THREE- BIOGRAPHY SUMMARY

DIMENSIONAL Paul Biju-Duval is a PhD Evaluating the stability of

FINITE ELEMENT student at UT Austin from Paris, curved and skewed I-girder
France. Before going back to bridges during construction is of
SOFTWARE FOR school, Paul worked four years critical importance to designers
STEEL BRIDGE as construction engineer in and erectors. The erection and
Dubai and then design engineer construction phases are often
ERECTION AND in Paris. Paul holds a Master’s the critical stage for stability of
CONSTRUCTION degree from Georgia Tech. the system; however obtaining a
good model of the bridge during
Colter Roskos is a PhD student
these phases can be very
at UT Austin from Kalispell,
difficult. The UT Bridge
MT. His research focuses on the
software, first released in 2010,
use of precast concrete panels
was developed with a user-
for curved bridges applications.
friendly pre-and post-processor
Victoria McCammon, P.E., is a to provide a valuable tool for
Master’s student at UT Austin. designers and erection engineers
She has worked several years specifically targeting these
for TxDOT Bridge Division. critical stages. The program
Todd Helwig is a Professor at allows an engineer to create a
UT Austin in the CAEE three-dimensional model of
Department. He is also Vice- straight and curved bridge
President of the SSRC. systems so that the behavior
during erection and construction
PAUL BIJU-DUVAL Eric Williamson is a Professor can be evaluated while
at UT Austin in the CAEE considering a wide variety of
Department. He also serves as erection or deck placement
the ASCE/SEI Technical scenarios. While the software
Administrative Committee has provided a valuable tool that
Chair for Bridges. has filled a void in the bridge
Oguzhan Bayrak is a Professor industry for analyzing straight
at UT Austin in the CAEE and curved steel bridge systems,
Department. there are a number of key
features of the software that are
Patricia Clayton is an Assistant being improved and expanded.
Professor at UT Austin in the This paper therefore focuses on
CAEE Department. an overview of the software
Michael Engelhardt is a capabilities and a discussion of
Professor at UT Austin in the the modifications currently
CAEE Department. He is also underway as well as planned
TODD HELWIG the Director of the Ferguson modifications.
Structural Engineering
Laboratory.
THREE-DIMENSIONAL FINITE ELEMENT SOFTWARE FOR
STEEL BRIDGE ERECTION AND CONSTRUCTION
Introduction and Background of 3D shell models and although it was believed
that such models would eventually replace these
The critical stage for the stability of many bridges simplified models, the complexity involved in
often occurs during construction when the loading creating 3D models as well as the post-processing
and support conditions of the bridge are widely
required to comprehend the results from such a
variable. Historically, the majority of bridge
failures have occurred during construction due to model have greatly hampered wide-spread use. UT
the unpredictable behavior of the structural system Bridge aims at filling this gap. An improved
from the perspective of loading, support modeling approach to conventional grid models
conditions and stability. The safety of the erection was considered by Chang (2), using thin-walled
scheme is dependent upon the evaluation of the beam theory. A program for simulating the
behavior with several critical stages involving the construction of curved steel I-girders was
partially erected structure when limited bracing proposed and used in NCHRP Report 725
may be available. In addition, a great deal of “Guidelines for Analysis Methods and
uncertainty on the structural behavior can occur Construction Engineering of Curved and Skewed
during the concrete deck placement in which the Steel Girder Bridges” (3). However, the large size
steel girder alone must support the full of the input files has also limited the use of such
construction load and the system is susceptible to a models. While historically the use of 3D
variety of potential instabilities. While most
modelling techniques were hampered by limited
commercial software programs focus on the
analysis of completed bridges in their fully- computational resources, technological
composite state, the authors are aware of no improvements have resulted in the ability for
commercial analysis programs that are currently simple laptop and desktop computers to carry out
available to model accurately in a fast way very sophisticated analyses. However the
partially erected steel I-girder bridges. Curved bottleneck that is slowing the use of 3D modelling
bridges in particular are difficult to model, and the in routine analyses is actually the necessary time
combined bending and torsion that occurs due to for users to become proficient in the modelling
the geometry can result in a flexible structural techniques as well as the required time to create
system in cases with a partial erected system. In these detailed models and analyze the results.
the development of a suitable erection and
concrete deck placement scheme, obtaining a good In an effort to fill the void of available analysis
prediction of the deformations and strength of the options for designers and erection engineers, the
system at the various construction stages can be program UT Bridge was developed (Stith, 2010)
critical to the success of bridge construction. A through funding provided by the Texas
structural analysis providing this valuable Department of Transportation (TxDOT). The
deformation and strength prediction will often software allows the user to efficiently develop a
result in the elimination of problematic conditions 3D model of steel I-girder bridges for the analysis
during construction and can therefore avoid costly during erection or deck placement. While such
construction delays or unsafe conditions for the models can be generated in several general-
construction personnel and travelling public. purpose finite element programs that are
commercially available, the creation of such
Most current commercial programs that are used
models can take several days or weeks even for
in practice for the analysis and design of curved the most experienced analyst. The UT Bridge
and skewed I-girder bridges (such as MDX or software consists of three modules that include a
DESCUS) perform a refined grid analysis. Zureick pre-processor, processor, and a post-processor.
(1) acknowledged that the most accurate means to The user defines the basic bridge geometry
analyze complex bridge geometries is with the use through a series of input forms in the pre-

