Engineering Structures 69 (2014) 72–82

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

A framework for statistical distribution factor threshold determination

of steel–concrete composite bridges under farm traffic
Junwon Seo a,⇑, Brent M. Phares b, Justin Dahlberg b, Terry J. Wipf c, Ahmad Abu-Hawash d
a
Department Civil and Environmental Engineering, South Dakota State University, Brookings, SD 57007, United States
b
Iowa State University, Bridge Engineering Center, Ames, IA 50010, United States
c
Iowa State University, Department of Civil, Construction, and Environmental Engineering, Ames, IA 50010, United States
d
Iowa Department of Transportation, Ames, IA 50010, United States

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a novel statistical framework to determine distribution factors (DFs) for steel–
Received 28 December 2012 concrete composite girder bridges subjected to agricultural vehicles. The framework consists of multiple
Revised 9 March 2014 parts including live load field testing, finite element simulations, and statistical analyses. For field testing,
Accepted 10 March 2014
strain sensors are installed at critical locations to monitor strain data resulting from passes of test agri-
Available online 4 April 2014
cultural vehicles. Measured strains are utilized to determine experimental DFs and also used to calibrate
finite element models. As part of the model simulation, a number of vehicles of interest are selected and
Keywords:
applied to the models to compute analytical DFs. Statistical thresholds for each group of interior and exte-
Steel–concrete composite girders
Agricultural vehicles
rior girders are calculated by performing a statistical analysis of the computed data. To demonstrate this
Field tests procedure, a sample application of interest is discussed. Findings indicate that the proposed framework is
Finite element simulations capable of reasonably estimate lateral live-load DFs for interior and exterior girders of the particular rural
Statistical analyses bridge under the effect of varying agricultural loads. The proposed framework is anticipated to provide a
more sophisticated live load distribution characteristics’ estimate on such bridges loaded.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction for bridge engineers. Most available technical documents include

Using appropriate lateral live-load distribution factors (DFs) is a
key process for reliable design and structural safety assessment of
bridges. DFs have been sometimes used to evaluate individual gir-
der damage in a bridge system [1]. Many bridges in the US are in
service on secondary roadways where heavy agricultural vehicles
travel often; these vehicles have quite different characteristics
from traditional highway-type vehicles resulting in rather unique
bridge responses [2–5]. Specifically, Seo et al. [2] demonstrated
that the single front axle of the agricultural vehicle led to greater
loads being carried by the center girder, resulting in a greater field
response compared to the highway vehicles. In the United States,
the structural adequacy of bridges has typically been evaluated
using DFs calculated based upon the American Association of State
Highway and Transportation Officials (AASHTO) specifications
[6,7]. Unfortunately, the AASHTO specifications do not address
farm vehicles. In addition to the specifications, limited studies
relating the impacts of agricultural implements on the structural
adequacy of bridges exist [1–5], resulting in few good resources
⇑ Corresponding author. Tel.: +1 605 688 5226; fax: +1 605 688 6476.
E-mail address: junwon.seo@sdstate.edu (J. Seo).
the examination of these effects on roads and pavements [8–11].
Therefore, a framework for determining agricultural vehicle-
induced DFs in an efficient manner is needed.
Most literature related to bridge load distribution focuses on
computing DFs for traditional road vehicles. The AASHTO specifica-
tions [6,7] provide the DF formulas that were primarily derived
from computational parametric studies for bridges subjected to
conventional trucks [12]. The AASHTO code DFs have been exper-
imentally evaluated by performing field tests on existing bridges
loaded by several test trucks [13–15]. These studies showed that,
in most cases, the AASHTO values were conservatively adequate.
Fortunately, this tendency is consistent with other studies [16–
18] comparing DFs determined from field testing, computational
models, and codified procedures. Cai [19] proposed new DF formu-
las to better calculate the DFs than the AASHTO formulas, yet was
still conservative. The conservative nature of the AASHTO formulas
raised concerns for bridge designers and rating engineers. These
concerns led to the initiation of NCHRP project 12–62 [20] where
the goal was to develop a simplified DF determination framework,
accounting for a broad range of bridge and diaphragm configura-
tions along with transverse truck positions. Although the frame-
work developed from the project was capable of efficiently

