Professional Documents
Culture Documents
EverFE Theory Manual
EverFE Theory Manual
February 2003
1
x
y 20-noded brick element
z
zero
thickness
2
used and all required element integration is performed numerically using 8-point (2x2x2) Gauss
quadrature. The initial public release of EverFE (version 1.02, released January 1998) used 27-point
(3x3x3) Gauss integration; however, subsequent internal studies showed that the higher-order integration
scheme added to the computational time without significantly improving accuracy.
2.3 Modeling of the dense liquid foundation
The dense liquid foundation can either support tension, or be tensionless. It is important to note that
the tensionless dense liquid does not account for any pre-compression due to dead load, i.e. the total
vertical deflection including the effect of dead load must be overcome before the dense liquid foundation
stress and stiffness become zero.
The 8-noded element illustrated in Figure 1 is used to discretize the dense liquid. This element was
formulated specifically for this application, and full details of the implementation can be found in Davids
(1998). The element incorporates standard quadratic shape functions for interpolation of vertical
displacements within the element (Zienkiewicz and Taylor 1994), ensuring that it displaces compatibly
with the 20-noded brick element with which it shares nodes. An isoparametric element formulation is
used, and all necessary element integrations are performed numerically using 9-point (3x3) Guass
quadrature to ensure accurate results when the tensionless option is selected.
The only constitutive parameter needed for this element is the distributed stiffness of the dense liquid
foundation [force/volume]. For the tensionless foundation, if tension occurs at an element integration
point during the solution process, the stress and stiffness at that point are set to zero during integration of
the element stiffness matrix and equivalent force vector. For the conventional, tension-supporting dense
liquid, the stiffness remains constant at all points.
kSB
x or y
o
Interface Base element x or y
Interface constitutive relationship elements transfer
shear stress
Figure 2: Modeling Separation and Shear Transfer at the Slab-Base Interface
Shear transfer between the slab and base can be important when analyzing pavements subjected to
thermal and/or shrinkage strains. Rasmussen and Rozycki (2001) overviewed the factors governing slab-
base shear transfer, noting that both friction and interlock between the slab and base play a role. In
addition, they calibrated a bilinear, elastic-plastic shear transfer model from results of push tests of slabs
3
on various base types. One conclusion of the their study was that the effect of slab-base shear transfer
should be incorporated in 3D analyses of pavement systems. Another study by Zhang and Li (2001)
focused on developing a one-dimensional analytical model for predicting shrinkage-induced stresses in
concrete pavements that accounts for slab-base shear transfer. Like the model developed by Rasmussen
and Rozycki, their model ultimately relied on a bilinear, elastic-plastic shear transfer model. Zhang and Li
concluded that the type of supporting base – and thus the degree to which it restrains slab shrinkage –
significantly affects slab stresses.
To capture slab-base shear transfer, EverFE employs 16-noded, zero-thickness quadratic interface
elements that are meshed between the slab and base as shown in Figures 1 and 2. The element
incorporates standard quadratic shape functions for interpolation of displacements (Zienkiewicz and
Taylor 1994), ensuring that it displaces compatibly with the 20-node brick elements with which it shares
nodes. The element tracks relative displacements between the slab and base in the vertical (z) and both
horizontal (x and y) directions. An isoparametric element formulation is used, and all necessary element
integrations are performed numerically using 9-point (3x3) Guass quadrature.
