TRANSPORTATION RESEARCH RECORD 1629 Paper No.

98-0313 41

EverFE
Rigid Pavement Three-Dimensional
Finite Element Analysis Tool

WILLIAM G. DAVIDS, GEORGE M. TURKIYYAH, AND JOE P. MAHONEY

The interactive and computational features of EverFE, a new rigid pave- At present, it is widely accepted that 3D finite element models are
ment three-dimensional (3D) finite element (FE) analysis tool, are pre- necessary to adequately capture the structural response of rigid
sented. To date, the use of 3D FE analysis has been hampered by (a) the pavement systems (7 ). However, there are still areas that must be
difficulty of model generation and result interpretation, (b) the inability of
many programs to adequately model joint shear transfer due to aggregate addressed if realistic 3D finite element models are to be routinely
interlock and dowel action, and (c) the large computational requirements used for the evaluation, retrofit, and design of rigid pavements.
of conventional solution techniques employed by available programs. The
development of EverFE is motivated by the desire to make 3D FE analy-
sis feasible for routine analysis of rigid pavements. The intuitive graphi- • Difficulty of model generation and result interpretation. The
cal user interface employed by EverFE, which greatly simplifies model
generation of realistic 3D finite element models is difficult and
generation and result interpretation, is demonstrated using a sample prob-
lem. A novel technique for modeling aggregate interlock joint shear trans- time-consuming. Even such common scenarios such as skewed
fer that rationally incorporates nonlinearities is developed and presented, joints, irregular dowel spacings, and extended shoulders can make
as is a new method for modeling dowel joint shear transfer. The solution model generation impractically tedious. Further, intuitive visual-
strategy employed by EverFE that allows realistic 3D models to be simu- ization software is mandatory to interpret stress and displacement
lated on desktop computers is briefly described, and its performance is results in 3D.
presented.
• Joint shear transfer modeling techniques. Aggregate interlock
and dowel shear transfer at transverse joints are complex phenom-
Many researchers have modeled the response of rigid pavement ena, and appropriate elements and/or constitutive models are not
systems to applied static, dynamic, and thermal loadings. Initially, present in available 3D finite element codes. This is a crucial aspect
two-dimensional (2D) finite element programs that rely on plate of rigid pavement modeling, since serviceability failures of jointed
elements to discretize the pavement slabs and base layers were rigid pavements often result from poor joint performance.
developed. These programs attempted to capture some of the more • Large computational requirements in terms of memory and CPU
complex aspects of the behavior of rigid pavement slabs such time. Most available 3D finite element codes employ direct solution
as dowel-concrete interaction, aggregate interlock, and loss of techniques that preclude the solution of large 3D models on com-
subgrade support. Tabatabaie and Barenberg (1) developed the monly available desktop computers, and present a further impediment
program ILLI-SLAB, which incorporates dowel-slab interaction to the routine use of 3D finite element models in practice.
springs, aggregate interlock springs, and the ability to model loss
of bond between the subgrade and foundation. The program JSLAB
(2) allows dowel-slab interaction to be modeled by treating the The rigid pavement analysis program EverFE has been developed
dowel as a beam on elastic foundation. in an attempt to make 3D finite element analysis more accessible to
These 2D programs have proved to be valuable tools, and since users in a broad range of settings. EverFE makes it simple and prac-
their inception have been extended beyond their original capabili- tical to explore the effects of various factors (dowel locations, gaps
ties and used in many studies [e.g., work by Guo et al. (3)]. How- around dowels, temperature effects, etc.) on pavement behavior, and
ever, the range of applicability of these models is limited, and they to perform parametric studies to evaluate different design and retro-
are not capable of capturing detailed local response, such as stress fit strategies. Within one package, EverFE incorporates graphical
distributions near joints. To overcome their limitations, researchers pre- and postprocessing capabilities tuned to the needs of rigid pave-
have resorted to the development of three-dimensional (3D) finite ment modeling and allowing transparent finite element model gen-
element models of rigid pavement using available finite element eration, innovative computational techniques for the modeling of
programs. Ioannides and Donelly addressed the problem of non- joint shear transfer, and efficient multigrid solution strategies. These
linear foundation support (4); Channakeshava et al. employed a non-
features permit realistic models with complex geometries to be gen-
linear constitutive model for the slab concrete and attempted to
erated in a matter of minutes, and solutions to be obtained on desk-
rationally incorporate dowel-slab interaction in a large 3D model
top personal computers in a reasonable amount of time. In this
(5). Another recent study (6) focuses on the response of rigid pave-
paper, the user interface and features of EverFE are first explored
ment systems to static loading, and comparison with available test
through the development and solution of a model problem. Second,
data and existing 2D programs.
details of the joint shear transfer modeling techniques and incorpo-
ration of slab lift-off are presented. Following this, the multigrid
Department of Civil Engineering, University of Washington, Box 352700, solution strategies are examined. The results are summarized in the
Seattle, WA 98195. final section.
42 Paper No. 98-0313 TRANSPORTATION RESEARCH RECORD 1629

