Evaluating Nonlinear Elastic Models

for Unbound Granular Materials

in Accelerated Testing Facility
Alvaro González, Mofreh Saleh, and Adnan Ali

The characterization of unbound granular materials is particularly For a more realistic simulation of the elastic response of the gran-
important in thin-surfaced pavements where most of the bearing capac- ular materials, nonlinear stress-dependent constitutive models have
ity is provided by the granular layers. For an accurate modeling of the been introduced by several authors. In addition, improved modeling
elastic response of these materials, several important models described tools, such as three-dimensional finite element methods (3D-FEM),
in the literature have been considered in a study. Three of the most- have been introduced to better simulate the conditions of the material
used nonlinear elastic models for unbound granular materials used by in the field. This work aims to compare the calculated nonlinear elas-
researchers and practitioners in the pavement engineering field are tic responses of UGM pavements with field measurements from an
evaluated. The nonlinear elastic models of the materials were especially accelerated testing facility.
implemented in ABAQUS, a general purpose, three-dimensional finite
element code (3D-FEM) that uses user-defined material subroutines. BACKGROUND
The 3D-FEM solutions were validated through a set of data collected in
the full-scale Canterbury Accelerated Pavement Testing Indoor Facility The stress pattern induced in a pavement by a moving wheel load is
(CAPTIF). Vertical, transverse, and longitudinal strains were measured quite complex. The elements of the pavement structure are subjected
at CAPTIF through an array of strain gauges buried at different depths to stress pulses, each consisting of vertical, horizontal, and shear
in the pavement sections. The vertical stresses were measured by load components. The deformation response of granular layers under
pressure cells at different locations in the pavement sections. In general, traffic loading can be described by a recoverable (elastic or resilient)
the results obtained from the nonlinear models are considered good for the deformation and a residual (permanent) deformation.
materials and structures studied in the research. Little difference in the Regarding the traditional theories of elasticity, the elastic proper-
accuracy of the nonlinear models was found. ties of a material are defined by the modulus of elasticity E and Pois-
son’s ratio ν. The same approach has been used for modeling the
granular materials, but instead of use of elastic modulus, the resilient
The appropriate characterization of granular materials response under modulus (Mr) has been adopted. For repeated load triaxial tests with
traffic loads is essential for the structural design of pavements. Tra- constant confining pressure, the resilient modulus and Poisson ratio
ditional pavement design methods for unbound granular materials are defined respectively by
(UGM) are highly empirical; however, good efforts have been carried
out to develop mechanistic models that account for the stress depen- Δ ( σ1 − σ 3 ) 
Mr = ; ν = − 3,r (1)
dency of these materials. In flexible pavements, especially when the 1,r 1,r
surface is a thin asphalt layer or chip seal, the role of granular layers
is very important in the general performance of the structure since the where Δ(σ1 − σ3) is the difference between the vertical and radial
load is distributed to the subgrade mainly through these layers. stresses (deviatoric stress), and 1,r and 3,r are the vertical and radial
The first attempts to introduce mechanistic analysis in pavement resilient strains, respectively.
design for UGM were based on linear elastic theory assuming lin- Among all the factors that affect Mr (density, grading, moisture con-
ear elastic behavior in all pavement layers (1). In these analyses, the tent, stress history, aggregate type), the most important factor is the
response of the pavement was calculated by using layered linear stress state (2, 3). It is possible to find a detailed description of factors
elastic software, assuming an elastic modulus for each layer. This affecting resilient response in the literature. Several models have been
approach is simple and easy to use for the practitioner but does not developed to address the effect of stress state on the resilient response.
match the real elastic behavior of granular layers. Furthermore, per- A brief description of the three models used in this work follows.
forming linear elastic analyses on pavements with a thin surface
layer and a thick granular layer leads to nonexistent high tensile
stresses within the granular layers. NONLINEAR MODELS

k-θ Model
School of Engineering, University of Canterbury, Private Bag 4800, Christchurch 8020,
New Zealand. Corresponding author: M. Saleh: [email protected]. The k-θ model is a power relationship between the sum of the first
stress invariant θ (bulk stress) and the resilient modulus:
Transportation Research Record: Journal of the Transportation Research Board,
No. 1990, Transportation Research Board of the National Academies, Washington, k2
⎛ θ⎞
D.C., 2007, pp. 141–149. Mr = k1 * Pa ⎜ ⎟ (2)
DOI: 10.3141/1990-16 ⎝ Pa ⎠