The bilinear element constitutive relationship is based on that given by Rasmussen and Rozycki
(2001) and Zhang and Li (2001). Figure 2 illustrates the constitutive relationship, which is characterized
by an initial distributed stiffness kSB>IRUFHYROXPH@DQGVOLSGLVSODFHPHQW o. While kSB has the same units
as the well-known dense liquid foundation modulus, kSB is a distributed stiffness in the x- and y-directions,
and the shear stresses in the x-y plane at the slab-base interface are caused by relative horizontal
displacements between the slab and base layer. This constitutive relationship is assumed to apply
independently in both the x- and y-directions as long as the slab and base remain in contact (i.e. a
compressive normal stress exists at the slab-base interface). The fact that there will be little or no shear
transfer when slab-base separation occurs is accommodated by setting the interface stiffness and shear
VWUHVVWR]HURZKHQHYHUWKHUHODWLYHYHUWLFDOGLVSODFHPHQW z > 0. Modeling this loss of shear transfer with
loss of slab-base contact can be important, especially when thermal gradients are simulated (Davids et al.
2003).
Note that unlike a frictional model, the shear stress does not depend on the magnitude of the normal
stress. However, for very large values of kSB, this model approaches Coulomb friction with a very large
friction coefficient, and for very small values of kSB, it is equivalent to a frictionless interface. An
advantage of the modeling scheme used by EverFE is that the symmetry of the system stiffness equations
is maintained, which allows the use of the highly efficient preconditioned conjugate-gradient solver
discussed in Section 7. Idealizing slab-base interaction with conventional Coulomb friction would destroy
this symmetry, requiring the use of more complex (and likely less efficient) solution techniques.
4
x ∆x
Original
position α
Misaligned r
Gap between position q
dowel and slab
Plan View
Slab
C.L. joint Original
x
position
z
∆z
β
s Misaligned q
Dowel-slab springs
position
Elevation View
(a) Dowel-Slab Interaction (b) Dowel Misalignment
5
material nonlinearity through the simple stiffness matrix transformation of Equation 3. Internally, the
contact conditions are checked and updated at each iteration of the nonlinear solver.
With the foregoing formulation, the embedded bending element has also been extended to permit the
inclusion of general bond-slip relations between the dowel and surrounding slab (Davids 2000). If the
incremental vector of relative displacements between the slab and dowel is denoted as d , and it is
assumed that the corresponding incremental force vector df can be computed as:
df = Dd∆ ( 4)
In Equation 4, D is a 3x3 constitutive matrix. It has been shown that the stiffness matrix of the embedded
dowel element with general bond-slip relations, Kdt, can then be computed as:
K dt = K de + ∫−1 B T DBhdη
1
(5)
In Equation 5, B is a matrix operator containing shape functions of both the embedding solid element and
the dowel element, and h is the length of the dowel element; integration is performed with respect to the
dowel element local coordinate, .
Physically, this formulation of the embedded element with a general bond-slip relationship between
the dowel and slab is analogous to the classic beam on elastic foundation, but differs in that forces in the
three coordinate directions can exist between the dowel and the slab. The magnitude of the forces depends
on the relative displacement between the dowel and the slab and the constitutive relations of the
dowel/slab interface incorporated in the matrix, D.
4.2 Implementation in EverFE
In EverFE, the 3-noded quadratic embedded dowel elements are used to model the portions of the
dowels embedded in the slabs. To ensure accurate results, 12 embedded elements are used to model the
embedded portion of the dowels on each side of each transverse joint, and a 2-noded shear beam is used
to model the portion of the dowel spanning the joint.
When dowel looseness is modeled, 10 of the 12 embedded dowel elements are used over the portion
of the dowel where there is a gap between the dowel and slab to ensure a sufficient number of potential
points of contact between the dowels and slabs. If the dowel is treated as a beam on a dense liquid
foundation, the 12 elements are evenly distributed along the embedded portion of the dowel. The diagonal
terms of D corresponding to the model y- and z- coordinates, D(2,2) and D(3,3), are the dowel-slab
support modulus specified in EverFE. The dowel-slab support modulus is computed as the product Kd,
where K is the commonly used modulus of dowel support [force/volume], and d is the dowel diameter.
