See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/274040288

Investigation of Thermal Gradient Effects in the I-35W St. Anthony Falls Bridge

Article  in  Journal of Bridge Engineering · September 2013

DOI: 10.1061/(ASCE)BE.1943-5592.0000438

CITATIONS READS
20 1,074

3 authors:

Brock Hedegaard Catherine. French

University of Minnesota Duluth University of Minnesota Twin Cities
17 PUBLICATIONS   116 CITATIONS    65 PUBLICATIONS   1,026 CITATIONS   

SEE PROFILE SEE PROFILE

Carol K. Shield
University of Minnesota Twin Cities
67 PUBLICATIONS   1,256 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Strength of PS Beam End Repairs View project

Monitoring and Modeling of St. Anthony Falls Bridge View project

All content following this page was uploaded by Brock Hedegaard on 23 November 2015.

The user has requested enhancement of the downloaded file.

 Investigation of Thermal Gradient Effects
in the I-35W St. Anthony Falls Bridge
Brock D. Hedegaard, S.M.ASCE1; Catherine E. W. French, M.ASCE2; and Carol K. Shield, M.ASCE3

Abstract: Thermal gradients were measured through the section of the I-35W St. Anthony Falls Bridge, a posttensioned concrete box girder
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.

bridge in Minneapolis, Minnesota, over the course of 3 years. The magnitudes and shapes of the measured thermal gradients were compared
with various design gradients, and a fifth-order curve was found to best approximate the shape of the gradients. The responses of the structure to
the largest measured thermal gradients were compared with stresses and deformations predicted by finite-element modeling given applied de-
sign gradients. The measured structural response was found to be best predicted when the finite-element model of the bridge was subjected to
a fifth-order design thermal gradient scaled to match maximum top surface temperature values proposed by AASHTO LRFD Bridge Design
Specifications for the region. Stresses and deformations predicted by finite-element modeling using the AASHTO LRFD bilinear design gra-
dients were found to be considerably lower than those derived from measured results. Recommendations for design thermal gradients are pro-
posed. DOI: 10.1061/(ASCE)BE.1943-5592.0000438. © 2013 American Society of Civil Engineers.
CE Database subject headings: Girder bridges; Bridge failures; Box girder; Temperature effects; Finite element method; Minnesota.
Author keywords: Bridges, box girder; Temperature effects; Monitoring; Finite-element method.

Background
Thermal gradients, temperature variations throughout a cross sec-
tion, in concrete bridge structures are caused by a combination of
solar radiation, conduction, and convection with the surrounding
atmosphere. The low thermal conductivity of concrete causes the
section to heat or cool nonuniformly during a daily cycle. These
gradients are typically most pronounced through the depth of the
cross section; the incident solar radiation heats the top surface of the
bridge deck, and the heat flows down through the superstructure.
Positive thermal gradients, defined as the top surface temperature
being higher than the temperature in the webs, are generally ob-
served in early summer on clear, sunny, hot afternoons, typically
between 2:00 and 4:00 p.m., with high solar radiation. Negative
thermal gradients, defined as the top surface temperature being
lower than the temperature in the webs, are generally found during
early mornings, usually between 5:00 and 8:00 a.m., throughout
the year.
Temperature differences in a section cause a structure to expand
or contract nonuniformly. Under positive thermal gradients, for
example, the top surface of the structure will expand more than the
bottom surface, causing the structure to deflect upward. Restraints
associated with boundary conditions induce axial and bending
stresses into the cross section. Furthermore, if the thermal gradient
profile is nonlinear, compatibility stresses are generated to satisfy
1 y
Ph.D. Candidate, Dept. of Civil Engineering, Univ. of Minnesota, Minne-
apolis, MN 55455-0116 (corresponding author). E-mail: hedeg003@umn.edu
2
CSE Distinguished Professor, Dept. of Civil Engineering, Univ. of Min-
nesota, Minneapolis, MN 55455-0116.
3
Professor, Dept. of Civil Engineering, Univ. of Minnesota, Minneapolis,
MN 55455-0116.
Note. This manuscript was submitted on February 16, 2012; approved on
September 11, 2012; published online on September 13, 2012. Discussion
period open until February 1, 2014; separate discussions must be submitted
for individual papers. This paper is part of the Journal of Bridge Engi-
neering, Vol. 18, No. 9, September 1, 2013. ©ASCE, ISSN 1084-0702/
2013/9-890–900/$25.00.
the assumption under Bernoulli beam bending that plane sections
remain plane. The stresses induced by thermal gradients can be
larger than those induced by vehicle live loading.
Literature Review
The heat flow problem in bridge structures was considered com-
putationally by Potgieter and Gamble (1983). The authors con-
structed a finite-difference heat flow model and complemented their
numerical study with field measurements from the Kishwaukee
River Bridge, located near Rockford, Illinois. The work of Potgieter
and Gamble was advanced by Imbsen et al. (1985), in what was later
adapted into the AASHTO LRFD Bridge Design Specifications
(AASHTO 2010). The AASHTO LRFD (AASHTO 2010) design
positive gradient is illustrated in Fig. 1. For the design gradients
presented within this paper, the plotted gradient temperatures are
given as the temperature difference of the cross section from the
temperature in the webs. The design negative gradient for structures
with plain concrete decks and no asphalt overlay is found by mul-
tiplying the design positive gradient temperatures by 20:3.
The design gradient from the New Zealand Code (Priestley
1978), shown in Fig. 2, is a fifth-order curve decreasing from
maximum gradient temperature T0 at the top of the deck to zero at
a depth of 1,200 mm (47.2 in.) defined by
 5
Tgrad ðyÞ ¼ T0 (1)
1,200
where y is in millimeters and is defined positive up from the point
1,200 mm (47.2 in.) below the top surface. The specified maximum
gradient temperature T0 for plain concrete decks in New Zealand
with no asphalt overlay is equal to 32°C (57.6°F). The fifth-order
curve is applied through the depth of the webs and for decks above
unenclosed air. For decks above enclosed air cells in box girders,
a linear gradient is prescribed with a top gradient temperature
equal to T0 and, for plain concrete decks with no asphalt overlay,

