Professional Documents
Culture Documents
Abstract: Flexible pavements are multilayer structures, typically with a viscoelastic asphalt layer followed by nonlinear unbound/bound
layers. Conventionally, multilayered elastic analysis is performed to obtain the response of flexible pavements for design and inverse analyses;
however, assuming asphalt pavement to be a linear elastic material is an oversimplification of its actual behavior. It is well known that the
responses of asphalt pavements are both rate and temperature dependent. In the present work, a computationally efficient model has been
developed to analyze flexible pavements, considering the top layer of linear viscoelastic asphalt concrete (AC), followed by a stress-
dependent (nonlinear) base layer, and an elastic subgrade. Constitutive equations are formulated for layered viscoelastic–nonlinear axisym-
metric systems. It is shown that the developed model can be used to simulate pavement response under stationary or transient loading. A
comparison between the model responses and results obtained using a finite-element (FE) model shows that the model could be used to
simulate both deflections and stress responses in multilayer viscoelastic nonlinear structures. The primary advantage of the model, as opposed
to FE models, is its computational efficiency. This makes it viable for use in backcalculation algorithms. DOI: 10.1061/(ASCE)EM.1943-
7889.0001095. © 2016 American Society of Civil Engineers.
Author keywords: Granular base; Pavement response; Finite-element method; Falling weight deflectometer; Asphalt pavement.
i¼0 i¼0
K s ðσi Þ modulus; and Dðt; σÞ = stress-dependent creep compliance. Typi-
cally, in many nonlinear materials, the shape of the relaxation
modulus of the material is preserved, even though the material
X
k X
k
uz ðtÞ ¼ exhibits stress or strain dependency (Shames and Cozzarelli 1997;
H−z ðt − τ i ÞΔσ þ
uve Δuin ð15Þ
i
i¼1 i¼0 Nekouzadeh and Genin 2013). Such NLV problems are solved by
assuming that time dependence and stress (or strain) dependence
Z Xk
can be decomposed into two functions:
t dσ
uz ðtÞ ¼ H−z ðt − τ Þ
uve dτ þ Δuin ð16Þ Dðt; σÞ ¼ hðσÞDt ðtÞ ð19Þ
τ ¼0 dτ i¼0
τ ¼0
g(σ )
0.8
component, dfðεðτ ÞÞ=dε, and the viscoelasticity derives from Et . 34 kPa
A direct extension of the concepts of the QLV model to develop 0.75 69 kPa
103 kPa
formulations for a viscoelastic nonlinear multilayer system, where 0.7 138 kPa
172 kPa
the unbound layer is nonlinear and the AC layer is linear viscoelas- 241 kPa
0.65 345 kPa
tic, leads to 483 kPa
Z t 0.6
965 kPa
R dfðεðτ ÞÞ dεðτ Þ 3 4 5 6 7 8
σðtÞ ¼ Et ðx; y; z; tR − τ Þ dτ ð22Þ 1 x 10 1 x 10 1 x 10 1 x 10 1 x 10 1 x 10
τ ¼0 dε dτ Relaxation modulus E(t) kPa
where Et ðx; y; z; tR Þ = relaxation function at location ðx; y; zÞ; and Fig. 4. Variation of gðσÞ with EðtÞ of AC layer at various stress load-
fðεðτ ÞÞ = a function of strain, εðτ Þ. Alternatively, to obtain vertical ings, illustrating dependency of gðσÞ on both loading stress and time
surface deflection in pavements, Eq. (22) can be expressed in terms
of a vertical deflection response to Heaviside step loading as
follows:
Z t Fig. 4 also illustrates that gðσÞ varies with different values of EðtÞ.
R This means that gðσÞ is not based solely on the stress, and as a
uve
z ðtÞ ¼ ueH−t ðtR − τ ; σ ¼ 1ÞgðσÞdσðτ Þ ð23Þ
τ ¼0 result, Fung’s model cannot be used in a layered pavement struc-
ture. This is meaningful since the change in the stress distribution
where uvez ðtÞ = surface (NLV) displacement; uH−t ðt; σ ¼ 1Þ = unit
e
within the pavement layers due to the viscoelastic effect [as EðtÞ
nonlinear elastic response due to a unit stress; and gðσÞ = a function varies] will impose changes in the behavior of the stress-dependent
of stress, which can be expressed as granular layer.
