Construction and Building Materials 122 (2016) 214230

Contents lists available at ScienceDirect

Construction and Building Materials

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

Numerical modeling of geogrid-reinforced flexible pavement and

corresponding validation using large-scale tank test
Fan Gu a,, Xue Luo b, Rong Luo c, Robert L. Lytton d, Elie Y. Hajj e, Raj V. Siddharthan f
a
Texas A&M Transportation Institute, Texas A&M University System, 3135 TAMU, CE/TTI Bldg. 501A, College Station, Texas 77843, United States
b
Texas A&M Transportation Institute, Texas A&M University System, 3135 TAMU, CE/TTI Bldg. 508B, College Station, Texas 77843, United States
c
Texas A&M Transportation Institute, Texas A&M University System, 3135 TAMU, CE/TTI Bldg. 503C, College Station, Texas 77843, United States
d
Zachry Department of Civil Engineering, Texas A&M University, 3136 TAMU, CE/TTI Bldg. 503A, College Station, Texas 77843, United States
e
Department of Civil and Environmental Engineering, University of Nevada Reno, 1664 N. Virginia St. MS257, Reno, NV 89557, United States
f
Department of Civil and Environmental Engineering, University of Nevada Reno, 1664 N. Virginia St. MS258, Reno, NV 89557, United States

h i g h l i g h t s

Geogrid-reinforced pavement structure is modeled using the finite element software ABAQUS.

An analytical model is developed to predict the cross-anisotropic resilient modulus of geogrid-reinforced granular material.

A user-defined material subroutine is programmed to simulate the nonlinear cross-anisotropic behavior of geogrid-reinforced granular material.

The influence of geogrid on pavement performance is quantified using the finite element results.

A comprehensive large-scale tank test program is designed to validate the developed geogrid-reinforced pavement model.

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

Article history: This study aimed to develop a finite element model to simulate the geogrid-reinforced flexible pavement
Received 22 March 2016 structure by taking into account the lateral confinement effect of geogrid layer, the interaction between
Received in revised form 7 June 2016 geogrid and aggregate/soil, and the nonlinear cross-anisotropy of geogrid-reinforced unbound granular
Accepted 14 June 2016
material (UGM). First, an analytical model was proposed to quantify the effect of the lateral confinement
Available online 9 July 2016
of geogrid layer on the resilient modulus of UGM. By comparing to the laboratory triaxial test results, the
developed analytical model was proven to accurately predict the resilient modulus of geogrid-reinforced
Keywords:
UGM. Second, the Goodman interface element model was used to characterize the contact behavior of
Geogrid-reinforced flexible pavement
Finite element model
geogrid-aggregate/soil interface. In order to simulate the nonlinear cross-anisotropic behavior of
Cross-anisotropy geogrid-reinforced UGM, a user-defined material (UMAT) subroutine was programmed using the secant
Large-scale tank test modulus approach. The accuracy of the developed UMAT was verified by comparing the numerical sim-
ulation results to the analytical solutions in a virtual triaxial test.
Two pairs of geogrid-reinforced and unreinforced pavement models were analyzed in this study. It was
found that the geogrid reinforcement is effective in mitigating the rutting damage of base course and sub-
grade, but cannot significantly extend the fatigue life of flexible pavement. The geogrid reinforced in the
middle of the base course is better at reducing the rutting damage of base course than that placed at the
base/subgrade interface. However, the geogrid reinforcement is much more effective in reducing the rut-
ting damage of the subgrade when it is placed at the bottom of the base course. A comprehensive large-
scale tank (LST) testing program was designed to record the critical pavement responses, including the
surface deflection, the tensile strain at the bottom of the asphalt concrete, and the vertical stresses in base
course and subgrade. The developed geogrid-reinforced and unreinforced finite element models were
finally validated by comparing the model predictions with those measurements from the LST test.
2016 Elsevier Ltd. All rights reserved.

Corresponding author.
E-mail addresses: tracygufan@tamu.edu (F. Gu), xueluo@tamu.edu (X. Luo),
rongluo@tamu.edu (R. Luo), r-lytton@civil.tamu.edu (R.L. Lytton), elieh@unr.edu
(E.Y. Hajj), siddhart@unr.edu (R.V. Siddharthan).
1. Introduction
Geogrids are commonly used in unbound aggregates as a means
of enhancing the performance of flexible base layer or the railroad
ballast layer [1,2]. Many studies introduced tests performed on

http://dx.doi.org/10.1016/j.conbuildmat.2016.06.081
0950-0618/ 2016 Elsevier Ltd. All rights reserved.
 F. Gu et al. / Construction and Building Materials 122 (2016) 214230 215