890 / JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013

J. Bridge Eng. 2013.18:890-900.

 temperature decreasing at a rate of 1°C per 20 mm (1°F per 0.44 in.).
The bottom gradient tail temperature is specified as 1.5°C (2.7°F),
decreasing linearly to zero over a height of 200 mm (7.9 in.) measured
up from the bottom of the section. A design negative thermal gradient
is not specified in the New Zealand Code (Priestley 1978).
A number of investigations regarding thermal gradients in con-
crete bridges have been conducted. Potgieter and Gamble’s (1983)
investigation of the Kishwaukee River Bridge, a continuous five-
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
Fig. 1. AASHTO LRFD Bridge Design Specifications (2010) design
positive thermal gradient
Fig. 2. New Zealand Code design positive thermal gradient as pre-
sented in Priestley (1978)
Fig. 3. Elevation view of St. Anthony Falls Bridge indicating overall dimensions: callout flags represent instrumented locations (Hedegaard et al. 2013,
© ASCE)
JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013 / 891
J. Bridge Eng. 2013.18:890-900.
span segmental box girder bridge with blacktop covering, revealed
that the shape, but not the top surface magnitude, of the New Zealand
gradient matched measured thermal gradients. Shushkewich (1998)
investigated the measured thermal gradients of the North Halawa
Valley Viaduct, a cast-in-place concrete box girder bridge in Hawaii.
Positive and negative thermal gradients were found to correspond
well with AASHTO (1998) proposals, which are identical to the
thermal gradient provisions in AASHTO LRFD (AASHTO 2010).
Thompson et al. (1998) considered the Ramp P structure, a curved
precast segmental concrete box girder bridge on highway US 183
in Austin, Texas. Gradients were measured both with and without
50-mm (2-in.) blacktop covering. Measured gradients were typically
lower than those specified in AASHTO LRFD (AASHTO 1994),
which had an identical positive design gradient to AASHTO LRFD
(AASHTO 2010) but used a multiplier of 20:5 instead of 20:3 for
defining the negative gradient. However, it was stated that more data
were needed to construct a sound statistical comparison. Roberts-
Wollman et al. (2002) investigated thermal gradients in precast
segmental concrete box girders in the San Antonio Y Project. They
concluded that typical positive gradients could be approximated by
a fifth-order curve similar to that presented in Priestley (1978). They
also stated that the AASHTO LRFD (AASHTO 1994) positive and
negative design gradients were conservative.
Structure and Instrumentation
In this study, thermal gradients were measured in the I-35W St.
Anthony Falls Bridge in Minneapolis, Minnesota. The bridge was
constructed as two parallel structures for northbound and south-
bound traffic and consisted of posttensioned concrete box girders.
The driving surface had no blacktop covering. The bridge in-
strumentation is summarized in Hedegaard et al. (2013). Fig. 3
shows the overall bridge geometry and span lengths. Fig. 4 shows the
elevation of the bridge with span and pier designations, as well as
typical cross-sectional shapes [Minnesota Department of Trans-
portation (DOT) 2008]. Minneapolis is located within solar radiation
Zone 2 for computing thermal gradients according to the AASHTO
LRFD (AASHTO 2010). All times listed in this paper are Central
Standard Time (CST) and have not been adjusted for daylight
savings time.
One segment of the southbound structure near the midspan of the
river span (Location 7) was heavily instrumented with thermistors to
explore the temperature distribution throughout the cross section.
Some characteristic dimensions and the thermistor layout at this
section, including those integral with the vibrating wire strain
gauges, are presented in Fig. 5. Insufficient instrumentation pre-
vented detailed investigation of thermal gradients at other loca-
tions within the northbound and southbound structures; however,
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Elevation view of St. Anthony Falls Bridge with typical cross sections (Hedegaard et al. 2013, © ASCE)

Fig. 5. Characteristic section dimensions and instrumentation layout in southbound structure Location 7

temperatures measured in association with the vibrating wire strain

gauges at other instrumented locations were spot checked to verify
consistency along the length of the bridge.
Temperature and strain data collected prior to September 17,
2009, were only sampled every 6 h (midnight, 6:00 a.m., noon, and
6:00 p.m.), whereas data collected after September 17, 2009, were
sampled hourly. For each set of measurements, five readings were
taken from each instrument over the course of 15 min. To average
out the effects of any transient loading such as traffic, presented
results represent the averages from those five readings.

Comparison of Measured and Design

Thermal Gradients

Measured thermal gradients from Location 7 of the southbound St.

Anthony Falls Bridge superstructure were compared with AASHTO
LRFD (AASHTO 2010) and New Zealand (Priestley 1978) design
gradients as presented previously. A graphical comparison of these
two design positive thermal gradients, both scaled to the same top
surface gradient temperature T0 , is presented in Fig. 6. Even with
equal maximum gradient temperature values, the area under the
Priestley gradient curve is considerably larger than that under the
AASHTO LRFD gradient.
The magnitude of the measured positive thermal gradients was
taken as the difference between (1) the average of the operational
topmost thermistors of all six-thermistor sets in the top flange and (2)
the minimum measured temperature along the centerlines of the
Fig. 6. Comparison of AASHTO LRFD (AASHTO 2010) and New
Zealand (Priestley 1978) design gradients
webs. Minimum temperatures were captured along the centerlines
of the webs to minimize the influences of gradients through the
thickness of the webs, which were obtained using sets of three
thermistors installed through the width of the webs. The thermal
gradients through the webs were typically not linear; often the
temperature at the centerline of the web was lower than temperatures