Page 1 of 11
processor (as described in Figure 1), which was The program is currently in Version 1.6 and has
developed using a Visual Basic (VB) interface. been well accepted and used by a number of
The user can select either an eigenvalue analysis designers and erection engineers. As with most
or a first-order structural analysis. Once the software packages, through the use, various bugs
geometry of the bridge is defined, the pre- and limitations in the software have been
processor also allows the user to specify the identified since its release in 2010. Because the
erection scenario or deck placement scheme software is free, the developers have not had the
considering temporary supports such as shore resources to continually update the software;
towers or holding cranes. When the user is however a major effort began in 2014 to update
satisfied with the geometry and erection/deck and improve the software.
placement scenario, the input file is sent to the
The expected release date of Version 2.0 of the
FORTRAN-based processor that is based upon the
software (henceforth referred to as V2.0 in the
pre-processor to carry out the structural analysis.
remainder of the document) is expected in mid- to
Upon completion of the analysis, the user can then
late- 2016. Although the basic framework of the
open and review the results in the C++ based post-
software is the same (pre-processor, processor,
processor.
post-processor), these individual components have
been totally changed from Version 1.6 (henceforth
referred to as V1.6). In addition to changing the
pre- and post-processors, V2.0 also includes new
element formulations with improved accuracy, a
new solver, and a new eigensolver. V2.0 will
result in significant improvements in the accuracy
and capabilities of the software. The program
provides a 3D model with improved modelling
accuracy of curved and skewed I-girders systems
using quadratic isoparametric shell elements for all
primary steel members (connecting plates and
stiffeners included) and truss or beam elements for
all brace members, using state-of-the-art solver
and eigensolver algorithms.
While UT Bridge is able to perform a deck
placement analysis, the scope of this paper is
limited to the erection sequence analysis of I-
girder systems in V2.0. The goal of this paper is to
outline the basic changes to the program and to
demonstrate the use of the program. Validation
results of V2.0 with commercially available
general-purpose finite element programs are
provided. In addition, a brief overview of
subsequent changes that are under consideration
for the software are also provided.

Model Generation
Modeling of the structure in UT Bridge is
achieved through a user-friendly interface using a
series of VB forms. The geometry is defined using
a minimum number of parameters, namely the
Figure 1 – Pre-processor (user-friendly VB forms)
bridge type (straight or curved), the number of
architecture of UT Bridge Version 2.0