large-scale or in-service geogrid-reinforced pavement sections shear stiffness ks will be determined by Eq. (2) using the pullout test
[3,4]. The test results indicated that geogrids are effective in data.
improving the stiffness and stability of the reinforced pavement
structures and reducing the accumulated permanent deformation. DP
ks 2
Beneficial effects of the geogrid layer have been identified on the 2l  Dur
responses of pavements under the traffic loading through two
major mechanisms [5]: where DP is the incremental applied pullout force, l is the embed-
ded length of geogrid, and Dur is the incremental relative displace-
a) Lateral confinement, which is produced by the interface fric- ment. To simulate the lateral confinement effect, Kwon et al. [13]
tional interaction and interlocking between the base course developed a geogrid-reinforced pavement model by empirically
aggregates and the geogrid layer. Significant tensile stress is assigning the additional confining stresses around the geogrid layer.
generated in the geogrid layer when a spreading motion is This approach was found to effectively capture the resilient modu-
created by the traffic loading, which in turn reduces the ver- lus of geogrid-reinforced base layer. Yang and Han [14] provided an
tical stress and shear stress dramatically due to the analytical model to predict the resilient modulus of geogrid-
increased base course stiffness. reinforced UGM. One assumption made in developing the analytical
b) Vertical membrane effect. The inward shear stress caused by model is that the additional confining stresses are uniformly dis-
the membrane deformation reduces the outward shear tributed in the base course. This assumption, however, ignores the
stress generated by the repetitive wheel loading. As a result, phenomenon that the influence of the geogrid reinforcement
the vertical stress is reduced and distributed widely around decreases with the distance of the aggregates from the geogrid,
the geogrid layer. and the geogrid reinforcement is negligible when the material is
distant from the geogrid. Therefore, it is desirable to improve the
To extend the use of geogrid in flexible pavement structures, analytical model by using a more realistic additional confining
there is a need to incorporate the geogrid material into the pave- stress distribution, and further to incorporate the improved analyt-
ment design. The efficient laboratory characterization of geogrid- ical model into the numerical modeling of geogrid-reinforced pave-
reinforced unbound granular material (UGM) is the first step for ment structure.
including the geogrid material in the pavement design, which
has been completed recently by Gu et al. [6]. The repeated load tri-
axial tests were used to quantify the characteristics of geogrid rein-
forcement in terms of the cross-anisotropic resilient moduli and
permanent deformation of the geogrid-reinforced UGM. It was
found that the geogrid reinforcement effectively increases both
the horizontal and vertical moduli of the UGM, meanwhile signifi-
cantly reduces the accumulated permanent deformation of the
UGM. The development of a numerical model to accurately simu-
late the geogrid-reinforced flexible pavement structure is the next
step to quantify the influence of a geogrid on flexible pavement
performance, and further to guide the geogrid-reinforced flexible
(a) Displacement Pattern of UGM Restraint by Geogrid
pavement design.
The numerical modeling of geogrid-reinforced pavement struc-
ture mainly focuses on the constitutive models of paving materials, After test
Before test
the geogrid-aggregate/soil interface model, and the lateral confine-
ment effect of the geogrid. The existing studies have shown that
modeling the nonlinear cross-anisotropy nature of UGM is crucial Deformed
to the accurate performance prediction of flexible pavement
[710]. However, limited studies have been found on modeling the
cross-anisotropic behavior of UGM for the geogrid-reinforced flexi-
ble pavement. Gu et al. [6] evaluated the effect of geogrid on the
cross-anisotropy of UGM in the laboratory, which provided a sound Geogrid
basis for modeling the geogrid-reinforced pavement structure. In Reinforcement Force T
Aggregate
the numerical modeling of geogrid-reinforced pavement, the inter-
action between geogrid and aggregate/soil interface is another
(b) Difference in Radial Movement of Geogrid and Aggregate
important aspect. When surfaces of geogrid and aggregate/soil are
in contact, they usually transmit shear and normal stresses across
their interface. The Goodman model has been widely used to charac-
terize such interface contact, which is shown in Eq. (1) [11].




ds ks 0 dur
1
drn 0 kn dv r

where s is the shear stress; rn is the normal stress; ur is the relative

shear displacement; mr is the relative normal displacement; ks is the
shear stiffness; and kn is the normal stiffness. The interface slippage
condition is quantified by the shear stiffness ks . If the geogrid-
aggregate/soil interface is fully bonded, the shear stiffness ks will (c) Equivalence of Reinforcement Force to Additional Stress 3
be assigned a large value, for example ks = 2.7  108 kPa/m [12]. If
the slippage occurs at the geogrid-aggregate/soil interface, the Fig. 1. Schematic Plot of Geogrid Reinforcement on UGM Specimen.
216 F. Gu et al. / Construction and Building Materials 122 (2016) 214230

300 To address the aforementioned research needs, this study aims

to develop finite element models to simulate the geogrid-
Predicted Horizontal Modulus (MPa)

250 reinforced pavement structures by taking into account the nonlin-

ear cross-anisotropy of geogrid-reinforced UGM, the interaction
between geogrid and aggregate/soil, and the lateral confinement
200
effect of geogrid. Specifically, a user-defined material subroutine
y = 1.0621x - 18.339 is developed to characterize the nonlinear cross-anisotropy of
150 R = 0.9569 UGM. The geogrid layer is defined by a membrane element to sim-
ulate the membrane effect. The Goodman model is used to charac-
100 terize the contact behavior of geogrid-aggregate/soil interface. One
analytical model is developed for quantifying the effect of lateral
50 confinement on nonlinear cross-anisotropy of the UGM. The devel-
oped analytical model is then used to determine the resilient mod-
0 ulus of geogrid-reinforced UGM. Finally, the pavement responses
0 50 100 150 200 250 300 predicted by the geogrid-reinforced pavement models are com-
Measured Horizontal Modulus (MPa) pared to the large-scale tank test measurements to validate the
developed numerical model.

(a) Predicted Horizontal Moduli Vs. Measured Horizontal Moduli

2. Analytical model for predicting resilient modulus of geogrid-
600 reinforced UGM
Predicted Vertical Modulus (MPa)

500 In this study, an analytical model is proposed to predict the ver-

tical and horizontal moduli of the geogrid-reinforced UGM. Fig. 1a
shows a schematic plot of a geogrid-reinforced UGM specimen in
400
the triaxial test. The geogrid-reinforced specimen is compressed
in the axial direction, and normally expands in the lateral direction
300 y = 0.9589x - 6.3422 due to the plastic and resilient deformation. It is demonstrated that
R = 0.9766
the lateral movement of UGM is restrained by the geogrid. The
200 shear stress is generated due to the relative lateral displacement
between geogrid and aggregate, which results in the stretch of
100 the embedded geogrid. Note that the lateral movement of aggre-
gate and geogrid cannot be identical. Fig. 1b shows the difference
0 of lateral movement between geogrid and aggregate. A coefficient
0 100 200 300 400 500 600 a is employed to account for the difference of radial displacement
Measured Vertical Modulus (MPa) between geogrid and aggregate, as shown in Eq. (3).

earr
a 3
(b) Predicted Vertical Moduli Vs. Measured Vertical Moduli errg
Fig. 2. Comparison of Resilient Moduli Predicted by Analytical Models with where earr is the aggregate radial tensile strain at the interface
Measured Values. between the geogrid and aggregate, errg is the geogrid radial tensile

(a) Unreinforced and Geogrid-Reinforced (b) Unreinforced and Geogrid-Reinforced

Pavement with a 15 cm Base Course Pavement with a 25 cm Base Course

Fig. 3. Unreinforced and Geogrid-Reinforced Pavement Structures Used in This Study.

 F. Gu et al. / Construction and Building Materials 122 (2016) 214230 217

Loading Area
Define the constitutive model inputs

15 cm Asphalt Concrete

15 cm Geogrid 25 cm Base Course Get initial stress state and incremental strains
Reinforced Zone
Geogrid Layer

Compute stresses from current strains

1.4 m Subgrade
Axis of
Symmetry Adjust vertical Compute vertical moduli from current stress state
moduli using
damping factor

Check the convergence

criteria
No
Fig. 4. Example of Finite Element Meshed Geogrid-Reinforced Pavement Model. Yes

Update the axisymmetric cross-anisotropic Jacobian matrix

Table 1
Selected laboratory tests for material characterization.