141
142 Transportation Research Record 1990

where k1, k2 are regression coefficients and Pa is the atmospheric This model was recommended in a study carried out by Andrei
pressure. et al. (12). In that study, 13 predictive models were evaluated, and
Among the first researchers to suggest this model were Seed this last was deemed the best compromise among accuracy, ease of
et al. (4), and it also was used by Dehlen and Monismith (5). The implementation, and computational stability.
simplicity of this model has made it extremely handy and extensively
accepted for analysis of stress dependency of material stiffness.
3D Finite Element Model
However, the disadvantage of this model is that it assumes a con-
stant Poisson’s ratio, and the effects of stress on resilient modulus are The FEM has a direct application in pavement analysis because it is
accounted for only by the sum of the principal stresses. Some authors possible to divide the layers into several finite elements. One of the
have reported the limitation of this model (6, 7). main advantages of this partition is that the stiffness of each element
can vary according to its stress state or stress path. For that reason,
Classical Universal Model the FEM has been extensively used in research to model pavement
structures (13–16).
May and Witczak noted that the in situ resilient modulus of a granular The main difficulty of implementing the nonlinear solution for
layer is a function not only of the bulk stress but also of the magnitude granular materials in finite element software is the iterative nature
of the shear strain induced mainly by shear, or deviator stress (8). Uzan of the equations that will require longer computation time. As the
included deviator stress in the k-θ model and wrote the equation as resilient modulus is not constant, this value must be modified and
follows (9): updated according to the stress conditions calculated by the software,
as is schematically illustrated in Figure 1. In this example, the initial
k2 k3 stiffness of a nonloaded granular material piecewise is Mr0. When
⎛ θ ⎞ ⎛σ ⎞
Mr = k1 * Pa * ⎜ ⎟ ⎜ d ⎟ (3) the element is loaded, the stress path changes and a new resilient
⎝ Pa ⎠ ⎝ Pa ⎠ modulus is reached. Once the material achieves a new stiffness, a
new stress path occurs and hence the resilient modulus changes. For
where k1, k2, k3 are regression coefficients and σd is the deviator an accurate resolution of this iterative finite element problem, this
stress. This model has been shown to be superior to the k-θ model procedure must be repeated until the resilient modulus of the element
in some studies (7, 10). converges.
For this research, ABAQUS 3D-FEM code was used to address
the described problem. ABAQUS is a general purpose finite element
Modified Universal Model
software used for research purposes and also by several companies
A variation of the classical universal model for base course was in various engineering fields. It has shown satisfactory results in
adopted by NCHRP (11): previous work on granular pavements (17 ).
ABAQUS/Standard uses the iterative Newton method to solve
k2 k3 nonlinear equations. The applied load is augmented incrementally,
⎛ θ ⎞ ⎛τ ⎞
Mr = k1 * Pa * ⎜ ⎟ ⎜ oct + 1⎟ (4) and at each increment the finite element package iteratively solves
⎝ Pa ⎠ ⎝ Pa ⎠ a system of equations to get an approximate solution. The iterations
continue on the basis of the previous solutions until it reaches a rea-
where τoct (octahedral shear stress) is expressed in terms of the prin- sonable convergence. Because ABAQUS is a general purpose finite
cipal stresses (σ1, σ2, σ3) by element code, it does not include specific nonlinear elastic models
for granular materials. Hence, nine different material models were
1
τ oct =
3
(σ 1 − σ 2 ) + ( σ 2 − σ 3 ) + ( σ 3 − σ1 )
2 2 2
(5) programmed by Ali (18), but only the three described earlier were
used in this research. The models were defined by internal subroutines

Mr Mr Mr Mr Mr
Mr1 Mr2
Mr0
θ θ θ θ θ
θ0 θ1 θ1 θ2 θ2
σd0 σd1 σd1 σd2 σd2

σc0 σc0 σc1 σc1 σc1 σc1 σc2 σc2 σc2 σc2
Mr0 Mr0 Mr1 Mr1 Mr2

σd0 σd1 σd1 σd2 σd2

Initial modulus of A stress path is induced Modulus must be Since the element Again the
the material. Unload on the element when updated. reaches a new modulus is
pavement. load (wheel) is applied. modulus, a new updated. The
stress condition is iteration continues
achieved. until Mri ~ Mri+1.

FIGURE 1 Example of incremental calculation of stiffness in granular element.