892 / JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013

J. Bridge Eng. 2013.18:890-900.

 from both the outermost thermistor [with 57 mm (2.25 in.) of con-
crete cover] and the innermost thermistor [with 57 mm (2.25 in.) of
concrete cover from the air inside the box]. Differences between any
two web thermistors rarely exceeded 2.0°C (3.8°F) and thus were
negligible compared with gradients through the depth of the box
section.
Because top surface temperature measurements were not taken,
comparisons made between measured and design gradient magni-
tudes required that the value of the design gradient at a depth of
50 mm (2 in.) below the top surface of the deck (where the topmost
thermistor was located) be used. Consequently, the AASHTO
LRFD gradient in Zone 2 [top surface gradient temperature equal
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
to 25.6°C (46°F)] would have a magnitude of 16.2°C (29.0°F) at
a depth of 50 mm (2 in.), and the Priestley fifth-order gradient scaled
to the same top surface gradient temperature (henceforth referred to
as the Priestley-Z2 gradient) would have a magnitude of 20.6°C
(37.1°F) at that same depth. These design gradient magnitudes at the
depth of the thermistors are compared with the measured positive
gradient magnitudes from September 1, 2008, until October 26,
2011, in Fig. 7(a). The break in the data was caused by an outage of
the static instrumentation system.
The measured positive gradients showed a strong seasonal trend,
as was expected. Maximum measured gradients prior to September
17, 2009, appeared to be smaller than those measured afterward, but
this was an artifact of the 6-h sample rate compared with the 1-h
sample rate (i.e., peak gradient effects were missed with the 6-h
sample rate). It was observed that the daily maximum gradients often
occurred around 2:00 to 3:00 p.m. (equivalently 3:00 to 4:00 p.m.
during Central Daylight Savings Time). The measured gradient
magnitudes exceeded the AASHTO LRFD magnitude regularly
during the summers and occasionally approached the Priestley-Z2
levels.
Negative gradient magnitudes were calculated in the same
manner as positive gradients, but the maximum temperature at the
centerlines of the webs was considered instead of the minimum.
Design negative gradients, including the Priestley-Z2 curve for
which no negative design gradient was specified, were considered
by scaling the positive gradients by 20:3 per AASHTO LRFD
(AASHTO 2010) specifications. Again, the design magnitudes were
considered at a depth of 50 mm (2 in.) below the deck surface to
enable comparison with the measured negative gradients, as shown
in Fig. 7(b).
The seasonal dependence of the negative gradient magnitudes
was not as pronounced as was observed for the positive gradients,
but the overall maximum negative gradients tended to occur during
Fig. 7. Measured (a) positive and (b) negative gradient magnitudes: design gradient magnitudes considered at a depth of 50 mm (2 in.) below the deck
surface for comparison against measured gradients
JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013 / 893
J. Bridge Eng. 2013.18:890-900.
the winter months. Maximum daily negative gradients most often
occurred between 5:00 and 8:00 a.m. The measured negative gra-
dient magnitudes often exceeded both the AASHTO LRFD and the
Priestley-Z2 interpolated magnitudes.
The five maximum measured positive gradients are plotted in
Fig. 8. Weather conditions taken from the Minneapolis-St. Paul
Airport [approximately 11 km (7 mi) from the bridge] during the
days of maximum measured positive gradients are presented in
Table 1. The measured gradients were compared with the AASHTO
LRFD (AASHTO 2010) design gradient for Zone 2, the Priestley-Z2
curve, and, for gradients in decks above enclosed air cells, the
Priestley linear gradient, all assuming no blacktop and with top
temperature equal to 25.6°C (46°F) according to AASHTO LRFD
Zone 2. In all cases, the measured gradients were better approxi-
mated by the Priestley-Z2 curve than the AASHTO LRFD gradient.
The Priestley linear gradient above the enclosed box was a poor fit
at the centerline of the boxes. In fact, the gradients at the centerline
of the boxes were not largely different from those measured along
the webs. Gradient tails in the bottom flange ranged from 0.5 to
1.5°C (0.9–2.7°F).
The five maximum measured negative gradients are plotted in
Fig. 9 with corresponding weather conditions taken from the
Minneapolis-St. Paul Airport presented in Table 2. The measured
gradients were compared with the AASHTO LRFD (AASHTO 2010)
design negative gradient assuming no blacktop, the Priestley-Z2 curve
scaled by 20:3 to remain consistent with the AASHTO LRFD top
surface gradient temperature, and the Priestley linear gradient for
decks above enclosed air cells as used for positive gradient com-
parison but scaled by 20:3. The measured negative gradients were
noted to be much more diverse in shape than the positive gradients and
were not consistently matched in shape by either design gradient.
However, the Priestley-Z2 scaled by 20:3 appeared to produce
reasonable negative gradient estimates in the top flange. Again,
gradients at the centerline of the boxes were not largely different from
those measured along the webs.
Only Location 7 was instrumented in a manner to investigate
gradient shapes, but the consistency of thermal gradient magnitudes
along the length of the structure was examined by considering the
temperature difference between top and bottom flange vibrating
wire strain gauges at each instrumented section. Regardless of sec-
tion geometry, the nominal locations of the top flange strain gauges
were consistently 140 mm (5.5 in.) below the deck top surface and
bottom flange gauges were 80 mm (3.0 in.) above the bottom fiber.
Measured temperatures in the top and bottom flanges along the
length of the southbound superstructure corresponding to the
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 8. Maximum measured positive gradients at Location 7 compared with design gradients through (a) centerline of exterior box, top flange only, and
(b) along centerline of west web of exterior box

Table 1. Dates, Times, and Weather Conditions for Maximum Measured Positive Thermal Gradients during the First 3 Years of Monitoring
Maximum gradient Daily maximum Daily minimum Daily mean wind Daily precipitation
Date time (CST; hrs.) temperature (°C) temperature (°C) speed (km=h) (mm) Weather events
5/16/2010 1400 23 12 5 0.0 Mostly cloudy morning, scattered clouds
in afternoon
6/17/2010 1500 32 19 16 0.3 Partly cloudy/scattered clouds in morning
and afternoon, thunderstorm at 1800 hrs.
6/6/2011 1500 36 21 6 0.0 Clear morning, partly cloudy afternoon
6/7/2011 1400 39 26 12 0.0 Clear
7/1/2011 1500 37 22 8 3.6 Mostly cloudy morning, scattered clouds
in afternoon, thunderstorm at 1900 hrs.