Material Constitutive Lab test Model input Update stresses and return to main program
type model
Asphalt Viscoelastic Dynamic modulus Prony series parameters Fig. 5. Flowchart of The Developed UMAT Subroutine.
concrete test (Gi, Ki, and si)
Base course Nonlinear cross- Rapid triaxial test Inputs of the developed
anisotropic subroutine
Geogrid Elastic Direct tension test Tensile sheet modulus Distance Away from the Load Center (m)
Subgrade Elastic California bearing Youngs modulus 0 0.2 0.4 0.6 0.8 1 1.2
ratio test 0
Surface Deflection (mm)

0.1

Table 2 0.2
Determined Prony-series model coefficients for asphalt concrete.

Prony-series coefficients
0.3

i Gi Ki si 0.4
1 0.362 0.362 4.09E06
2 0.363 0.363 2.56E04 0.5
3 0.1765 0.1765 7.71E03
4 0.074 0.074 2.10E01 0.6
5 0.0165 0.0165 3.88E+00
Control 15 cm Base Control 25 cm Base
6 0.0057 0.0057 6.53E+01
Geogrid-Bottom 15 cm Base Geogrid-Middle 25 cm Base
Elastic parameters: instantaneous modulus = 2630 ksi; Poissons ratio = 0.35.
Fig. 6. Surface Deflections of Geogrid-Reinforced and Unreinforced Pavement
Structures Subjected to a 565 kPa Load.
Table 3
Cross-anisotropic properties of geogrid-reinforced and unreinforced UGM.
of the aggregate. Eq. (4) is an implicit equation for the coefficient
Material Type Nonlinear cross-anisotropic parameters a. The stretch of the geogrid generates a reinforcement force T to
k1 k2 k3 n m mxy mxx confine the UGM specimen through the aggregate particle interlock
Unreinforced UGM 1281 0.81 0.08 0.45 0.35 0.17 0.43 and interface friction [14]. Fig. 1c shows that the reinforcement
Geogrid-reinforced UGM 1733 0.72 0.02 0.45 0.35 0.17 0.43 force T is equivalent to a triangularly distributed additional confin-
ing stress Dr3 , which only acts on a 15 cm geogrid-reinforced influ-
strain. Note that the value of a is normally larger than 1, which rep- ence zone [15]. This distribution takes into account the
resents that the aggregate has a larger lateral movement than the phenomenon that the geogrid reinforcement influence decreases
geogrid. The analytical solution to determine the coefficient a is with the distance of the aggregate from the geogrid, and the geogrid
shown in Eqs. (4) and (5) [14]. reinforcement is negligible when the material is distant from the
    geogrid.
D 2 D Under an axisymmetric plane-stress condition, the reaction
b  J0 b   J1 b r3 4
2 D 2 force T is determined by Eq. (6).

" #1=2
2Ga a  11  m2g
b 5
dM
where Ji x is the Bessel function of order i, D is the diameter of
the aggregate specimen (i.e. D = 15 cm), Ga is the shear modulus
M g
T  errg mg ehh 6
1  m2g
where M is the geogrid sheet stiffness, mg is the Poissons ratio of the
geogrid, errg is the geogrid tensile strain in the radial direction, and
ehhg is the geogrid tensile strain in the circumferential direction. By
218 F. Gu et al. / Construction and Building Materials 122 (2016) 214230

250 800
)

Average Compressive Strain in Base

Tensile Strain at the bottom of AC (

200
600

Layer ( )
150
400
100
200
50

0 0
Control 15 cm Geogrid-Bottom Control 25 cm Geogrid-Middle
Control 15 cm Geogrid-Bottom Control 25 cm Geogrid-Middle
Base 15 cm Base Base 25 cm Base
Base 15 cm Base Base 25 cm Base
Fig. 9. Comparison of Average Compressive Strain in Base Layer between
Fig. 7. Comparison of Tensile Strain at the Bottom of Asphalt Concrete between
Geogrid-Reinforced and Unreinforced Pavement Models.
Geogrid-Reinforced and Unreinforced Pavement Structures.

Vertical Compressive Stress Distribution (kPa) 800

0 20 40 60 80 100 120

Compressive Strain at the Top of

0
600
Depth of Base Layer (cm)

Subgrade ( ) 400
6

9 200

12
0
Geogrid Location Control 15 cm Geogrid-Bottom Control 25 cm Geogrid-Middle
15 Base 15 cm Base Base 25 cm Base
Control 15 cm Base Geogrid-Bottom
Fig. 10. Comparison of Compressive Strains at the Top of Subgrade between
Geogrid-Reinforced and Unreinforced Pavement Structures.
(a) Vertical Stress Distribution within a 15 cm Base Layer

Vertical Compressive Stress Distribution (kPa)

0 20 40 60 80 100 120 140
0
Depth of Base Layer (cm)

10
Geogrid Location
15

20

25
Control 25 cm Base Geogrid-Middle

(b) Vertical Stress Distribution within a 25 cm Base Layer

Fig. 8. Vertical Stress Distribution in Geogrid-Reinforced and Unreinforced Base
Layer.

Fig. 11. Large-scale tank test Setup for Geogrid-Reinforced Flexible Pavement.
assuming the geogrid expands uniformly in both the radial and the
circumferential directions, Eq. (6) is simplified as, 2T 2M
Dr3 max  eg 8
d 1  mg d rr
M
T  eg 7 where d is the thickness of the influence zone (i.e. d = 15 cm).
1  mg rr
Substituting Eq. (3) into Eq. (8) yields,
Since the equivalent additional confining stress Dr3 is triangu-
2M
larly distributed in the influence zone, the maximum additional Dr3 max  ea 9
confining stress Dr3 max can be calculated by Eq. (8). 1  mg da rr
 F. Gu et al. / Construction and Building Materials 122 (2016) 214230 219