http://dx.doi.org/10.1016/j.engstruct.2014.03.005
0141-0296/Ó 2014 Elsevier Ltd. All rights reserved.
 J. Seo et al. / Engineering Structures 69 (2014) 72–82
estimating DFs for most US highway bridges [20–22], various farm
implements, which have vastly different geometries, suspensions,
and other attributes [1–5], have not been considered. In the mean-
time, a recent study [2] has showed that the AASHTO formulas
overestimated DFs for the bridges under most two axle wheel farm
vehicles, but the AASHTO DFs were just permissive for the terrerator
with a front single wheel system. In the European Union, simplified
DF formulas have not been introduced in Eurocode [23] for design-
ing bridges because they may be too conservative or permissive to
be put in practice [24]. In lieu of the DF formulas, a linear elastic
analysis has been predominantly adopted to determine live load
girder distribution that can be of practical interest in the bridge
design and assessment [23,25]. The effects of farm vehicles on live
load distribution characteristics of bridges have not been ad-
dressed in European practice [23–25].
Recent studies have been conducted to examine live load DFs
of bridges subjected to non-conventional highway vehicles such
as military trucks [26,27] and special overload vehicles [28].
Kim et al. [26,27] investigated the flexural live load distribution
characteristics and corresponding load rating of composite
steel–concrete bridges loaded with military trucks. Over one hun-
dred military loading scenarios combining axle spacing, axle num-
ber, and weight were considered to investigate their effects on the
load distribution of representative bridges. It was concluded that
the load distribution was dominated by the weight and spacing
of the military trucks and the AASHTO LRFD formulas conserva-
tively predicted DFs of the bridges. Later a simplified formula
was also proposed for assessing the critical weight of military
trucks on such bridges [27]. Bae and Oliva [28] established a
framework to identify flexural load DFs for multi-girder bridges
under overload vehicles, including single-lane and dual-lane/trai-
ler vehicles, by means of a linear elastic analysis. It was reported
that DF formulas involving vehicle configurations and bridge char-
acteristics were developed to examine their effects on the bridges.
In addition to the non-highway vehicles, other recent DF studies
for different bridge types, including posttensioned box-girder
[29], reinforced concrete T-beam [30], steel I-girder [31–33] have
been carried out using typical highway trucks. As stated above,
significant efforts have been made to modify and implement
design guidelines and specifications related to DFs considering
various vehicle configurations and bridge types. However, techni-
cal information specific to evaluation of bridges under farm traffic
loads is scarce.
The objective of this work is to develop a framework for the
statistical DF threshold determination of steel–concrete composite
bridges loaded by various agricultural vehicles and to study its
application to an actual bridge for substantiating its feasibility.
The framework includes carrying out multiple field tests, rigorous
finite element simulations, and statistical analyses. To demon-
strate the proposed framework, a rural composite steel–concrete
bridge was used in this study. Strain data collected as various
vehicles crossed the structure were used to calculate experimental
DFs and calibrate a finite element model. Utilizing different con-
figurations of farm vehicles commonly used in the United States,
model simulations were conducted using the calibrated model
to determine an ensemble of analytical DFs. Thresholds for each
group of exterior and interior girders were calculated by perform-
ing a statistical analysis of the analytical DFs. To verify the validity
of the statistical thresholds for the select bridge, these thresholds
were compared to those from the field tests and the AASHTO
specifications. It is anticipated that the proposed and validated
framework is capable of more accurately and efficiently comput-
ing DFs for the bridges subjected to agricultural vehicles with
different configurations and weights. Further, the framework can
be used for reliable design and structural integrity evaluation on
such bridges.
73
2. Proposed statistical distribution factor threshold
determination framework
The proposed framework for the statistical determination of
steel–concrete composite bridge DF thresholds is discussed herein
and is illustrated in Fig. 1. Step 1 is to perform multiple field tests
on a selected bridge using full-scale farm vehicles with known
characteristics. The goal of the field tests is to collect actual data
which can be used for both experimental DF calculation and ana-
lytical model calibration. Several factors should be considered
and evaluated prior to performing field tests on the target bridge.
These factors include the selection of rural steel–concrete compos-
ite bridges, development of instrumentation plan, selection of test
vehicles, and determination of loading paths. Representative com-
posite girder bridges located on secondary roadways where farm
vehicles travel frequently should be selected through the consider-
ation of their accessibility and proximity to bridge engineers
performing field testing. The field tests also require a detailed
instrumentation plan so that accurate model calibration may be
achieved. As part of developing the plan, sensor type, sensor num-
ber, and sensor location should be determined for collection of
information. Typical agricultural vehicles frequently found on sec-
ondary roadways should be selected and then their paths should
be determined prior to field tests. During field tests, field response
(e.g., strain time histories) resulting from each vehicle pass is
obtained and used for model calibration.
Step 2 in Fig. 1 is the calibration of analytical models based
upon the field collected data. Calibrating analytical models with
field data is a vital process in the overall framework of determining
statistical DF thresholds of rural bridges. Model calibration can be
made through an iterative process of minimizing errors between
field and analytical data by systematically altering the model until
little further improvement can be made. The calibration process
consists of three major sub-steps: selection of calibration parame-
ters, adjustment of parameters’ values, and statistical comparison.
Calibration parameters affecting structural behavior of rural
bridges should be initially selected. A past study of a typical girder
bridge for its model calibration [34] indicated that model proper-
ties, which include moments of inertia for critical bridge compo-
nents, moduli of elasticity for materials, and restraints at
supports, were considered critical to model calibration. These cal-
ibration parameters are used in this framework. Their initial values
for the parameters are described using information obtained from
bridge plans, inspection history records, and/or field measured
geometries. These values for each calibration parameter can be
adjusted, within reasonable upper and lower limits, until the
Fig. 1. A flowchart of rural bridge distribution factor determination.
74 J. Seo et al. / Engineering Structures 69 (2014) 72–82

