Professional Documents
Culture Documents
net/publication/266387629
CITATIONS READS
11 254
2 authors:
All content following this page was uploaded by Haifang Wen on 17 November 2014.
Abstract: Bottom-up fatigue cracking is one of the major distresses for asphalt pavements. Accurate prediction of fatigue cracking for
Downloaded from ascelibrary.org by WA STATE UNIV LIBRARIES on 11/17/14. Copyright ASCE. For personal use only; all rights reserved.
asphalt pavement is of paramount importance for a cost-effective pavement design. A fatigue model based on mechanistic-empirical pave-
ment design is modified from an Asphalt Institute model. However, there are some controversies about the effectiveness of the mechanistic-
empirical pavement design fatigue model. The major concern exists on the use of dynamic modulus as a key parameter and there is no damage
property of asphaltic mix to predict fatigue, which is induced by damage to the material. This study developed a damage-based phenom-
enological fatigue model. The pavements at the Federal Highway Administration’s Accelerated Loading Facility (ALF) were used to test the
effectiveness of existing models, including the mechanistic-empirical pavement design fatigue model, and validity of the damage-based
fatigue model. The data used in this study included dynamic modulus, critical strain-energy density of hot-mix asphalt (HMA), tensile strain
at the bottom of HMA layer, and the fatigue life of ALF pavements. It was found that the damage-based model significantly improved the
accuracy of the prediction, when compared with the mechanistic-empirical pavement design fatigue model and other conventional models.
DOI: 10.1061/(ASCE)MT.1943-5533.0000573. © 2013 American Society of Civil Engineers.
CE Database subject headings: Fatigue; Strain; Asphalt pavements; Damage; Cracking.
Author keywords: Fatigue; Critical strain-energy density; Dynamic modulus; Tensile strain; Fatigue model.
the bottom-up fatigue-cracking model of asphalt pavement in bottom of asphalt layer (Roque and Buttlar 1992), as illustrated in
the MEPDG (Muthadi and Kim 2008; Ker et al. 2007; Gramajo Fig. 2. Considering that HMA is an anisotropic material, selecting a
et al. 2007). The accuracy of prediction is found not to be as good test mode that simulates the field conditions will produce results
as other performance models, such as rutting or thermal cracking. that are more reliable.
Observation of the MEPDG fatigue model indicates that when two Another important factor is the thickness of the HMA layer. The
asphalt pavements have same tensile strain at the bottom of HMA Asphalt Institute model does not include HMA layer thickness.
layer, dynamic modulus of HMA controls the fatigue performance However, for pavements with the same tensile strain at the bottom
of asphalt pavements. Dynamic modulus is measured within the of HMA layer and dynamic modulus of HMA, the pavement with
linear viscoelastic range of HMA. However, fatigue occurs as a thicker HMA layer will have longer fatigue life, as it takes more
result of damage by repeated stresses by traffic loads, which is traffic repetitions for cracks to propagate through the HMA layer
in the damage domain of HMA. Currently, the fatigue model does and appear in the surface, as illustrated in Fig. 3. In Fig. 3, the pave-
not include a damage property of HMA. There is a need to assess ment structure on the left has thicker or stronger base, but thinner
current MPEDG and possibly develop a new fatigue model for HMA layer than those of the pavement structure on the right.
bottom-up fatigue cracking. Assuming that the same HMA is used, the tensile strain at the
In addition to these mechanistic-empirical models, many bottom of HMA layer could be the same for both structures,
researchers worked on advanced modeling to characterize namely, crack initiation and propagation rate. However, the pave-
asphalt—pavement fatigue. In a previous study, the viscoelasticity ment on the right will carry more traffic loads than the left one to
and continuum damage (VECD) constitutive model was success- reach the same level of cracking, due to the longer path for crack to
fully applied to characterize the fatigue behavior of asphalt appear in the surface of pavement. The MEPDG fatigue model
concrete in uniaxial tensile testing mode (Kim et al. 1997). In includes HMA layer thickness, but in the formula of k1 .
the VECD theory, pseudostrain is an essential parameter for
applying Schapery’s correspondence principle to the hysteretic
stress-strain behavior of asphalt concrete (Schapery 1984). The
use of pseudostrain essentially accounts for the viscoelasticity of
the material and allows for the separate characterization of damage
within the specimen. This approach features fundamental charac-
terization and has been demonstrated to be effective in predicting
the fatigue behavior of asphaltic mixtures in the laboratory. The
HMA fracture model developed by Roque et al. (2004) describes
discontinuous crack growth by increasing crack length. The crack
will advance if the accumulated dissipated creep-strain energy
exceeds the dissipated creep-strain energy limit of the mixture.