González, Saleh, and Ali 143

TABLE 1 Description of Models and Input conditions were set to “pinned” (U1 = U2 = U3 = 0) in the bottom
Parameters Required by Finite Element Code (in Figure 2, surface FBE), side (GDE), and rear (FCG) surfaces;
X-symmetry (U1 = UR2 = UR3 = 0) in surface FBAC; and Y-symmetry
Model Name
(U2 = UR1 = UR3 = 0) in surface BEDA.
Universal Universal To perform the 3D-FEM model, three finite element types were
Input # Linear k-θ Modified Classic used: a linear hexahedron (eight nodes), a quadratic hexahedron
(20 nodes), and a quadratic wedge (15 nodes). The linear hexa-
1 1 2 4 5 hedron was used to model the thin asphalt layer (25 mm) by using
2 ν k1 k1 k1 3,120 cubic linear elements of 25-mm dimension. The quadratic
3 E k2 k2 k2 hexahedron was adopted to model the cubic elements of the base
4 ν k3 k3 course and subgrade layers. A total of 5,025 cubic quadratic elements
5 Emin ν ν were used considering element sizes of 100, 50, and 25 mm. Finally,
6 Pa Emin Emin 105 quadratic wedges were implemented to fit the accelerated
7 Pa Pa pavement facility geometry into the 3D-FEM model.
8
9 Verification of 3D-FEM Code
k1, k2, k3 = material parameters obtained from laboratory tests To verify the usage of the 3D-FEM code, some preliminary compar-
ν = Poisson’s ratio
E = elastic modulus (MPa) for linear elastic case isons were carried out in this research. The comparisons consisted in
Pa = normalized pressure (atmospheric pressure) evaluating the vertical stresses (σv) and vertical displacements (δv) of
Emin = minimum elastic modulus accepted by nonlinear a pavement by using the 3D-FEM model and two known linear elastic
elastic models
layer softwares: EVERSTRESS, developed by the Washington State
Department of Transportation (19), and MePads, developed by
CSIR Transportek of South Africa (20). Linear elastic elements were
or UMATs (user-defined material models) programmed by using assumed in the 3D-FEM case. The geometry used for the comparison
FORTRAN. is a section of the accelerated test track facility, and a simple pavement
Table 1 shows the description of the models programmed into the structure was analyzed for this purpose. The structure consisted of a
UMAT subroutine and the input parameters required by ABAQUS. 200-mm base course (Mr = 300 MPa, ν = 0.35), a 1,300-mm subgrade
(Mr = 50 MPa, ν = 0.35), and a stiff layer (Mr = 10,000 MPa, ν = 0.35)
3D-FEM Model Geometry to simulate the boundary conditions of the test track facility by using
the linear elastic layer software. A single wheel load of 20 kN using a
The geometry adopted by the 3D-FEM model corresponds to pave- contact pressure of 500 kPa was considered in the analysis.
ment sections available in the accelerated test track (Figure 2) and To evaluate the differences between the models objectively, two
is described further in this paper. Four different parts were assem- statistical parameters were used: the absolute relative difference
bled to create a quarter-model, to reduce the computational effort by (ABS) and the root mean square (RMS). Both are defined by the
making use of the symmetry in the geometry and the loading. The following equations:
boundary conditions of the model surfaces were adapted to create a
1 n
⎛ d − d mi ⎞
quarter-model. If the displacements and rotations are Ui and URi, ABS = × ∑⎜ c × 100 (6)
respectively, and i = 1,2,3 correspond to X,Y,Z axis, the boundary n i =1 ⎝ d mi ⎟⎠

z
y 4000mm
x
(Inner wheelpath) (Outer wheelpath)
G
C 25mm thick
m m 150
4000m 1300m
asphalt layer
0 mm
D
Quarter
A section
base course
used for
Emu coils finite element
1500mm

method
1500mm

@75mm c/c
F subgrade

E
mm
550
B
2500mm

(a) (b)

FIGURE 2 Scheme of accelerated pavement testing sections: (a) 3-D view of quarter section and (b) cross section of CAPTIF pavement.
144 Transportation Research Record 1990

⎛ 1 n ⎛ d − d ⎞2 ⎞ of four sections were used in the validation of the nonlinear models.

