Professional Documents
Culture Documents
Three-Dimensional Finite Element Analysis of Flexible Pavements
Three-Dimensional Finite Element Analysis of Flexible Pavements
OF FLEXIBLE PAVEMENTS
BY
Jia Wang
B.S. North Industry University, China, 1996
A THESIS
Submitted in Partial Fulfillment of the
Requirements for the Degree of
Master of Science
(in Civil Engineering)
Advisory Committee:
William G. Davids, Assistant Professor of Civil Engineering, Advisor
Eric N. Landis, Associate Professor of Civil Engineering
Vincent Caccese, Associate Professor of Mechanical Engineering
By Jia Wang
Thesis Advisor: Dr. William G. Davids
An Abstract of the Thesis Presented
in Partial Fulfillment of the Requirements for the
Degree of Master of Science
(in Civil Engineering)
May, 2001
Flexible pavements or roads surfaced with asphalt have been in use for the past 100
years. Currently, the design of flexible pavements is largely based on empirical methods.
However, there is currently a shift underway towards more mechanistic design
techniques. While layered elastic analysis and two-dimensional finite element (FE)
methods have been generally been used to deternline stresses, strains and displacements
in flexible pavements, they suffer several severe limitations. To overcome these
difficulties, three-dimensional (3D) FE analysis must be used to analyze pavement
structqres. This study focuses on exploring the use of 3D finite-element methods to
examine the response of flexible pavements.
For this study, an efficient 3D FE meshing tool was developed. This meshing tool
allows us to develop models of layered system, inter-layer debonding and slip, various
wheel and axle loadings.
The 3D FE models were tested by comparing predicted results with experimentally
measured field data. Critical finite-element model dimensions were determined, and
ACKNOWLEDGEMENTS
I would first like to acknowledge the sponsors of this project. Financial support was
provided by the Washington State Department of Transportation and the University of
Maine.
Special thanks to my advisor, Dr. William Davids, for accepting me into the master's
program and providing direction for this project. His time and effort with the teaching,
thesis corrections, and many meetings despite his increasingly busy schedule are greatly
appreciated. Dr. Davids was very helpful in offering guidance, encouragement, and
recommendations for my scholastic and professional interests as well.
Special thanks go to members of my advisory committee: Prof. Eric Landis, Prof.
Vincent Caccese. They were always available to answer my questions and comment on
my work. I also received much guidance on this project from Prof. Dana Hamphrey.
Finally, I would like to my family and friends for their encouragement and
understanding throughout my graduate career.
TABLE OF CONTENTS
..
ACKNOWLEDGEMENTS...................................................................11
LIST OF TABLES...............................................................................vi
LIST OF FIGURES ............................................................................. vii..
CHAPTER 1. INTRODUCTION............................................................ 1
1.1 Background............................................................................. 1
1.2 Objectives of Study .....................................................................3
1.3 Organization of the Thesis............................................................. 4
CHAPTER 2 . LITERATURE REVIEW............................................................................6
2.1 Introduction...................................................................................................................... 6
2.2 Layered Elastic Analysis ................................................................................................. 7
2.2.1 Two-layered System..........................................................................................8
2.2.2 Examples of Layered Analysis Programs........................................................9
2.2.3 Back-calculation Using Layered Elastic Analysis..................................... 10
2.2.4 Advantages and Disadvantages ..................................................
11
2.3 Two-dimensional Finite Element Methods......................................... 12
2.3.1 Axis-symmetric vs . Plane Strain ..................................................
13
2.3.2 Currently Available 2D FE Models for Flexible Pavement Analysis .......14
2.3.2.1 ILLI-PAVE Model ...................................................... 14
2.3.2.2 MICH-PAVE Model ...................................................
14
2.3.2.3 Commercially available FE programs ...............................15
2.3.3 Advantages and Disadvantages of 2D FE Analysis ............................15
2.4 Three-dimensional (3D) Finite Element Methods................................. 16
2.4.1 Cho et a1 (1 996).................................................................... 17
2.4.2 Hsien H . Chen, Hurt M . Marshek, and Chhote L . Saraf (1990) ........ 17
2.4.3 K. D . Hjelmstad, J . Kim and Q. H . Zuo (1997)..............................................18
2.4.4 Sam Helwany, John Dyer, and Joe Leidy (1998) ...............................18
2.4.5 Samir N . Shoukry and Gergis W . William (1999) ..............................19
2.4.6 A Bensalem, A . J . Brown, M.E. Nunn, D . B. Merrill,
and Wyn G. Lloyd (2000)................................................
20
2.4.7 Ronald Blab, and John T. Harvey (2000) ........................................ 20
2.5 Summary................................................................................ 21
LIST OF TABLES
Table 4.1
Table 5.1
Table 5.2
Table 5.3
Table 5.4
Table 6.1
Table 6.2
LIST OF FIGURES
Figure 1.1
Figure 2.1
Figure 2.2
Figure 3.1
Figure 3.2
Figure 3.3
Figure 3.4
Figure 3.5
Figure 3.6
Figure 3.7
32
Building Grid Sort Element Block .................................................
Figure 3.8
Figure 3.9
35
A 3D View of Fully Merged Mesh .................................................
Figure 3.10
Figure 3.1 1
38
Plan View of Merged Meshes .......................................................
Figure 4.1
Figure 4.2
Figure 4.3
Figure 4.4
Figure 4.5
Figure 4.6
Figure 4.7
Figure 4.8
Figure 4.9
Figure 4.18 The Best Pair of E2 and E3. for z = 1940 mm .....................................60
60
Figure 4.19 Displacement Basin Comparison. z = 1940mm...................................
Figure 4.20 Principal Stress Contour at the bottom of the AC with E2 = 46 MPa.
E3 = 28 MPa for North Yarrnouth Control Section. z = 1940mm............61
Figure 4.21
Figure 4.25 Principal Stress Contour at the Bottom of the AC with E2 = 49 MPa.
E3 = 76 MPa for TWP-3 1MD Control Section. z = 2240mm .................64
...
Vlll
Figure 5.1
Figure 5.2
Figure 5.3
Figure 5.4
Figure 5.5
Figure 5.6
Figure 5.7
Figure 5.8
Figure 5.9
79
Figure 5.10 Displacement Basin Comparison. Mid-layer Method ...................................
Figure 5.1 1 The Best Pair of KI and K2 in the Least Square Sense. Integration
Point Method............................................................................................... 80
Figure 5.12 Displacement Basin Comparison. K I = 10 MPa and K2 = 0.78, Integration
Point Method................................................................................................ 81
Figure 5.13 Stresses Contour at the Bottom of the AC with KI = 1OMPa. K2 = 0.78
for North Yarmouth Control Section. Integral Points Method .....................81
Figure 5.14 Maximum Principal Stresses at the Bottom of the AC with K1 =
30 MPa. K2 = 0.53 for North Yarmouth Control Section. Mid-layer
Method................................................................................................................
82
Figure 6.1
Figure 6.2
Figure 6.3
Figure 6.4
Figure 6.5
Figure 6.6
Integration Domain...................................................................................... 91
Figure 6.7
Figure 6.8
Figure 6.9
. .
. .
102
Figure 6.18 Model Descnption. Plan View......................................................................
Figure 6.19 Model Description. Elevation View............................................................103
Figure. 6.20 Max . Tensile Stresses at bottom of AC (thickness = 75mm. E2 =
35MPa).....................................................................................104
Figure. 6.21
Figure. 6.22
Figure. 6.23
Figure. 6.24
Figure. 6.25
Figure. 6.26
Figure. 6.27
Figure. 6.28
Figure 6.29
Figure 6.30
Figure 6.3 1
Figure 6.32
Figure 6.33
Figure 6.34
Figure 6.35
Figure 6.36
Figure 6.37
xii
CHAPTER 1. INTRODUCTION
1.1 Background
The first asphalt road was constructed in the US about 100 years ago in New Jersey.
There are currently about 2.2 million miles of roadway surfaced by asphalt concrete
pavements (Huang, 1993). Flexible pavements are made up of bituminous and granular
materials. A typical flexible pavement section can be idealized as a multi-layered system
consisting of asphalt layers resting on soil layers having different material properties as
shown in Figure 1.1.
Asphalt Concrete
Sub-base (often gravel)
Sub-grade (existing soil)
In general, there are three approaches that can be used to compute the stresses and
strains in pavement structures: layered elastic methods, two-dimensional (2D) finite
element modeling, and three-dimensional (3D) finite element modeling.
