, Electron. J. Assoc. Asphalt
models used in this study would be materially improved with a Paving Technol. 65 (1996) 1–39.
greater number of data points on the pavement sections used. [13] F. Zhou, S. Hu, X. Hu, T. Scullion, M. Mikhail, L.F. Walubita, Development,
calibration, and verification of a new mechanistic-empirical reflective cracking
model for HMA overlay thickness design and analysis, J. Transp. Eng. 136 (4)
Conflict of interest (2010) 353–369.
[14] Y. Zhang, F. Gu, B. Birgisson, R.L. Lytton, Modelling cracking damage of asphalt
mixtures under compressive monotonic and repeated loads using pseudo J-
None. integral Paris’ law, Road Mater. Pavement Des. 19 (3) (2018) 525–535.
[15] R.A. Schapery, A Theory of Crack Growth in Visco-Elastic Media (Rep. No. MM
2764-73-1), Mechanics and Materials Research Center, Texas A&M University,
Acknowledgement College Station, Texas, 1973.
[16] R.L. Lytton, J. Uzan, E.G. Fernando, R. Roque, D. Hiltunen, S.M. Stoffels,
Development and Validation of Performance Prediction Models and
The authors gratefully acknowledge the funding support of Specifications for Asphalt Binders and Paving Mixes, Strategic Highway
NCHRP Project 01-52: A Mechanistic–Empirical Model for Top- Research Program, Washington, DC, 1993.
[17] E.G. Fernando, D. Musani, D.-W. Park, W. Liu, Evaluation of Effects of Tire Size
Down Cracking of Asphalt Pavement Layers. and Inflation Pressure on Tire Contact Stresses and Pavement Response (Rep.
No. FHWA/TX-06/0-4361-1), Texas Transportation Institute, College Station,
Texas, 2006.
References [18] USDOT, 2016. Traffic Monitoring Guide., Washington, DC.
[19] M. Ling, X. Luo, F. Gu, R.L. Lytton, An inverse approach to determine complex
[1] ARA, Inc, Guide for Mechanistic-Empirical Design of New and Rehabilitated modulus gradient of field-aged asphalt mixtures, Mater. Struct. 50 (2) (2017)
Pavement Structures (Final Report, NCHRP Project 1-37A), Transportation 138.
Research Board, National Research Council, Washington, DC, 2004. [20] M. Ling, X. Luo, F. Gu, R.L. Lytton, Time-temperature-aging-depth shift
[2] R. Roque, J. Zou, Y.R. Kim, C.M. Baek, S. Thirunavukkarasu, B.S. Underwood, M. functions for dynamic modulus master curves of asphalt mixtures, Constr.
N. Guddati, Top-Down Cracking of Hot-Mix Asphalt Layers: Models for Build. Mater. 157 (2017) 943–951.
Initiation and Propagation (Final Report, NCHRP 1-42A), National [21] X. Luo, F. Gu, R.L. Lytton, Kinetics-based aging prediction of asphalt mixtures
Cooperative Highway Research Program, Washington, DC, 2010. using field deflection data, Int. J. Pavement Eng. (2017) 1–11.
[3] H. Wang, I.L. Al-Qadi, Near-surface pavement failure under multiaxial stress [22] J. Bari, M.W. Witczak, Development of a new revised version of the Witczak E*
state in thick asphalt pavement, Transp. Res. Rec. 2154 (1) (2010) 91–99. predictive model for hot mix asphalt mixtures, J. Assoc. Asphalt Paving
[4] M. De Beer, C. Fisher, F.J. Jooste, Determination of pneumatic tyre/pavement Technol. 75 (2006).
interface contact stresses under moving loads and some effects on pavement [23] M.W. Kumara, M. Gunaratne, J.J. Lu, B. Dietrich, (2014) Methodology for
with thin asphalt surfacing layers Seattle, in: Proceedings of the 8th random surface-initiated crack growth prediction in asphalt pavements, J.
International Conference on Asphalt Pavements, 1997, pp. 179–227. Mater. Civ. Eng. 16 (2) (2004) 175–185.
[5] M. Ling, X. Luo, S. Hu, F. Gu, R.L. Lytton, Numerical modeling and artificial [24] D.R. Hiltunen, R. Roque, A mechanics-based prediction model for thermal
neural network for predicting J-Integral of top-down cracking in asphalt cracking of asphaltic concrete pavements (with discussion), J. Assoc. Asphalt
pavement, Transp. Res. Rec. 2631 (2017) 83–95. Paving Technol. (1994) 63.

S-ar putea să vă placă și