MODEL GENERATION AND RESULT ful feature if no base layers are considered and the user wants to cap-
INTERPRETATION ture the effects of slab lift-off. The material properties assumed for the
present model problem are as follows: for the slab, E = 28 000 MPa,
In this section, the interactive and computational features of EverFE ν = 0.25, density = 2400 kg/m3, α = 1.1 ×10−05°C; for the elastic base
will be illustrated by the generation and solution of a model prob- layer, E = 175 MPa, ν = 0.25; for the dowels, E = 200 000 MPa,
lem consisting of two doweled rigid pavement slabs subjected to ν = 0.3. The Winkler foundation was specified to have a modulus of
axle and temperature loadings. The interactive, graphical post- 0.040 MPa/mm.
processing capabilities of EverFE will also be demonstrated. The
focus of this section is not on verification and interpretation of the
results from the model problem, but on the features of EverFE. Specification of Doweled Joints

EverFE allows the user to quickly specify dowels placed in common

Specification of Slab and Foundation Model patterns, such as equally spaced along transverse joints or located
only within the wheelpaths. The dowels are assumed to be located
Figure 1 shows the input panel of EverFE where the slab and foun- at mid-thickness of the slab and may be specified as bonded or
dation geometry are specified. EverFE permits the modeling of one debonded. All other aspects of the dowel locations and embedments
or multiple slabs with transverse joints at any orientation. Elastic are user-controlled, as shown in Figure 2, with immediate visual
base layers below the slab may be explicitly modeled, and the foun- feedback in the plan and elevation views of the system.
dation below the elastic base layers is treated as a dense liquid The dowel locations shown in Figure 2 are those used in the
(Winkler) foundation; extended shoulders may also be considered. model problem. Once the dowel configuration and spacings have
Immediate visual feedback is provided to the user as parameters been selected, EverFE computes the locations of individual dowels
and dimensions are changed. and presents them in a spreadsheet-type format than can be edited
As shown in Figure 1, the model problem consists of two slabs if the user wishes to specify individual dowel locations. This is a
3350 mm long × 3660 mm wide × 254 mm thick with a skewed joint useful feature for modeling as-built rigid pavement systems in the
and a single base layer with an extended shoulder. A separate input event that the dowels have been mislocated during construction. In
panel (not shown) is used to specify the material properties of the slab, addition, dowel looseness may be modeled by specifying a gap
base layers, foundation, and dowels. At present, EverFE assumes that between the dowels and the slab. The gap is assumed to vary lin-
the slab and foundation are linearly elastic; a possible future extension early from a maximum value at the face of the joint to zero at a
would be the inclusion of more realistic nonlinear material models. specified distance along the embedded portions of the dowel, as
The Winkler foundation may be specified to carry no tension, a use- shown in Figure 2.

FIGURE 1 EverFE geometry panel and model geometry.

Davids et al. Paper No. 98-0313 43

FIGURE 2 Definition of dowel locations in EverFE.

More details on the modeling of gaps between the dowels and the • Axle loads. Axle loads, consisting of two wheels, may also be
slab and the resulting contact nonlinearities are presented in a later directly specified. The wheels of each axle may be represented as
section. point loads or patch loads, and the transverse spacing of the wheels
(axle width) is also controlled by the user.