The layered elastic approach is the most popular and easily understood procedure. In
this method, the system is divided into an arbitrary number of horizontal layers (Vokas et
al. 1985). The thickness of each individual layer and material properties may vary from
one layer to the next, but in any one layer the material is assumed to be homogeneous and
linearly elastic. Although the layered elastic method is more easily implemented than
finite element methods, it still has severe limitations: materials must be homogenous and
linearly elastic within each layer, and the wheel loads applied on the surface must be
axi-symmetric. Those shortcomings make it difficult to simulate realistic scenarios. For
example, it is very hard to rationally accommodate material non-linearity and incorporate
spatially varying tire contact pressures, which can significantly affect the behavior of the
pavement systems (de Beer et al. 1997; Bensalem et al, 2000).
For 2D finite element analysis, plane strain or axis-symmetric conditions are
generally assumed. Compared to the layered elastic method, the practical applications of
this method are greater, as it can rigorously handle material anisotropy, material nonlinearity, and a variety of boundary conditions (Zienkiewicz and Taylor, 1988).
Unfortunately, 2D models can not accurately capture non-uniform tire contact pressure
and multiple wheel loads.
To overcome the limitations inherent in 2D modeling approaches, 3D finite element
models are becoming more widespread. With 3D FE analysis, we can study the response
of flexible pavements under spatially varying tirefpavement contact pressures, capture the
effect of non-linear materials or the effect of combination of loads, including unsymmetric or different loading types. However 3D FE models can be very difficult and
time-consuming to use. In particular, problems that often arise when using 3D FE
analysis include:
Choice of element type, how to form a individual mesh, how to merge smaller meshes
together to form a complete, locally refined mesh, including constraints, and other topics
will be explored.
Chapter 4 covers general issues regarding modeling of the flexible pavements.
Critical dimensions in three directions will be determined from parametric studies and
the model will be verified through conlparison with experimentally measured data.
Chapter 5 discusses the effect of material non-linearity in the sub-base. First,
background on and implementation of the K-0 material model will be reviewed. Next,
issues regarding convergence will be studied. Finally, the best combination of non-linear
model parameters will be back-calculated by comparing with the same experimentally
measured data used in Chapter 4.
Chapter 6 focuses on the spatially varying tirelpavement contact prssures and their
effects on the response of flexible pavement structures. A method for capturing tirepavement contact pressure variations will be developed based on published data (de Beer
et al, 1997). Parametric studies will be performed to examine the effects of load, tire
pressure, asphalt thickness, and sub-base stiffness on pavement response.
Chapter 7 summarizes this study and presents conclusions and recommendations for
future research.
2.1 Introduction
In United States, the first asphalt roadway constructed of bituminous and granular
materials was built in the late 1800's. From the very beginning of flexible pavement
construction history to early 1900's, experience dominated pavement design and
construction. Through this experience gained over the years, many design methods were
developed for detenning some critical features, like thickness of the asphalt surface. As
of 1990, there were millions of miles of paved road in US, 94% of which are topped by
asphalt (Huang, 1993). A typical flexible pavement cross section is shown in Figure 2.1.
r Wheel Load
Asphalt Concrete
Base
Sub-grade
Three types of flexible pavement constructions have been used: conventional flexible
pavement, full-depth asphalt pavement, and contained rock asphalt mat (CRAM). As
knowledge increased and other technologies developed, a composite pavement made up
of hot mix asphalt concrete (HMA) and portland cement concrete (PCC, beneath the
HMA) came into being with the most desirable characteristics. However, the CRAM
construction is still relatively rare and composite pavement is very expensive, and hence
seldom used in practice (Huang, 1993).
As mentioned in Chapter 1, various empirical methods have been developed for
analyzing flexible pavement structures. Due to limitations of analytical tools developed
in the 1960's and 19707s,the design of flexible pavements is still largely empiricallybased. The empirical method limits itself to a certain set of environmental and material
conditions (Huang, 1993). If the conditions change, the design is no loner valid. The
mechanistic-empirical method relates some input, such as wheel loads to some output,
such as stress or strain. The mechanistic method is more reliable for extrapolation from
measured data than empirical methods. However, the effectiveness of any mechanistic
design method relies on the accuracy of the predicted stresses and strains. Due to their
flexibility and power, three-dimensional (3D) finite element methods are increasingly
being used to analyze flexible pavements.
This chapter will review in detail the background and use of three major approaches
to the structural modeling of flexible pavements: layered elastic analysis, and twodimensional (2D) finite element and 3D finite element methods.
homogeneous and linearly elastic with an elastic modulus, E and a Poisson's ratio, v.
Figure 2.2 shows a 3-layered system.
To apply the layered elastic method, assumptions need to be made, such as:
1. Each layer is homogeneous and linearly elastic with a finite thickness hi.
system with any arbitrary number of layers. The stresses in a two-layered system depend
on the modulus ratio EI/E2, and thickness-radius ratio hl/a (a is the radius of loading
area). Vertical surface deflections for two layered systems and the critical tensile strain
can be computed via analysis charts.
The case of a two-layered system is the full-depth construction in which a thick layer
of hot mixed asphalt (HMA) is placed on the sub-grade. This type of construction is
quite popular in areas where good local sub-base materials are not available. The threelayered or multiple-layered system employs one or more sub-base layers. The basic
assumptions and construction of three-layered system are very similar to the two-layered
system. Three-dimensional strains can be determined from analysis charts, such as those
found in (Huang, 1993). Using superposition, the analysis of multiple layered systems
can incorporate various loading cases in terms of different number of wheel loads, such
as single wheel, dual wheels, and dual-tandem wheels.
KENLAYER can be applied to layered systems under single, dual, dual-tandem wheel
loads with each layer's material properties being linearly elastic, non-linearly elastic or
visco-elastic. To deal with material non-linearity, KENLAYER divides the granular base
layer into a number of sub-layers and the stresses at the mid-height are used to compute
the modulus of each layer. Based on the computed stresses, a new elastic modulus is
determined; this iterative process is repeated until the modulus converges.
The DAMA computer program can be used to analyze a multiple-layered elastic
pavement structure under a single- or dual-wheel load. The number of layers can not
exceed five. In DAMA, the sub-grade and the asphalt layers are considered to be
linearly elastic and the untreated sub-base to be non-linear. Instead of using iterative
method to determine the modulus of granular layer, the effect of stress dependency
(Hwang and Witczak, 1981) is included by an effective elastic modulus computed
according to in Eq. 2.1 :
-0 139 0287
E2 = 10.447hl-0.471 h2-0041
. El . E3. K ; . ~ ~ ~
(2.1)
where, El, E2, E3, are the modulus of asphalt layer, granular base, and sub-grade,
respectively; hl, h2 are the thickness of the asphalt layer and granular base. Kl, and Kz
are parameters for K-8 model with K2 = 0.5.
2.2.3 Back-calculation Using Layered Elastic Analysis
The inverse problem of determining material properties of a flexible pavement
structures from its response to surface loading (often called back-calculation) is a
common application for layered analysis method. In back-calculation, iterative schemes
are employed where the material properties are varied until the theoretical solutions
match the measured deflections. Since Poisson's ratio of pavement layers do not affect
results by much, generally only layer moduli are varied (Maestas, et a1 1994).
The PC-compatible computer programs, including ELSDEF (Bush, 1980),
MODCOMP2 (Irwin 1983), MODULUS4 (Scullion et al, 1990) and ISSEM4
(Stubstad 1988) were used to perform the back-calculation study by Maestas et al.
(1994). Deflections measured from falling weight deflectometer (FWD) field were used
to approximate layer moduli of all pavement sections. A total of 29 flexible pavement
sections were chose from different traffic loading and sub-grade soil types in Arizona.
ELSDEF uses a linear elastic analysis technique to perform the back-calculation. It
updates the layer moduli iteratively until the computed deflections match the measured
ones within an error tolerance. Similarly, MODCOMP2, MODULUS4 and ISSEM4
(Maestas, et a1 1994) all use iterative strategies when performing the back-calculation.
The results from various back-calculation programs are not significantly different except
for the sub-grade. Maestas et al. (1994) state that back-calculation is not reliable for the
use in routine design. Further research is needed to refine the process to reduce the
effects of extraneous errors.
entire layer. This assumption makes it difficult to analyze layered systems consisting of
non-linear granular materials, such as untreated sub-bases and sub-grade. This difficulty
can be overcome by using the finite element method. Secondly, all wheel loads applied
on the top of the asphalt concrete have to be axi-symmetric, which is not true for actual
wheel loads. Limitations like these do not allow layered elastic analysis to capture the
response of pavement systems and account for all aspects of behavior.
(2.2)
where, E is the strain vector; S is a suitable linear operator and U is nodal displacement
vector. The element stiffness matrices are computed using:
dependent K-8 model to characterize the resilient modulus of soils through a stressdependent n~odulusand constant Poisson's ratio.