⎜⎝ n ∑
RMS = ⎜ × ⎜ c mi
⎟ × 100 (7) A diagram of project PR3-0805 is showed in Figure 3. The pavements
i =1 ⎝ dmi ⎟⎠ ⎟⎠ tested were thin-surfaced (25-mm asphalt for all projects) with one
unbound granular layer (200-, 275-, and 300-mm depth) and three
where different clayey subgrades. The properties of the materials and
sections are shown in Figure 4.
n = number of data or points evaluated at different depths
The same base course material was used for both projects. The
between the wheels,
base course material is a crushed alluvial gravel from a quarry located
dmi = value calculated by using the linear elastic layer software, and
near the test track facility. The rock from the quarry was angular,
dc = value calculated by using the FEM model.
hard, light bluish-grey sandy medium gravel with minor silt. The
For the vertical stress analyses, an ABS value of 0.53% between material is well graded with nil plastic fines (plasticity index = 0)
ABAQUS and EVERSTRESS was obtained, and the calculated RMS with only 3% fines (i.e., mass passing the 75-micron sieve).
was 0.92%. For vertical deformations, values of 8.53% for ABS and Two subgrade soil types with a thickness of ranging from 1,200 to
10.53% for RMS were found. Between ABAQUS and MePADS, 1,350 mm were used during the tests. The subgrade was compacted
values of 4.63% and 5.62% were found for ABS and RMS for ver- in layers 150 mm thick to ensure the uniform compaction over the
tical stresses. For the vertical displacements, an ABS value of 5.23% full depth of the layer. The subgrade for the PR3-0610 on all pave-
and an RMS value of 7.04% were calculated. ment sections was a silty clay, 1,200 mm thick, which has a nominal
The results obtained for the comparison provided a reasonable in situ California bearing ratio value of 11%. The upper subgrade
match, and the differences found are mainly caused by the different for the CAPTIF PR3-0805 test on all pavement sections was a clayey
boundary conditions of the 3D-FEM model compared to the layered subgrade with different moisture contents. A silty clay was used for
software. In this last case, the layers are considered infinite in the the lower half of the pavement (depth 750 to 1,500 mm).
horizontal direction, whereas the 3D-FEM model has constrained A total of three different dual wheel loads were applied on each
boundary conditions. Another factor that affects the results is the section: 40 kN and 50 kN for PR3-0805 and 40 kN and 60 kN for
shape of the wheel print, which was assumed circular for the linear PR3-0610.
elastic software and rectangular for the finite element analysis.

Modeling Pavement Sections

CANTERBURY ACCELERATED TESTING FACILITY
The loaded area dimensions were determined from a tire print test
The Canterbury accelerated pavement testing indoor facility (CAPTIF) at the CAPTIF facility. The tire imprint shapes were approximately
in Christchurch, New Zealand, is a 58-m-long (on the centerline) rectangular, having dimensions of 200- to 250-mm width and 200-mm
circular track contained within a concrete tank 1.5 m deep by 4.0 m length. To match the contact area with the actual applied loads, a con-
wide so that the moisture content of the pavement materials can be tact pressure of approximately 700 kPa was assumed. The separation
controlled and the boundary conditions are known. A center platform between the wheels was set to 350 mm.
carries the machinery and electronics needed to drive the system. The data used for this research project were collected at
Mounted on this platform is a sliding frame that can move horizon- 100,000 cycles, at which the postcompaction period of the granular
tally by 1 m. This radial movement allows the wheelpaths to be layers was considered completed. After this stage, it is possible to
varied laterally and can be used to have the two “vehicles” operating assume that little variation in the elastic response occurs. The speed
in independent wheelpaths. At the ends of this frame, two radial of the SLAVE units during the data collection was set to 45 km/h.
arms connect to the simulated loading and vehicle emulator (SLAVE) Because the main focus of this research is the comparison of
units. These arms are hinged in the vertical plane so that the SLAVE nonlinear models for base course materials, the subgrade was con-
units can be removed from the track during pavement construction sidered linear elastic in all the 3D-FEM analyses. Therefore, an
or profile measurement and are hinged in the horizontal plane to elastic modulus was determined for this particular layer. Measure-
allow vertical movement of the vehicles. A more detailed description ments made by using the dynamic cone penetrometer, light falling
of the CAPTIF and its systems is given by Steven (21). weight deflectometer, and falling weight deflectometer were used to
The response of the pavement during the CAPTIF tests was define the stiffness of the subgrade layer. The results were consis-
monitored by using instruments installed in each pavement section. tent, and values of 89 MPa and 121 MPa were adopted for PR3-0805
Each pavement section was instrumented with strain coils (fabricated Sections D and E, respectively, and a uniform value of 35 MPa was
at CAPTIF) and pressure cells. The relay boards, triggering systems, assumed for both PR3-0610 sections.
and software were developed at the University of Canterbury. All Three repeated triaxial laboratory tests were carried out on the
the strain coil pairs are calibrated and tested before their installation. base course material to obtain the coefficients used in the nonlinear
In addition, sensors to measure transversal strains, longitudinal strains, models using AUSTROADS standards developed by Vuong and
and vertical stresses are installed in some sections. Hazell (22). The results of these tests were fitted to the k-θ model,
resulting in an adjusted R2 value of 0.99 with k1 = 112.5 and k2 = 0.70.
The same procedure was performed for the classical universal model
Projects Analyzed for Study (getting an adjusted R2 = 0.99 and k1 = 101.9, k2 = 0.78, k3 = −0.17)
and for the modified universal model (adjusted R2 = 0.99 and k1 =
For this research, data from Transit New Zealand projects PR3-0805 1277.28, k2 = 0.78, k3 = −0.30). These values were introduced in the
and PR3-0610 were used. The pavements tested in these projects UMATs functions of ABAQUS. Also, a linear elastic base course
correspond to typical New Zealand low-volume roads structures. material was included in the analysis. An elastic modulus of 300 MPa
From each project two different sections were constructed. A total was defined for the whole layer for this case.
González, Saleh, and Ali 145