In Eq. (9), the aggregate radial tensile strain earr is the summa- ea3;p 0:85ea1;p 13
tion of the radial elastic strain ea3;r and the radial plastic strain
Substituting Eqs. (10), (11) and (13) into Eq. (9) yields,
ea3;p . The radial elastic strain ea3;r is calculated by the Generalized
Hookes law, as shown in Eq. (10). 2M
Dr3 max
1  mg da
r3 Dr3 max m13 r1 m33 r3 Dr3 max

ea3;r   10 r3 Dr3 max m13 r1 m33 r3 Dr3 max
EH EV EH   
EH EV EH

where r1 is the axial stress applied to the specimen, r3 is the initial
N
q b p m
confining pressure, m13 is the Poissons ratio to characterize the 0:85e0 e J 2 aI1 Kn 14
effect of the axial strain on the lateral strain, m33 is the Poissons
ratio to characterize the effect of the lateral stress on the lateral In Eq. (14), the only unknown parameter is the maximum addi-
strain, EH is the horizontal modulus of the specimen, and EV is the tional confining stress Dr3 max . An iteration method is utilized to
vertical modulus of the specimen. Eq. (11) is used to calculate the solve for this parameter.
axial plastic strain ea1;p [16]. The calculated maximum additional confining stress Dr3 max is
used to determine the distribution function of equivalent
p
ea1;p e0 eN J2 aI1 Kn
q b m
11 additional confining stress Dr3 z along the depth z of the specimen.
The determined equivalent additional confining stress distribution
where J2 13 r1  r3 Dr3 max 2 , I1 r1 2r3 Dr3 max , and e0 , Dr3 z is then input into Eq. (15) to calculate the modified vertical
q, b, m and n are permanent deformation properties of the unrein- modulus of the base course EVModified z in the influence zone.
forced specimen. The relationship between the radial plastic strain
k2  k3
I1 Dr3 z soct
ea3;p and the axial plastic strain ea1;p is shown in Eq. (12) [17]. EVModified z k1 Pa 15
  Pa Pa
1 1 sin w
ea3;p ea1;p 12 where I1 is the first invariant of the stress tensor; soct is the octahe-
2 1  sin w
dral shear stress; Pa is the atmospheric pressure; k1 , k2 and k3 are
where w is the dilation angle of the specimen. Assuming that the regression coefficients. The effective vertical modulus of the entire
dilation angle w is 15, Eq. (12) is simplified as, geogrid-reinforced UGM specimen EVEffectiv e is calculated by Eq.

(a) Flexible Pavement with a 15 cm Base Course

(b) Flexible Pavement with a 25 cm Base Course

Fig. 12. Location of Instruments in Flexible Pavement Structures.
220 F. Gu et al. / Construction and Building Materials 122 (2016) 214230

0.80 400
LVDT 1
Soil Tank Measurements (mm)

)
Soil Tank Measurements (
0.60 300

Control Control
0.40 Geogrid-Bottom 200 Geogrid-Bottom
Line of Equality Line of Equality
LVDT 2
10 % Equality + 10 % Equality
0.20 LVDT 4 100
LVDT 3 20 % Equality + 20 % Equality

0.00 0
0.00 0.20 0.40 0.60 0.80 0 100 200 300 400
FE Simulations (mm) FE Simulations ( )

(a) Pavement Structures with a 15 cm Base Course (a) Pavement Structures with a 15 cm Base Course
1.00
400
Soil Tank Measurements (mm)

)
0.80

Soil Tank Measurements (

LVDT 1
300
0.60 Control
Control
Geogrid
200 Geogrid-Middle
0.40 Line of Equality
Line of Equality
10 % Equality
LVDT 4 LVDT 2 + 10 % Equality
0.20 20 % Equality 100
LVDT 3 + 20 % Equality
0.00
0.00 0.20 0.40 0.60 0.80 1.00 0
0 100 200 300 400
FE Simulations (mm)
FE Simulations ( )
(b) Pavement Structures with a 25 cm Base Course
(b) Pavement Structures with a 25 cm Base Course
Fig. 13. Comparison of Measured and Predicted Surface Deflections for Geogrid-
Reinforced and Unreinforced Pavement Structures. Fig. 14. Comparison of Measured and Predicted Tensile Strains at the Bottom of
Asphalt Concrete for Geogrid-Reinforced and Unreinforced Pavement Structures.

(16), which takes into account the variation of the location of the
geogrid in the UGM specimen.
8 Rd
>
> EVUGM hd E zdz d
>
> 0 VModified
< l < h  2d
>
> h 2
< R dl
EVEffectiv e EVUGM h2dl 2 EV Modified zdz  
0
l < 2d
>
> h
>
> R dhl
>
> 
: EVUGM l2d 02 EVModified zdz
h
l > h  2d
where EVUGM is the vertical modulus of the unreinforced base
course, h is the thickness of the base course, and l is the distance
between the geogrid layer and the bottom of the base course. The
effective horizontal modulus of the geogrid-reinforced UGM speci-
men EHEffectiv e is calculated by Eq. (17).
EHEffectiv e n  EVEffectiv e
where n is the ratio of the horizontal modulus to the vertical mod-
ulus, which is determined from the repeated load test.
Fig. 2 shows the comparison of the resilient moduli of the
geogrid-reinforced UGM predicted by the proposed analytical
models above and those measured from the laboratory triaxial
tests [6]. The horizontal and vertical resilient moduli predicted
by the analytical models match the measured values with R-
squared values of 0.96 and 0.98, respectively. This indicates that
the proposed analytical models can accurately predict both the
horizontal and vertical moduli of the geogrid-reinforced UGM.
3. Development of geogrid-reinforced pavement model
3.1. Modeling techniques for geogrid-reinforced pavement