The comparison of results from MICH-PAVE and those from ILLI-PAVE illustrates
that MICH-PAVE yields more reasonable outcome than the latter with measured data
(Chen et al, 1995).
strain model is limited to the use of line loads and the axis-symmetric modeling approach
models wheel truck loads as circular. Many investigations (Weissman, 1997; Cunagin
et al, 1991) have illustrated that the contact area between the pavements and tire is
essentially rectangular, and that titelpavement contact stresses can vary significantly over
the tirelpavement contact area (Tielking et al, 1995).
3D finite element methods are increasingly being used to model the response of flexible
pavements. 3D finite element modeling is increasingly viewed as the best approach to
understand pavement performance. A pavement system is typically modeled as a multilayered structure with different material properties in each layer. Interface elements or
springs can be used to transfer the shear between layers, and well-controlled boundary
conditions are critical to analyze behavior of entire pavement system (Blab et al, 2000).
While 3D modeling method can generate more realistic results than 2D modeling, it
generally requires more intensive pre-processing procedures. Further, the number of
elements increases, the total computational time and the amount of memory will increase
tremendously. Fortunately, as computer industry develops exponentially and appearance
of high-performance algorithms, more elements can be used and accordingly the results
will be more and more accurate.
Further, a number of researchers have shown that spatially varying contact pressures
between the tire and pavement can significantly affect response (Tielking et al, 1995;
Weissman 1997). 2D models cannot capture this effect.
elastic and visco-elastic. The 3D FE computer program NIKE3D was used for this
analysis. The results indicate that measured tensile stains at the bottom of the AC
layer are very sensitive to vehicle speed and only longitudinal strains at the top and the
bottom of the AC layer were sensitive to tire pressure. It was found that FE modeling of
modeling of flexible pavements, if validated, can be very useful, because it can be used to
predict primary response parameters without resorting to field experiments, which may
be costly. Further, 3D FE analysis has some advantages such as: it may substitute for
full-scale testing; the analysis may be used to form the basis for generalized mechanistic
design procedures; and the analysis may used to verify results from simpler 2D analyses.
authors tried to predict the probability that failures will occur. It was found that the most
common circular load model with uniform vertical stress equal to the tire inflation
pressure is not sufficient to account for distresses occurring close to the surface. The
VRSPTA directly measures and quantifies the contact stresses between the passing tire an
pavement surface, therefore, it is more realistic. The described FE model validation was
performed using linear elastic layer theory. The graphic results illustrate that the FE
model gives very reasonable results. However$ requires that complex constitutive
relationships be used to accurately predict the rutting test results.
2.5 Summary
Significant progress in both modeling strategies has been made in past years. In this
Chapter, theoretical background, advantages and limitations of layered elastic model, 2D
FE methods and 3D FE methods were reviewed. Layered elastic models can not
accurately capture non-uniform tire contact pressure and multiple-wheel loads and have
difficulty incorporating material non-linearity. 2D finite element models do not allow the
realistic modeling of applied loads. To overcome these problems, 3D FE models have
been developed.
Although a nunlber of studies employing 3D finite element analysis have been
performed, there are some import topics that need further research:
The automatic and efficient generation of locally refined meshes for 3D FE
analysis of flexible pavement systems.
Verify that the models can adequately predict measured response. This must
include studies on the interaction between adjacent wheel loads and an examination
of required model extents.
3.1 Introduction
As mentioned in chapter 2, both two-dimensional (2D) and three- dimensional (3D)
methods can be employed to capture the structural response of flexible pavements. Many
researchers have investigated the response of flexible pavements and 2D programs have
been developed specifically for flexible pavements such as MICHPAVE and
ILLI-PAVE. However, 2D analysis cannot accommodate non-uniform tire contact
pressure and multiple-wheel loads; 3D finite element models must be used to properly
capture these effects.
However, in order to solve the relatively large 3D problems on the desktop
computers, the meshing code has to be efficient and automatic. This chapter focuses on
the development of effective meshing tool. The meshing code was developed using the
object-oriented programming language, C++.
First of all, an overview of the finite element models will be presented, where
issues such as localized mesh refinement, model size, and boundary conditions will be
discussed. A brief discussion of element types and implementation will be presented.
Secondly, the overall procedure for generating the meshes will be covered. This will
be followed by a discussion of the specific steps taken to efficiently generate and
combine individual portions of the meshes. Finally, a brief summary concludes the
chapter.
I-
Wheel Load
X
Surface
nterface
terface
Sub-grade
The finite element pavement structure used in this study is made up of three layers,
including asphalt concrete (AC) layer, sub-base and sub-grade layers. Each layer of the
model has different material properties, and is of uniform thickness.
'
conditions, model size, material properties and so on) the same, the case with a relatively
large number of FE elements will yield more accurate results than the one with less FE
elements. Yet, the more FE elements used, the more computer resources required.
Therefore, it is effective to refine the region around the wheel loads and use fewer, larger
FE elements away fiom the load.
Because the layers of pavement systems consist of different materials that may be unbonded, we would like to model slip between layers. To accommodate this need, two
interface layers (one between AC and sub-base, the other between sub-base and subgrade) were introduced to capture slippage. Basically, two possibilities, fully bonded and
un-bonded will be considered. For the fully bonded case, nodes on the surface of base
layer are perfectly bonded to nodes at the bottom of AC layer (i.e. the matched nodes
have same displacement in all three (x, y, z ) directions). For the un-bonded case, the
matched nodes will be allowed to separate under tension. Therefore, the interface layer
will be tensionless, and have stiffness constants Kx and Ky (x and y direction stiffness per
unit area, respectively) to capture shear transfer between layers.
There are several other types of brick elements that could be used, including 8nodded linear elements, 27-noded quadratic elements, 32-noded cubic elements and so
on. In general, the more nodes per element, the more accurate the element, but the more
difficult mesh generation and application becomes. The 20-noded brick element achieves
a balance between the difficulty of meshing and accuracy of results. Each model layer is
meshed with 20-noded brick elements; a single 20-noded brick element is shown in
Figure 3.3.
axis directions corresponding to global coordinates system. Figure 3.5 illustrates how
coordinates and local connectivity will be computed. Note that while curved edges or
faces are not considered, the 6-sided region does not need to be a rectangular prism.
NY
Figure 3.5 Brick Element Mesh
The nodes are numbered first along the z axis and then they axis and eventually the x
direction. The interface elements can be meshed in the same way (first go along the zaxis, y-axis and x-axis in that order). The order of brick element mesh in plan will
determine how the interface elements are meshed.
merge adjacent meshes. For example, two meshes can be merged together to form a
bigger mesh, then a 3rdblock of elements can be merged to form an even larger mesh,
etc., until all individual meshes are merged together. If distance between any two nodes
is effectively zero, these two nodes take the same number. For simplicity, a 2D example
can be used here to illustrate the situation as shown in Figure 3.6. The connectivity of
the remaining, unmatched nodes on the second mesh are added to new whole mesh
sequentially.
After Merge
Figure 3.6 Merging the Meshes
mesh
nodes instead of thousand of them (of course, the number of nodes in each grid-sort
element depends on total number grid sort elements).
When another mesh (called mesh 11) needs to be merged to the existing one (mesh I),
all nodes on the mesh I1 will be looped over. For each node in mesh 11, it is easy to figure
out which grid sort element it belongs to by checking the coordinates nodal coordinates
against those of the grid sort cells. Once we have done that, we can compute the distance
between this node (on mesh 11) to all nodes only in that grid sort element which belong to
mesh I (see Figure 3.8).
Mesh I
For example, it is obvious that node 1 on mesh I1 is in the grid element 6 and mesh I
only contributes nodes 14, 17 and 18 to grid element 6. Therefore, we can calculate the
distance between node 1 and nodes 14, 17, and 18 as opposed to looping sequentially
over all of the nodes in mesh I. After repeating the process of checking by corresponding
grid cell and tracking the matching nodes, it is simple to re-number the nodes in mesh I1
and re-define the element connectivities. The small example shown here does not seem
to save much time, but if the number of total nodes are about several thousand, the run
time will be decreased tremendously as shown in the next section. The algorithm of grid
sorting procedure is shown as follows:
Xmin,Y m i Zmin
, = minimum x, y, z coordinates of mesh - 1 respectively
X,, Y,, Z , = maximum x, y, z coordinates of mesh * 1.05 respectively
Ndr, N+ Nd, = number of divisions of grid in x, y, z directions
1 get [ X m , Ymm,
x m i n , ymin, Zmin];
2 6~ = (Xmm- Xmjn)1 Ndx;
3 6y = (Y- - Ymjn) I N4;
4 6~ = (Zmm - Zmjn) 1 Ndr
5 build-sort ( )
6 loop over all grid sort elements
7 for i = 1 to number of elements (Nm* N&,* Nm )
store 8 connectivities and coordinates of each element in order;
8
9 loop over all nodes in mesh I
10 for j = 1 to number of nodes
11
store node number in list of each grid-sort element;
12 End
13 End
14 Loop over all solid elements
15 for i = 1 to number of elements
16 loop over all nodes in mesh 11
17
for j = 1 to number of nodes
determine which grid-sort element the node lies in;
18
loop over all nodes in that grid-sort element;
19
20
for k = 1 to number;
21
find a match, two nodes become one;
22
End
23
End
24 End
25 re-number all other nodes in mesh 11;
26 END
,
$
I
I I
I :
i i
I :
/
...'