Section D

Section C

Section E

CAPTIF PR3-0805 Test

Section B

Section A

Transition Section

Position of the EMU

strain coils

FIGURE 3 Top view of CAPTIF Project PR3-0805—asphalt surface, 25 mm; granular

depth, 150/300 mm; subgrade, 1,325/1,175 mm. (OPGA  open-graded asphalt;
AC  asphalt concrete; OMC  optimum moisture content.)

25 mm 25 mm 25 mm 25 mm

200 mm
300 mm 300 mm 275 mm

Clay Clay Silty Clay Silty Clay

MDD = 1.76 ton/m3 MDD = 1.76 ton/m3 MDD = 1.87 ton/m3 MDD = 1.87 ton/m3
MC = 30.0% MC = 22.0% MC = 9.6% MC = 9.6%
OMC = 22.0% OMC = 22.0% OMC = 14.0% OMC = 14.0%

(a) (b) (c) (d)

FIGURE 4 Properties of pavement sections used for analysis: (a) PR-0805, Section D;
(b) PR-0805, Section E; (c) PR-0610, Section C; and (d) PR-0610, Section D.
(MDD  maximum dry density; MC  moisture content; OMC  optimum moisture content.)
146 Transportation Research Record 1990

The asphalt layer was modeled as a linear elastic material with a modified universal and k-θ models, respectively. Again the RMS
modulus of 5,500 MPa. This value was verified in the laboratory by value for the linear elastic model is the highest (RMS = 28.6%).
Steven (21). The longitudinal strains depict a similar tendency compared to
As mentioned, CAPTIF sections are provided with a set of coils transversal strains results. There is little difference among the
that are buried in the base course. Different types of sensors (vertical, nonlinear models, but the general inaccuracy is higher compared to
longitudinal, and transversal) were installed in each section at dif- transversal strains (maximum and minimum RMS values of 25.6%
ferent depths. For this research project the measurements were taken for the classical universal and 23.5% for the modified universal).
only between the wheels. Thus, the 3D-FEM calculations were Once again the linear elastic model gets a higher RMS value of
carried out at the same location of the sensors buried in the pavement. 53.0% RMS.
A detailed description of the position (depth from surface) and type The best match between predicted and measured responses was
of sensors used for this research is presented in Table 2. obtained for vertical stress calculations. The nonlinear models show
similar good results (RMS values close to 11%) and linear elastic
models reach the highest RMS of 27.6%. The vertical stresses from the
ANALYSIS OF RESULTS analysis in the finite element code and from CAPTIF measurements
for one section at different depths are illustrated in Figure 5.
For the evaluation of the different models, the ABS and RMS val- It is important to notice that the lowest general match between
ues were calculated to measure the overall performance of each the predicted and measured values was the vertical strains. One
model. Once again Equations 6 and 7 were used for this purpose, but possible reason could be the incorporation of a linear elastic sub-
for this analysis dmi is the measured value from the testing facility grade in the model instead of a more realistic nonlinear behavior. In
and dmc is the value calculated by using the nonlinear models. In addition, the assumption of a uniform tire contact pressure is not
addition, a linear elastic model was incorporated in the analysis. The representative as has been extensively studied (23). In contrast, the
detailed results of all the RMS values are presented in Table 3. The vertical stresses achieve the best match with measured values since
ABS values were not included since they showed the same trend as the stresses are not very sensitive to stiffness changes.
RMS results.
Figure 5 shows the results for predicted and measured vertical
strains in one pavement section at different depths from the surface. APPLICATION TO PAVEMENT DESIGN
From this figure it is possible to get an idea of a reasonably good pre-
diction achieved by the modified universal model (RMS = 24.5%) Two statistical parameters (RMS and ABS) have been presented
and a less accurate prediction achieved by the linear elastic model for the comparison of measured and calculated values in this research.
(RMS = 56.8%). These parameters are used to get an idea of the general accuracy of
A general better fitting for all the models was obtained for the pre- the nonlinear and linear models for the structural analysis of unbound
diction of the transversal strains. In this case, little difference was granular materials. However, the RMS and ABS values do not clearly
found between the average values of the nonlinear models with a show, in terms of load repetitions, the practical effect of using
maximum and minimum RMS value of 14.2% and 13.1% for the nonlinear models in the design of unbound granular pavements.