The paper by Dei Poli et al. (1992) discusses the basis for the development of K from the properties of the
slab concrete and dowel; additionally, the EverFE help manual includes the results of a parametric study
showing the effect of the parameter Kd on load transfer efficiency for a simple two-slab model. D(1,1)
applies in the x-direction, and is the dowel-slab restraint modulus that controls the degree of bond
between the dowels and slabs. A large value of the dowel-slab restraint modulus implies a high degree of
bond between the dowels and slabs, and a value of 0 implies no bond. For an example illustrating the
effect of this parameter, see Davids et al. (2003).
Transverse ties are modeled in the same manner as the dowels, although only the dense liquid support
option is available, and fewer elements are used to discretize the ties since their shears and moments are
of secondary interest.
4.3 Simulation of Dowel Misalignment/Mislocation
EverFE also allows the simulation of dowel misalignment and/or mislocation through the specification of
IRXUSDUDPHWHUV x z WKDWVKLIWDQLQGLYLGXDOGRZHODORQJWKHx- and z-axes and define its angular
misalignment in the horizontal and vertical planes (see Figure 4(b)). The dowel support and restraint moduli
coincide with the local dowel coordinate axes (q,r,s), which are rotated from the global (x,y,z) axes by the
6
DQJOHV DQG 7KHPHVKLQJDOJRULWKPSUHFLVHO\ORFDWHVLQGLYLGXDOIOH[XUDOHOHPHQWVZLWKLQWKHVROLG
elements by first solving for the intersection of each dowel with solid element faces, and then subdividing
each dowel into at least 12 individual quadratic embedded flexural elements on each side of the joint face as
discussed previously.
7
Figure 5: EverFE Screen Shot of Aggregate Interlock Example
1.5
1.4
Peak Slab Stress (MPa)
1.3
1.1
0.9
0 10 20 30 40 50 60 70 80 90 100
Load Transfer Efficiency (%)
8
5.2 Nonlinear Aggregate Interlock Load Transfer
The nonlinear aggregate interlock load transfer option allows the consideration of both the effect of
relative vertical joint displacement and joint opening on aggregate interlock load transfer effectiveness.
EverFE relies on a two-phase aggregate interlock model developed by Walraven (1981, 1994) to generate
nonlinear aggregate interlock crack constitutive relations. The crack is assumed to follow the aggregate
particle boundaries, and the aggregate particles bearing on the cement paste are taken to be at the point of
slip. Walraven’s model also assumes that the aggregate particles are graded according to a Fuller
distribution, and the maximum particle diameter, Dmax, and the aggregate volume fraction, pk, are model
parameters.
Given an ultimate strength of the cement paste, pu, a coefficient of friction between the paste and
DJJUHJDWHRI DQGFRPSXWLQJWKHSURMHFWHGFRQWDFWDUHDVEHWZHHQWKHDJJUHJDWHDQGSDVWHXVLQJWKH
deformed geometry, the forces required for equilibrium of a single aggregate particle
diameter/embedment combination can be computed. Using the probability of occurrence for a particular
embedment/diameter combination derived by Walraven (1981), the likely forces on all particles are then
summed to give the total forces acting on a crack plane for a given relative slip displacement and joint
opening.
Typical crack shear stress-displacement relations predicted by the model are shown in Figure 7,
where each curve corresponds to a specific joint opening. Although only 3 curves are shown, EverFE
internally generates 40 curves over a range of joint openings between 0 and 20 mm to give a very
complete definition of the shear stress-displacement relations. The majority of these curves apply for joint
openings between 0 and 2 mm, where the most rapid changes in load transfer effectiveness with joint
opening occur. As with the linear model, shear stress is transferred only in the vertical direction.