.',..'"
...-.--
,.-'
I:
I
1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0
Tim t o m e model (seaxrls)
3.6 Implementation
3.6.1 Typical Procedure for Generating a Layered Pavement System
All model layers have exactly the same configuration or plane dimensions. In
general, each portion of an individual layer will be meshed first and then merged together
to fornl the entire layer. Once all model layers and interface layers are meshed, they will
be merged together again to form whole pavement system. The algorithm is shown as
following:
Loop over all member layers
for i = 1 to number of layers
mesh portion (Inear
) the wheel loads;
mesh portions of the s a l e layer in far field and then merge to portion 1;
End
Loop over 2 interface layers
for i = 1 to 2;
mesh portion (I-inter) near the wheel loads;
mesh portions of the same layer in far field and then merge to portion I-inter;
End
Merge the interface layers to pavement member layers;
END
N , and ti, represent number of sub-layers and thickness of the ithmember layer;
The plan view is shown in Figure 3.1 1
Plan View
All control parameters are basically the same for all possibilities except the loading
part, which deals with different traffic loads, for example, to handle circular patch loads,
the solver needs know the diameter of this circular area, etc.
3.7 Summary
In this chapter, an overview of modeling strategies and element types necessary for
the effective 3D finite element modeling of pavement systems were presented. Some
topics related to mesh generation were covered as well, including local refinement of the
region around wheel loads, modeling interface slippage, and incorporation of finite
boundaries.
Basically, each model layer forms a block of the mesh. Individual layer and interface
layer meshes are merged to other meshes of their own type to generate a bigger mesh.
These meshes of brick elements or interface elements are then merged together to form
the entire pavement system. The interface layers can allow slip between layers with
horizontal shear transfer; by specifying different shear stiffness constants for the interface
elements, we can simulate different degrees of slip.
Two types of nodal constraints are considered at layer interface. For the bonded
case, the matched nodes are fully bonded together so that they have the same
displacement in three directions; for the un-bonded case, the matched nodes separate
under tension in the vertical direction, and can slip in the horizontal plane.
To assist in generating the mesh efficiently, a grid sorting algorithm was developed,
and shown to be very effective using a numerical example.
740mnl
Sub-grade
Plan View
Elevation View
Layer
As~haltConcrete
Base
Sub-grade
Dimensions(mm)
125
635
740
E (MPa), v
1400.0.35
50,0.29
20,0.29
Figures 4.4 - 4.6 show the maximum principal stresses at the bottom of the asphalt
concrete under a single dual wheel load (loading case I), single axle load (loading case I
+ loading caseIII), and dual axle load (loading case I-IV) respectively.
Figure 4.4 Maximum Principal Stress at the bottom of AC (single dual wheel,
loading case I)
Figure 4.5 Maximum Principal Stress at the bottom of AC (single axle, loading case
I+II)
45
Figure 4.6 Maximum Principal Stress at the bottom of AC (dual axle, loading case IIV)
The maximum principal tensile stresses at the bottom of AC due to all single dualwheel, single axle and all 8 wheel loads are 0.8865 MPa, 0.8569 MPa and 0.8536 MPa,
respectively. The largest difference among them is 3.85%. Clearly, it is reasonable and
accurate enough that we study the response of pavement system under single dual-wheel
loads.
Asphalt Concrete
E2 = SOMPa, v = 0.29
Base
Sub-grade
faces will be restricted ftom moving in those directions which are perpendicular to the
faces. For simplicity, the model will be square.
Due to the bi-axial symmetry of the loading and boundary conditions, we can use
a quarter-symmetric mesh. Figures 4.8,4.9 and 4.10 show the finite element model and
mesh used here. Each layer will be assumed perfectly bonded for the purpose of
determining critical dimensions.
Layer 3 Subgrade
Lz restraint
\
x restraint
layers. The loading area is assumed to be a rectangular region and no moving loads are
taken into consideration.
y restraint
restraint
4 v'
restraint
the critical plan dimensions have been determined, the vertical dimension will be varied
to determine its effect on stresses and displacements. The pavement will be modeled as a
square region and the longitudinal (x) and transverse (y) dimensions will be varied
simultaneously. Figure 4.1 1 shows the effect of plan dimensions on the maximum tensile
stresses in the asphalt concrete for different values of the sub-grade depth, z.
Plan Dimensions(mm)
Clearly, Figure 4.1 1 illustrates that the stresses converge to a constant value and that
the critical dimensions in plan are about 2000 mm X 2000 mm irrespective of z. The
change in tensile stress in asphalt concrete layer for plan extents beyond 2000 mm X
2000 mm is negligible. Interestingly, stresses peaked for z = 740 mm and then went
down slightly when the sub-grade vertical dimension was increased to 1640 mm.
However, the effect of z on the tensile stress is quite small. Based on the results, it seems
reasonable to fix the quarter-symmetric model dimensions at 2000 mm X 2000 mm in
plan and keep z larger than or equal to 740 mm. Equivalently, if a quarter symmetric
model is not used, we could employ a model with plan dimensions 4000 mm X 4000 mm.
be selected. The only variable here remains unknown is the depth of sub-grade, z. For
different values of z, different pairs of Ez, E3 (nlodulus of elasticity of sub-base, subgrade, respectively) can be computed to give a good fit to the experimentally measured
deflections. In this section, we will examine the effect of varying z, and back-calculating
different E2, E3 combinations to give good agreement. From previous studies, z =
740 mm is the critical vertical dimension, and then z will be varied from 1040mm, 1940
mm to 2040 mm. In general, z can be any value larger than or equal to 740 mm. For the
North Yarmouth control section, we pick up z = 740 mm (total depth = 1500 mm, comes
from previous work done in section 4.2) and z = 1940 mm (total depth = 2700 mm). For
the TWP3 1-MD control section, models with z = 1040 mm (total depth = 1800 mm) and z
= 2240 mm
(total depth = 3000 rnrn) are employed, since the thicker the asphalt concrete
layer, the thicker the base and sub-grade layer or bigger difference between the two
models are expected.
The boundary conditions and loading cases will be the same as in section 4.2. A
quarter-symmetric mesh will be used. All corresponding nodes between layers will be
fully bonded (no slip) and the soil will be treated as linearly elastic. No self-weight of any
layer is taken into consideration.
Wheel load
Wheel load
* A
North Yarmouth
point attached to a reference arm resting in back of the influence of the load. The
maximum deflection underneath the dual wheels is taken by the wheel probe mounted 16
A grid system encompassing eight points per 100 fl test section was laid out
for each project.
2.
A single rear axle dump truck loaded to an 18 kip rear axle load was
positioned on the desired location and the wheel probe inserted between the
dual rear tires.
3.
Initial readings were taken for the gages (for our use only).
4.
The truck was driven away and the pavement was permitted to rebound for
about 1 minute.
5.
Final dial gage readings were taken and the BB was moved to the next point.
According to Figure 4.14, the rebound of the pavement directly under the wheel load,
A , is computed using:
A = +dial
- dial,)
where dial, dialo are the final and initial dial gage readings respectively and A is the
corrected deflection value.
(4-2)
-1
Rear of
DY/k
Figure 4.13 Plan View of Location of Benkelman Beam Probe between Dual Truck
Tires
Height
A d justmen t
Note:
N o t t o scole
A formal optimization algorithm was not used for this procedure; more simply, 0 was
computed for different combinations of E2 and E3,and the pair that gave the best 0 was
selected. Figure 4.15 shows selected results from the simulations. The best fit to the
experimental data was seen with E2 = 67 MPa and E3 = 10 MPa. Figure 4.16 shows good
agreement between predicted and measured displacen~ents.Figure 4.1 7 shows the
maximum principal stresses at the bottom of the asphalt concrete layer. The largest
maximum principlestress is 0.836 MPa.
'f,
15
E2 =67 MPa
E2 = 100 MPa
20
25
30
35
40
45
Sub-grade Modulus of Elasticity in soil, E3 (MPa)
Figure 4.17 Principal Stress Contour at the bottom of the AC with E2 = 67 MPa, E3
= 10 MPa for North Yarmouth Control Section, z = 740 mm
Figures 4.1 8 and 4.19 present the results of the back-calculations of resilient moduli
that give the best comparison with the experimentally measured data in the least square
sense with z = 1940 rnrn. The best combination is E2 = 46 MPa and E3 = 28 MPa; and the
corresponding stress contour at the bottom of asphalt concrete is given in Figure 4.18.