TABLE 2 Location of Strain and Stress Measurements

PR3-0805, Section D PR3-0805, Section E PR3-0610, Section C PR3-0610, Section D

Strain Stress Strain Stress Strain Stress Strain Stress

Depth
(mm) V T L V V T L V V T L V V T L V

75.0 X X X X X
100.0 X
112.5 X
125.0 X X
150.0 X X X
162.5 X X
187.5 X
200.0 X X
225.0 X X X X X
237.5 X
250.0 X
262.5 X
300.0 X X
312.5 X X

V = vertical, T = transversal, L = longitudinal

González, Saleh, and Ali 147

TABLE 3 Calculated RMS for All Cases

Calculated RMS (%)

Project Section Load (kN) Linear k-θ Classical Modified

Vertical strain
0805 D 40 56.8 32.7 42.3 24.5
0805 D 50 57.1 39.7 53.6 22.7
0805 E 40 55.5 35.5 46.3 24.9
0610 C 40 64.3 43.4 29.9 41.2
0610 C 60 46.5 25.6 36.8 12.8
0610 D 40 66.8 27.6 43.9 38.1
0610 D 60 65.3 16.1 19.9 20.1
Average 58.9 31.5 39.0 26.3
Transversal strain
0610 C 40 10.4 3.4 1.1 8.5
0610 C 60 21.0 9.8 6.5 14.4
0610 D 40 35.2 17.0 17.9 16.8
0610 D 60 47.7 22.3 29.3 17.2
Average 28.6 13.1 13.7 14.2
Longitudinal strain
0610 C 40 58.5 15.2 9.2 22.8
0610 C 60 41.9 46.6 53.3 39.2
0610 D 40 45.0 20.3 29.8 8.3
0610 D 60 66.5 14.7 10.0 23.7
Average 53.0 24.2 25.6 23.5
Vertical stress
0805 E 40 32.8 20.3 19.1 21.8
0610 C 40 21.0 15.1 18.8 10.1
0610 C 60 42.7 16.1 13.4 19.9
0610 D 40 22.5 2.0 4.7 1.6
0610 D 60 18.9 4.3 6.8 0.8
Average 27.6 11.6 12.6 10.9

To undertake this purpose, two simple low-volume pavement The number of maximum load repetitions for the granular layers
structures were analyzed for load repetitions by using the South was calculated by using the permanent deformation equations pro-
African mechanistic pavement design method (1). For simplicity, vided by the South African method. These equations use the max-
Section D of project PR-0805 and Section D of project PR-0610 imum (vertical) and minimum (horizontal) stresses in the middle of
were selected for the analysis, using the same properties adopted for the granular layer as the critical design parameters. (For this example,
the FEM modeling (Pavements 1 and 4 in Figure 4). the critical stresses were calculated under the center of wheels.)

1200 160

140
1000
120
Vertical Strain (x10-6)

Vertical Stress (kPa)

800
100

600 80

60
400
40
200
20

0 0
125.0 162.5 200.0 312.5 100 250
Depth (mm) Depth (mm)

CAPTIF Linear ktheta Classical Modified CAPTIF Linear ktheta Classical Modified

(a) (b)

FIGURE 5 Measured and calculated vertical strains and stresses for two sections: (a) PR-0805, Section D, and (b) PR-0610, Section C.
148 Transportation Research Record 1990

TABLE 4 Calculated Stresses in Middle of Base Course and Maximum

Load Cycles

PR-0805 Section D PR-0610 Section D

σmax σmin Number of σmax σmin Number of

(kPa) (kPa) Cycles (106) (kPa) (kPa) Cycles (106)

Linear 105.3 −20.8 23.5 71.8 −85.4 4.00

k-θ 118.8 −8.2 5.8 91.8 −19.8 16.4
Classic 120.3 −7.2 5.6 93.1 −18.9 15.9
Modified 116.7 −9.8 6.0 90.2 −20.9 17.0

NOTE: Tensile stresses have been considered negative in this table.