Specification of Aggregate Interlock Shear Transfer Figure 3 illustrates the model problem subjected to two axle loads
with each wheel represented by a rectangular patch. Again, visual
The user can model aggregate interlock using linear spring ele- feedback is provided in both the plan and elevation views of the
ments whose stiffness is specified in units of force per unit length rigid pavement system. Any number of loads may be specified, and
per unit area. Alternatively, a more realistic aggregate interlock any wheel or axle load may be interactively deleted from the model.
model based on work by Walraven (8,9) may be used; details of this In addition to wheel loads, a linearly varying temperature gradient
model are presented in a later section. If a Walraven-type method through the thickness of the slab can be specified. This temperature
is chosen to model aggregate interlock, several joint constitutive gradient can be applied concurrently with any combination of wheel
models based on typical parameters are available. Alternatively, loading and is assumed to be identical for each slab. For the model
appropriate parameters may be specified and a new aggregate inter- problem currently being specified, the top surface of the slab is set at
lock constitutive model developed. In the present model problem, −5°C and the bottom surface at +5°C. The user may choose to model
slab/base layer separation, which is important when considering tem-
aggregate interlock shear transfer is modeled using linear springs,
perature loading; details of how this effect is incorporated may be
with a stiffness of 0.5 MPa/mm.
found in a later section. It should be noted that if a temperature gra-
dient is specified and loss of contact is modeled, the top of the upper-
most base layer will be subjected to the same temperature change as
Specification of Loads
the bottom of the slab.
A direct manipulation interface allows users to interactively locate,
move, and specify the magnitude of various types of vertical loads: Meshing and Solution

• Point loads. Used to represent a single wheel load. EverFE gives the user control over the following solution and mesh-
• Circular patch loads. Used to represent a single wheel load. ing options:
The user specifies the total load and the radius of the circular area
over which the load is uniformly distributed. • Solution technique. The user has control over refinements in the
• Rectangular patch loads. Similar to circular patch loads, with solution technique and is able to select from multigrid variants that
the load distributed over a rectangular area. optimize either memory usage or solution time.
44 Paper No. 98-0313 TRANSPORTATION RESEARCH RECORD 1629

FIGURE 3 EverFE loading panel and model loads.

• Contact modeling. The user may choose whether or not to The sample model has been solved using a fine mesh with 19,155
model the slab separating from the upper base layer using a nodal degrees of freedom, and slab lift-off considered. The faster multi-
contact approach; details of the incorporation of the nodal contact grid variant multigrid preconditioned conjugate gradient (CG)/
conditions are presented in a later section. It is assumed that there is direct was used in the solution, which required 9 Newton iterations
no horizontal transfer of load between the slab and subgrade when and approximately 1 hr to solve on a Pentium 166 PC. EverFE
loss of contact is modeled (i.e., frictionless contact is assumed). required approximately 75 MB of RAM to solve this problem.
• Mesh generation. An automatic mesh generator is available in
the system. It produces hexahedral elements for the slabs and base
layers, surface elements for the subgrade, and beam and interface Visualization of Results
elements for modeling joint shear transfer. The user only needs to
specify a desired level of refinement for which the mesh generator EverFE allows the displaced shape of the system to be viewed as
produces the appropriate discretization. a wireframe in three dimensions. Figure 5 shows the displaced
shape of the current model subjected to axle and temperature load-
A typical mesh generated by EverFE is shown in Figure 4. Note ings. Note that the slabs have lifted off the base layer (not shown
that this mesh need not be visible to the user. for clarity) due to temperature curling. Several options are avail-
able to the user when viewing the displacements. The displaced
EverFE employs an isoparametric, 20-noded quadratic hexahedral and/or undisplaced shapes of any combination of the slabs may be
element (10), shown to be superior to an 8-noded linear brick element viewed, as well as the displaced and/or undisplaced shapes of the
for modeling rigid pavement systems (6). The dense liquid founda- subgrade. In addition, the user can zoom in or out and rotate the
tion is meshed using isoparametric, 8-noded quadratic elements. All 3D wireframe interactively using mouse controls. The value by
elements are numerically integrated using Gaussian quadrature. By which the displacements are scaled for plotting purposes is also
default, the horizontal boundary conditions are the minimum required user-definable. Color maps of principal stresses can be viewed in
to prevent rigid body motion of the system. If slab/base separation is user-specified x-y, x-z, and y-z planes corresponding to mesh divi-
specified, each slab has three zero-displacement constraints (enforced sions. Either the maximum (largest tensile) or minimum (largest
at two nodes on the bottom of each slab) that prevent horizontal dis- compressive) stress can be selected; the color scale may be based
placements and rotations of the slabs. One set of similar constraints is on the magnitudes of stresses over the entire model (global), or
used for the base as a whole; no lateral restraints are imposed on the only those in the selected plane (by slice). Detailed numerical val-
base layers. If slab/base loss of contact is not considered, only the base ues for all stress and displacement components may be retrieved
as a whole is restrained, as the slabs are considered bonded to the by selecting a point in the color plot and may be viewed in the
uppermost base layer and need not be individually restrained. input panel.
Davids et al. Paper No. 98-0313 45

FIGURE 4 Finite element idealization of two-slab system.