As shown in Fig. 3, two pairs of unreinforced and geogrid-
reinforced pavement structures are used in this study. Considering
16
the practical pavement construction, the geogrid layer is placed at
the base/subgrade interface when the thickness of the unbound

base course is 15 cm, and placed in the middle of base course when
the thickness of the unbound base course is 25 cm. The pavement
structures are subjected to dynamic loading cycles with a loading
amplitude of 565 kPa. Both the loading time and the unloading
time are 0.05 s. The rest period for each loading cycle is 0.9 s.
The loading zone is applied with a circular loading foot with a
radius of 15 cm. This load configuration is used to simulate the
traffic load on the in-service pavement section.
The finite element models are developed using the software
ABAQUS [18] to simulate the unreinforced and geogrid-
17
reinforced pavement structures. Fig. 4 presents one example of
the finite element meshed geogrid-reinforced pavement model.
The cylindrical pavement structure is simplified as an axisymmet-
ric model. Fine mesh is used in the vicinity of the load. The asphalt
concrete, base course and subgrade are represented as 8-node
biquadratic homogeneous solid elements with reduced integration.
The geogrid layer is represented by the 3-node quadratic mem-
brane element. The interface between the geogrid layer and the
aggregate/soil layer is characterized by the Goodman model, which
is defined in Eq. (1). In this study, the lateral confinement effect of
geogrid is simulated by assigning a modified resilient modulus for
the geogrid-reinforced base material, which is shown as the
 F. Gu et al. / Construction and Building Materials 122 (2016) 214230 221

shaded area in Fig. 4. McDowell et al. [19] and Schuettpelz et al. In ABAQUS, Prony-series models are used to characterize the
[15] showed that the geogrid provided the reinforcing effect in time-dependent viscoelastic behavior of the asphalt concrete,
an area that is typically 15 cm thick. Gu et al. [6] reported that which are shown in Eqs. (18) and (19) [18].
placing the geogrid layer in the middle or at one quarter height !
X
n
below the middle of base material affects its horizontal and vertical Gt G0 1  Gi 1  et=s 18
modulus, while placing the geogrid layer at the bottom exerts a i1
minor influence on the modulus of the base material. This finding
!
indicated that the lateral confinement of the geogrid is effective X
n
t=s
only when the geogrid layer is placed within the base course. For Kt K 0 1  K i 1  e 19
this reason, the range of the geogrid influence zone is herein i1

assumed to be 15 cm in height when the geogrid is placed in the where Gt and Kt are relaxation shear modulus and bulk modu-
middle of the base course, and assumed to be zero when the geo- lus; G0 and K 0 are instantaneous shear modulus and bulk modulus;
grid is placed at the bottom of the base course. The modified resi- Gi , K i and si are the input coefficients. Table 2 lists the Prony-series
lient modulus properties of the geogrid-reinforced base material model inputs used in this study.
can be determined by the proposed analytical model. The generalized Hookes law is used to define the cross-
anisotropic behavior of geogrid-reinforced and unreinforced UGM
3.2. Determination of model input material properties for an axisymmetric problem, which is shown in Eq. (20).
2 3
2 3 2
ex 3
m
In this study, the asphalt concrete is characterized as a vis- 1
Ex
 Exyx  mExxx 0 rx
coelastic material, the base layer with and without geogrid is 6 mxy 7
6 E 1 mxy
 Ex 7
0 76 7 6 ey 7
6 Ey 6 ry 7 6 7
76 20
x
defined as a nonlinear cross-anisotropic elastic material, and the 6 mxx 76 7
6 m
 Exyx 1
07 4 rx 5 4 ex 5
geogrid and subgrade are assumed to be linear elastic. Table 1 pre- 4 Ex Ex 5
sents the selected laboratory tests to characterize the paving mate- 0 0 0 1 sxy cxy
Gxy
rials. The dynamic modulus test is used to characterize the
viscoelastic behavior of asphalt concrete [20]. The nonlinear where Ex is the horizontal modulus; Ey is the vertical modulus; Gxy is
cross-anisotropic properties of geogrid-reinforced and unrein- the shear modulus; mxy is the Poissons ratio to characterize
forced base material are determined using the rapid triaxial test the effect of vertical strain on horizontal strain; mxx is the Poissons
[21]. The tensile sheet modulus of geogrid material is measured ratio to characterize the effect of horizontal stress on horizontal
by the direct tension test [22]. The elastic modulus of subgrade is strain.
estimated by the California bearing ratio test [23]. In ABAQUS, this constitutive model needs to be rewritten as a
strain-stress relationship, which is expressed as,

2 2 3
rx 3 n1  nm2yx nmyx a nmxx nm2yx 0
2
ex 3
150 6r 7 E 6 76 ey 7
6 y7 y 6 nmyx a 1  m2xx nmyx a 0 776 7
Soil Tank Measurements (kPa)

6 7 6 6 7
120 4 rx 5 ab 64 nmxx nm2yx nmyx a n1  nm2yx
7
0 54 ex 5
Control
1
sxy 0 0 0 mab c
2 xy
90 Geogrid
P2 Line of Equality 21
60
P4 P3 10 % Equality Gxy
P5 where n EEyx ; m Ey
; a 1 mxx ; b 1  mxx  2nm2yx . This strain-
30 20 % Equality
stress relationship is used to compute the incremental stress for a
P1
given incremental strain in ABAQUS. In Eq. (21), the vertical modu-
0
0 30 60 90 120 150 lus Ey is a stress-dependent term, which is calculated by Eq. (22).
FE Simulations (kPa)  k2  k3
I1 soct
Ey k1 Pa 22
Pa Pa
(a) Pavement Structures with a 15 cm Base Course
where I1 is the first invariant of the stress tensor; soct is the octahe-
dral shear stress; Pa is the atmospheric pressure; k1 , k2 and k3 are
150
regression coefficients. Table 3 presents the nonlinear cross-
Soil Tank Measurements (kPa)

P6 anisotropic properties of the geogrid-reinforced and unreinforced

120
Control UGM determined from the rapid triaxial test.
P7 In ABAQUS, the input tensile sheet stiffness of geogrid material
90 Geogrid
is 455 kN/m, which is measured from the direct tension test. The
Line of Equality
60
input resilient modulus of subgrade is 56 MPa, which is estimated
P2 10 % Equality
P5 P4
by the California bearing ratio test.
20 % Equality
30 P3
P1 3.3. Development of nonlinear cross-anisotropic user-defined material
0 subroutine
0 30 60 90 120 150
FE Simulations (kPa) A user-defined material (UMAT) subroutine is developed to
characterize the nonlinear cross-anisotropic behavior of UGM in
ABAQUS. The UMAT subroutine adopts the direct secant modulus
(b) Pavement Structures with a 25 cm Base Course
approach to determine the nonlinear resilient modulus solution
Fig. 15. Comparison of Measured and Predicted Vertical Stresses within the Base in each iteration. The trial vertical modulus is computed using
and Subgrade for Pavement Structures with and without Geogrid. Eq. (23) in each iteration [24,25].
222 F. Gu et al. / Construction and Building Materials 122 (2016) 214230

a. Axisymmetric plot of triaxial test b. Vertical moduli distribution (Unit MPa)

c. Horizontal strains distribution d. Vertical strains distribution

Fig. 16. Simulation of Triaxial Load test in ABAQUS.