9
0.02 mm Joint Opening
8
7
0.12 mm Joint Opening
6
Shear Stress (MPa)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Relative Vertical Joint Displacement (mm)
9
The implementation of Walraven’s model in EverFE is consistent with the general formulation of a
materially nonlinear finite-element analysis. The necessary tangent moduli are computed numerically
from the constitutive relations, and stresses are numerically interpolated from the constitutive relations for
a given joint opening and relative vertical displacement. It is important to note that the finite-element
implementation accounts for the variation in joint opening with vertical position on the joint face that
develops under loading of the pavement system. One limitation of the model, however, is that it does not
account for the sawcut at the top of a typical contraction joint where no aggregate interlock load transfer
would normally occur.
The parameters necessary to define the nonlinear aggregate interlock model in EverFE are the
maximum paste strength, pu, the paste-DJJUHJDWHFRHIILFLHQWRIIULFWLRQ WKHDJJUHJDWHYROXPHIUDFWLRQ
pk, and the maximum aggregate diameter, Dmax. Walraven (1994) suggested the following relationship
between the compressive strength of a 150 mm concrete cube, fccDQG pu, where both are in MPa:
σ pu = 8.0 f cc ( 6)
The compressive strength of a 150 mm concrete cube can be assumed to be approximately 1.25 times the
strength of a standard concrete cylinder, f c’ (Wang and Salmon 1985). In addition, when weak aggregate
is used that is prone to fracture, Walraven suggested proportioning pu downward by a fracture index, Cf <
1.0. Recent research on the topic of aggregate interlock joint shear transfer (Jensen and Hansen 2003;
Wattar et al. 2001) suggests that the basic concepts underlying Walraven’s model are sound. However, its
accuracy in predicting pavement joint load transfer may vary with specific characteristics of aggregate
shape and type, as well as the degree of damage at the joint. In addition, a study by Davids and Mahoney
(1999) has shown good qualitative and reasonable quantitative agreement between existing experimental
data and predictions of aggregate interlock load transfer efficiency using Walraven’s model.
10
element method, using a rectangular rule to numerically integrate the tire contact pressure over its area of
application.
Clearly, the critical parameters in this calculation are nx and ny. To examine their effect on solution
accuracy, consider the following example where a single slab founded on a dense liquid is subjected to a
single 40 kN wheel load at its edge with L = W = 200 mm. The slab is 4400 mm long, 3600 mm wide, and
254 mm thick with E = 27,6003DDQG 7KHVODELVPHVKHGZLWK[HOHPHQWVLQSODQDQG
elements through the slab thickness, and is shown in Figure 8. The model was solved using values of nx =
ny ranging between 1 and 20. When nx = ny = 1, the wheel load is treated as a single point load. Table 1
gives the x-direction stresses at the top and bottom of the slab for different values of nx = ny, and shows
clear convergence by nx = ny = 10. Since the equivalent nodal force calculation presented here is not
computationally expensive, EverFE uses a fixed value of nx = ny = 20 for all simulations to ensure
accuracy for a wide range of wheel load sizes and levels of mesh refinement. It should be noted that other
researchers have developed different methods of handling the problem of wheel load contact stresses
(Hjelmstat et al. 1997); however, the method presented here is conceptually simple, easily implemented,
and accurate.
Difficulties can arise when one or more of the sub-area loads falls outside the slab boundary, and
cannot be located in an element. EverFE handles this by moving the point of application of pi around a
circle with an ever-increasing radius until pi falls within a solid element, and the calculation then proceeds
as detailed. This may degrade the accuracy of the method, however, and wheel loads having portions of
their contact areas outside slab boundaries should be avoided.
11
Table 1: Effect of nx and ny on Slab Stresses
12
achieved when r is sufficiently close to zero. If the model is linear, Kk is constant, and only a single
iteration is required.
r=F
while ||r||/||F|| > 10-04
update contact constraints
update Kk
solve KkdUk = r
Uk = Uk-1 + dUk
update r
end
Algorithm 1: Nonlinear Solution Strategy
It is worth noting that for models with a base layer, the solution of KkdUk = r for dUk involves a sub-
iteration using a technique called Uzawa’s method to satisfy the contact constraints that is not detailed
here. This is necessary to avoid ill conditioning of Kk and maintain the efficiency of the MG-PCG solver.