Summary: Model Generation and Result Interpretation as the slab nodes are required to coincide with the dowel locations.
To circumvent this problem, EverFE models the portion of the dow-
In this section, the graphical user interface (GUI) of EverFE, allow- els within the slab using an embedded quadratic beam element as pre-
ing users to rapidly and intuitively generate and solve finite element sented by Davids and Turkiyyah (11). As shown in Figure 6, this
models of rigid pavement systems and interpret their results, has been allows the dowel to be meshed independently of the slab. The dowel
illustrated. The efficiency and ease of use of the GUI allow users to nodes are constrained to displace compatibly with the solid element
easily test different designs, perform parametric studies, and analyze the dowel lies within by a simple transformation of the dowel stiff-
as-built or retrofitted configurations. EverFE is currently being tested ness matrix. This ensures no loss in solution accuracy over a con-
by pavement engineers at the Washington State Department of ventional formulation and permits the dowel to be debonded by
Transportation (WSDOT), where it has been favorably received. releasing the dowel’s axial degrees of freedom. The portion of the
Underlying this GUI is a set of powerful computational features dowel spanning the joint is modeled with a conventional shear beam.
that rely on recently developed approaches to modeling loss of con- Also of concern to pavement designers is the presence of gaps
tact and dowel and aggregate interlock shear transfer. Further, solu- between the dowels and slab that are the result of repeated loading
tions on modest desktop computers are possible only because of the (12). Recent attempts to rigorously model dowel-slab interaction have
unique iterative solution strategies employed by EverFE. The re- been limited to 2D models (3,13), and gaps between the dowels and
mainder of this manuscript presents details of contact and joint shear slab are explicitly considered only by Zaman and Alvappillai (13).
transfer modeling and briefly describes solution strategies. The embedded dowel formulation employed by EverFE allows these
gaps to be explicitly modeled. At the start of the solution, the dowel
nodes in the gap region are permitted to displace independently of the
JOINT AND CONTACT MODELING solid element they lie within. At each solution iteration, the displace-
ments of the dowel nodes relative to the solid embedding elements are
Dowel Modeling monitored; when the displaced dowel node comes into contact with
the embedding element, constraints preventing interpenetration are
In 3D finite element models of rigid pavements, dowels are typically activated. Preliminary studies on the effect of these gaps on load trans-
represented by beam elements meshed explicitly with the slab ele- fer efficiency and further details of the embedded dowel formulation
ments. This places severe limitations on the slab mesh development, may be found elsewhere (11).

FIGURE 6 Plan of single slab (left); embedded dowel element

FIGURE 5 Displaced shape of slabs. (right).
46 Paper No. 98-0313 TRANSPORTATION RESEARCH RECORD 1629

Aggregate Interlock Shear Transfer τ = σ pu ( Ay + µ Ax ) (3)