analytical response matches the field response. These limits can be necessary to ensure the accuracy and reliability of the model and
defined based on engineering judgment combined with field to allow its use to yield a wide range of DFs for farm vehicles with
response and inspection records. It is typical practice for bridge varying characteristics. To validate the analytical DFs for test vehi-
engineers to use the range of 75% and 125% of the initial values, un- cles, the DFs for all girders of selected bridges are compared to
less other field behavior shows that a greater or smaller range may those from field tests. Specifically, a graphical comparison between
be more appropriate [34,35]. The model is loaded with each test experimental and analytical DFs for different transverse vehicle
vehicle during its calibration. To minimize errors between field positions can be made for this purpose. The comparison plots in-
and analytical responses, statistical comparisons are produced. As clude all girder distribution characteristics resulting from each
shown in the following equations, the errors, including Percent passage of the individual test vehicles. Step 4 in Fig. 1 is to calcu-
Error (PE); Percent Scale Error (PSE); and Correlation Coefficient late an ensemble of analytical DFs. These DFs are computed from
(CC), are used for the statistical comparison [34,35]: the model simulations using varying farm vehicles. For this work,
P ! these vehicles are obtained from searchable databases of farm
ðef  ea Þ2 vehicle specifications from several manufacturers in the United
PE ¼ 100  P ð1Þ
ðef Þ2 States, which include both historical and current vehicle models
characterized in terms of weights, spacings, and widths. Note that
P  the established vehicle pool includes the test vehicles which served
max jef  ea j
PSE ¼ 100  P ð2Þ as the basis for the validation of analytical DFs. Each calibrated and
max jef j
validated model is individually loaded by vehicles at different
0 1 transverse locations considered significant for capturing the most
P critical DFs. Strain response to each vehicle is then used to com-
B ðef  ef Þðea  ea Þ C
CC ¼ @P qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA ð3Þ pute analytical DFs. Repetition of this process for varying vehicles
2 2
ðef  ef Þ ðea  ea Þ produces the sought DFs required for the determination of statisti-
cal thresholds.
in which ef and ea are field and analytical strain responses; and ef Step 5 is to define statistical thresholds of DFs computed in step
and ea denote sample mean values of ef and ea . Detailed information 4. The thresholds are calculated for interior and exterior girders
on the model calibration can be found elsewhere [32]. separately in order to make them comparable to the AASHTO
Step 3 begins with the validation of DFs computed from the code-specified DFs and for consistency with past studies [12–22].
model calibrated in the previous step. This validation process is In this work, thresholds are set as the 95% confidence threshold,

(a)

South curb North curb

(b) Steel girder Concrete girder

1.0m 9@0.69m=6.21m 1.0m

G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12
Strain gauges at mid- and end-span

0.3m 0.012m 0.012m

(c) 0.12m

0.006m 0.022m
0.51m 0.47m

0.41m 0.012m
0.81m 0.012m
0.15m
0.16m
G2 and G11 G3 to G10
0.28m

0.47m
G1 and G12

Fig. 2. Tested bridge: (a) overview, (b) cross section with sensor locations, and (c) girder details.
 J. Seo et al. / Engineering Structures 69 (2014) 72–82 75

Table 1
Test Vehicle Characteristics.