The failure criterion adopted by the authors for this example is a too sophisticated for practitioners, although some linear elastic soft-
permanent deformation of 10 mm on the top of the granular layer. ware (19) and pavement design methods such as AUSTROADS
The load used for the calculations of the number of repetitions is (24) have incorporated simple and practical tools to take into
a dual wheel load of 40 kN (contact pressure = 700 kPa with a sep- account the nonlinear behavior of these materials. However, some
aration of 350 mm). simple 3D-FEM nonlinear software especially designed for pave-
The nonlinear analysis was carried out by using the FEM models ments has been developed and appears to provide realistic and reli-
from CAPTIF. Also, the analysis was performed by using a layered able solutions for thin granular pavement structures (25).
linear elastic software (20). The results of the analysis are presented
in Table 4 and show that there is little difference between the non-
linear elastic models for both pavements in terms of maximum and ACKNOWLEDGMENTS
minimum stresses and hence in the number of maximum load repe-
titions allowed in the base course layer. However, the layered linear The authors acknowledge the help from Transit New Zealand for
elastic analysis shows higher tensile horizontal stresses, especially making CAPTIF data available and from David Alabaster and
for the PR-0610 pavement, probably because of the weak subgrade the CAPTIF staff of Transit New Zealand, Bruce Steven of the Uni-
underneath the base course layer. versity of Canterbury, and Sabine Werkmeister of the University
For maximum number of load cycles, the estimation for the linear of Canterbury.
elastic analysis for the PR-0805 pavement is approximately three times
higher than that of the nonlinear estimation. For PR-0610, the esti-
mation of the linear elastic analysis is about four times lower than REFERENCES
the nonlinear analysis. It is important to notice that in the calculation
of the number of repetitions, the Mohr’s circle has been shifted, 1. Theyse, H. L., M. De Beer, and F. C. Rust. Overview of South African
Mechanistic Pavement Design Method. In Transportation Research
following the South African approach to avoid tensile stresses in
Record 1539, TRB, National Research Council, Washington, D.C., 1996,
granular layers (1). pp. 6–17.
2. Hicks, R. G., and C. L. Monismith. Factors Influencing the Resilient
Response of Granular Materials. In Highway Research Record 345,
CONCLUSIONS HRB, National Research Council, Washington, D.C., 1971, pp. 15–31.
3. Lekarp, F., U. Isacsson, and A. Dawson. State of the Art I: Resilient
Response of Unbound Aggregates. Journal of Transportation Engineer-
The present work demonstrates the feasibility of using advanced ing, Vol. 126, No. 1, 2000, pp. 66–75.
structural analysis tools such as 3D-FEM software and nonlinear 4. Seed, H. B., F. G. Mitry, C. L. Monismith, and C. K. Chan. Prediction
elastic models for the analysis of thin-surfaced pavement structures of Flexible Pavement Deflections from Laboratory Repeated-Load Tests.
typically associated with low-volume roads. Results showed that National Research Council, Washington, D.C., 1967.
nonlinear models used in FEM are an accurate solution in this type 5. Dehlen, G. L., and C. L. Monismith. Effect of Nonlinear Response
Material Response on the Behavior of Pavements Under Traffic. In High-
of pavement structure.
way Research Record 310, HRB, National Research Council, Washington,
In general, the results obtained from the nonlinear models are con- D.C., 1970, pp. 1–16.
sidered good for the materials and structures studied in this research. 6. Sweere, G. T. H. Unbound Granular Bases for Roads. PhD thesis.
Little difference in the accuracy of the nonlinear models was found Technische University of Delft, Netherlands, 1993.
under this project. 7. Kolisoja, P. Simple Automatic Stress Path Control System for Triaxial
Testing. Proc., 13th International Conference on Soil Mechanics and
One important limitation of this study is that only one set of mea-
Foundation Engineering, Part 1, New Delhi, India, 1994.
surements from CAPTIF (pavement response between the wheels) was 8. May, R. W., and M. W. Witczak. Effective Granular Modulus to Model
used. However, this is one of the most critical locations to analyze Pavement Responses. In Transportation Research Record 810, TRB,
since most of the current pavement design methodologies recog- National Research Council, Washington, D.C., 1981, pp. 1–9.
nize this point as one of the most critical positions in the pavement 9. Uzan, J. Characterization of Granular Material. In Transportation
structure. Research Record 1022, TRB, National Research Council, Washington,
D.C., 1985, pp. 52–59.
It also was shown that for thin pavement structures it is important 10. Lade, P. V., and R. B. Nelson. Modeling the Elastic Behavior of Granular
to use 3D-FEM analysis to avoid the calculation of nonexistent ten- Materials. International Journal for Numerical and Analytical Methods
sile stresses in the granular layers. However, these methods are still in Geomechanics, Vol. 11, No. 5, 1987, pp. 521–542.
González, Saleh, and Ali 149