ueH ðtR ; σÞ
gðσÞ ¼ ð24Þ
uH−t ðtR ; σ ¼
e
1Þ Proposed Model: LAVAN
Boltzmann’s superposition integrals [e.g., Eq. (1)] assume that the
where ueH ðtR ; σÞ = nonlinear elastic unit displacement due to a
system is linear, i.e., it is not stress-dependent. This assumption
given stress ðσÞ. The LAVA algorithm was modified to implement
may be violated to approximate the nonlinearity of the system if
an iterative nonlinear solution for the granular base, which was as-
stress-dependent relaxation functions (i.e., the stress-dependent
sumed to follow Eq. (7). The ueH ðtR ; σÞ in Eq. (24) was calculated in
unit response functions) of the multilayered structure under many
a range of stress values from 0.7 to 965.3 kPa (0.1 to 140 psi) and
stress levels are precomputed and used in the modified superposi-
using EðtÞ values (for AC) in a range of times, ranging from 10−8 to
tion integral as follows:
108 s. Then, ueH−t ðtR ; σ ¼ 1Þ was calculated for unit stress. Sub-
Z t
sequently, using Eq. (24), gðσÞ values were computed for different R dIðτ Þ
stress levels and E(t) values. For Fung’s theory [i.e., Eqs. (22)–(24)] Rve ðx; y; z; tÞ ¼ ReH ½x; y; z; Iðτ Þ; tR − τ dτ ð25Þ
τ ¼0 dτ
to hold, gðσÞ must be purely a function of stress and not a function
of EðtÞ. The gðσÞ values were computed for the pavement section where Rve ðx; y; z; tÞ = NLV response of layered pavement struc-
shown in Fig. 3 and plotted in Fig. 4. Fig. 4 shows the variation in ture; ReH ½x; y; z; Iðτ Þ; tR − τ = unit response function that is a func-
gðσÞ, where the gðσÞ values decrease with increasing stress ðσÞ. tion of both time and input, Iðτ Þ, which in this study is the stress
This is expected behavior for a nonlinear material since, as the applied at the surface of the pavement. The reader is cautioned that
stress increases, the unbound layer moduli will increase. However, this approach is an approximation and usually can be accomplished
where uvez ðr; z; tÞ = vertical deflection at time t and location ðr; zÞ;
5
10
and ueH−z ðr; z; σðτ Þ; tR − τ Þ ¼ uez ½r; z; σðτ Þ; tR − τ =σðτ Þ, where
uez is the nonlinear response of the pavement at a loading stress
Downloaded from ascelibrary.org by Indian Institute of Technology, Tirupati on 03/01/23. Copyright ASCE. For personal use only; all rights reserved.
Note: FB = finite boundary; IB = infinite boundary. Deflections are in micrometers (i.e., 10−6 m).
domain size is subsequently analyzed with all the vertical as well deflections predicted by LAVAN were found to lie between the
as bottom boundaries supported using infinite elements. For the FE results predicted by the two boundary conditions.
selected domain size, a FE mesh refinement of 10 mm in the Satisfactory performance of the model would require predicting
AC layer and 25 mm in the base layer is used. The model response a comparable time response by the model. The surface deflection
under transient loading is compared with results obtained from the history computed by LAVAN and Abaqus (analysis with finite
general-purpose FE software Abaqus. For this purpose, haversine boundaries) for the Control and CRTB mixes are plotted in Figs. 6
loading in an axisymmetric setup is used. Abaqus consumed ap- and 7, respectively. In the figures, the curves labeled AS show re-
proximately 17 min in analyzing a haversine loading of sults obtained using Abaqus, whereas the curves labeled LS show
951.5 kPa (138 psi) and 35 ms, whereas LAVAN could generate results obtained using LAVAN. A comparable response is visible in
the results in 3.6 min on the same desktop computer. the figures, which has been further quantified using Eqs. (31) and
The surface deflection obtained using FE for the two boundary (32). As expected, under the same geometric and loading condi-
conditions, (1) finite boundaries: roller support on the vertical tions, the stiffer Control mix generated lower deflections compared
boundaries and fixed support on the bottom and (2) infinite ele- to the softer CRTB mix. Fig. 6(a) shows the results when stress at
ments at boundaries, are compared with LAVAN in Table 1. It r ¼ 0 is used in LAVAN for nonlinearity computation, and, for
should be noted that both boundary conditions do not strictly re- comparison purposes, Fig. 6(b) shows the results when stress at r ¼
present the semi-infinite geometry of the problem. In Abaqus the 3.5a is used in LAVAN. Note that S1, S2, S3, S4, S5, S6, S7, and
solution in the infinite element is considered to be linear, which is S8 in the figures correspond to surface deflection at Sensor-1
assumed to match the material properties of the adjacent finite through Sensor-8 spaced at 20.3, 30.5, 45.7, 61.0, 91.4, 121.9,
element. Hence the infinite elements provide stiffness to the boun- and 152.4 cm (0, 8, 12, 18, 24, 36, 48, and 60 in.) from the
dary assuming the deflection at r ¼ ∞ to be zero. It can be seen centerline of the load. In general a better match for the deflection
from Table 1 that, for both the Control mix and the CRTB mix, the basin between the FE and LAVAN results can be found when a
surface deflection predicted by the FE method using finite boun- stress state at r ¼ 3.5a is used while incorporating nonlinearity.