Page 2 of 11
girders, the length of each girder, the girder
spacing (which may be non-uniform), the radius of
curvature (in case of a curved bridge), and the
support skew. Nodes are defined at the mid-plane
of the shells. The program uses Cartesian
coordinates for straight bridges and cylindrical
coordinates for curved bridges (using the center of
curvature of the bridge as the origin of the
coordinate system), which allows for easy and
exact nodes coordinates calculation. The accuracy
of the node generation algorithm results in a
proper shell orientation, which is critical to the
definition of the shells stiffness matrices.
Therefore, the nodes are defined in the local
isoparametric coordinates and must be
transformed to global coordinates. Shell
kinematics is also kept accurate, as a set of local Figure 2 – 8-noded isoparametric finite element
orthogonal vectors at each node is required. Applying boundary conditions to curved girders
Separate sets of orthogonal vectors are defined, requires special care. While the overall stiffness
depending on whether the shell represents part of a matrix is defined in global Cartesian coordinates
flange, a web, a connecting plate or a stiffener. As (even for curved girders), the restraints at each
a single node may be shared by a flange and/or a support, however, are specified in local
web and/or a connecting plate finite element, coordinates as demonstrated in Figure 3. For
several local sets of vectors may be required for example, restraining the out-of-plane displacement
individual nodes. for a curved girder defining a 90° angle in plane
The finite element that was selected for V2.0 is an means restraining the x-displacement for a node
isoparametric eight-noded quadratic general shell located at the start of the girder, but restraining the
element such as that depicted in Figure 2. The y-displacement for a node located at the end of the
element uses five degrees of freedom per node girder. Such restraints may be accomplished either
(three displacements, two rotations, with no numerically, by adding springs at the
“drilling” degree of freedom). This element is corresponding degrees of freedom, or through the
commonly recommended to model thin shells. The definition of local degrees of freedom that are then
element is very similar to the most commonly used restrained. Adding springs is often not the optimal
elements in many general-purpose FEA programs. solution, since a large stiffness value may result in
The program that is being used to validate Version an ill-defined stiffness matrix, while a low
2.0 is ABAQUS (5), which has the S8R5 element stiffness value may lead to insufficient restraint.
that is similar to the new element. Comparisons However, working on local degrees of freedom for
between V2.0 and the ABAQUS models are the stiffness matrix assembly and solving, and then
subsequently presented in this paper. Full back transforming the local displacements to
integration on the element was adopted, using overall Cartesian displacements, proves to be the
three integration points in each direction. most accurate method of applying boundary
Integration through the thickness is done using conditions to a curved girder.
two Gauss points. Overall, a series of eighteen Specifically, the restrained degrees of freedom are
integration points are therefore considered. the following: in the case of a single girder (no
cross-frame), a pin support requires restraints in
the three displacements in the bottom flange to

Page 3 of 11
web node, as well as the out-of-plane displacement The resulting cross-section consists of two
of the top flange to web node; a roller support elements for each flange and four for each web.
requires restraints in the vertical and out-of-plane Having two elements per flange was considered
displacements of the bottom flange to web node, enough, especially when using a quadratic shell
as well as the out-of-plane displacements of the element. Based on an aspect ratio that is targeted
top flange to web node. In the case of multiple to be close to one, or in the worst case under five,
girders braced together, the pin or roller restraints having four elements for the web was considered
are applied only at the bottom flange to web adequate.
support nodes. By restraining only the translations
Although UT Bridge V1.6 models the cross frames
of only the flange to web nodes, twist can be
as tension-only diagonal systems, V2.0 allows the
adequately restrained, but warping is still allowed.
user to specify either compression system x-
frames or k-frames (regular or inverted). Another
new addition to V2.0 is the ability to specify a
lateral truss at the top and/or bottom flange. The x-
frames are modeled using three-dimensional truss
bars connected to the mid-edge nodes of the outer
web shells. The k-frames are connected to the
same nodes but use beam elements for the top and
bottom chords and truss bars for the diagonals.
Finally, the lateral trusses are connected on the
flanges exterior nodes and are modeled using truss
bars. The truss and beam element stiffness is
computed in the local coordinates and then
converted to global Cartesian coordinates.
Defining braces is easily achieved in the VB forms
by specifying their location on the girders. For
curved girders, working on cylindrical
Figure 3 – Global vs. local degrees of freedom coordinates, and specifically, on the curvilinear
coordinate along the girder, results in quick and
One of the strengths of the UT Bridge software
accurate modeling.
has been the ability to model temporary supports
such as shore towers or holding cranes to Meshing of the model is achieved automatically
determine the impact on the behavior of the after the call to a separate FORTRAN program
partially erected system. Shore towers are that stores a series of “key points” along each
temporary restraints that physically behave similar girder. Those key points may correspond to
to roller supports, restraining the bottom flange to splices, supports, stiffeners, or any other location
web node vertical and out-of-plane displacements, that the user selects based from the bridge framing
and the top flange to web node out-of-plane plans. Exporting framing plans to actual models is
displacement. Therefore, it is assumed that the therefore made relatively simply. For each girder,
girder will be braced at the shore tower location. once the necessary parameters are defined, the
Holding cranes are modeled as an upward reaction program automatically meshes the model, based
at the point of support (on the top flange) – but on a mesh size specified by the user. Modeling the
twist and lateral deformation are not restrained. structure and meshing the model can often be
carried out in a matter of minutes. The automatic
At each section along the length of a girder, a
meshing does not prevent the user from selecting
series of seventeen nodes are defined, which
different mesh sizes for different segments. For
results in eight elements as depicted in Figure 4.
example, to minimize the size of the stiffness and