For nonlinear problems, the contribution of the 20-noded solid elements used to discretize the linearly
elastic slabs and base to Kk is not updated. This saves significant computational time, since much of the
work involved in forming Kk arises from the numerical integration of these large element stiffness
matrices.
13
e = U* − U (8)
Ke = r (9)
A small number of Gauss-Seidel iterations are performed for the finer meshes to remove high-
frequency error components, and the low frequency error components are approximated via a direct – and
inexpensive – solution on the coarsest mesh (Brandt 1977). Figure 9 presents the algorithm and a
conceptual overview for a two-mesh sequence. EverFE relies on a V-cycle multigrid scheme where r is
sequentially restricted from the finest to the coarsest mesh with symmetric Gauss-Seidel smoothing
performed at each step. Following the solution on the coarsest grid, the approximated error vector is
sequentially interpolated and smoothed from the coarsest to the finest mesh. EverFE’s use of a single
multigrid V-cycle to precondition a conjugate gradient iteration takes advantage of the symmetric positive
definiteness of the system stiffness equations.
One of the primary difficulties in implementing a multigrid scheme for un-nested mesh sequences of
spatially inhomogeneous domains such as those found in layered foundations is defining appropriate
restriction and interpolation operators. Restriction can be viewed as computing a force vector on a coarse
mesh that is statically equivalent to the known force vector on a finer mesh. This process is often
expressed in matrix form as follows, where rc denotes a coarse mesh residual force vector, rf denotes the
known fine mesh force vector, and R is the restriction operator:
r c = Rr f (10)
Similarly, interpolation is defined as the process of approximating the fine mesh error in the displacement
vector, ef, from a known coarse mesh error, ec using the interpolation operator, T:
e f = Te c (11)
subroutine MultiGrid(r f ,e f )
( )
smooth e f
smooth
rf =r f −K fef
restrict r c = Rr f
e f
= Te c
r c = Rr f
e c = K −1r
interpolate
e f = Te c
( )
smooth e f
The multigrid implementation used by EverFE relies on the element shape functions to define R and
T, which has been shown to be advantageous (Davids and Turkiyyah 1999; Fish et al. 1996). This allows
the restriction and interpolation operations to be performed on a node-by-node basis, provided that for
each fine mesh node, the coarse mesh element it lies within and the corresponding coarse mesh element
coordinates are known. Efficiently establishing this information is critical, and is achieved using
geometric search procedures detailed by Davids and Turkiyyah (1999). This general approach also
permits the easy meshing of solid and bending elements within the same model, since all calculations are
14
performed at the element level. This is critical, since the dowels are explicitly modeled as flexural
elements as discussed in Section 4.
15
generates nonlinear aggregate interlock constitutive relationships; and the user interface, which is written
in the scripting language Tcl/Tk.
Because of computational requirements, it was necessary to develop the finite-element, meshing, and
aggregate interlock model definition code using a low-level compiled language such as Fortran or C/C++.
Ultimately, C++ was chosen for two reasons: its flexibility (and thus the ease with which it allows the
implementation of the complex solution algorithms and specialized elements detailed in this manual), and
its effective dynamic memory allocation capabilities, which are crucial when solving large problems on
desktop computers. The C++ code is extensive, and details of its architecture are not presented here; for
more information, see Davids (1998).
The user interface is written in Tcl/Tk version 8.3, a freely available scripting language that runs on
multiple Unix, Windows and Macintosh platforms. Many of the higher-level interface widgets (dialog
boxes, checkboxes, menus, etc.) used by EverFE were taken from the additional Tcl/Tk package Tix
version 8.2, which is also freely available. Visualization of stresses, displacements, and dowel shears and
moments is accomplished with the Visualization Toolkit (vtk), freely available visualization software that
can be used with both Tcl/Tk and C++ (in EverFE it is called directly from Tcl/Tk). All of the EverFE
source code written in Tcl/Tk is installed with EverFE and used in uncompiled form, and these files must
never be modified.