The nmximum principal stress is 0.936 MPa, which is about 10.7% larger than the
0.836 MPa determined with z = 740 mm.
Figure 4.20 Principal Stress Contour at the bottom of the AC with E2 = 46 MPa, E3
= 28 MPa for North Yarmouth Control Section, z = 1940mm
4.4.4 Back-calculation for TWP31MD Pavement Control Section
As shown in Figure 4.12, the control section of TWP3 1-MD has a thicker asphalt
concrete layer. Loads, boundary conditions, and the meshes remain unchanged. The
same procedure will be repeated as done for the control section of North Yarrnouth.
Here, two models with different vertical dimensions z = 1040 mm and z = 2240 rnm
(total depth 1800 mm, 3000 rnm respectively) will be examined. The best pair of
Young's nlodulus, E2 and Ej ,and displacement basins and stress contours for the two
models are illustrated in Figures 4.21-4.23 and Figures 4.24-4.25.
I*
1
:
.......*.......
- ".....
0.8 -
.-
0.6
0.5
E2 = 10 MPa
E2=70MPa
E2=100Mf'a
0.9
0.7
k...
.....
"*-
............
.....'%.................
*-.- *.-- *
................
*.
............. .....................
...................
.*....................
600
I /
&measured
data
%E2 = 68MPa, E3 = 30MPa
1000
1400
1800
Distance(mm)
Figure 4.23 Principal Stress Contour at the bottom of the AC with E2= 68 MPa, E3
= 30 MPa for TWP31MD Control Section, z = 1040 mm
Figure 4.25 Principal Stress Contour at the Bottom of the AC with E2 = 49 MPa, E3
= 76 MPa for TWP31MD Control Section, z = 2240mm
Two models with different sub-grade depth, z, were taken into considerations for
TWP-3 1MD. The best sub-base and sub-grade moduli were back-calculated to give the
best comparison with the experimentally measured deflections. The largest maximum
tensile stress at the bottom of asphalt concrete is 0.3968 MPa the model with z = 1040
and 0.372 MPa for the model with z = 2240 mm. The error between them is about 7.5%
and illustrates that with all other conditions the same, the finite element model can give
good comparisons with experinlentid results. In particular, as long as a reasonable value
of z (sub-grade depth) is employed, soil properties can be back-calculated to fit the
displacements. Varying z does not significantly affect the stresses.
4.5 Summary
First, the effect of single dual-wheel load, single axle and dual wheel tandem axle
loads was discussed. The results presented in Section 4.1 illustrate that it is reasonable to
build a model around a single dual-wheel load. Secondly, the effective model size (2000
rnrn X 2000 mm X 1500 mm for a quarter symmetric model) was determined on the basis
E=~ , 0 ~ '
where K, and K2 are experimentally derived constants and 8 is the first stress
invariant. Note that 8 is positive if compressive.
6 = o ,+02
+03
= o x+oy+or
(5.2)
Because the sub-base (the 2ndlayer) affects the results more than the sub-grade (the
3rdlayer) does, the sub-base will be treated by using a K-8 model and sub-grade will
be assumed to behave linearly for the models employed in this chapter.
(5.3)
In Eq. 5.3, B is a matrix of linear operators (derivatives of shape functions) and D is the
elasticity matrix defined as follows:
the plan dimensions of the soil layer. This approach will be considered and will be
referred to the "integration point method", which can be regarded as the rigorous way
to implement the K-8 model.
There is a potential difficulty with the integration point method, however. In
particular, we may get negative or zero values for 8 at some locations in the soil layer,
leading to small or zero values for E at those integration points. This can cause
difficulties for iterative solvers like those used in this study, which are sensitive to poor
conditioning of the system stiffness matrix, K. To avoid this, a minimum value, Emin
needs to be set for E. This is an important parameter which will likely affect results.
Another possibility for implementing a K-8 model would be to sanlple the stresses at
the center of the soil layer, computing E based on the value of 8 at this point, and taking
this E as applying through the entire layer thickness. This approach will likely avoid
some of the problems with the integration point method. This second approach will also
be considered, and will be termed the "mid-layer method" for the remainder of this
chapter.
Both the integration point and mid-layer methods will be implemented using a direct
iteration approach, which can be expressed as follows
Until convergence
Update stresses
Re-computed E at each location
Form K
Solve KU = P
end
In this algorithm, K is the secant stiffness matrix based on the element moduli, U is the
total displacement vector, and P is the vector of nodal forces. Convergence will be
defined as the point at which stresses and displacements do not change significantly with
successive iterations.
It should be noted that the K-8 model has been implemented in 2D layered analysis
and 2D FE analysis as well (Huang, 1993). Two methods have been developed to
incorporate the material non-linearity. In first one, the non-linear granular layer is
subdivided into a arbitrary number of layers and the stresses at the mid-depth of each
layer are used to determine the modulus; in the second method, the granular materials are
considered as a single layer and the modulus will be computed at a single point of the
layer. As it is well known that most granular materials do not take tension, the horizontal
stress will be set to zero if it is negative or in tension. Although in method 2, the point
selected is usually in upper half of the layer and tensile stresses occur very rarely, an
arbitrary or minimum modulus Eminmay still be needed to ensure convergence.
base. For the mid-layer method, n~odulusof elasticity is picked at the mid-height of subbase and used through the entire thickness of sub-base. Therefore, elasticity matrix D
formed from the two methods will be significantly different. As a result, nodal
displacements and stresses will be different as well. In addition to examining the two
different implementations, the influence of de-bonding will also be considered. One
model assumes that the asphalt and sub-base are fully bonded. The second model does
not permit tension at the asphalt sub-base interface, and allows free horizontal sliding
between the asphalt and sub-base.
Maximum deflections and tensile stresses for the fully bonded and un-bonded
condition are shown in Figures 5.1 through Figure 5.6. Table 5.1 summarizes the
maximum tensile stress at the bottom of the asphalt and the largest vertical displacement.
Number of iterations
Figure 5.1 Maximum Vertical Deflections vs. Number of Material Iterations, Fully
Bonded, Integration Point Method
8
10
12
14
Number of Material Iterations
16
18
Figure 5.2 Max. Principal Tensile Stresses in Asphalt vs. Number of Material
iterations, Fully Bonded Integration Point Method
0.825
8
10
12
14
Number of material iterations
16
18
Figure 5.3 Maximum Vertical Deflections vs. Number of Material iterations, Fully
Bonded, Mid-layer Method
8
10
12
14
Number of material iterations
16
18
20
Figure 5.4 Max. Tensile Stresses vs. Number of Material iterations, Fully Bonded,
Mid-layer Method
Figure 5.5 Maximum Vertical Deflections vs. Number of Material iterations, Unbonded at Asphalt / Sub-base Interface, Integration Point Method
1I
2
8
10
12
14
Number of material iterations
16
18
20
Figure 5.6 Max. Principal Tensile Stresses vs. Number of Material iterations, Unbonded, Integration Point Method
0.9424
0.9423
Emin = 1OMPa
0.9422
0.9421
h
E
E
-0.942
Q
V)
0.9419
0.9418
0.9417
0.9416
0.9415
Number of material iterations
Figure 5.7 Maximum Vertical Deflections vs. Number of Material iterations, UnBonded, Mid-layer Method
Figure 5.8 Max. Tensile Stresses vs. Number of Material iterations, Un-Bonded,
Mid-layer Method
Table 5.1 Summary of Max. Tensile Stresses and Displacements for two Methods
Emin=
2MPa
Emin=
5MPa
Emin =
1OMPa
Mid-layer Method
(5 iterations)
Fully
Unbonded
Bonded
0.8221
0.9663
0.9423 0.8588
0.8221
0.9663
0.8588
0.9423-0.8221
0.9663
0.8588
0.9423
Results shown in Figure 5.2 illustrate that the maximum tensile stresses in the asphalt
become nearly constant after about 10 material iterations for the integration point method
irrespective of the value of Emin.Figure 5.1 shows that while displacements are still
increasing with further iterations, the rate of increase has slowed significantly. The
magnitude of stresses and displacements are, as expected, dependent on the value of Emin,
For example, decreasing Eminfrom 10 MPa to 5MPa increases the maximum tensile
stress from 1.37 MPa to 1.55 MPa, a change of 13%. However, deflections increase by
only 6.7% as Eminis decreased from 10 MPa to 5MPa. Also, it appears that convergence
properties are similar when interlayer slip and debonding is considered. Interestingly,
debonding does not affect stresses, as much as it affects displacements.
Based on the results shown in Figure 5.5 and 5.6, integration point implementation
of the k-8 model for the remainder of this Chapter will rely on a value of Emin= 5 MPa
and use 10 material iterations. All models will assume full bond between the asphalt and
the base.