Eiy 1  kEyi1 kEiycomputed 23 4. Finite element analysis results and discussion

Based on the aforementioned modeling techniques, the geogrid-

where Eiy is the vertical modulus output from the ith iteration; Ei1
y is
reinforced and unreinforced pavement models are developed to
the vertical modulus output from the (i  1)th iteration; k is the
simulate the pavement structures shown in Fig. 3. The critical
damping factor; Eiycomputed is the vertical modulus computed from pavement responses are computed when they are subjected to
Eq. (22) at the ith iteration. The convergence criteria are shown in the traffic load, which include:
Eqs. (24) and (25) [26].
a) Surface deflection,
jEiy  Eyi1 j b) Tensile strain at the bottom of asphalt concrete, and
Error i 6 1% 24 c) Compressive strains and stresses in base and subgrade
jEiy j
These computed critical strains are used to evaluate the perfor-
n
X
2 mance of geogrid-reinforced and unreinforced pavement
Eiy  Ei1
y structures.
i1 Fig. 6 compares the surface deflections of the geogrid-reinforced
Error c 6 0:2% 25
X
n
2 pavement models with those of the unreinforced pavement models
Eiy
i1
when they are subjected to a 565 kPa load on a circular area with a
radius of 15 cm It is seen that placing the geogrid in the middle of
where Errori is the individual error for each node; Error c is the the base course slightly decreases the surface deflections nearby
cumulative error for the entire model; n is the number of nodes the load area, while placing the geogrid at the bottom of the base
in the model. Fig. 5 is the flowchart of the developed UMAT subrou- course cannot reduce the pavement surface deflections. This is
tine. The UMAT subroutine code is presented in the Appendix. because placing the geogrid in the middle of base course increases
 F. Gu et al. / Construction and Building Materials 122 (2016) 214230
Table 4
Comparison of Computed Specimen Reponses Using Analytical Solution and Finite
Element Approach.
Stress Analytical solution Finite element approach
state (kPa)
ry rx Ey (MPa) ey (le) ex (le) Ey (MPa) ey (le)
70 40 208.8 191 116 206.7 190
130 60 294.9 287 91 294.2 287
250 120 469.9 275 147 465.1 274
the resilient modulus of the base course, while placing the geogrid
at the base/subgrade interface cannot reinforce the base layer.
Meanwhile, the slippage occurs between the geogrid surface and
the aggregate/soil surface when the geogrid is placed at the bottom
of the base course, which significantly decreases the shear stiffness
of the geogrid-aggregate/soil interface.
The fatigue life of flexible pavement depends on the tensile
strain at the bottom of the asphalt concrete. Fig. 7 shows the ten-
sile strain at the bottom of the asphalt concrete for the geogrid-
reinforced and unreinforced pavement models. It is seen that the
tensile strain increases by 8% when the geogrid is placed at the bot-
tom of the base course, and reduces by 5% when placing the geo-
grid in the middle of the base course. This indicates that placing
the geogrid in the middle or at the bottom of the base course can-
not significantly affect the fatigue life of flexible pavement.
The critical pavement responses related to the rutting damage
include the vertical compressive stress and strain in each layer.
The rutting damage of the base course is dependent on the vertical
compressive stress and strain in the base layer. The distributions of
vertical stress in the base layer beneath the load center are also
plotted in Figs. 8a and b. Fig. 8a shows that placing the geogrid at
the bottom of the base course effectively reduces the vertical stres-
ses around the geogrid layer. The decrease of vertical stresses
becomes negligible when the base material is 10 cm away from
the geogrid layer. Fig. 8b indicates that placing the geogrid in the
middle of base course significantly decreases the vertical stresses
in the entire 25 cm base course. This demonstrates that placing
the geogrid in the middle of the base course is much more effective
in reducing the vertical stresses in the base layer than placing the
geogrid at the bottom of the base course. It is known that the
decrease of vertical compressive stresses in the base layer is bene-
ficial for reducing the permanent deformation of base material
[16]. Therefore, placing the geogrid in the middle of base course
is better at reducing the rutting damage of the base course.
Fig. 9 compares the average compressive strain in the base layer
between geogrid-reinforced and unreinforced pavement models. It
is seen that the average compressive strain in base layer reduces by
17% when the geogrid layer is placed in the middle of the base
course. Placing the geogrid at the bottom of base course cannot
decrease the average compressive strain in base layer. The rutting
model in the mechanistic-empirical pavement design guide
(MEPDG) shows that a smaller average vertical compressive strain
in the base course yields less rutting damage [27]. Hence, it is con-
firmed that the geogrid is effective in reducing the rutting damage
of the base course when it is placed in the middle of the base layer.
The compressive strain at the top of the subgrade determines
the rutting damage of subgrade. Fig. 10 shows the compressive
strain at the top of the subgrade for the geogrid-reinforced and
unreinforced pavement models. It is shown that placing the geo-
grid at the base/subgrade interface reduces the compressive strain
at the top of the subgrade by 38%, while placing the geogrid in the
middle of the base course only decreases it by 10%. This indicates
that the geogrid is much more effective in reducing the
compressive strain at the top of the subgrade when it is placed
223
at the base/subgrade interface. The MEPDG rutting model also
shows that a smaller compressive strain at the top of the subgrade
yields less rutting damage in the subgrade [27]. Therefore, placing
the geogrid at the base/subgrade interface is better at reducing the
rutting in the subgrade than placing it in the middle of base course.
ex (le)
116
91 5. Validation of numerical models using large-scale tank test
146
In order to validate the developed finite element models, a well-
designed large-scale tank (LST) testing program with extensive
instrumentations was conducted on the geogrid-reinforced and
unreinforced flexible pavements using a circular steel tank with
2.4 m diameter and 1.8 m height. The setup of LST test program
is presented in Fig. 11.
The implemented testing program differs from previous studies
which traditionally assessed the influence of geogrid on pavement
performance under repetitive surface loading until failure
[3,2830]. Instead, the experimental program focused on recording
the critical pavement responses, including the surface deflection,
the tensile strain at the bottom of the asphalt concrete, and vertical
stresses within the base course and subgrade. Fig. 12 illustrates the
location of the instruments in geogrid-reinforced and unreinforced
flexible pavements. L1L5 represent the linear variable differen-
tial transformers (LVDTs). The tensile strain gauge is denoted as
ACSG. The pressure sensors embedded in the base course and
subgrade are shown as P1P5 in Fig. 12a and P1P7 in Fig. 12b,
respectively. The measured pavement responses are compared to
the finite element simulation results to validate the developed
geogrid-reinforced and unreinforced pavement models.
Figs. 13a and b show the comparison of the surface deflections