Test vehicles GVW (kN) Length (m) Front Axle Width (m) Rear axle width (m)
Tractor with one tank 340.4 12.4 2.5 2.4
Tractor with two tanks 306.6 19.4 2.7 2.9
Terragator 193.4 7.8 1.0 2.4
Tractor with grain wagon 236.6 10.7 2.7 2.4
Five-axle semi-truck 355.0 15.9 2.1 2.1

indicating a high probability that 95% of all vehicles will induce a representative test agricultural vehicles, including the tractor with
distribution that will not exceed the limit. These thresholds are one tank and the tractor with a grain wagon. The schematics of all
determined based on statistical analyses of all data points five test agricultural vehicles can be seen in Fig. 4.
computed from the model simulations. Note that, for practical Each vehicle was driven at a crawl speed centered on the bridge
applications, analytical DF distribution created from the data and also 0.6 m from centerline, which positioned the vehicle close
points is assumed to be a normal distribution based upon the cen- to the north rail (Fig. 2). Strain time history data were obtained for
tral limit theorem [36] unless the data population is an extremely each vehicle passage; sample plots for field strains resulting from
non-normal distribution function. the tractor with one tank driven at the centerline and at the eccen-
tric position are shown in Fig. 5(a) and (b), respectively. In Fig. 5(a),
the test vehicles traveled at the centerline, the magnitude of strain
3. Application of the framework to rural bridge girders peaks and overall strain responses between each pair of girders
with respect to the centerline such as G1 and G12 are nearly iden-
A sample application of the framework is detailed in the follow- tical. As can be seen in Fig. 5(b) and as expected from the fact that
ing subsections: field testing, model calibration, analytical DF the vehicles were moved across the bridge at the eccentric position
validation, analytical DF computation, and statistical DF threshold (close to the north rail), the overall strain magnitudes at the
determination. bottom flange of the north girders (e.g., G9, G10, G11, and G12),
appeared to be larger than those of the south girders. These field
3.1. Field testing data were used to calculate experimental DFs for each girder along
the length of the bridge using Eq. (4). This equation, seen in past
Since agricultural vehicles with a variety of weights and config- studies of typical composite steel girder bridges [14,37], accounts
urations have historically been used on US rural bridges, multiple for the different section moduli of girders.
field tests were performed on an existing composite steel–concrete  !  !
 ESt;i et;i  M t;i
bridge using different farm vehicles. The simply supported bridge,  
g i ¼ max Pj ¼ max Pj ð4Þ
located on a rural roadway in Boone County, Iowa, has zero skew 8ðt;iÞ  ESt;i et;i 8ðt;iÞ  M t;i
i¼1 i¼1
abutments, a length of 12.1 m, and a deck width of 8.2 m between
the edges of curb. The superstructure is composed of a 0.19 m thick where g i is the girder distribution factor of the ith girder; E the
concrete deck, ten steel interior girders spaced at 0.69 m center-to- Young’s modulus; St;i the section modulus of the ith girder at time
center, and two concrete exterior girders 1.0 m from the adjacent t; et;i the strain quantity at time t at the ith girder; Mt;i the flexural
girder. The first interior steel girders are slightly deeper than the moment at time t at the ith girder and j the total number of girder.
other interior steel girders. The bridge considered typical of US rur- Fig. 6(a) and (b) shows the experimental girder DFs with the
al roadways was selected for this study because the specified vehicles centered on the bridge and located 0.6 m from the
bridge characteristics often found in the rural bridge inventory in-
clude the large number of girders with their narrow spacing. Note
that typical highway bridges have the small number of girders with
their wider spacing as compared to rural bridges. A photograph
and a cross section of the bridge as well as the dimensions of the
girders can be seen in Fig. 2. A network of detachable strain sensors
was installed on the bottom of the girders at mid- and end-span
and was connected to a data logging system managed by a laptop
computer. The complete testing system, which also included a
vehicle tracking component, was used to collect strain data while
test vehicles crossed the bridge.
The test vehicles, including four farm vehicles with different
characteristics and one highway-type truck, were used for this
testing. The farm vehicles consisted of a terragator, a tractor with
a grain wagon, a tractor with one liquid manure applicator tank,
and a tractor with two liquid manure applicator tanks. A five-axle
semi-truck was also employed during testing to experimentally
compare the response to a highway vehicle to that from agricul-
tural vehicles. The farm vehicles’ gross vehicle weights (GVWs)
varied from 193.4 kN to 340.4 kN and lengths ranged from 7.8 m
to 19.4 m measured from front to rear axles; the range of front axle
widths is from 1.0 m to 2.7 m and of the rear axle widths is from
2.4 m to 2.9 m. The five-axle truck had a GVW of 355 kN and a
length and width of 15.9 m and 2.1 m, respectively. Table 1 lists Fig. 3. Sample pictures of representative test agricultural vehicles: (a) a tractor
these vehicle characteristics. Fig. 3 shows sample pictures for with one liquid manure applicator tank and (b) a tractor with a grain wagon.
76 J. Seo et al. / Engineering Structures 69 (2014) 72–82