Fig. 9. Maximum measured negative gradients at Location 7 compared with design gradients through (a) centerline of exterior box, top flange only, and
(b) along centerline of west web of exterior box

894 / JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013

J. Bridge Eng. 2013.18:890-900.

 Table 2. Dates, Times, and Weather Conditions for Maximum Measured Negative Thermal Gradients during the First 3 Years of Monitoring
Maximum gradient Daily maximum Daily minimum
Date time (CST; hrs.) temperature (°C) temperature (°C)
1/2/2010 0800 217 226
1/3/2010 0800 214 226
1/21/2011 0600 216 227
2/8/2011 0700 216 223
9/15/2011 0600 14 2 3
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
maximum measured positive gradient on July 1, 2011, at 3:00 p.m.
are presented in Fig. 10. It was noted that the top surface temperature
was nearly uniform along the length of the bridge, but the bottom
flange temperature dropped significantly near the piers. This was
presumed to be caused by the large thermal mass of the pier dia-
phragms, the increase in thickness of the bottom flange near the
piers, and possibly the additional shade and cover provided by the
decorative fins at the tops of the piers. The top flange temperatures
reinforce the idea of consistent thermal gradients along the length
of the structure caused by solar radiation, whereas the bottom tem-
peratures indicate the complexities of the heat transfer problem given
varying geometry and thermal boundary conditions. Top and bottom
flange temperatures during other positive and negative gradients on
both southbound and northbound structures showed similar trends.
Finite-Element Modeling
Finite element models of the southbound structure were constructed
using ABAQUS version 6.10 to allow comparison between the
measured structural behavior and computational estimations. The
construction of the three-dimensional (3D) finite-element model
(FEM) and validation of the model with static truck testing were
presented by Hedegaard et al. (2013). For thermal investigations, it
was assumed that thermal gradients could be reasonably approxi-
mated as one dimensional (that is, temperature varied only with
depth from the top surface) so that a two-dimensional (2D) FEM
could accurately predict the structural response.
The 2D model was limited to the continuous spans, Spans 1–3, of
the southbound bridge. Piers 2–4 were included. The concrete was
modeled using eight-node quadratic shell elements with reduced
integration. Out-of-plane deflection of the entire model was con-
strained, such that the shell formulation emulated a plane-stress
problem. Individual elements were typically 600 mm (24 in.) in
length, meaning that the superstructure contained approximately
500 elements along the length of the bridge. To more accurately
model the shape of the temperature gradient, the mesh was refined in
the top flange with minimum element lengths of 100 mm (4 in.).
Between 10 and 18 elements were modeled through the depth of the
section. Elements were assumed to have out-of-plane thickness
corresponding to the geometry of the section. Thus, although the box
geometry was not explicitly captured, the section moment of inertia
was consistent with the physical structure. The posttensioning
strands were approximated as two-node linear truss elements em-
bedded within the top and bottom flanges of the concrete box. These
truss elements had appropriate axial stiffness and no bending
stiffness to emulate the stiffness of the strands. Mild steel was not
explicitly modeled but was instead assumed to be uniformly dis-
tributed throughout the sections and taken into account by adjusting
the modulus of elasticity of the concrete. The concrete barrier rail on
the interior edge of the bridge and the curb and rail on the exterior
J. Bridge Eng. 2013.18:890-900.
Daily mean wind Daily precipitation
speed (km=h) (mm) Weather events
4 0.0 Clear morning, partly cloudy afternoon
3 0.0 Partly cloudy morning, clear afternoon
5 0.3 Clear early morning, overcast with light
snow in afternoon
10 0.0 Clear morning, partly cloudy afternoon
0.0 Clear early morning, mostly cloudy
afternoon, overcast evening
Fig. 10. Measured temperatures in top and bottom flanges along the
length of the southbound superstructure from maximum measured
positive gradient on July 1, 2011 at 3:00 p.m.
edge of the bridge were not modeled, as instrumentation was not
provided to investigate possible changes in the temperature distri-
bution in and around the rails, and the stiffness provided by the rails
was found to not greatly contribute to the global stiffness of the
structure. For the boundary conditions, all parts of the disk-bearing
assemblies, including the elastomeric (polyether urethane) disks and
the steel bearing plates, were modeled using eight-node quadratic
shell elements. Disk bearing assemblies for the expansion joints at
Abutment 1 and where Span 3 attached to Pier 4 were allowed to
slide longitudinally without friction. Disk bearings at Piers 2 and 3
were constrained to have no longitudinal sliding to model the pin-type
connection between the piers and superstructure at these locations.
Material properties used in the 2D model are listed in Table 3.
The selection of modulus properties for the concrete was based on
experimental cylinder strength data (Hedegaard et al. 2013). Values
for concrete density and Poisson’s ratio were assumed to be equal to
nominal material values, as were all values for the posttensioning
steel. Material properties for the elastomeric disks were obtained
through correspondence with the bearing assembly supplier (R. J.
Watson, Inc., personal communication, September 15, 2010). The
coefficients of thermal expansion (CTE) used for the modeled
concrete were derived from measurements of laboratory samples,
strain gauge measurements on the St. Anthony Falls Bridge under
uniform temperature changes, and linear potentiometer readings at
the bridge expansion joints.
Concrete samples from the superstructure and piers were col-
lected and cast into beams and cylinders for measuring the CTE
in the laboratory. The specimens were stored in a temperature-
controlled room where the temperature was varied between 215
and 25°C (5 and 77°F) over 6 weeks. The tests took place 16 months
JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013 / 895
 Table 3. Material Properties Used in FEM
Modulus of elasticity,
Material name Description
Concrete-superstructure Concrete in superstructure 28-day
design strength 5 44.8 MPa (6.5 ksi)
Concrete-pier Concrete in piers 28-day design
strength 5 27.6 MPa (4.0 ksi)
Steel Posttensioning steel (axial stiffness only)
Elastomer Incompressible elastomeric disk bearing
a
Coefficient of thermal expansion and density for Elastomer material not specified in FEM
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
after casting so that shrinkage strains could be assumed negligible
during the test. The laboratory CTE was computed as the slope of the
least-squares line relating the measured strains to the temperature.
The average CTE measured in this fashion was 10:3 mɛ=°C
ð5:72 mɛ=°FÞ for the superstructure concrete specimens and
8:73 mɛ=°C ð4:85 mɛ=°FÞ for the pier concrete, with coefficients of
variation of 6.3 and 3.5%, respectively.