predicted by the finite element models and the LST test measure-
ments when the pavement structures are subjected to a 565 kPa
load. The model-predicted surface deflections are in agreement
with the LST test measurements from LVDTs 1, 2, and 3. The devi-
ation between the measured surface deflection by LVDT 4 and that
predicted by the finite element model exists because the surface
deflection at this location is too small and thus may not have been
accurately captured by the LVDT. The comparison of surface deflec-
tion results indicate that the developed geogrid-reinforced and
unreinforced pavement models have the capability to predict the
pavement surface deflections.
The comparisons of the tensile strains at the bottom of the
asphalt concrete are plotted in Fig. 14. It is seen that the developed
finite element models accurately predict the tensile strains in the
geogrid-reinforced and unreinforced pavement structures. The
LST test measurements show that placing the geogrid in the middle
of the base course or at the base/subgrade interface only slightly
change the tensile strain at the bottom of the asphalt concrete. It
is confirmed that placing the geogrid within the base course cannot
effectively extend the fatigue life of flexible pavement.
Figs. 15a and b present the comparisons of predicted vertical
pressures within the base and subgrade layer between the finite
element model predictions and the LST test measurements. Most
of the measured pressure values are captured by the developed
finite element models, except the measurement of pressure cell
P1. The measured vertical stresses from pressure cell P1 are
much lower than the finite element model predicted results. There
are a number of possible explanations for these discrepancies,
including that the stress dependent behavior of the subgrade was
not taken into account by the numerical modeling or the role of
soil arching effect over the pressure sensor in the pavement
construction.
In general, the finite element simulation results are in good
agreement with the LST test measurements for both the geogrid-
reinforced and unreinforced pavement structures. The proposed
224 F. Gu et al. / Construction and Building Materials 122 (2016) 214230
modeling techniques, the characterization of paving materials, and
the developed UMAT subroutine are important to develop
accurate numerical models for the geogrid-reinforced pavement
structures.
6. Summary and future work
This study developed the finite element models to simulate the
geogrid-reinforced pavement structures by taking into account, the
lateral confinement effect of the geogrid, the interaction between
geogrid and aggregate/soil, and the nonlinear cross-anisotropy of
UGM. The developed numerical models are validated by comparing
the model predicted pavement responses to the LST test measure-
ments. The major findings of this study are summarized as follows.
An analytical model was proposed to predict the cross-
anisotropic resilient modulus of geogrid-reinforced UGM based
on material properties of the UGM and the geogrid. The lateral con-
finement force generated by the stretched geogrid is equivalent to
a triangularly distributed additional confining stress distribution,
which acts on the geogrid-reinforced influence zone. The equiva-
lent additional confining stress distribution is used to determine
the modified vertical and horizontal resilient moduli of geogrid-
reinforced UGM. The comparison of the analytical model predicted
resilient moduli with the laboratory test measurements indicate
that the proposed analytical model can accurately predict both
the horizontal and vertical moduli of the geogrid-reinforced UGM.
Two pairs of geogrid-reinforced and unreinforced pavement
structures were simulated by using the software ABAQUS. It was
found that the placement of the geogrid significantly reduces the
vertical compressive stress and strain in the base course and at
the top of the subgrade, but cannot affect the tensile strain at the
bottom of the asphalt concrete. This indicated that the geogrid
reinforcement is effective in mitigating the rutting damage of the
base course and subgrade, but cannot extend the fatigue life of
flexible pavement. The geogrid reinforced in the middle of the base
course is better at reducing the rutting damage of the base course
than that placed at the base/subgrade interface. While the geogrid
reinforcement is much more effective in reducing the rutting dam-
age of the subgrade when it is placed at the bottom of the base
course.
A comprehensive LST testing program was designed to monitor
the critical pavement responses, including the surface deflection,
tensile strain at the bottom of the asphalt concrete, and the vertical
stresses in base course and subgrade. The measured pavement
responses were compared to the finite element simulation results.
It was shown that the finite element simulation results are in good
agreement with the LST test measurements for both the geogrid-
reinforced and unreinforced pavement structures. The proposed
modeling techniques, the characterization of paving materials,
and the developed UMAT subroutine are important to develop
accurate numerical models for the geogrid-reinforced pavement
structures.
The future numerical modeling studies will focus on the sensi-
tivity analysis of the validated geogrid-reinforced pavement mod-
els. The developed finite element models will be used to quantify
the influence of a variety of factors, including the layer thickness,
layer modulus, sheet stiffness of the geogrid, and geogrid location
on the performance of geogrid-reinforced flexible pavements.
Acknowledgments
The authors acknowledge the financial support provided by the
National Cooperative Highway Research Program (NCHRP). Special
thanks are given to the research team from University of Nevada,
Reno for conducting the large-scale tank test.
Appendix A
The ABAQUS user subroutine UMAT for nonlinear cross-
anisotropic material is presented here. It is written in FORTRAN
language.
To verify the accuracy of the developed UMAT, a triaxial load
test is simulated using ABAQUS under various stress states.
Fig. 16a is a schematic plot of the simulated triaxial load test in
the axisymmetric condition. The inputted nonlinear cross-
anisotropic properties of unreinforced granular material are shown
in Table 3. Fig. 16bd show the distribution of vertical moduli, hor-
izontal strains and vertical strains of the granular material speci-
men under a stress state with 70 kPa vertical stress and 40 kPa
confining pressure. The computed vertical moduli, horizontal
strains at point A and vertical strains at point B under various
stress states are also compared to the analytical solutions calcu-
lated by Eqs. (20) and (22), which are shown in Table 4. It is seen
that the simulation results provide good agreement with the ana-
lytical results calculated from the constitutive models. This indi-
cates that the developed UMAT subroutine can accurately
characterize the nonlinear cross-anisotropic behavior of UGM.
F. Gu et al. / Construction and Building Materials 122 (2016) 214230 225
226 F. Gu et al. / Construction and Building Materials 122 (2016) 214230
F. Gu et al. / Construction and Building Materials 122 (2016) 214230 227
228 F. Gu et al. / Construction and Building Materials 122 (2016) 214230
F. Gu et al. / Construction and Building Materials 122 (2016) 214230 229
230 F. Gu et al. / Construction and Building Materials 122 (2016) 214230