The remainder of this section provides details on directory-file structure and the interaction of
EverFE’s constituent programs.
8.1 Directory-File Structure and Program Interaction
EverFE, like most software, is installed in a user-specified directory (the default location is
C:\Program Files\EverFE2.23). Figure 10 shows EverFE’s top-level directory structure.
All of the Tcl/Tk scripts are located in the scripts subdirectory. The FE-solver subdirectory
contains the files driver.exe (the executable that generates the finite-element input files) new_fe.exe
(the main finite-element executable), and agg_int.exe (a separate executable program that generates
and saves the nonlinear aggregate interlock constitutive model information used by the finite-element
code).
Basic model data is stored in the data subdirectory, which for each project contains a single file with
a .prj extension and a subdirectory that is necessary to store the project definition an analysis results.
The name of both the .prj file and the subdirectory corresponding to an EverFE project are identical to
the user-specified name the project is saved under. Within each project subdirectory, EverFE saves a file
called model_params.dat, an ASCII text file containing the information necessary to define a project
(model geometry, material properties, dowel information, meshing parameters, loading information, etc.).
16
The Tcl/Tk scripts read this file when a project is opened, and it also serves as the sole input source for
the program driver.exe, which generates the input files defining the finite-element mesh needed by
new_fe.exe. Note that the multigrid-preconditioned conjugate gradient solver detailed in Section 7
utilizes three finite-element meshes with decreasing levels of refinement, and thus driver.exe actually
generates three separate input files each time an analysis is executed. To save disk space, these files are
deleted after the analysis is completed. When an EverFE analysis executes successfully, additional files
are written to the project subdirectory that contain the model stresses, displacements and dowel results.
These output files are read directly by Tcl/Tk source code when the user enters any component of the
visualization panel; they are stored until the model is saved without re-analysis, at which point they are
deleted to ensure that a project never has a finite-element model definition that is inconsistent with the
saved output.
The agg_int directory contains input and output data for each nonlinear aggregate interlock model
that is generated and saved. There are two subdirectories within agg_int: one for models with metric
units and one for models with English units. The documentation directory contains two files:
EverFE_Help.chm, the compiled interactive help file developed with Microsoft’s HtmlHelp utility that
is accessed directly from the EverFE2.23 Help menu, and the file theory_manual.pdf. Finally, the
tcl_bins directory contains all of the binary code and libraries necessary to run Tcl/Tk/Tix/vtk.
When pre-defined projects are run in batch mode, the Tcl/Tk code simply puts the programs
driver.exe and new_fe.exe in a loop, sequentially analyzing each project.
9. References
Brandt, A. (1977). “Multi-Level Adaptive Solutions to Boundary-Value Problems.” Mathematics of
Computation, 31(138):333 – 390.
Brill, D.R., Hayhoe, G.F. and Lee, X. (1997). “Three-Dimensional Finite-Element Modeling of Rigid
Pavement Structures.” Aircraft Pavement Technology: In the Midst of Change, ASCE, pp. 151-165.
Davids, W. and Turkiyyah, G. (1997). Development of Embedded Bending Member to Model Dowel Action.
Journal of Structural Engineering, ASCE, 123(10):1312 – 1320.
Davids, W.G. (1998). Modeling of Rigid Pavements: Joint Shear Transfer Mechanisms and Finite Element
Solution Strategies. PhD Dissertation, University of Washington, Seattle, WA.
Davids, W.G. and Mahoney, J. (1999). “Experimental Verification of Rigid Pavement Joint Load Transfer
Modeling with EverFE.” Transportation Research Record 1684, TRB, National Research Council,
Washington, D.C., pp. 81-89.