For all in situ estimations of CTE, month-long data sets begin-
ning at least 1 full year after bridge opening were analyzed to min-
imize time-dependent strains. Only readings with measured top flange
and bottom fiber temperature differences less than 0.5°C (0.9°F)
were considered to minimize the effects of thermal gradients.
For the estimation of the concrete CTE from vibrating wire strain
gauges, axial expansion of the superstructure was assumed to be
unrestrained because of the flexibility of the piers. The total mea-
sured change in strain was equal to the sum of the thermal strain,
strains caused by restraint, and time-dependent strains. By necessity,
it was assumed that traffic loading, the magnitude and positioning of
which could not be measured, was not captured by the vibrating wire
strain gauges, although it is likely that traffic might impact the static
strain readings. Data sets collected over short time periods mini-
mized time-dependent strains. Strains due to restraint or traffic
loading were minimized by examining vertical or transverse gauges,
or longitudinal gauges at the neutral axis applying the unrestrained
axial deformation assumption. Therefore, the total strain under these
conditions was assumed to be purely thermal strain, and the slope of
the least-squares fit line between thermal strains and temperature
changes yielded the CTE. The average CTE value computed by this
method was 9:83 mɛ=°C ð5:46 mɛ=°FÞ for the superstructure, with
computed values from all locations along the bridge within 2.6% of
the average. This method was not used for pier concrete.
For the estimation of CTE from expansion joint movements,
linear potentiometers were used to measure the relative movement of
the superstructure and the bearing. Piers were flexible in bending, so
all spans could expand axially without significant restraint. Because
Span 3 was effectively connected to Pier 4 by a roller connection,
and because Span 4 was cast integral with Abutment 5 and pinned to
the top of Pier 4, which was assumed to be highly flexible compared
with the abutment, it was further assumed that Pier 4 would deflect
by an amount equal to the total unrestrained thermal elongation
of Span 4. The linear potentiometer measurements from Span 3
connected to Pier 4 included both expansion of the continuous
three-span superstructure, and the deflection of Pier 4 caused by the
expansion of Span 4. Therefore, the estimated change in length of
the continuous Spans 1–3 was equal to
DLSpans2123 ¼ DLP1 þ DLP3 2 aconc LSpan 4 DT (2)
where DLP1 and DLP3 5 changes in the linear potentiometer
readings at Abutment 1 and at Span 3 connected to Pier 4, re-
spectively, over change in uniform temperature DT, LSpan4 5 length
896 / JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013
J. Bridge Eng. 2013.18:890-900.
Coefficient of
thermal expansion, Poisson’s ratio Density (nominal),
GPa (ksi) mɛ=°C ðmɛ=°FÞ (nominal) kg=m3 ðlb=ft3 Þ
34.5 (5,000) 10.1 (5.60) 0.2 2,400 (150)
27.6 (4,000) 8.73 (4.85) 0.2 2,400 (150)
196.5 (28,500) 12.2 (6.78) 0.3 7,700 (480)
0.069 (10) N/Aa 0.499 N/Aa
of Span 4 equal to 44.3 m (145 ft), and aconc 5 CTE of the su-
perstructure concrete to be determined. The slope of the least-
squares fit line between DLSpans-123 and DT divided by the total
length of Spans 1–3, equal to 327 m (1,072 ft), was equal to the CTE
of the superstructure concrete. This was a recursive problem, so
a value of aconc was first assumed for computing DLSpans-123 , then
recalculated by the least-squares fit line, and then used again to
compute a new DLSpans-123 . This process was repeated until aconc
converged. The average CTE value computed by this method was
10:1 mɛ=°C ð5:60 mɛ=°FÞ for the superstructure, and values mea-
sured for each box individually did not differ from the average by
more than 2%. This method was not applicable for the pier concrete.
Material coefficients of thermal expansion used in the model
were averaged from the described test methods. Therefore, a CTE
of 10:1 mɛ=°C ð5:60 mɛ=°FÞ was chosen for the superstructure and
8:73 mɛ=°C ð4:85 mɛ=°FÞ was chosen for the pier. Thermal ex-
pansion of the elastomeric disk material was not considered to be
critical for the analysis and was ignored. The CTE of steel was
assumed to be a typical value of 12:2 mɛ=°C ð6:78 mɛ=°FÞ.
Thermal gradients applied in the model varied only with depth, as
would typically be used for design purposes. Thus, despite changing
sectional properties along the length, and especially the great in-
crease in thermal mass at the diaphragms, the gradient shape was
assumed constant along the length of the structure. Thermal gra-
dients through the thickness of the webs were ignored, and the tail
gradient in the bottom flange was assumed to be negligible (i.e., T3
as shown in Fig. 1 for the AASHTO LRFD gradient was set equal to
zero for simplification).
The 2D model used to investigate thermal effects was validated
with static truck test data and results from the 3D model (French et al.
2012).
Validation of Model with Respect to Measured
Thermal Gradients
The global behavior of the 2D FEM was compared with the mea-
sured behavior of the bridge under the maximum measured positive
thermal gradients as described in Table 1 and Fig. 8. To isolate the
measured behavior caused by the thermal gradient, initial strain
readings were selected within a short time interval (,2 months)
from the gradient readings to minimize time-dependent creep and
shrinkage strains. Initial readings were chosen when the temperature
distribution through Location 7 was approximately uniform, and the
web temperatures were nearly unchanged between the initial and
gradient measurements. Because of continual changes in ambient
temperature and weather conditions, a perfectly uniform tempera-
ture through the section, especially in the top flange, was never
realized. Thermal gradients were input into the model by fitting the
change in measured gradients (gradient measurement minus initial
readings through the depth) with a curve of the form
  n
y strains caused by unrestrained thermal expansion, creep, and
Tgrad ðyÞ ¼ T0
1,200
where y is in millimeters and the values of T0 and n were chosen to
minimize the sum of squares error between modeled and measured
thermal gradients. Using Eq. (3), temperatures were specified at each
node of the model. Temperatures within each element were com-
puted according to the isoparametric mapping of the quadratic shell
elements.
Because the temperature distribution was unknown between the
top surface down to the top thermistor 50 mm (2 in.) below the top
surface, three variations of the fitted thermal gradient were tested
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
(3) shrinkage as follows:
Dɛ m ¼ Dɛ 2 aconc × DT 2 Dɛ cr 2 Dɛ sh (4)
where Dɛ 5 change in measured total strain, DT 5 change in
temperature, aconc 5 coefficient of thermal expansion of the con-
crete, and Dɛ cr and Dɛ sh 5 change in creep and shrinkage strains
over the same time period, respectively. Because the initial readings
were chosen to minimize the time-dependent strains, Dɛ cr and Dɛ sh
were assumed to be zero.
Experimentally derived stresses had variations up to 2 MPa
(280 psi) across the width of the deck for the same thermal gradient,