References [15] C. Schuettpelz, D. Fratta, T.B. Edil, Evaluation of the zone of influence and
stiffness improvement from geogrid reinforcement in granular materials,
Transp. Res. Rec. 2116 (2009) 7684.
[1] S.W. Perkins, M. Ismeik, A synthesis and evaluation of geosynthetic-reinforced
[16] F. Gu, Y. Zhang, C.V. Droddy, R. Luo, R.L. Lytton, Development of a new
base course layers in flexible pavements: Part I Experimental work, Geosynth.
mechanistic-empirical rutting model for unbound granular material, J. Mater.
Int. 4 (6) (1997) 549604.
Civ. Eng. 04016051 (2016).
[2] Y. Qian, D. Mishra, E. Tutumluer, H. Kazmee, Characterization of geogrid
[17] F. Tatsuoka, Discussion on the strength and dilatancy of sands by Bolton, M.
reinforced ballast behavior at different levels of degradation through triaxial
D., Geotechnique 37 (1) (1987) 219226.
shear strength test and discrete element modeling, Geotext. Geomembr. 43
[18] ABAQUS, ABAQUS Standard Users Manual, 6.10, ABAQUS Inc., Providence,
(2015) 393402.
Rhode Island, 2010.
[3] R. Hass, J. Walls, R.G. Carroll, Geogrid reinforcement of granular bases in
[19] G.R. McDowell, O. Harireche, H. Konietzky, S.F. Brown, N.H. Thom, Discrete
flexible pavements, Transp. Res. Rec. 1188 (1988) 1927.
element modelling of geogrid-reinforced aggregates, PI Civil Eng. Geotec. 159
[4] S.W. Perkins, Evaluation of Geosynthetic Reinforced Flexible Pavement
(1) (2006) 3548.
Systems Using Two Pavement Test Facilities, in: Report No. FHWA/MT-02-
[20] ASTM, Standard Test Method for Dynamic Modulus of Asphalt Mixtures.
008/20040, 2002.
D3497-79, West Conshohocken, PA, 2003.
[5] B.R. Christopher, R.R. Berg, S.W. Perkins, Geosynthetic reinforcement in
[21] A. Adu-Osei, D.N. Little, R.L. Lytton, Cross-anisotropic characterization of
roadway sections, in: NCHRP Synthesis for NCHRP Project 207, Task 112.
unbound granular materials, Transp. Res. Rec. 1757 (2001) 8291.
Final Report,, 2001.
[22] ASTM, Standard Test Method for Determining Tensile Properties of Geogrids
[6] F. Gu, Y. Zhang, X. Luo, R. Luo, R.L. Lytton, Impact of geogrid on cross-
by the Single or Multi-Rib Tensile Method. D6637-15, West Conshohocken, PA,
anisotropy and permanent deformation of unbound granular materials,
2015.
Transp. Res. Rec. 2580 (1) (2016). In Press.
[23] ASTM, Standard Test Method for California Bearing Ratio of Laboratory-
[7] E. Tutumluer, M.R. Thompson, Anisotropic modeling of granular bases in
Compacted Soils. D1883-14, West Conshohocken, PA, 2014.
flexible pavements, Transp. Res. Rec. 1577 (1997) 1826.
[24] E. Tutumluer, Predicting behavior of flexible pavements with granular bases
[8] J. Oh, R.L. Lytton, E. Fernando, Modeling of pavement response using nonlinear
(Ph.D. dissertation), Georgia Institute of Technology, Atlanta, 1995.
cross-anisotropy approach, J. Transp. Eng. 132 (6) (2006) 458468.
[25] K.D. Hjelmstad, E. Taciroglu, Analysis and implementation of resilient modulus
[9] H. Wang, I. Al-Qadi, Importance of nonlinear anisotropic modeling of granular
models for granular solids, J. Eng. Mech. 126 (8) (2000) 821830.
base for predicting maximum viscoelastic pavement responses under moving
[26] M. Kim, E. Tutumluer, J. Kwon, Nonlinear pavement foundation modeling for
vehicular loading, J. Eng. Mech. 139 (1) (2013) 2938.
three-dimensional finite element analysis of flexible pavements, Int. J.
[10] F. Gu, H. Sahin, X. Luo, R. Luo, R.L. Lytton, Estimation of resilient modulus of
Geomech. 9 (5) (2009) 195208.
unbound aggregates using performance-related base course properties, J.
[27] AASHTO, Mechanistic-Empirical Pavement Design Guide: A Manual of
Mater. Civ. Eng. 27 (6) (2014) 04014188.
Practice., AASHTO Designation: MEPDG-1, Washington, D.C., 2008.
[11] R.E. Goodman, R.L. Taylor, T.L. Brekke, A model for the mechanics of jointed
[28] S.W. Perkins, Mechanical response of geosynthetic-reinforced flexible
rock, J. Soil Mech. Found. Div. 94 (3) (1968) 637659.
pavements, Geosynth. Int. 6 (5) (1999) 347381.
[12] J. Kwon, E. Tutumluer, M. Kim, Development of a mechanistic model for
[29] I.L. Al-Qadi, S.H. Dessouky, J. Kwon, E. Tutumluer, Geogrid in flexible
geosynthetic-reinforced flexible pavements, Geosynth. Int. 12 (6) (2005) 310
pavements: validated mechanism, Transp. Res. Rec. 2045 (2008) 102109.
320.
[30] Y. Qian, J. Han, S.K. Pokharel, R.L. Parsons, Performance of triangular aperture
[13] J. Kwon, E. Tutumluer, H. Konietzky, Aggregate base residual stresses affecting
geogrid-reinforced base courses over weak subgrade under cyclic loading, J.
geogrid reinforced flexible pavement response, Int. J. Pavement Eng. 9 (4)
Mater. Civ. Eng. 25 (8) (2013) 10131021.
(2008) 275285.
[14] X. Yang, J. Han, Analytical model for resilient modulus and permanent
deformation of geosynthetic-reinforced unbound granular material, J. Geotech.
Geoenviron. Eng. 139 (9) (2013) 14431453.