Page 4 of 11
geometric stiffness matrices, the user may opt to stiffness matrix, U the displacement vector, and F
use a relatively coarse mesh. But at segments the load vector, the equation K*U = F, is solved
where the bridge behavior requires more precision, using the latest PARDISO solver available in the
for example around negative moment regions, the INTEL FORTRAN compiler. The PARDISO
user may define a finer mesh. subroutine allows for a quick and accurate solution
of a set of sparse linear equations. It should be
noted that the stiffness matrix, as well as the
geometric stiffness matrix, are stored in the
Compressed Row Storage (CSR) format. This
means that only the non-zero matrix coefficients
are stored, which in the case of the highly sparse
matrices defined in structural mechanics represent
less than 1% of all the matrix coefficients for
simple models, or even less than 0.1% or 0.01%
for large or very large models. This allows for
optimized memory storage and faster solving.

Figure 4 – Mesh and node orientation

Furthermore, by defining different steps and
turning on user-specified segments for each girder
at each step, UT Bridge is able to perform a full
erection sequence analysis. Based on the selected
segments, the program automatically activates or
deactivates the corresponding nodes, elements,
braces, loads and supports. The user is also
prompted to specify when to activate and
deactivate temporary supports. Altogether, this
enables a quick analysis of a structure at different
steps along the corresponding erection sequence
instead of having to work on a different model for
each step. Figure 5 – Processor architecture
Processor Once the linear elastic solution is found, the Von
A flow chart of the solution process is shown in Mises stress invariant is calculated at each
Figure 5. Defining K as the overall structural integration point of all the shells. The values at the