Traditionally, aggregate interlock shear transfer has been modeled where

with simple linear spring elements (6,14). EverFE also permits a
linear aggregate interlock model, which is implemented with a σpu = strength of cement paste,
16-noded, “zero-thickness” quadratic interface element meshed µ = coefficient of friction between paste and aggregate,
–
between two quadratic hexahedral elements. The element “stiff- Ax = x-projection of sum of most probable contact areas, and
–
ness” is specified in units of force per length per unit area, analogous Ay = y-projection of sum of most probable contact areas.
to the modulus specified for a Winkler foundation. – –
The determination of Ax and Ay is based on the statistical distri-
Although this approach is computationally simple, it does not bution of the aggregate in the concrete matrix, and the geometry of
allow the complex mechanism of aggregate interlock shear transfer
the spherical aggregate particles intersecting the crack plane at a
to be accurately modeled. Further, the rational choice of an appro-
given tangential and normal displacement. Walraven gives typical
priate spring stiffness is difficult. To overcome these limitations,
values for σpu and µ that correlate well with laboratory test data.
EverFE allows detailed two-phase constitutive relations for aggre-
From these basic assumptions, Walraven’s model can be used to
gate interlock shear transfer as developed by Walraven (8,9) to be
generate curves relating shear stress and displacement for a given
incorporated in the finite element model.
joint/crack opening (Figure 8). Such curves are used as constitutive
Walraven formulated the basic relationships between stress and
relations to model nonlinear aggregate interlock shear transfer in
displacement across a discrete crack based on the representation
EverFE’s 16-noded quadratic interface element. This permits the
of concrete as constituted by two separate materials: the cement
paste, and the aggregate. The weakest link in the concrete matrix is variation in shear transfer through the thickness of the joint due to
assumed to be the bond between the cement paste and the aggre- differing horizontal joint displacements to be captured. The authors
gate, and cracking thus occurs along the aggregate boundaries. note that EverFE generates these curves transparently to the user,
Because of the large plastic deformations of the cement paste due who needs to specify only the strength of the cement paste and the
to pore-volume reduction, the cement paste is idealized as obeying paste-aggregate coefficient of friction.
a rigid-plastic stress-strain law.
The aggregate particles are idealized as incompressible spheres
of varying size, distributed according to a Fuller curve. The aggre- Loss of Contact Between Slab and Base
gate particles intersect the crack face at various depths depending on
their statistical distribution within the concrete matrix (Figure 7). As Modeling the loss of contact between a slab and an unbonded base
a result of these assumptions, the crack may be reduced to a 2D layer is critical when considering temperature-induced curling; it
problem of finite width. The shear and normal stresses at the crack may also be significant when examining the effects of wheel load-
result from the tangential and normal projections of the stresses pro- ing at the joint (Figure 9). EverFE permits the user to model slab lift-
duced when the cement paste deforms plastically as it bears on the off using a nodal contact approach. The slab and the base layer are
aggregate particles (Figure 7). meshed separately but in such a way that the locations of the bottom
These stresses are related by assumption that the contact areas are nodes of the slab coincide with the locations of the top nodes of the
about to slip, and thus base. Stress and displacement conditions at each coinciding pair of
slab/base nodes are monitored during the solution, and the nodes are
τ pu = µσ pu (1) appropriately constrained or released. Note that at the joint, a single
base node is shared with two slab nodes—that is, one slab may be
The equilibrium of a crack plane of unit area requires that the net in contact with the unbonded base, while the other slab separates
forces be balanced as follows: from the base layer at the same point (see Figure 9). In this manner,
differential vertical joint displacements may be properly captured,
σ = σ pu ( Ax − µ Ay ) (2) even when modeling the base as an elastic continuum. If no base

FIGURE 7 Distribution of aggregate (left) and stresses (right) on spherical particle.

FIGURE 8 Shear stress versus displacement for various joint openings.

FIGURE 9 Loss of contact between slab and subgrade.

48 Paper No. 98-0313 TRANSPORTATION RESEARCH RECORD 1629

layers are modeled, a no-tension Winkler foundation may be speci- unknowns (degrees of freedom), drastically increasing the required
fied directly below the slab to model the loss of contact. computational effort as the finite element model is refined. This
It must be noted that nodal contact modeling requires the incor- combination of memory and computational requirements makes
poration of inequality constraints and an appropriate nonlinear direct methods impractical on modest desktop computers when
analysis. An efficient method for solving constrained problems solving realistic 3D models of rigid pavement systems.
using iterative solution techniques may be found in work by Davids To circumvent this problem, iterative methods for solving Equa-
and Turkiyyah (15), which outlines the solution strategies employed tion 6 must be employed in which the matrix factorization for K is
by EverFE. Next, the solution strategy is briefly described. never computed. EverFE uses multigrid methods to solve Equation
6, which at present are the most efficient iterative techniques avail-
able. Specifically, EverFE uses a multigrid preconditioned CG
HIGH-PERFORMANCE SOLUTION STRATEGIES method, which takes advantage of the symmetry of K. The use of
multigrid methods to solve problems in structural mechanics has
Another ingredient for making finite element modeling a practical been actively researched in recent years (16,17 ). Theoretically,
tool for rigid pavement analysis is computational efficiency. A prac- multigrid methods require an amount of work only proportional to
tical tool must be able to simulate realistic problems on typical desk- the number of unknowns to solve Equation 6 if certain conditions
top machines in a reasonable amount of time using a reasonable are met (16). Although achieving this linearity is difficult in prac-
amount of memory. EverFE employs Newton’s method to solve the tice, the authors’ studies indicate that the gains in efficiency pro-
system equilibrium equations given by vided by multigrid methods over direct solution techniques are quite
dramatic. Details of the multigrid methods employed by EverFE and
P−F=0 ( 4) their performance may be found elsewhere (15).