0.4m
(a)
0.4m 0.7m
2.5m 2.5m 2.4m 2.4m 2.4m
0.6m

0.7m
3.3m 5.6m 1.8m 1.8m

(b) 0.5m 0.7m

2.7m 2.7m 2.9m 2.9m 2.9m 2.9m

0.7m

3.9m 6.4m 1.9m 5.2m 1.9m

(c)
1.0m 2.4m

1.0m

5.8m 1.9m

(d) 0.5m
1.2m

2.7m 2.7m 2.4m

0.8m

3.4m 7.3m

0.2m 0.2m
(e)

2.1m 1.8m 1.8m

0.4m
3.4m 1.3m 9.7m 1.2m
Fig. 4. Schematics of all five test agricultural vehicles: (a) tractor with one tank, (b) tractor with two tanks, (c) terragator, (d) tractor with grain wagon, and (e) five-axle semi-
truck.

centerline, respectively. Each figure includes five lines representing
the DFs for the different test vehicles and two lines indicating the
AASHTO Standard and LRFD DFs. As shown in these figures,
the AASHTO code-compliant DFs appear to be larger than those of
the experimental girder DFs resulting from each of all the test vehi-
cles. It appears that the DFs are fairly sensitive to different vehicle
characteristics and transverse positions for the exterior and interior
girders. In Fig. 6(a), the DFs for the central girders (i.e., G5, G6, G7,
and G8) directly loaded by the wheels of each of the test vehicles
appear to be larger than those of the intermediate girders between
central and exterior girders. Specifically, the terragator has a single
front wheel resulting in a greater percentage of loads being carried
by G6 and G7; thus, the corresponding DFs are much greater than
those of the other girders. In addition to the vehicle characteristics
and transvers positions, the concrete exterior girder DFs are higher
than those from some interior girders because the stiffness of the
concrete exterior girders with edge curb stiffening effects is much
greater than that of interior girders. As for the effect of the eccentric
vehicle loadings on experimental DF characteristics, the DFs of
north and south girders in Fig. 6(b) are, in most cases, greater and
smaller than those of the tests loaded with the concentric vehicles
in Fig. 6(a), respectively. Overall, the load-sharing characteristics
 J. Seo et al. / Engineering Structures 69 (2014) 72–82
Fig. 5. Typical strain response time history: (a) concentric loads and (b) eccentric
loads.
Fig. 6. Envelopes of field-based distribution factors for all girders: (a) concentric
loads and (b) eccentric loads.
77
Fig. 7. Analytical model schematic: (a) overall model loaded with grain wagon
tractor and (b) model cross-section.
for the test agricultural vehicles are similar to those for the highway
trucks, excluding the terragator with the single front wheel system.
3.2. Model calibration
Model calibration was performed to replicate the response to
the test vehicles and then used to calculate analytical DFs for a
large number of farm vehicles with different characteristics. The
calibration process started with modeling the target bridge with
various structural elements, including beam, shell, and rotational
spring elements available in WinGEN, which is a finite element
model software developed by Bridge Diagnostics, Inc. (BDI) [35].
The model was initially generated based upon the bridge configu-
ration and sectional details gained from the as-built plans. Fig. 7a
shows the analytical model for the tested bridge loaded with the
test tractor and grain wagon. Interior girders were modeled with
BDI beam elements. Exterior girders with bridge rails were mod-
eled using the same beam elements as the interior girders, but each
of the rail cross-sections was additionally considered in the model.
The concrete deck was idealized using BDI shell elements. A total of
324 beam and 864 shell elements were used for the model crea-
tion. As shown in Fig. 7b, the deck was located on top of the girder,
and associated centerlines of the girder and the deck were con-
nected with rigid link elements. Theoretical sensors, which are
functionally equivalent to the gauges used in the testing, are also
located at the same positions of field testing in the model. To
reasonably simulate rotational restraints at abutments, support
conditions were modeled using BDI rotational springs. All girders
were modeled as composite cross-sections. Specifically, a total of
five different cross-sections of the composite girders (i.e., G1 G2,
G3–G10, G11, and G12) were separately defined for the model.
Again, section properties and nominal material characteristics for
78 J. Seo et al. / Engineering Structures 69 (2014) 72–82

Table 2
Calibration parameters with model accuracy.