Page 5 of 11
integration points are extrapolated to the shell was selected. The FEAST eigensolver was
nodes, using the shape functions of a nine-noded developed in the late 2000s at the University of
isoparametric quadratic element, as there are nine Massachusetts (6). As stated by its author, it “takes
integration points per plane through the shell its inspiration from a density-matrix representation
thickness. Finally, the values at the nodes are and a contour-representation technique” originally
averaged to achieve a smooth stress field. As developed in the field of quantum mechanics. This
bending and torsional stresses dictate the behavior technique enables the quick solution of almost any
of the girders, it was chosen to extrapolate the Von eigenvalue problem, particularly in the case of
Mises stress invariant through the shell thickness, symmetric real sparse matrices, which is the case
based on the line defined by the stress values at the in structural mechanics. Each node is connected
two integration points. For the top and bottom to a maximum of ten other nodes in an overall
flanges, the stress is displayed at the respective model that contains thousands of nodes, while the
upper and lower surface, whereas for the webs and stiffness and geometric stiffness matrices are
the stiffeners, the stress is displayed at the mid- symmetric due to the virtual work principle.
plane. Finally, to comply with the shell finite
For the purpose of finding critical buckling modes,
element theory, which assumes that there is no
the user is prompted to enter a search interval
normal stress orthogonal to the shell surface, a
within which all the modes will be encountered.
total of three actual stress fields are represented:
For the sake of keeping a limited number of output
one for the flanges, one for the webs, and one the
files, a maximum of five modes is retained for
stiffeners and connecting plates. For example, for
each step, but this may be increased in the future.
a node that is shared between a web element and a
The eigensolver is able to capture either global or
stiffener element around mid-span of a girder, two
local (flange or web) buckling modes.
values are actually stored and displayed: one for
the web, large due to the in-plane bending stress,
and one for the stiffener, low as the in-plane
Examples
bending stress for the web actually results in a An important step in program development is the
zero out-of-plane stress for the stiffener. Typically, validation of the program. A number of systems
only the shear stress components contribute to the have been evaluated to date to ensure that proper
Von Mises stress invariant for stiffeners and modelling decisions have been made in the
connecting plates. development of V2.0. For the purpose of
validating the program, a series of benchmark
For buckling analyses, once the static solution is problems have thus far been considered. The
obtained, the structural geometric stiffness matrix systems consist of full bridges with both straight
is assembled and the program performs an and curved geometry. Cases with and without
eigenvalue buckling analysis of the structure. This support skew have been modeled. For validation
means that the equation (K + λ*Kg)*X = 0 is purposes companion models have been created
solved, where Kg is the geometric stiffness matrix, using the general-purpose FEA program
X an eigenvector and λ an eigenvalue. In the field ABAQUS. Comparisons have been made with
of structural stability, the geometric stiffness numerical comparisons of deflections, stresses,
matrix is computed from the stresses derived from eigenvalues, etc. Some of these numerical
the linear elastic static solution, an eigenvalue is a comparisons are presented later in this paper in a
multiplier on the applied loads that leads to table. Comparisons have also been made with the
buckling of the structure, as defined by a graphical output using contour plots from the UT
bifurcation from the linear elastic solution, and an Bridge V2.0 and ABAQUS. The modeling
eigenvector is the deformed shape of the structure parameters in ABAQUS were selected to mirror as
at that point. In V2.0, the FEAST eigensolver much as possible those assumed in UT Bridge. In
available in the Intel Math Kernel Library (MKL) particular, the ABAQUS S8R5 general shell was

Page 6 of 11
adopted since the element that has been Post-Processor of UT Bridge no longer requires
incorporated into UT Bridge is based upon the opening separate program modules, which was
same formulation. One of the few modelling required in the past. This allows the user to move
differences between V2.0 and ABAQUS are how around the program and switch between input
the cross-frames are actually attached to the screens and screens showing the model geometry
girders. Since selecting the mid-edge nodes of much quicker.
quadratic elements is not possible in ABAQUS, an Curved Girder Example
additional number of four elements were generated
In order to validate the accuracy of V2.0 with
along the webs in order to maintain the same
ABAQUS, a highly curved prismatic girder was
stiffness for the cross-frames, which results in a selected. The girder has a radius of curvature of
slight flexibility introduced into the connection. 26.5' and a span of 41.67' long. The respective
A number of screen captures are presented on the flange and web dimensions are 12.0"x0.75" and
following pages that show comparisons between 60.0"x0.75". The tight radius of curvature was
the UT Bridge program viewer and the ABAQUS selected to validate the program and is not meant
model. Significant changes have been incorporated to reflect the curvature of actual road or railway
into the V2.0 post-processor compared to V1.6. bridges. However, such geometries may not be
The graphics of the viewer is very similar to the uncommon for other applications such as a
ABAQUS output. As with the other forms, the pedestrian bridge. As the girder defines a 90°
viewer is encoded in Visual Basic. The model angle in plan, this benchmark problem can be
itself is displayed using an algorithm encoded in potentially considered as a worst case scenario as
OpenGL, which is with DirectX one of the two far as checking boundary conditions. Figures 6 and
main standards used in three-dimensional graphics 7 show the displacement contours in the x- and y-
display. However, as OpenGL is specifically directions with maximum values of 7.602". An
targeted at addressing C++ based codes, the Open interesting aspect of this problem is that the x- and
Toolkit (OpenTK) library was added to the VB y- displacements are symmetric due to the
code in order to be able to use some of the already symmetry of the problem. The contour of the
defined OpenGL functions. Generation of the maximum resultant deformation from V2.0 is
model, its geometry, displaced shape, buckled shown in Figure 8 with a maximum value of
shape, and Von Mises stress distribution, however, 21.175". The corresponding contour of the
is achieved through a “manual” specific 3D maximum resultant displacement from ABAQUS
rendering algorithm that allows for control of for this problem is shown in Figure 9, with a
advanced parameters. These parameters include maximum value of 21.210”, which is almost
the magnification factor of the displaced shape and identical. It should be noted that there is a slight
mode shapes, the rainbow pattern used to represent difference between the contour from V2.0 and
the magnitude of those displacements, the offset ABAQUS. Whereas V2.0 displays a continuous
between the shells edges and their interior surface (rainbow) contour, the contour graphs from
(in order to avoid the so-called “stitching” effect ABAQUS are discretized.
that may affect the overall quality of the Although an eigenvalue buckling analysis is not
rendering). The output files required for the very meaningful for a girder with significant
rendering are generated at the end of the main horizontal curvature, the analysis was carried out
FORTRAN calculations and are formatted in the two programs. The eigenvectors (buckled
similarly to the well-known .obj format shapes) for the first mode from the V2.0 and
encountered in three-dimensional graphics display. ABAQUS models are shown in Figures 10 and 11,
The viewer form allows for a rapid display of the respectively. As noted in the captions, there is
model, boundary conditions, temporary supports, excellent agreement between the corresponding
displaced shape, buckling eigenvalues, mode eigenvalues for the two programs. Similar
shapes and Von Mises stress distribution for each agreement was also achieved for higher modes.
step of the erection sequence. One of the major
improvements between V1.6 and V2.0 is that the