The dissipated creep-strain energy is obtained from the indirect
Fig. 1. Schematic diagram of critical strain-energy density
tensile (IDT) test. In a previous study, the concept of dissipated
pseudostrain energy (DPSE) was used, with the DPSE being bro-
ken down into three parts to consider the change in phase angle
between load cycles, change in the phase angle within a load cycle,
and change in stiffness between load cycles (Branco et al. 2008).
However, these advanced mechanistic models are yet to be
validated with field performance. Currently, mechanistic-empirical
models are still needed to predict field fatigue performance of an
asphalt pavement. This study describes the development of a new
mechanistic-empirical fatigue model.
Materials
Six types of asphalt binders were used at ALF, including PG70-22
Two new damage-based fatigue models are proposed to account as control, air blown, linearly grafted styrene-butadiene-styrene
for fracture property, CSED in this case, and the HMA layer thick- elastomeric polymer (SBS-LG), crumb rubber-TB (CR-TB), ter-
ness as a stand-alone parameter, as shown in Eqs. (7) and (8), as polymer, and fiber. The same aggregates and binder content were
follows: b used. Coarse-graded diabase with a nominal maximum aggregate
1 size of 12.5 mm was mixed with different binders to produce the
Nf ¼ a ðCSEDÞd he ð7Þ
εt mixtures. The binder content was 5.3%. Detailed description about
b c the HMA material used, such as aggregate gradation, could be
1 1 found elsewhere (Qi et al. 2008).
Nf ¼ a ðCSEDÞd he ð8Þ
εt E
and then become interconnected (Qi et al. 2008). Therefore, the and modulus of HMA and base varied significantly. As a result,
transverse strains were used in this study. the tensile strains for different pavement sections differed signifi-
cantly from each other, ranging from 475 to 1,524 microstrains.
Field Performance
CSED from IDT Strength Tests
For fatigue cracking, cracks were manually recorded as they ap-
peared at the surface of the pavements. The percentage of the area The CSEDs were determined from the IDT strength tests for differ-
cracked in the loaded area was calculated (Qi et al. 2008). The ent HMAs. Past studies have shown that CSED is an indicator to
pavement sections exhibited different degree of fatigue distresses fatigue cracking in the field. Higher CSED indicates higher fatigue
at the end of trafficking, ranging from 0% to 100%. For the purpose resistance. It was found that HMA with SBS-LG had the highest
of modeling, the numbers of loads corresponding to 3% fatigue CSED, followed by control PG70-22, fiber, CR-TB, terpolymer,
cracking were used in this study. This is because 3% was the mini- and air blown, as shown in Fig. 5.
mum fatigue cracking for the pavement sections that exhibited dis-
tresses and more pavement sections can be included in the analysis. Field Performance of ALF Pavements
The fatigue-cracking growth curves at the ALF sites did not cross
each other for different sections. Therefore, the performance rank- The fatigue performance of ALF pavement varied significantly.
ing of these sections is independent of the percentage of cracking The control 150 mm_PG70-22 pavement showed the best perfor-
selected to use. mance, in terms of number of passes needed to reach 3% fatigue
Because of the availability of field performance and material cracking, followed by 100 mm_fiber, 100 mm_SBS-LG,
testing, only six sections had complete data set of material proper- 150 mm_air blown, 100 mm_terpolymer, and 100 mm_crumb rub-
ties and field performance, as shown in Table 2. These six sections ber, as shown in Fig. 5.
were used to study the fatigue models.
Evaluation of the Fatigue Model
Experimental Results Various existing and proposed fatigue models were evaluated based
on the ALF pavement performance and material properties. It is
expected that for well-controlled test sections, such as ALF
Uniaxial Dynamic Modulus pavements, an effective fatigue model would be able to predict
The dynamic moduli of HMA are shown in Table 2. The dynamic or correlate well with the pavement performance. Because only
modulus from the uniaxial test is a parameter needed in the the ALF pavement performance was used in this study, the primary
purpose was to test the effectiveness of the form of model, instead
Table 1. Construction Quality Data of evaluating the accuracy of prediction by these models. The
validation of these models can only be performed when data sets
Air Effective binder
other than those of ALF are used.
Materials PG grade voids (%) content by volume (%)
Lane 4/SBS-LG 70-23 5.9 10.9
Lane 5/CR-TB 74-28 6.6 11.1 Strain Model
Lane 6/terpolymer 74-28 6.4 10.3 The strain model was evaluated, based on the tensile strain at the
Lane 7/fiber 79-28 7.2 11 bottom of HMA layer from the insrumentation. The k1 and k2
Lane 8/PG70-22 74-31 5.1 11.6
parameters were obtained from nonlinear regression. The regressed
Lane 10/air blown 73-23 7.1 9.6
strain model for ALF pavement is shown in Eq. (9)
No of Passes to 3% Fatigue
4E+05 y = 0.8781x
Cracking (E Model)
4E+05 R² = 0.4861
3E+05
3E+05
2E+05
2E+05
1E+05
5E+04
0E+00
0E+00 1E+05 2E+05 3E+05 4E+05 5E+05
Downloaded from ascelibrary.org by WA STATE UNIV LIBRARIES on 11/17/14. Copyright ASCE. For personal use only; all rights reserved.