As expected, there are large differences between the mid-layer method and the
integration point method. First, the mid-layer method converges much more quickly than
the integration point method, reaching nearly constant stresses and strains after only 5
material iterations. Further, the results are insensitive to the assumed value of Emin,
indicating that the value of 8 at the middle of the sub-base does not become negative.
Perhaps most striking, however, is the large difference in predicted stresses and
displacements, with the mid-layer method giving predictions that are approximately half
of the values predicted when using the integration point method. Based on these results,
the mid-layer implementation of K-8 model used for the remainder of the studies of this
chapter will be based on results after 10 material iterations and Eminwill be fixed at 5
MPa as for the integration point method.
The largest difference in results for the integration point method and the mid-layer
implementation does raise an interesting question: can different Kl and K2be backcalculated using each method to give a good fit to field-measured deflections? Further,
what are the asphalt stresses predicted using each these back-calculated values of Kl and
K2? These issues will be explored in the remainder of this chapter.
Material
Layer #
E(MPa)
ness(mm)
Iteration
1
2
Asphalt
Sub-base
1 Sub-grade
1400
K-8
10
740
A previous study conducted at U. Maine (Nickels, 1995) back-calculated the nonlinear parameters Kl and Kr for the North Yarmouth control section considered here by
using MICH-PAVE, a 2D computer finite element program (discussed in Chapter 2).
Load parameters and material properties used in this prior study are presented in Tables
5.3 and 5.4 respectively. Nickels (1995) showed that Kl
gave
a good fit to field-measured data, and these values will be taken as a starting point here.
Comparison of measured and predicted displacements using integration point method and
the mid-layer method are shown in Figures 5.9 and 5.10.
Property
Static Load (kips)
Tire Pressure (psi)
Radius of Load area (in.)
Value
5.35
&
Properties
Behavior
Asphalt Concrete
Linear Elastic
Modulus (psi)
Poisson's Ratio
Unit Weight (pcf)
E = 200000
0.35
150
+
- -- -- - - - -
Material
Subbase
Nonlinear
Cohesionless
KI=4480,
0.29
130
Subgrade
Nonlinear
Cohesionless
K1=3000, ~ ~ = 0 . 5 7 ~
0.29
125
-.-,-.
measured data
K1 = 30.9MPa, K2 = 0.53
/'
For the integration point method, Kl and K2 obtained from the 2D approach do not
give displacements that fit the experimental data very well. That is not surprising, since
the 2D MICH-PAVE model bases updates to E for an entire layer on the average value
computed across elements (Harichandran et al., 1993). In contrast, the integration-point
method approach, which uses values for E at each integration point, captures the variation
in E over layer thickness and horizontal dimensions affected by the load. For the midlayer method, results yielded by using published data give a good comparison with the
experimentally measured data as shown in Figure 5.10.
0.9
0.8 -
measured data
K1 = 30.9Wa, K2 = 0.53
Figure 5.11 The Best Pair of KI and K2in the Least Square Sense, Integration Point
Method
A back-calculation will now be performed which gives (Kl,,Kz) providing the best
comparison with measured data in the least square sense. Figures 5.1 1 and 5.12 clearly
show that, when Kl
well with the measured data best for the integration-point method. Figure 5.13 shows
the maximum principle stresses at the bottom of the asphalt concrete.
'
0.9
+
0.8 -
measud data
K1= I-, K2 = 0.8
0.7 -
Figure 5.13 Stresses Contour at the Bottom of the AC with KI = lOMPa, K2 = 0.78
for North Yarmouth Control Section, Integration Point Method
= 30.9
MPa and K2 =
0.53 were used. Figure 5.13 shows the stresses at the bottom of the asphalt larger
predicted using the mid-layer method and Kl
= 30.9
Figure 5.14 Maximum Principle Stresses at the Bottom of the AC with Kl = 30MPa,
K2 = 0.53 for North Yarmouth Control Section, Mid-layer Method
It was found that the best combination of Kl and K2 from the mid-layer method was
very similar to the published data (Nickels, 1995). However, the maximum principal
stresses at the bottom of asphalt is about 30% less than results obtained from the
integration point method. The direction of maximum principal stresses are mainly in y
direction. These results indicate that there are still open questions regarding the use of K8 models in 3D finite element analysis.
Real tire pavement contact stresses are not constant and experimental data illustrates
that the contact area between the tire and pavement will vary under different wheel loads
and inflation pressures (de Beer et al., 1997; Tielking et al, 1995). Analytical studies have
shown that spatially varying contact pressures can affect the response of a flexible
pavement system significantly (Bensalem et al., 2000). This chapter will examine the
effect of spatially varying tire contact pressure on asphalt concrete displacement and
stresses. For simplicity, we will consider only static loading, although in general the
truck suspension system and vehicle speed do affect stresses in real pavements
(Lee, et al 1993; Cunagin et al, 1993).
First, this chapter will cover a strategy for modeling spatially varying pressures.
Contact patterns will be based on laboratory-measured values taken from de Beer et a1
(1997). A convergence study will be performed to determine the required level of
discretization of the contact area. Next, parametric studies will be performed using the
3D finite element modeling strategies developed in this study to examine the effect of
spatially varying tire contact pressures on pavement response. A variety of loading
conditions and model material and geometric properties will be considered. Finally, a
brief summary finishes this chapter.
q,, = 0.86P
+ 175
(6.1)
where 460 is the average tire pavement contact stress at the tire center [@a], and P is tire
inflation pressure in kPa (de Beer et al, 1997). The maximum vertical stress at the tire
edge is controlled by loads, and the relationship is given in Eq. 6.2:
q, = - 0 . 5 3 ~ ~57.46L - 534.05
(6.2)
where q, is the vertical tire pavement contact stress at the tire edge, and L is tire load in
Figure 6.2 Load, Inflation Pressure Relationship on the Vertical Contact Stress
Distribution (de Beer et al, 1997)
Z 1eJns!d
VldStlA SSOWV MQWnN Nld
OZ6l~L.LlQlSLPLELZLLIOL6 0 L 9 S b E 2 L
. . . . . . .
...
11 @JnBkJ
V U S W SSOWV MQlrYflNNld
v
W
Plan View
1-2a
Section A-A
w-2d
n
Section B-B
Transverse Direction
The value of q60, the stress at tire center obtained from Eq. 6.2 is an average
pressure over the inner 60 per cent of the tire width (see Figure 6.6). Hence, q,, the
vertical contact pressure at the tire center needs to be determined so that 460 is truly the
average pressure on interior.
To ensure the average holds, the following integral needs to be satisfied:
where,
qc
To ensure that the average pressure, 960, exists over the entire 60% of the contact area,
the interior area (Ainkrior)is defined as follows:
This gives the value of a for any w, I, and d to ensure that AinIerior= 0 . 6 ~ 1
Figures 6.7 and 6.8 show typical vertical and transverse stress distribution for L = 40 kN
(total vertical load), P = 620 kPa (inflation pressure).
1800 '1800
Transverse Direction
Traffic Direction
Figure 6.7 Typical Vertical Contact Pressure Distribution (tire load, L = 40W,
inflation pressure, P = 620KPa)
..
. . . . . ..
'
2200
Transverse Direction
Traffic Direction
Figure 6.8 Typical Transverse Contact Pressure Distribution (tire load, L = 40 kN,
inflation pressure, P = 620 kPa)
Generally, total vertical load yielded fiom integration of the vertical pressures defined
in the previous section is lower than the actual value of L used to determine q, and q,. As
a result, the length of the contact area, I, must be adjusted so that the total applied load
due to the contact pressures is L. Mathematically, this can be expressed as:
where p is the vertical contact pressure which varies over the contact area, A. To
numerically integrate pressures over the contact area, the contact area is divided into a
grid as shown in Figure 6.9. the dimensions of refined regions have been chosen as 2d
and 2a in x, y directions, respectively. For simplicity, the number of divisions in refined
regions as well as in coarse regions is identical and will be called n for the purpose of
discussion.
Fine (2a)
Coarse
w4Number of Divisions
Fine (2a)&
=n
To do the integration over the entire contact area, we have chosen to generate a
Delaunay Triangulation of the grid shown in Figure 6.9, which is easily implemented.
The mesh of triangles will be generated using the computer program, "triangle"
(Shewchuk, 1999) and a typical mesh is shown below in Figure 6.10:
I
0
50
100
150
200
250
300
where p is the pressure acting on the contact area, n is the number of triangles, pi is the
average pressure acting on an individual triangle element and Ai is the area of an
individual triangle. Using Delaunay triangulation and Eq. 6.7 to compute the total force
due to the contact pressures, Eq. 6.6 can be solved for the particular length of the contact
area, I, that yields vertical equilibrium. This was accomplished using the bisection
method, and a general computer program was written to generate vertical pressures and
contact areas for a variety of wheel loads and tire inflation pressures.