Page 7 of 11
Figure 6 – Curved girder deformed shape (UT Figure 9 – Curved girder deformed shape
Bridge, Ux,max = -7.602”) (ABAQUS, Umax = 21.21”)

Figure 7 – Curved girder deformed shape (UT Figure 10 – Curved girder 1st mode shape (UT
Bridge, Uy,max = 7.602”) Bridge, λ = 11.438)

Figure 8 – Curved girder deformed shape (UT Figure 11 – Curved girder 1st mode shape
Bridge, Umax = 21.175”) (ABAQUS, λ = 11.411)

Page 8 of 11
Skewed bridge of 18"x0.945", bottom flange dimensions
20"x2.165", and web dimensions 56.1"x0.5625".
Similar to the relatively severe case of the As noted above, a framing plan taking advantage
horizontally curved girder from the last example, of lean-on bracing is considered. Four steps are
an equally severe case was also considered for a defined for the erection sequence, with one full
straight bridge with skewed supports. A plan view girder lifted at each step. Nearly exact correlation
of the girder system is shown in Figure 12. was achieved between the V2.0 and ABAQUS
models for each stage of the erection. Figures 13
and 14 show a comparison of the contour graph of
the result displacement for V2.0 and ABAQUS.
The maximum value from each analysis was 2.8".
The results from the eigenvalue buckling analyses
from each stage were also nearly exact between
the two programs. The eigenvectors (buckled
shapes) from the two programs are shown in
Figures 15 and 16, with the corresponding
eigenvalues being 2.6 and 2.5.

Figure 13 – Skewed bridge deformed shape, step4

(UT Bridge, Umax = 2.8”)

Figure 12 – Skewed bridge framing plan

The structure is a straight, four-girder system with
a relatively extreme support skew angle of 70°.
The bridge had 12 intermediate cross frames and 6
locations where “lean-on” bracing was used where Figure 14– Skewed bridge deformed shape, step4
only the top and bottom struts were provided. The (ABAQUS, Umax = 2.8”)
girders were prismatic with top flange dimensions

Page 9 of 11
 shown in Figures 18 and 19 with the maximum
displacement of 4.1" in V2.0 and 4.3" in
ABAQUS.