Fig. 5. Number of passes for 3% fatigue cracking and CSED Fig. 7. Effectiveness of the E model
1.2857
1 dynamic modulus of HMA is measured from the laboratory-
N f ¼ 8 × 108 ð9Þ
εt fabricated specimen. These two parameters can only be obtained
after the construction is completed. The regressed MEPDG model
The regressed ALF pavement performance, in terms of number for ALF pavements is shown in Eqs. (11)–(14):
of passes to 3% fatigue cracking is compared with the measured −0.0879
1
number of passes in the field, as shown in Fig. 6. It could be seen N f ¼ Ck1 ðEÞ1.1196 ð11Þ
that, although reasonable, the data are very scattered, with an R2 of εt
0.1 and the correlation is poor for such well-controlled pavements.
0.003602
k1 ¼ 1 0.000398 þ ð12Þ
E Model 1 þ e11.02−3.49hac
The E model includes dynamic modulus of HMA. The dynamic
moduli of HMA and the tensile strain were used to perform the C ¼ 10M ð13Þ
nonlinear regression to determine the values of k1 , k2 , and k3 .
The regressed E model for ALF pavements is shown in Eq. (10).
Vb
The regressed and measured fatigue lives are plotted in Fig. 7. The M ¼ 4.84 − 0.69 ð14Þ
E model improved the correlation, with an R2 of 0.48, when Va þ Vb
compared with the strain model.
As shown in Fig. 8, the MEPDG model further improved the
−0.407
1 correlation between the measured and regressed pavement perfor-
N f ¼ 2.56 ðEÞ1.582 ð10Þ mance. It indicates that the MEPDG model is a more effec-
εt
tive fatigue model, when compared with the strain model and
where E = modulus of material (MPa). the E model. However, the effectiveness of such a model is still
questioned for such well-controlled pavements. In other words,
MEPDG Model the MEPDG model failed to rank the field fatigue performance
of some of the mixtures, as shown in Fig. 8.
A nonlinear regression was conducted to obtain the k2 and k3 val-
ues in the MEPDG fatigue model, based on the tensile strain,
dynamic modulus, HMA thickness, in situ air void, and effective CSED Model
binder content. It is noted that the air void and effective binder The tensile strain, CSED from IDT, and thickness of HMA were
content are those of in situ HMA and used to account for the used to develop the CSED model. The dynamic modulus of
material variation from production and construction, as the
5E+05
5E+05
Regressed No. of Passes to 3%
5E+05
Fatigue (MEPDG Model)
5E+05
No of Passes to 3% Cracking
4E+05
4E+05
4E+05 y = 0.9577x
4E+05
(Strain Model)
R² = 0.8728
3E+05
3E+05
3E+05 3E+05
2E+05 2E+05
2E+05 2E+05
1E+05 1E+05
5E+04 5E+04
0E+00 0E+00
0E+00 1E+05 2E+05 3E+05 4E+05 5E+05 0E+00 1E+05 2E+05 3E+05 4E+05 5E+05
Field No of Passes to 3% Cracking Field No of passes for 3% Cracking
Fig. 6. Effectiveness of the strain model Fig. 8. Effectiveness of the MEPDG model
No of Passes to 3% Cracking
HMA is not included in this model. The HMA thickness is included
as a stand-alone parameter, instead of a built-in parameter in k1 4E+05 y = 0.9899x
(CSED+E Model)
of the MEPDG model. The construction quality data, such as in R² = 0.9714
situ air void and effective binder content in the MEPDG model 3E+05
were not used in this CSED model. This is because the CSED
2E+05
of HMA were determined using field cores of HMA. The variation
of air void and effective binder content have been accounted for by 1E+05
the CSED. The developed CSED model for ALF pavements is
shown in Eq. (15): 0E+00
0E+00 1E+05 2E+05 3E+05 4E+05 5E+05
1.758
1 Field No of Passes to 3% Cracking
N f ¼ 31020.05 ðCSEDÞ1.030 h−0.24 ð15Þ
εt Fig. 10. Effectiveness of the CSED+E model
Downloaded from ascelibrary.org by WA STATE UNIV LIBRARIES on 11/17/14. Copyright ASCE. For personal use only; all rights reserved.
y = 1.0004x
4E+05 R² = 0.9942 HMA layer thickness and fatigue life is not expected.
4E+05 • Combining CSED and dynamic modulus in the fatigue model
(CSED Model)