Davids, W.G. and Turkiyyah, G.M. (1999). “Multigrid Preconditioner for Unstructured Nonlinear 3D FE
Models.” Journal of Engineering Mechanics, ASCE, 125(2):186-196.
Davids, W.G. (2000). “Effect of Dowel Looseness on Response of Jointed Concrete Pavements.” Journal of
Transportation Engineering, ASCE, 126(1):50-57.
Davids, W.G., Wang, Z.M., Turkiyyah, G., Mahoney, J. and Bush, D. (2003). “3D Finite Element Analysis of
Jointed Plain Concrete Pavement with EverFE2.2.” Transportation Research Record, TRB, National
Research Council, Washington, D.C. (in press).
Dei Poli, S., Di Prisco and Gambarova, P.G. (1992). “Shear Response, Deformations, and Subgrade Stiffness
of a Dowel Bar Embedded in Concrete.” ACI Structural Journal, 89(6):665-675.
Fish, J., Pan, L., Belsky, V. and Gomaa, S. (1996) “Unstructured Multigrid Method for Shells.” International
Journal for Numerical Methods in Engineering, 39:1181 – 1197.
Hjelmstat, K.D., Kim, J. and Zuo, K.H. (1997). “Finite Element Procedures for Three-Dimensional Pavement
Analysis.” Aircraft/Pavement Technology: In the Midst of Change, pp. 125-137, ASCE.
17
Ioannides, A.M. and Korovesis, G.T. (1990). “Aggregate Interlock: A Pure Shear Load Transfer Mechanism.”
Transportation Research Record 1286, TRB, National Research Council, Washington, D.C., 2001, pp. 14
– 24.
Ioannides, A.M. and Korovesis, G.T. (1992). “Analysis and Design of Doweled Slab-on-Grade Pavement
Systems.” Journal of Transportation Engineering, 118(6):745-768.
Jensen, E.A. and Hansen, W. (2003). “A New Model for Predicting Aggregate Interlock Shear Transfer in
Jointed Concrete Pavements.” Proceedings of EM2003-The 16th ASCE Engineering Mechanics
Conference, Seattle, WA, July 16-18, (CD-ROM).
Kuo, C., Hall, K. and Darter, M.. “Three-Dimensional Finite Element Model for Analysis of Concrete
Pavement Support.” Transportation Research Record 1505, TRB, National Research Council,
Washington, D.C., 1996, pp. 119 – 127.
Rasmussen, R.O. and Rozycki, D.K. (2001). “Characterization and Modeling of Axial Slab-Support
Restraint.” Transportation Research Record 1778, TRB, National Research Council, Washington, D.C.,
pp. 26 – 32.
Saad, Y. (1996). Iterative Methods for Sparse Linear Systems. PWS Publishing Co., Boston, MA.
Walraven, J.C. (1981). “Fundamental Analysis of Aggregate Interlock.” Journal of the Structural Division,
ASCE, 107(ST11):2245 – 2270.
Walraven, J.C. (1994). “Rough Cracks Subjected to Earthquake Loading.” Journal of Structural Engineering,
120(5):1510 – 1524.
Wang, C-K. and Salmon , C.G. (1985). Reinforced Concrete Design (4th Ed.). Harper and Row, New York.
Wattar, S.W., Hawkins, N.M. and Barenberg, E.J. (2001). “Aggregate Interlock Behavior of Large Crack
Width Concrete Joints in PCC Airport Pavements.” Technical Report, Department of Civil and
Environmental Engineering, University of Illinois at Urbana-Champaign.
Zhang, J. and Li, V.C. (2001). “Influence of Supporting Base Characteristics on Shrinkage-Induced
Stresses in Concrete Pavements.” Journal of Transportation Engineering, ASCE, 127(6);455 – 642.
Zienkiewicz, O.C. and Taylor, R.L. (1994). The Finite Element Method, Volume 1 (4th Ed.). McGraw Hill
Book Company, London.
18