Calibration parameters Components Initial values Calibrated values Percent change (%)
4
Moment of inertia (cm ) G1 6.7E+06 8.1E+06 21.9
G2 9.3E+04 7.3E+04 21.9
G3-G10 1.5E+05 1.7E+05 13.5
G11 9.3E+04 7.3E+04 21.9
G12 8.1E+06 9.9E+06 21.9
Young’s Modulus (MPa) Deck 2.2E+04 2.7E+04 21.9
Rotational spring (cm-kN/rad) Exterior Abutment 0.0E+00 1.1E+06 N/A
Interior Abutment 0.0E+00 4.2E+05 N/A
Statistical errors Percent 15.6% 2.8% 82.1
Scale 5.8% 2.8% 51.7
Correlation Coefficient 91.0% 98.7% 8.5

Note: N/A is referred to as ‘‘Not Applicable’’.

Fig. 8. Comparison between experimental and analytical DFs for all girders: (a) tractor with one tank, (b) tractor with two tanks, (c) terragator, (d) tractor with grain wagon,
and (e) five-axle semi-truck.

each of the girders obtained from the as-built plans were initially
assigned to all elements. Note that the analytical strain responses
were directly obtained from the theoretical sensors after running
the model with each test vehicle.
The generated model was calibrated by allowing model calibra-
tion parameters (including girder moments of inertia, modulus of
elasticity of deck, and rotational restraint at the supports) to vary
between pre-set limits. As stated earlier, the limits were identified
by reviewing the bridge plans, historical inspection records, and
load ratings determined based on field data and engineering judg-
ment regarding the bridge load-carrying capacity. The calibrated
parameters were determined using an iterative process, which
minimized the statistical errors [34] between the computational
and field responses. The model accuracy between initial and
 J. Seo et al. / Engineering Structures 69 (2014) 72–82
Fig. 9. Statistical distributions of the agricultural vehicle characteristics: (a) gross
vehicle weight, (b) vehicle length, and (c) vehicle gage width.
calibrated models is summarized in Table 2. Though the girder
sizes for both G1 and G12 were identical as shown in Fig. 2, their
section properties in the table were slightly different due to the
difference in the neutral axis locations during the passages of the
test vehicles. The difference was attributed to the fact that the neu-
tral axis locations slightly varied depending on the contribution of
railings and vehicle positions to each of the girders. The statistical
errors for the calibrated model, including percent, scale, and corre-
lation coefficient, are 2.8%, 2.8%, and 98.7%, respectively. The
percent changes between the calibrated and initial models are
82.1%, 51.7%, and 8.5%; thus, the calibrated model was able to pro-
duce more reasonable predictions of the bridge behavior when
subjected to the test vehicles than the initial model. For further
discussion of the calibration procedure for minimizing statistical
errors between field and analytical strains of a composite girder
bridge the reader is referred to Seo et al. [34].
3.3. Analytical distribution factor validation
To validate analytical DFs for all girders of the bridge, these DFs
were compared against those of field tests. The graphical
79
Fig. 10. Envelop of analytical distribution factors for 121 agricultural vehicles
driven at each transverse position.
comparison between experimental and analytical DFs for the con-
centric and eccentric vehicle loads is shown in Fig. 8. These plots
for all five test vehicles, including a tractor with one tank, a tractor
with two tanks, a terragator, a tractor with grain wagon, and a five-
axle semi-truck, are given in Fig. 8(a)–(e). The AASHTO Standard
and LRFD-based DFs are also included in these figures. Overall,
the analytical DFs caused by each test vehicle are smaller than
those obtained from the AASHTO codes. It appears that the analyt-
ical DFs for most interior girders subjected to the test vehicles are
somewhat smaller than those of field tests, while the DFs for most
exterior girders are slightly greater than the experimental DFs.
From the comparison, it was concluded that the calibrated model
reasonably produced the experimental DFs in most cases, although
differences between them in some cases occurred.
3.4. Analytical distribution factor computation
It is common for varying agricultural vehicles, which are differ-
ent from normal highway trucks, to be driven over rural bridges
along various travel paths. To take into account the variability in
the existing vehicles, information regarding numerous farm vehi-
cles was collected from several farm implement manufacturers.
In total, 121 unique configurations were established and used to
analytically determine DFs using the model calibrated with field
test data. The 121 vehicles were divided into five groups in terms
of number of axle ranging from two to six axles’ vehicles. Statistical
distributions were generated to better understand the vehicles’
overall characteristics. The gross vehicle weight (GVW), vehicle
length, and vehicle gage width are shown in Fig. 9. It appears that
the mode, which is the value that occur most frequently in a data-
set, of the GVW, vehicle length, and vehicle gage width are 486 kN,
11.5 m, and 3.0 m, respectively. A variance of the GVW of 90 kN to
666 kN, the vehicle length of 3.5–19.5 m, and the gage width of
1.9–3.7 m can be found in this figure. Although the initial develop-
ment of the framework focused on the single bridge structure for
the varying vehicles, it is envisioned that this will be adapted to
parametric bridges considering a variance of geometric factors,
including span length, girder spacing, stiffness parameters, and
deck thickness.
DFs were calculated for all girders for various transverse vehicle
positions using Equation 4. The analytically determined strain
response to each vehicle served as the basis for computing the
DFs. With the AASHTO DFs, envelopes of the girder DFs that were
determined when the vehicles were placed at all transverse
positions from the south rail are shown in Fig. 10. Note that the
particular vehicle placements (i.e., 0.6 m and 4.6 m) were
80 J. Seo et al. / Engineering Structures 69 (2014) 72–82