in the FEM. The baseline fitted gradient used Eq. (3) and fitting
parameters presented in Table 4 for the entire cross section. For the
two modified gradients, the temperature distribution throughout
most of the section was identical to the fitted gradient, and only the
gradient in the top 50 mm (2 in.) was changed. For the lower bound
estimate, the top surface temperature extrapolated from the best fit
curve was decreased by 20%, with a linear temperature distribution
between the decreased top surface temperature and the fitted tem-
perature at depth 50 mm (2 in.). Likewise for the upper bound
estimate, the top surface temperature extrapolated from the best fit
curve was increased by 20%, with a linear temperature distribution
between the increased top surface temperature and the fitted tem-
perature at depth 50 mm (2 in.).
In situ changes in stress were approximated by multiplying the
measured change in mechanical strains by the elastic modulus of the
concrete (equal to that used for the FEM, but not corrected according
to the mild steel reinforcement ratio). The change in mechanical
strain Dɛ m , sometimes called load-related strain, does not include
which might be attributed to errors in the positions of the gauge,
variations in deck thickness, or cross-sectional issues such as shear
lag. Large variations such as these were not unexpected, as stresses
in the top flange would vary rapidly with depth in sync with the
quickly decreasing temperature distribution though the top flange;
thus, small positional errors in gauge location were likely to prop-
agate as large errors in the derived stress estimates. Therefore, the
changes in experimentally derived longitudinal stresses at each of
the three instrument depths were averaged for each of the five
maximum measured positive gradients and compared with the
computed values of stress from the FEM using the three variations of
the fitted gradients as documented previously. These comparisons
using the baseline fitted gradients for Location 7 of the southbound
bridge are presented in Fig. 11. It was found that the measured and
modeled results corresponded well. The FEM results were found to
be insensitive to variations in the precise top surface temperature; the
upper and lower bound modifications to the input gradient altered the
FEM stresses by approximately 2% at all instrumented depths.