where P is a constant vector of external load effects, and F repre-

sents the vector of element nodal forces, which vary as a function of SUMMARY AND CONCLUSIONS
the nodal displacement vector, U. EverFE employs Newton’s
method for the solution of Equation 4, which requires the formation EverFE, a new rigid pavement 3D finite element analysis tool, has
of the tangent system stiffness matrix, K, defined as been developed. It incorporates an intuitive Windows-based GUI
for model generation and result interpretation with powerful com-
K = ∂F/ ∂U (5) putational tools and solution strategies. These features have been
demonstrated via the generation and solution of a realistic model
For both linear and nonlinear finite element models, there are two problem consisting of two rigid pavement slabs on a base layer and
primary sources of computational effort: (a) assembly of K, and dense liquid foundation subjected to wheel and temperature loadings.
(b) solution of the system equilibrium equations Specifics of the joint and contact modeling techniques employed
by EverFE have also been examined. Dowel modeling is accom-
K∂U = r (6) plished using a newly developed embedded bending element that
permits independent meshing of the dowels and modeling of dowel
for the displacement increment, ∂U, corresponding to the current looseness. Aggregate interlock may be modeled using linear spring
vector of residual forces, r. For both linear and nonlinear problems, elements or using a two-phase model that rationally accounts for
K is initially assembled prior to entering the Newton iteration, and nonlinear joint response and the variation in shear transfer with joint
the amount of work required to assemble K increases only linearly opening. Loss of contact between the slab and base may be modeled
with the number of elements. The amount of effort required to using a nodal contact approach.
update K at each Newton iteration of a nonlinear problem has been For the solution of the system equilibrium equations, EverFE
minimized by not recomputing the contributions of the slab and sub- employs a high-performance multigrid preconditioned CG strategy
grade, which are linearly elastic. For large problems, it is the solu- that is much more efficient than conventional direct solution meth-
tion of Equation 6 that consumes the most resources, and its efficient ods in terms of both computational time and memory use. This per-
solution is of primary importance. mits realistic 3D models to be run on modest desktop computers
The method used in most finite element pavement analysis pro- using EverFE.
grams for solving Equation 6 is a direct method requiring that the Comparison of results from EverFE with available experimental
matrix K be factored into lower and upper triangular matrices, L and data is currently under way. The scale-model laboratory studies of
U, such that Hammons (18) have been chosen for this purpose. These tests will
permit the comparison of computed and measured deflection basins
K = LU ( 7) of corner-loaded slabs with doweled joints that are founded directly
on a rubber mat (dense liquid) as well as on bonded and unbonded
Unfortunately, large 3D models of rigid pavement systems lead to cement-treated bases above a rubber mat. The preliminary results of
systems of equations for which the factors L and U typically require these comparisons show good agreement. A detailed study on the
three to five times as much storage as K itself, even when sparse data verification of EverFE will be a topic of future research.
structures are employed. If the symmetry of K is accounted for, only Combining an intuitive user interface, rational joint and contact
one factor need be stored since L = UT. However, the required stor- modeling features, and high-performance solution strategies,
age is still clearly significant; for example, the sample problem EverFE makes routine 3D finite element analysis of rigid pavement
solved earlier would require approximately 60 MB of RAM for the systems feasible in design and research settings. EverFE is being
storage of L or U alone. In addition, the amount of work needed to tested by WSDOT, where it has been well received. EverFE is avail-
solve Equation 6 grows faster than the square of the number of able for Windows95/NT and UNIX machines. Documentation and
Davids et al. Paper No. 98-0313 49