Figure 15 – Skewed bridge 1st mode shape, step 2

(UT Bridge, λ = 2.6)

Figure 17 – Curved bridge framing plan

Figure 16 – Skewed bridge 1st mode shape, step 2

(ABAQUS, λ = 2.5)
Figure 18 – Curved bridge deformed shape, step 4
Curved bridge (UT Bridge, Umax = 4.1”)
An additional validation analysis was carried out
on the four-girder system depicted in Figure 17.
The two-span bridge had a radius of curvature of
the inner girder equal to 288' and was braced with
63 cross-frames oriented in a radial manner. The
cross-section was kept uniform for the sake of
faster modelling in ABAQUS. The cross-sectional
dimensions were equal to 28"x2.5" for both the top
and bottom flanges, and 114"x1.25" for the web.
Four steps were defined in the erection sequence,
with two girders lifted at each step. Temporary
supports were modeled for steps 1, 2 and 3 to
prevent excessive deflections. Good correlation
was achieved between both programs at all steps
of the erection. Contours of the maximum
Figure 19 – Curved bridge deformed shape, step 4
displacement in the fully erected structures are
(ABAQUS, Umax = 4.3”)

Page 10 of 11
Validation  The ability to perform a large
displacement second-order analysis
The erection conditions of three bridges with
relatively severe geometries were analyzed using  Modeling of dapped ends and/or tapered
UT Bridge and ABAQUS. The geometries girders
included severed curvature as well as significant  Modeling of other types of bracing, such
support skew. In all of the analyses, the UT Bridge as diaphragms
and ABAQUS models had excellent correlation in
the results. A summary of the displacements and  Display of moment and shear diagrams.
eigenvalues are provided in Table 1. Although
severe geometrical conditions were modeled, the Summary and Conclusions
maximum difference between UT Bridge and a A 3D finite element program able to capture the
widely used general purpose FEA program was linear elastic behavior and the stability of straight,
approximately 6%, while for the majority of the curved, or skewed steel I-girder bridges was
results, the percent difference was less than 1%. presented. The program was validated with a
The differences are likely due to the fact that UT series of tests on ABAQUS, using similar
Bridge performs a full integration versus the modeling assumptions. The user-friendliness and
reduced integration in ABAQUS. Another accuracy of the UT Bridge program make it a
potential source of the differences is the reduced reliable tool for designers and erectors. Modeling
number of shell elements for the webs for UT time in UT Bridge is significantly reduced. Even
Bridge. However, even for the extreme very complex structural systems can be quickly
geometries, UT Bridge V2.0 has excellent modeled.
agreement with the ABAQUS solutions.
Step UT Bridge ABAQUS ∆
References
Ux -7.602" -7.614" -0.2 % (1) Zureick, A. & Naqib, R., 1999, “Horizontally
Uy 7.602" 7.614" -0.2 % Curved Steel I-Girders State-of-the-Art
Uz -19.675" -19.70" -0.1 %
Analysis Methods”, Journal of Bridge
Highly curved girder 1 Engineering, vol. 4, no. 1, pp. 38-47.
Umax 21.175" 21.21" -0.2 %
λ1 11.438 11.411 0.2 % (2) Chang, C. 2006, Construction simulation of
λ2 58.277 56.398 3.3 % curved steel I-girder bridges, PhD Thesis,
1 Umax 2.427" 2.430" -0.1 %
Georgia Institute of Technology, Atlanta,
Georgia.
1 λ1 1.121 1.117 0.4 %
2 Umax 2.498" 2.501" < -0.1 % (3) NCHRP Report 725 “Guidelines for Analysis
Skewed bridge Methods and Construction Engineering of
2 λ1 2.575 2.513 2.5 %
Curved and Skewed Steel Girder Bridges”,
3 Umax 2.651" 2.661" -0.4 %
2012, Transportation Research Board.
4 Umax 2.798" 2.796" < 0.1 %
1 Umax 0.183" 0.185" -1.1 % (4) Stith, J. 2010, Predicting the behavior of
2 Umax 0.173" 0.175" -1.1 %
horizontally curved I-girders during
Curved bridge construction, PhD Thesis, The University of
3 Umax 0.301" 0.311" -3.2 %
Texas at Austin, Austin, Texas.
4 Umax 4.069" 4.325" -5.9 %
(5) ABAQUS Version 6.12, Dassault Systèmes
Table 1 – Results summary Simulia Corporation, 2012.

Potential Modifications Under (6) Polizzi, E., 2009, “Density-matrix-based

algorithm for solving eigenvalue problems”,
Consideration Physical Review B, vol. 79, no. 11.
Although V2.0 is already very promising, the
authors are considering further improvements,
which would include the following features:

Page 11 of 11