Fig. 11. Statistical envelopes with multiple distribution factor histograms: (a) interior and (b) exterior girders.

established following the transverse positioning requirements in
the AASHTO specification [7]. This figure further illustrates that
DFs, in some cases, have a substantial variation due to vehicle char-
acteristics. Since the vehicles were positioned close to the south or
north exterior girder whose section modulus was much greater
than interior girders, the largest values occur for G1 and G12 over
the range of all vehicle characteristics. As a result of this tendency,
the exterior girder envelopes exceed the AASHTO values.
3.5. Statistical limit determination
Statistical limits were determined for interior and exterior
girder DFs, respectively. Both limits were defined to be the 95% con-
fidence thresholds so that the probability of observing DFs outside
of the limits is less than or equivalent to 5%. To determine a thresh-
old for the interior girders, the DFs computed from the model
simulations for all interior girders at the considered transverse
 J. Seo et al. / Engineering Structures 69 (2014) 72–82
positions were collected and analyzed to present statistics of the
data in terms of discrete probability density functions (PDFs) for
each position. A system PDF that covers the interior DFs was then
constructed. The system PDF consisted of multiple histograms rep-
resenting the number of observations within each DF interval for
each transverse vehicle position. The 95% confidence threshold
was calculated for the interior girders using a mean and standard
deviation that can be obtained from the system PDF. The same
process to determine a threshold for the exterior girders was also
completed. The thresholds along with multiple PDFs can be seen
in Fig. 11(a) for the interior girders and Fig. 11(b) for the exterior
girders. These figures indicate that no greater than 5% of cases
exceed the thresholds. In both figures, the pattern of multiple PDFs
for the interior and exterior girders appears to be symmetric with
respect to transverse vehicle position. It can be inferred from the
pattern that the agricultural implement loads are distributed pro-
portionally to all girders with respect to the centerline of the bridge.
The symmetric shape can be attributed to the symmetric bridge
geometry with no skewed supports. Fig. 11(a) shows that the
ranges for the interior girder PDFs are consistent for all positions.
The DF ranges for the interior and exterior girders occur due to
variability in the vehicle characteristics. To validate the determined
limits, DFs obtained from field testing were included for the trans-
verse positions of 2.1 m and 2.7 m in both figures, indicating these
points, in most cases, are within the range of the simulation-based
DFs.
4. Comparison with AASHTO codes
A significant lack of information exists regarding rural bridge
load distribution behavior resulting from atypical farm load pat-
terns. The proposed framework associated with variability in farm
vehicles also results in different DFs and different statistical
thresholds, emphasizing the need for another test of their reason-
ableness. Hence, the statistically determined thresholds were com-
pared to those from the AASHTO Standard [6] and LRFD [7]
specifications. As stated by the AASHTO Standard Specifications
[6] for composite steel–concrete bridges, the flexural DFs for mul-
tilane bridges with four or more steel girders can be computed for
interior and exterior girders as follows:
S for agricultural purposes was collected from several manufacturers
g int ¼
1:67
Se
g ext ¼
1:67
in which g int is the interior girder distribution factor; S is girder
spacing (m); g ext is exterior girder distribution factor; and Se is dis-
tance in meter between outside and adjacent interior girders (m).
Note that the AASHTO Standard Equations for DFs are specified
for a wheel load. Hence, the DFs were divided by 2 for comparison
to the LRFD code and field test results which are for a full vehicle
[13–15].
The AASHTO LRFD Specifications for interior girder DFs are a
function of span length, girder spacing, stiffness parameters, and
deck thickness parameters. The exterior girder DFs can be deter-
mined based upon the function of overhang distance and interior
DFs. The flexural DFs specified for a full vehicle in the specifications
can be determined in the following equations.
 0:6  0:2 !0:1
S S Kg
g int ¼ 0:075 þ
2900 L Lt3s
 