Table 4. Initial Strain (i.e., Uniform Temperature) Dates and Times and Fitting parameters for Eq. (3) for Maximum Measured Positive Gradients
Maximum gradient Maximum gradient Uniform temperature Uniform temperature Fitting parameter Fitting parameter
date time (CST; hrs.) date time (CST; hrs.) T0 (°C) for Eq. (3) n for Eq. (3)
5/16/2010 1400 4/14/2010 0800 26.76 6
6/17/2010 1500 6/8/2010 1600 28.63 6
6/6/2011 1500 8/3/2011 0800 31.20 6
6/7/2011 1400 8/2/2011 1200 28.54 6
7/1/2011 1500 8/2/2011 1200 28.58 5

Fig. 11. Changes in longitudinal stresses approximated from measured strain at southbound Location 7 and longitudinal curvature along length of
southbound bridge caused by five maximum measured positive thermal gradients compared with FEM-computed stresses and curvatures using baseline
fitted thermal gradients [no adjustment to top 50 mm (2 in.) of gradient]

JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013 / 897

J. Bridge Eng. 2013.18:890-900.

 The maximum measured positive thermal gradients induced
larger experimentally derived longitudinal stresses than were ob-
served during truck load testing (Hedegaard et al. 2013). Positioning
eight trucks with total weight of 1,770 kN (398 kips) at Location 7
induced experimentally derived longitudinal stress of 1.8 MPa
(260 psi) in the bottom flange. This stress was exceeded for all the
considered maximum positive gradients, whereby the experimen-
tally derived longitudinal stress in the bottom flange ranged from
3.1 (450 psi) to 4.7 MPa (680 psi).
Measured curvatures along the length of the southbound struc-
ture caused by the maximum measured positive gradients were
compared with FEM-computed curvatures using the three variations
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
of the fitted gradients. These comparisons for the baseline fitted
gradients are presented in Fig. 11. FEM curvatures were computed
using the slope of the least-squares fit line of the strain profile
through the full section depth. The presented measured curvatures
for locations with vibrating wire strain gauges were calculated from
the slope of the line through the section depth fitting the short-term
changes in total strain, equal to the mechanical strain from Eq. (4)
plus the thermal strain equal to aconc DT, averaged at only the bot-
tom flange and web gauge depths. Top flange vibrating wire strain
gauges were excluded in computation because large variations in the
change in total strain across the width of the top flange were ob-
served, which did not appear to be related to 3D or transverse be-
havior induced by the thermal gradient. For each considered positive
thermal gradient, the range of measured strains across the width of
the top flange at Location 7 was typically larger than 30 mɛ. Strains
from gauges in the bottom flange and webs were much more
consistent, with ranges at these two depths across the width of any
given cross section typically less than 5 mɛ for the considered
gradients. When observing changes in total strain between two
readings with very low thermal gradient (i.e., uniform temperature
changes of the bridge), the wide range of strains across the width of
the top flange was not observed in the measured data. In addition to
the expected differences caused by gauge resolution, gauge posi-
tions, and 3D effects caused by the thermal gradient, the top flange
total strain variations were believed to be caused by the rapidly
changing temperature in the deck and perhaps slightly different
thermal conditions inside the vibrating wire gauge than from the
surrounding concrete. An error in concrete temperature of only
1.0°C (1.8°F) would result in a change in estimated total strain of
11:5 mɛ, and although less than the observed range in the top flange,
would still significantly alter the computed curvature through a
section. Curvature estimates at locations with fiber optic strain gauges
used the changes in total strain from both top and bottom flange
gauges. Points omitted from the curvature plots in Fig. 11 indicate
times when critical gauges for computing curvature at the given lo-
cation were malfunctioning.
At Location 7 and near Piers 2 and 3, the measured curvatures
were typically higher than those predicted by the FEM for all
considered gradients. According to FEM results, the strain profile
through the depth of the sections near Piers 2 and 3 was nonlinear,
likely because of the proximity of the thickened diaphragm and
boundary conditions. Plotted FEM curvatures in Fig. 11 represent
values computed using strains through the full section depth.
Examining the strains in the FEM results only at the same
instrumented depths used for computing the measured curvatures
increased the magnitude of the FEM curvature predictions by
approximately 50% near the piers, more accurately describing the
observed curvatures. At Location 3, measured curvatures were
opposite in sign from FEM estimates, although the absolute errors
between measured and FEM results were not substantially dif-
ferent than observed for Location 7. This sign reversal was be-
lieved to be caused by slight discrepancies between boundary
898 / JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013
J. Bridge Eng. 2013.18:890-900.
conditions supplied by the modeled elastomeric disk bearings and
the physical boundary condition. Results at other locations were
predicted well by the model. Overall, it was believed that the
model reasonably captured the behavior of the physical structure.
Similar to the stress results, the FEM curvature results were found
to be insensitive to variations of the temperature distribution in the
top 50 mm (2 in.).
The presented results do not include the tail gradient in the bot-
tom flange. However, it was found that the global FEM behavior was
insensitive to the inclusion of a 1.5°C (2.7°F) tail gradient as was
typical of the maximum positive gradients shown in Fig. 8(b).
Changes in the FEM behavior caused by inclusion or exclusion of
the tail gradient were of similar magnitude to those observed from
the upper and lower bound modifications to the top surface tem-
perature as described previously. It would be expected that the tail
gradient would have a larger impact on the local deformations of the
box cross section than on the global response, but this was not
investigated for this study with the 2D model.
Modeled Bridge Behavior Using Design Gradients
The changes in experimentally derived longitudinal stresses for
each of the five maximum measured positive and negative gra-
dients at Location 7 were compared with the computed values of
stress from the FEM using the following design gradients: the
AASHTO LRFD (AASHTO 2010) gradient for Zone 2 with top
surface temperature equal to 25.6°C (46°F), the Priestley (1978)
fifth-order curve with top temperature equal to 32°C (57.6°F) as
stated in the New Zealand Code (henceforth called Priestley-
NZC), and the Priestley-Z2 gradient as described previously.
All negative design gradients were scaled by a factor of 20:3 from
their respective positive design gradient. These comparisons for
Location 7, which according to FEM results was the instrumented
location of maximum compressive and tensile longitudinal
stresses and very near the location of absolute maximum stresses
for each design gradient, of the southbound bridge are presented in
Fig. 12. Considering positive gradients, the changes in experi-
mentally derived stresses in the bottom flange were consistently
greater than those predicted by the AASHTO LRFD design
gradient applied to the FEM and always less than the Priestley-
NZC gradient. Averaging the experimentally derived longitu-
dinal stresses from the five maximum positive thermal gradients
showed that the Priestley-Z2 gradient was the best fit. For neg-
ative gradients, the changes in stresses approximated from
measured strains at the neutral axis of the section were typically
higher than stresses computed by the FEM using design gradient
values and lower in the bottom flange. This discrepancy was
believed to be caused by the difference in shape of the measured
negative gradients compared with the design gradients as seen in
Fig. 9(b), whereby the measured gradient temperature at a depth
of 1,000 mm (39.4 in.) far exceeded design values. All presented
negative gradients provided reasonable estimates of the experi-
mentally derived stresses.
Measured curvatures along the length of the southbound struc-
ture caused by the maximum measured positive gradients were
compared with FEM-computed curvatures caused by the three
aforementioned design gradients. Curvatures were computed in the
same manner as used for Fig. 11. This comparison for the south-
bound bridge is presented in Fig. 13. At Location 7, the measured
curvatures were typically nearest FEM predictions using the
Priestley-NZC gradient. At most other instrumented locations,
measured curvatures appeared to follow predictions from the
Priestley-Z2 gradient. Overall, it was decided that the Priestley-Z2
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
Fig. 12. Changes in longitudinal stresses approximated from measured strains (averaged at each depth) at Location 7 caused by five maximum
measured (a) positive and (b) negative thermal gradients compared with FEM-computed stresses from design gradients