The error in the numerical solution to Eq. 6.6 is defined as follows:
error =
cP A
L
5 tolerance
Transverse Direction
1800 - -1800
Traftic Direction
Transverse Direction
1800 '1900
Traftic Direction
Transverse Direction
1800 - '1600
Traftic Direction
Transverse Direction
1800 '1800
Traffic Direction
n (see Figure 6.9), a finite element model was constructed with the following properties
as shown in Figures 6.15 and Table 6.2:
Plan View
Sub-base
1740nm
Elevation View
Figure 6.15 Model Description
Layer
Asphalt Concrete
Base
Sub-grade
Thickness(mm)
125
635
740
E (MPa), v
1400,0.35
67,0.29
10,0.29
The dual wheel load was modeled with spatially varying pressures acting at the
center of the pavement with L = 40kN and P = 620KPa. The finite element mesh used is
shown in Figure 6.16. The results are shown in Figure 6.17:
Figure 6.16 3D View of Meshed Model (entire pavement system, 3024 solid
elements)
h
1.16
1.15 -
2 1.15 5
A
I
I
10
15
Number of Divisions, n
20
1.495 U)
%
E
U)
.-P )
1.49
U)
8'
x
10
15
Number of Divisions, n
20
It is clear that when number of divisions equals 15, the stresses and displacements
tend to converge. This study will use n = 10, however, since the difference between
predicted stresses and displacements is negligible for larger values of n. For example,
increasing n fiom 10 to 15 increases stresses by only 0.208% and displacements by only
0.198%.
The thicknesses of asphalt concrete that will be examined is 75 mm, 125 mm and 200
mm and Young's modulus of the sub-base is 67MPa (determined via back-calculation, in
Chapter 4), 37 Mpa, or 135 MPa. The sub-grade's modulus of elasticity will be fixed at
10MPa. In each combination, we will compare three different loading cases: I means
spatially varying tirelpavement contact pressures as detailed previously; I1 means the
uniform rectangular patch load with the same contact area as spatially varying pressures \
for the same value of L; and III means the uniform rectangular patch load with a vertical
contact stress equal to the inflation pressure, and a different contact area
-Wheel
Loads
7
Wheel loads
Asphalt Concrete
+ X
f~,=75,125,2OOnun
635mm, Ez
Sub-base
Sub-grade
P0nun
The stresses are the maximum principal stresses taken at the bottom of the asphalt.
The location where the maximum principal stresses occur varies, but remains within the
two longitudinal center lines of the two contact areas. The directions of those maximum
stresses are mainly in they (transverse direction) with small x (traffic direction)
components.
According to Figures 6.20-6.28, the spatially varying loads tend to generate larger
tensile stresses at the bottom of asphalt concrete than the uniform pressure loads, with a
maximum difference of about 30%, which implies that assuming a constant pressure over
the entire tire contact area tends to underestimate the response. Generally, the uniform
loads with the same contact area as the spatially varying loads (case 11) yield smaller
results than the loads with the same inflation pressure (case 11).
Under the same loads, the thinner the asphalt concrete, the more pronounced effect of
the spatially varying tirelpavement contact pressures. If we consider the same thickness
of asphalt concrete, subjected to the same vertical load, lower inflation pressures tend to
increase differences in tensile stresses between I and (I1 and 111). Similarly, decreasing
the stiffness of the sub-base (decreasing EZ)tends to somewhat increase the differential
in tensile stresses between case I and (I1 and 111).
1
20
r"
30
40
Vertical Load (kN
lnflation pressure. P = 7 4 0 K ~ a
';;;'
rn
1 1.2
2
tj
Q,
1 1.2 .
f3 111
ti
.-Q,
g
1.
'r"
x
20
30
40
Vertical Load (kN)
50
0.6
20
30
40
50
Vertical Load (kN
lnflation pressure. P = 7 4 0 ~ ~ a
1.6
Q,
.-g
0.8 .
'r"
--__---__
0.8 :---------I::
f3 111 .
P!
0.6 .
x
20
+.+=J
30
40
Vertical Load (kN)
I
11
111
4'
.
50
-$
1
20
30
40
Vertical Load (kN)
50
r"
1
20
30
40
Vertical Load (kN)
50
Figure 6.23 Max. Tensile Stresses at bottom of AC(thickness = 125mm, E2= 35MPa)
.3
0.8
20
30
40
50
Vertical Load (kN
Inflation pressure, P = 7 OKPa
1
30
40
50
I 20
0.8
20
30
40
Vertical Load (kN)
50
Figure 6.24 Max. Tensile Stresses at bottom of AC(thickness = 125mm, E2= 67MPa)
1.2
3
V)
%
V)
I-
al
*
L
= 111
20
30
40
Vertical Load (kN)
50
20
30
40
Vertical Load (kN)
50
Figure 6.26 Max. Tensile Stresses at bottom of AC(thickness = 200mm, E2= 35MPa)
3 1.4
n
E
u,
2u,
1.2
1
fj
al
.-
0.8
:
8
Z
Vertical Load (kN)
0.6
20
30
40
Vertical Load (kN)
50
Figure 6.27 Max. Tensile Stresses at bottom of AC(thickness = 200mm, E2= 67MPa)
lnflation pressure, P = 420KPa
0.2 1
20
20
20
30
40
Vertical Load (kN)
50
30
40
50
Vertical Load (kN
lnflation pressure. P = 7 1 0 ~ ~ a
30
40
Vertical Load (kN)
50
Figure 6.29 Max. Tensile Stresses at top of AC (thickness = 75mm, Ez= 35MPa)
109
0.3 .
Figure 6.30 Max. Tensile Stresses at top of AC (thickness = 75mm, E2= 67MPa)
a
Figure 6.31 Max. Tensile Stresses at top of AC (thickness = 75mm, E2= 135MPa)
w
6'
I
.
0.05
20
30
40
Vertical Load (kN)
50
Vertical Load (kN)
Figure 6.36 Max. Tensile Stresses at top of AC (thickness = 200mm, E2= 67MPa)
20
30
40
Vertical Load (kN)
50
Vertical Load (kN)
Figure 6.37 Max. Tensile Stresses at top of AC (thickness = 200mm, E2= 135MPa)
As shown in Figures 6.29-6.31, for the smallest pavement thickness, the principal
stresses are largest. Increasing E2 tends to decrease the stresses, and the differences
between the non-uniform and uniform pressures are relatively small. The largest
differential occurs for E2 = 135 MPa. Increasing inflation pressures for ZI = 75 mnl does
tend to increase the stresses slightly.
For ZI = 125 mm, the differences between uniform and non-uniform contact stresses
increases significantly with increasing E2, However, the value of E2 does not seem to
have a very large effect on the maximum principal stress predicted for lonading I, and
inflation pressure has a small effect.
For the thickest pavement section, the differential between I and (I1 and 111) grows
with increasing vertical load for all values of E2. Increasing inflation pressure tends to
slightly decrease the differential between I and (I1 and 111) in general. Again, increasing
E2 does not significantly affect the maximum principal tensile stresses at the top of the
asphalt.
Overall, with the exception of the thinnest pavement section with the softest sub-bse,
layers (ZI = 75 rnrn and E2 = 35 MPa), the maximum predicted tensile stresses in the top
of the asphalt tend to remain fairly constant with increasing pavement thickness. This is
important -it implies that irrespective of pavement thickness, the contribution of tensile
stresses to top-down cracking remains fairly constant.
6.6 Summary
In this chapter, a model was developed to simulate the spatially varying tire contact
pressures based on the study by de Beer et al, (1 997). The kernel of the model is
developing a mathematical expression for tire contact pressures as a function of tire load
and inflation pressure.
Parametric studies were perfomled to examine effecs of tire load, inflation pressure,
asphalt thickness and sub-base modulus on pavement response.
It was found that the regular rectangular patch loads tend to underestimate tensile
stresses in the bottom of the asphalt by 10%-30%, especially in thin flexible pavement
structures with low elastic n~odulusof sub-base material; the thicker the asphalt concrete
layer, the less sensitive the pavement in terms of maximum principal stresses at the
bottom of asphalt. Tehnsile stresses at the top of he asphalt are largest for the 75 mnlthick asphalt sections, but remained fairly constant between for a given wheel load
irrespective of inflation pressures and sub-grade stiffness for the thicker pavement
sections. The tensile stresses in the top of the asphalt predicted using spatially varying
tire contact pressures are significantly larger than the stresses predicted assuming a
uniform load.
developed. With this tool, we can efficiently build a relatively large 3D FE model of
a pavement system that incorporates multiple layers, inter-layer debonding and slip,
and a variety of load types.
widely used to analyze non-linear response of granular materials. Two options for
incorporating the K-0 model were considered in detail: a rigorous integration-point
method and a technique where material properties are updated based on mid-layer
stresses. The convergence properties of these two approaches were studied in detail.