download information for the latest version are available from 7. Hammons, M. I., and A. M. Ioannides. Developments in Rigid Pave-
http://www.ce.washington.edu/∼everfe. ment Response Modeling. Technical Report GL-96-15. U.S. Army
Corps of Engineers, Aug. 1996.
8. Walraven, J. C. Fundamental Analysis of Aggregate Interlock. Journal of
the Structural Division, ASCE, Vol. 107, No ST11, 1981, pp. 2245–2270.
ACKNOWLEDGMENTS 9. Walraven, J. C. Rough Cracks Subjected to Earthquake Loading. Jour-
nal of Structural Engineering, Vol. 120, No. 5, 1994, pp. 1510–1524.
This work was supported in part by WSDOT and by a fellowship 10. Zienkiewicz, O. C., and R. L. Taylor. The Finite Element Method,
Vol. I. McGraw-Hill Book Company, 1994.
from the Valle Scholarship and Scandinavian Exchange Program at 11. Davids, W., and G. Turkiyyah. Development of Embedded Bending
the University of Washington. The authors also wish to thank Ross Member to Model Dowel Action. Journal of Structural Engineering,
TenEyck for his efforts in developing the user interface. Vol. 123, No. 10, 1997, pp. 1312–1320.
12. Buch, N., and D. G. Zollinger. Development of Dowel Looseness Pre-
diction Model for Jointed Concrete Pavements. In Transportation
Research Record 1525, TRB, National Research Council, Washington,
REFERENCES D.C., 1996, pp. 21–27.
13. Zaman, M., and A. Alvappillai. Contact-Element Model for Dynamic
1. Tabatabaie, A. M., and E. J. Barenberg. Finite-Element Analysis of Analysis of Jointed Concrete Pavements. Journal of Transportation
Jointed or Cracked Concrete Pavements. In Transportation Research Engineering, Vol. 121, No. 5, 1995, pp. 425–433.
Record 671, TRB, National Research Council, Washington, D.C., 1978, 14. Ioannides, A. M., E. J. Barenberg, and M. R. Thompson. Finite Element
pp. 11–19. Model with Stress Dependent Support. In Transportation Research
2. Tayabji, S. D., and B. E. Colley. Analysis of Jointed Concrete Pave- Record 954, TRB, National Research Council, Washington, D.C., 1984,
ments. Technical Report FHWS-RD-86-041. FHWA, U.S. Department pp. 10–16.
of Transportation, 1986. 15. Davids, W., and G. Turkiyyah. A Multigrid Preconditioner for Unstruc-
3. Guo, H., J. A. Sherwood, and M. B. Snyder. Component Dowel-Bar tured 3D FE Models Incorporating Contact Nonlinearities. Journal of
Model for Load Transfer Systems in PCC Pavements. Journal of Trans- Engineering Mechanics, ASCE (in press).
portation Engineering, Vol. 121, No. 3, 1995, pp. 289–298. 16. Parsons, I. D., and J. F. Hall. The Multigrid Method in Solid Mechan-
4. Ioannides, A. M., and J. P. Donelly. Three-Dimensional Analysis of ics: Part I—Algorithm Description and Behavior. International Journal
Slab on Stress-Dependent Foundation. In Transportation Research for Numerical Methods in Engineering, Vol. 29, 1990, pp. 719–737.
Record 1196, TRB, National Research Council, Washington, D.C., 17. Muttin, F., and J.-L. Chenot. On a Conjugate-Gradient Two-Grid
1988, pp. 72–84. Method for Three-Dimensional Elasticity. Engineering Computations,
5. Channakeshava, C., F. Barzegar, and G. Voyiadjis. Nonlinear FE Analy- Vol. 12, 1995, pp. 3–20.
sis of Plain Concrete Pavement with Doweled Joints. Journal of Trans- 18. Hammons, M. I. Development of an Analysis System for Discontinuities
portation Engineering, Vol. 119, No. 5, 1993, pp. 763–781. in Rigid Airfield Pavements. Technical Report GL-97-3. U.S. Army
6. Kuo, C.-M., K. T. Hall, and M. I. Darter. Three-Dimensional Finite Ele- Corps of Engineers, April 1997.
ment Model for Analysis of Concrete Pavement Support. In Trans-
portation Research Record 1505, TRB, National Research Council, Publication of this paper sponsored by Committee on Rigid Pavement
Washington, D.C., 1996, pp. 119–127. Design.