daries was higher compared to results when infinite elements It is worth noting that, for the structure in Fig. 3, Huang’s
were used at the boundaries. Further, for both mixes the surface (2004) procedure for the location of stress used in the
1000 1000
AS1* AS1*
AS2 AS2
AS3 AS3
AS4 AS4
800 AS5 800 AS5
AS6 AS6
AS7 AS7
Surface deflection (μm)
AS8 AS8
LS1* LS1*
600 LS2 600 LS2
LS3 LS3
LS4 LS4
LS5 LS5
LS6 LS6
400 LS7 400 LS7
LS8 LS8
200 200
0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
(a) Time (ms) (b) Time (ms)
Fig. 6. Surface deflection comparison of Abaqus and LAVAN for Control mix (*AS1= Abaqus Sensor 1; *LS1 = LAVAN Sensor 1)
LS8 LS8
400 400
200 200
0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
(a) Time (ms) (b) Time (ms)
Fig. 7. Surface deflection comparison of Abaqus and LAVAN for CRTB mix (*AS1 = Abaqus Sensor 1; *LS1 = LAVAN Sensor 1)
nonlinear elastic analysis leads to r ¼ 2.8a, when a trapezoidal Next, the stress states ðσzz ; σrr ; σzr Þ obtained from LAVAN were
stress distribution with a 0.5 horizontal slope and 1 vertical slope compared to those computed by Abaqus. Figs. 8–10 show a com-
is assumed. It can be observed from Figs. 6 and 7 that the proposed parison between the flexible pavement stress state obtained for the
modified Boltzmann superposition works well close to the loading. CRTB mix using LAVAN and Abaqus at t ¼ 0.0175 s (for 35 ms
However, response prediction by the model gets less accurate as we haversine loading). It can be seen from the figures that the two sets
move away from the load. of contours closely follow each other. Furthermore, the difference
The difference between the FE results and LAVAN was quan- between the modulus predicted by the two models in the middle of
tified using the following two variables: the base layer under the center of the loading is less than 3.5%. It
can also be seen that the results obtained from LAVAN were more
jupeak peak
ABAQUS − uLAVAN j continuous (fewer breaks in the curves) compared to the results
PEpeak ¼ 100 ð31Þ
upeak
ABAQUS
obtained using Abaqus. This is due to the fact that the FE-based
solution poses difficulties at locations where there is an abrupt
−690
−6
−897 −897
90
−483 −6.9 −483 −6.9
10 10
−138 −138
−55
.2 −48.3
20 20
−4
4.
−4 85
8.3
30 30
−4 −4
4.8 1.4
z (cm)
5
z (cm)
−4
40 1.4 40
−3
4.5
−13.8
−13.8
− 20
50 −3 50
4.5
−20
.7
−2
.7
7.
−2
6
60 60 7.
6
Downloaded from ascelibrary.org by Indian Institute of Technology, Tirupati on 03/01/23. Copyright ASCE. For personal use only; all rights reserved.
70 70
80 80
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) r (cm) (b) r (cm)
Fig. 8. Vertical stress distribution (σzz , kPa) in pavement: (a) LAVAN; (b) Abaqus
.9
.9
8 9
−6 4.5
4.
−6
5
10 −34 10 −34
−6 .5 − .5 −69
−138
9 1 38
−6.9 −13.8
20 20
30 30
−13
z (cm)
z (cm)
.8
40 40 −13.8
50 50
−6.
9
60 −6.9 60
70 70
80 80
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) r (cm) (b) r (cm)
Fig. 9. Radial stress distribution (σrr , kPa) in pavement: (a) LAVAN; (b) Abaqus
0 0
7
3
20
.6
.7
276
48
276
20
27
483 34 34 69 20
10 5 138 69 34.5 10 5 138 27.6 .7
13.8 207 34.5
5.52 13.8
5.52
20 20
30 30
5.52
5.52
z (cm)
z (cm)
40 40 5.52
5.52
50 50
60 60 52
2
5.
5.5
70 70
80 80
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) r (cm) (b) r (cm)
Fig. 10. Shear stress distribution (σrz , kPa) in pavement: (a) LAVAN; (b) Abaqus