The best combination of non-linear parameters Kl, K2,was back-calculated to give
good comparison to the experimental data.
4. The effect of spatially varying tirelpavement contact pressures on pavement
response was studied. Previous investigations (Yue and Seve, 1995; Bensalem et a1
2000, etc) illustrate that assuming a constant stress distribution over the entire
tirelpavement contact area may not be accurate, and that spatially varying pressures
can affect the pavement stresses significantly. A mathematical model has been
developed to approximate the stress distribution variation in space (assuming a
smooth tire), and then comparison with uniform rectangular patch loads was
presented for a variety of asphalt thicknesses, sub-base stiffnesses, wheel loads and
tire inflation pressures.
7.2 Conclusions
The following conclusions were drawn fiom this study as listed below:
1. When including a three-layer flexible pavement system, the critical model size in
three directions is 4000 mm X 4000 mm sub-grade depth of at least 740 mm.
Modeling only a single dual wheel load is accurate enough to simulate real truck
loads.
2. No matter what the size of the model and material properties of the pavement
structures, the resilient moduli of member layers can generally be back-calculated to
give a good comparison to the corresponding experimentally measured
displacements. For the two control sections in this study (one with an asphalt
thickness of 125 mrn and one with an asphalt thickness of 230 mm), different pairs
of sub-base and sub-grade moduli were computed for different total sub-grade depths
to give good fits to experimentally measured data. The maximum tensile stresses
predicted varied by less than 11%.
3. The back-calculation of material parameters of the K-8 model was performed to
give a good agreement with the experimental displacements using two methods, the
integration point method and the mid-layer method. The K-8 model did not give a
better fit to the experimental data when compared to models using linearly elastic
materials. The best combinations of Kj and K2 are different from two methods and
difference of the corresponding maximum principal stresses at the bottom of asphalt
is about 50%, but the computer predicted displacement basins can fit the field
measured displacement very well (see Figures 5.10 and 5.12). This is not surprising,
due to the limitations of K-8 model and the way we implement the two methods as
discussed in Chapter 5. Further, it takes much longer to achieve solutions when using
K-8 model than when using linearly elastic materials.
compared to the rectangular patch loads with the same contact area, the spatially
varying tirelpavement pressures yield stresses at the bottom of asphalt that are 10%30% larger. The largest differential occurs in thin flexible pavement structures with a
sub-base having low elastic modulus.
2. Implementation of better material models. In this thesis, a K-8 model was employed
to incorporate material non-linearity. Unfortunately, the K-8 model has several
deficiencies. Better constitutive models for the sub-base and sub-grade are required.
3. Extension of current spatially varying pressure proqam. The computer program
developed in this study assumed that the pavements were subjected to static wheel loads
with smooth tires. However, actual tires with tread grooves should be taken in
consideration in future studies. Note that the process of triangulating the pressures and
adjusting total contact areas to presented here would be directly appliciable to any
distribution of contact pressures.
BIBLIOGRAPHY
Bensalem, A, A. J. Brown, M.E. Nunn, D. B. Merrill, and Wyn G. Lloyd, (2000), "Finite
Element Modeling of Fully Flexible Pavements: Surface Cracking and Wheel
Interaction", Preproceedings of the Second International Symposium on 3D Finite
Element for Pavement Analysis, Design, and Research, October 11- 13,2000 pp 103-122.
Blab, Ronald, John T. Harvey, (2000), "Modeling Measured 3D Tire Contact Stresses in
a Viscoelastic FE Pavement Model", Preproceedings of the Second International
Symposium on 3D Finite Element for Pavement Analysis, Design, and Research, October
11-13,2000, pp123-148.
Burmister, D. M. (1943). "The Theory of Stresses and Displacements in Layered Soil
Systems and applications to the Design of Airport Runways," Proceedings, Highway
Research Board, Vol. 23, pp. 126-144.
Burmister, D. M. (1945). "The General Theory of Stresses and Displacements in Layered
Soil Systems," Journal of Applied Physics, Vol. 16, pp. 84-94, 126-127,296-302.
Bush, A. J. 111. Nondestructive Testing for light Aircraft Pavements, Phase 11,
Development of Nondestructive Evaluation Methodology. Final Report FAA-RD-80-9-11.
FAA, Washington, D.C., 1986.
Chen, Dar-Hao, Musharraf Zaman, Joakim Laguros, and Alan Soltani (1995),
"Assessment of Computer Programs for Analysis of Flexible Pavement Structure",
Transportation Research Record, No. 1482, pp. 123-133.
Chen, Hsien H., Kurt M. Marshek, and Chhote L. Saraf, (1990), "Effects of Truck Tire
Contact Pressure Distribution on the Design of Flexible Pavements: A ThreeDimensional Finite Element Approach", Transportation Research Record, No. 1095 pp.
72-78.
Cho, Yoon-Ho, B. Frank McCullough, and Jose Weissmann, (1996), "Considerations on
Finite Element Method Application in Pavement Structural Analysis", Transportation
Research Record, No. 1539, pp. 96-101.
Cunagin, Wiley D. and Albert B. Grubbs, (1991), "Automated Acquisition of Truck Tire
Pressure Data", Transportation Research Record, No. 1322, pp. 112-121.
Davids, William G., George M. Turkiyyah, and Joe P. Mahoney (1998), "Modeling of
Rigid Pavements: Joint Shear Transfer Mechanisms and Finite Element Solution
Strategies", Report No. SGEM 98-6, Department of Civil Engineering, University of
Washington, Seattle, Washington.
Davids, William G., George M. Turkiyyah, and Joe P. Mahoney (1998), "EverFE Rigid
Pavement Three-Dimensional Finite Element Analysis Tool", Transportation Research
Record, No. 1629 pp. 41-49.
de Beer, M., C. Fisher and Fritz J. Jooste, (1997), "Determination of Pneumatic
Tirepavement Interface Contact Stresses Under Moving Loads and Some Effects on
Pavements with Thin Asphalt Surfacing Layers", Proceedings,Volume I, Eighth
International Conference on Asphalt Pavements, Seattle,Washington, August 10-14,
1997, pp 179-227.
De Jong, D. L., M. G. F. Peatz, and A. R. Korswagen, (1973), "Computer Progranl
System Under Normal and Tangential Load", Koin Klijke Shell-Laboratorium,
Amsterdam, External Report AMSR.0006.73.
Deusen D. V. Selection of Flexible Back-calculation Software for the Minnesota Road
Research Project. Report MN/PR-96/29. Minnesota Department of Transportation, St.
Paul, 1996.
Harichandran, Ronald S., Ming-Shan Yeh, and Gilbert Y. Baladi, (1990), "MICH-PAVE:
A Nonlinear Finite Element Program for Analysis of Flexible Pavements",
Transportation Research Record, No. 1286 pp. 123-131.
Helwany, Sam, John Dyer, and Joe Leidy, (1998), "Finite-Element Analysis of Flexible
Pavements", Journals of Transportation Engineering, Vol. 124 No. 5, pp. 491-498.
Hibbitt, Karlsson, and Sorensen, (1992), ABAQUS, version 5.2. Pawtucket, N.Y.
Hjelmstad, K. D., J. Kim and Q. H. Zuo, (1996), "Finite Element Procedures for ThreeDimensional Pavement Analysis", Transportation Research Record, No. 1539 pp. 66-71.
Huang, Yang H., (1993), "Pavement Analysis and Design", Prince-Hall, Inc.
Hwang, D., and M. W. Witczak, (1981), "Program DAMA (Chevron), User's Manual",
Department of Civil Engineering, University of Maryland.
Irwin, L. H. (1983), User's to MODCOMP2. Cornell Local Roads Program Report 83-8.
Cornell University Local Roads Program, Ithaca, N.Y., Nov. 1983.
Kopperman, S., G. Tiller, and M. Tseng, (1986), "ELSYM5, Interactive Microcomputer
Version, User's Manual", Report No. FHWA-TS-87-206, Federal Highway
Admimistration.
Lee Ying-Haur, Alaeddin Mohseni, and Michael I. Darter, (1 993), "Simplified Pavement
Performance Models", Transportation Research Record, No. 1397, pp7- 14.
005 10, proceeding of the 7gthAnnual Meeting of the Transportation Research Board,
Washington, D.C., Jan. 10-14, 1999.
Yue ZhongQi and Svec, Otto J., (1995), "Effect of tire-pavement contact pressure
distribution on the response of asphalt concrete pavements", Canadian Journal of Civil
Engineering, v22 n5 Oct. 1995, pp 849-860 03 15-1468.
Zienkiewicz, 0. C. and R. L. Taylor, (1988), "The Finite Element Method", Fourth
Edition, Volume I McGraw-Hill International (UK) Limited.