Fig. 13. Comparison of measured curvatures caused by five maximum
measured positive thermal gradients and computed curvatures from
design gradients
gradient best predicted measured curvature results along the entire
structure.
Recommendations for Design Gradients
In terms of thermal gradient shape and applied stresses and curva-
tures, the Priestley-Z2 gradient was found to best approximate the
measured response of the St. Anthony Falls Bridge, and the
Priestley-NZC was found to provide an upper bound estimate. No-
tably, as shown in Fig. 12, the FEM-computed stresses induced by
the Priestley-Z2 gradient better matched experimental data and were
considerably larger in the bottom flange than those computed using
the AASHTO LRFD gradient. In the top flange, changes in stress
were dominated by compatibility stresses (i.e., stresses induced to
satisfy Bernoulli beam theory that plane sections remain plane) and
were largely the same among the considered design gradients. FEM-
computed compressive stresses at the top fiber of Location 7 using
the AASHTO LRFD positive gradient were 8%, or 0.6 MPa (90 psi),
greater than those computed using the Priestley-Z2, but were 15%,
nearly 1.2 MPa (170 psi), less than those computed from the
Priestley-NZC gradient. These differences in compressive top-fiber
stresses were not significant for design purposes. FEM-computed
tensile stresses at the bottom fiber at Location 7 were 80%, or
JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013 / 899
J. Bridge Eng. 2013.18:890-900.
1.7 MPa (250 psi), greater for the Priestley-Z2 positive gradient and
126%, or 2.7 MPa (390 psi), greater for the Priestley-NZC positive
gradient than those computed using the AASHTO LRFD curve.
Increases in tensile stress of the stated magnitudes were on the order
of the tensile strength of typical concretes, meaning that this in-
crease could be significant if, for example, design was controlled by
tensile stress limits under service conditions. Similarly, the FEM-
computed curvatures were 50–95% greater for the Priestley-Z2 and
85–145% greater for the Priestley-NZC gradient than those from
the AASHTO LRFD curve. Because negative design gradients
were scaled versions of their respective positive gradient, relative
percentile differences between the different negative design gra-
dients were the same as for the positive design gradients, and
differences in magnitude as presented previously were multiplied
by 20:3. The design implications of using the Priestley negative
gradient curves as opposed to the AASHTO LRFD negative
gradient are minor.
It was concluded that the Priestley-Z2 gradient best fit the mea-
sured positive gradient results from the St. Anthony Falls Bridge.
The observation of multiple similarly large positive gradients
throughout the duration of measurements indicated that the Priestley-
Z2 gradient would likely represent a typical annual maximum gradi-
ent (i.e., return period of ∼1 year).
The maximum daily positive gradients consistently occurred
around 2:00 to 4:00 p.m. CST, which implies a possible correlation
between live loading around rush hour and maximum thermal gra-
dient effects. If large live loading and thermal effects are correlated,
then the thermal gradient load factor equal to 0.5 for service limit
states with live load as specified in Section 3.4.1 of the AASHTO
LRFD Bridge Design Specifications (AASHTO 2010) might be
unconservative. Similarly, the live load factor equal to zero for ser-
vice limit states with the full design gradient might also be uncon-
servative. Further statistical investigation is required to resolve these
issues.
Although the shape of the negative gradient could not be con-
sistently captured by the considered design gradients, multiplying
the Priestley-Z2 by a factor of 20:3, as suggested in the AASHTO
LRFD, produced stress results that reasonably approximated the
experimentally derived stresses. Per the observations in Fig. 7(b),
negative gradients with magnitudes greater than the Priestley-Z2
magnitude scaled to the level of the top gauges were observed during
44 days in the 3 years of data collection, resulting in a return period of
approximately 1 month.
 Summary and Conclusions
Temperature gradients were measured over the course of 3 years at
the I-35W St. Anthony Falls Bridge, a posttensioned concrete box
girder structure in Minneapolis, Minnesota. The shapes of the ther-
mal gradients through the depth of the section were found to be
nearest the shape of the fifth-order curve found in the New Zealand
Code (Priestley 1978). The maximum measured top surface gradient
temperatures for positive and negative thermal gradients were best
matched by the top surface temperature specified in AASHTO
LRFD (AASHTO 2010) for solar radiation Zone 2.
A 2D FEM was constructed for the prediction of the structural
Downloaded from ascelibrary.org by University Of Wisconsin-Madison on 01/07/15. Copyright ASCE. For personal use only; all rights reserved.
response subjected to thermal effects. The model was validated
using the five maximum measured positive thermal gradients and
was then applied to compute the structural response caused by the
investigated design gradients. It was found that experimentally de-
rived stresses and curvatures best matched, on average, the pre-
dictions using the Priestley-Z2 gradient, a fifth-order curve scaled to
match AASHTO LRFD Zone 2 top surface temperatures, whereas
the Priestley-NZC gradient, the fifth-order curve with magnitudes as
specified in the New Zealand Code, provided an upper-bound esti-
mate. The predicted curvatures and bottom fiber tensile stresses
caused by the Priestley-Z2 positive gradient were nearly 80% larger
than those using the AASHTO LRFD (AASHTO 2010) design
thermal gradient.
As shown by the variety of conclusions drawn by various authors
in differing studies of thermal gradients, it can be surmised that the
thermal response of structures can be highly variable, dependent not
only on location and climate but also on properties of the structure
such as blacktop covering and cross-sectional shape. For example,
Roberts-Wollman et al. (2002) found that measured gradients in
San Antonio could be approximated by a fifth-order curve with
top surface gradient temperature less than that presented in the
AASHTO LRFD, whereas Shushkewich (1998) stated that gradients
in Hawaii were much like the AASHTO gradients. This study shows
that the AASHTO LRFD (AASHTO 2010) design thermal gradients
are not necessarily conservative for structures in all regions. Specifi-
cally, the Priestley-Z2 scaled fifth-order gradient, which best approx-
imated measured results on average from the St. Anthony Falls Bridge,
was found to cause substantially larger deformations and bottom-
flange tensile stresses than the AASHTO LRFD (AASHTO 2010)
design gradient.
900 / JOURNAL OF BRIDGE ENGINEERING © ASCE / SEPTEMBER 2013
View publication stats J. Bridge Eng. 2013.18:890-900.
Acknowledgments
The authors acknowledge the support of the Minnesota DOT.
Numerical computations were performed using resources provided
by the University of Minnesota Supercomputing Institute. The opin-
ions expressed herein represent those of the authors and not neces-
sarily those of the sponsors.
References
AASHTO. (1994). AASHTO LRFD bridge design specifications, 1st Ed.,
Washington, DC.
AASHTO. (1998). Guide specifications for design and construction of
segmental concrete bridges, 2nd Ed., Washington, DC.
AASHTO. (2010). AASHTO LRFD bridge design specifications, 5th Ed.,
Washington, DC.
ABAQUS 6.10 [Computer software]. Providence, RI, Dassault Systèmes.
French, C. E. W., Shield, C. K., Stolarski, H. K., Hedegaard, B. D., and Jilk,
B. J. (2012). “Instrumentation, monitoring, and modeling of the I-35W
bridge.” Rep. MN/RC 2012-24, Minnesota Dept. of Transportation,
St. Paul, MN.
Hedegaard, B. D., French, C. E. W., Shield, C. K., Stolarski, H. K., and Jilk,
B. J. (2013). “Instrumentation and modeling of I-35W St. Anthony Falls
Bridge.” J. Bridge Eng., 18(6), 476–485.
Imbsen, R. A., Vandershaf, D. E., Schamber, R. A., and Nutt, R. V. (1985).
“Thermal effects in concrete bridge superstructures.” NCHRP Rep. 276,
Transportation Research Board, National Research Council, Washington, DC.
Minnesota Department of Transportation (DOT). (2008). “As-built construction
plan for Bridge Nos. 27409 and 27410.” State Project 2783-120, Federal
Project E.R.MN07(300), St. Paul, MN.
Potgieter, I. C., and Gamble, W. L. (1983). “Response of highway bridges to
nonlinear temperature distributions.” Rep. No. FHWA/IL/UI-201, Univ.
of Illinois at Urbana-Champaign, Champaign, IL.
Priestley, M. J. N. (1978). “Design of concrete bridges for temperature
gradients.” ACI J., 75(5), 209–217.
Roberts-Wollman, C. L., Breen, J. E., and Cawrse, J. (2002). “Measurements
of thermal gradients and their effects on segmental concrete bridge.”
J. Bridge Eng., 7(3), 166–174.
Shushkewich, K. W. (1998). “Design of segmental bridges for thermal
gradient.” PCI J., 43(4), 120–137.
Thompson, M. K., Davis, R. T., Breen, J. E., and Kreger, M. E. (1998).
“Measured behavior of a curved precast segmental concrete bridge erected
by balanced cantilevering.” Rep. No. FHWA/TX-98/1404-2, Univ. of
Texas, Austin, TX.