11. Development of the 2002 Guide for the Design of New and Rehabilitated In Transportation Research Record: Journal of the Transportation
Pavement Structures: Phase II. Project 1-37-A. TRB, National Research Board, No. 1823, Transportation Research Board of the National
Research Council, Washington, D.C., 1997. www4.trb/crp.nsf/All+ Academies, Washington, D.C., 2003, pp. 153–162.
Projects/NCHRP+1+−37A. Accessed July 2004. 18. Ali, A. Finite Element Modelling of Road Pavement to Address Nonlinear
12. Andrei, D., M. Witczak, C. Schwartz, and J. Uzan. Harmonized Behaviour of Unbound Materials Under Axle Loads. MS thesis. Univer-
Resilient Modulus Test Method for Unbound Pavement Materials. In sity of Canterbury, New Zealand, 2006.
Transportation Research Record: Journal of the Transportation Research 19. Everstress, Layered Elastic Pavement Analysis Program. Washington
Board, No. 1874, Transportation Research Board of the National Acad- State Department of Transportation, Olympia, 1999.
emies, Washington, D.C., 2004, pp. 29–37. 20. MePads, Layered Elastic Pavement Analysis Program. Council for
13. Werkmeister, S., R. Numrich, A. R. Dawson, and F. Wellner. Design of Scientific and Industrial Research, South Africa, 2001.
Granular Pavement Layers Considering Climatic Conditions. In Trans- 21. Steven, B. The Development and Verification of a Pavement Response
portation Research Record: Journal of the Transportation Research and Performance Model for Unbound Granular Pavements. PhD thesis.
Board, No. 1837, Transportation Research Board of the National Acad- University of Canterbury, New Zealand, 2005.
emies, Washington, D.C., 2003, pp. 61–70. 22. Vuong, B., and D. Hazell. Developments of Performance-Based Speci-
14. Loizos, A., and A. Scarpas. Verification of Falling Weight Deflectom- fications for Unbound Granular Materials. Australian Road Research
eter Backanalysis Using a Dynamic Finite Elements Simulation. Board, Vol. 12, No. 4, 2003, pp. 13–25.
International Journal of Pavement Engineering, Vol. 6, No. 2, 2005, 23. de Beer, M. Measurement of Tyre/Pavement Interface Stresses Under
pp. 115–123. Moving Wheel Loads. Heavy Vehicle Systems, Vol. 3, No. 1–4, 1996,
15. Jiang, J., S. Yun-kang, and Z. Zhi-qing. Analysis of Asphalt Pavement
pp. 97–115.
with Granular Base by Nonlinear Finite Element Method. Journal of
24. Pavement Design: A Guide to the Structural Design of Pavements.
Beijing University of Technology, Vol. 32, No. 2, 2006, pp. 182–186.
Austroads, Sydney, Australia, 2004.
16. Habiballah, T., and C. Chazallon. An Elastoplastic Model Based on
25. Oeser, M. Numeric Simulation of the Non-Linear Behaviour of Flexible
the Shakedown Concept for Flexible Pavements Unbound Granular
Materials. International Journal for Numerical and Analytical Methods Pavement Structures. PhD thesis. University of Technology, Dresden,
in Geomechanics, Vol. 29, No. 6, 2005, pp. 577–596. Germany, 2004.
17. Saleh, M. F., B. Steven, and D. Alabaster. Three-Dimensional Nonlinear
Finite Element Model for Simulating Pavement Response: Study at The Full-Scale and Accelerated Pavement Testing Committee sponsored publication
Canterbury Accelerated Pavement Testing Indoor Facility, New Zealand. of this paper.