de
g ext ¼ 0:77 þ g ð8Þ
2800 int
81
where S is girder spacing (mm); L is span length (mm); t s is depth of
concrete deck (mm); K g is nðI þ Ae2g Þ; n is modular ratio between
girder and deck materials; I is moment of inertia of the girder
(mm4); A is area of the girder (mm2); eg is distance between the
center of gravity of the girder and deck (mm); and de is distance
from the centerline of the web of exterior girder to the interior edge
of curb (mm).
The AASHTO code-compliant DFs for the bridge were calculated
using the listed equations and are included in Fig. 11. The AASHTO
values were compared to those obtained from the field testing and
model simulations. As shown in Fig. 11(a) for interior girders, the
AASHTO values are greater than the field and analytical DFs for
all transverse locations. For exterior girders the simulation-based
DFs are much greater for the positions close to either curb than
those from the AASHTO codes. This tendency was attributed to
the extra stiffness of exterior girders and edge curbs at each side
that is much larger than that of interior girders. However, the
DFs for vehicles driven at the center lines are slightly lesser than
the AASHTO values. In both figures, the AASHTO Standard and
LRFD provided 68% and 117% larger values than the threshold for
interior girders, whereas 27% and 34% smaller values than the
threshold for exterior girders. From these comparisons, one might
conclude that the AASHTO codes for this specific bridge loaded by
different agricultural vehicle types are too conservative for the
interior girders and, in some cases, unsatisfactory for the exterior
girders.
5. Summary and conclusions
This paper summarizes a proposed statistical framework for
determining flexural distribution factors (DFs) of rural bridges sub-
jected to agricultural implements. To demonstrate the application
of the proposed procedure, an existing steel girder bridge with con-
crete decking located on the secondary roadway network in Iowa
was selected. Several field tests with different types of farm vehicles
and a typical five-axle truck were performed to capture the strain
response; two different vehicle positions were considered. Analyt-
ical finite element model of the bridge was created with structural
frame and shell elements and was then calibrated with the field
strains. Information concerning a number of existing vehicles used
ð5Þ
and used as vehicular input loads in the calibrated model. DFs were
computed for all girders after performing structural analyses of the
theoretical vehicle loaded models. Statistical thresholds were
ð6Þ
determined for interior and exterior girders based-upon statistical
analysis of simulation data. These values were also compared to
DFs calculated following two AASHTO codes. Specific conclusions
from these applications can be made as follows:
 Both field and analytical results indicated that the experimental
and analytical DFs were sensitive to different farm vehicle char-
acteristics and transverse positions for the exterior and interior
girders of the select bridge. In addition to vehicle related param-
eters, the stiffness of the exterior girders significantly affected
the live-load DFs. This emphasizes the significance of variability
in farm vehicle characteristics and girders’ stiffness to more rea-
sonably estimate DFs of this bridge type.
 Analytical results revealed that the DFs from the calibrated
model enabled reasonable prediction of actual response of the
bridge, which closely matched the experimental DFs for each
ð7Þ test vehicle. Multiple histograms also showed that analytical
interior girder DFs for all possible transverse vehicle positions
were smaller than those obtained from the AASHTO codes. For
the exterior girders, the simulated DFs were larger for some
vehicle positions, relative to the codified AASHTO values.
82 J. Seo et al. / Engineering Structures 69 (2014) 72–82
Statistical framework proposed in this study was able to better
determine lateral live-load DFs for interior and exterior girders of
the particular rural bridge under the effect of varying agricultural
loads. The statistical thresholds obtained from the framework were
smaller for interior girders, but were larger than those from the
AASHTO codes. This single application may be insufficient to ade-
quately validate this framework in order to generalize live-load
DF trends for the class of all rural bridges. Therefore, further study
of more rural bridges with varying geometries should be carried
out to substantiate the framework’s reliability of determining
DFs. In future work, the framework will be used to develop new
AASHTO code-style DF formulas reflecting variability in rural
bridge characteristics and agricultural loads.
Acknowledgements
The authors wish to acknowledge the support of this work
through a pooled fund project administered by the Iowa Depart-
ment of Transportation. Other participants/sponsors include the
Minnesota DOT, Illinois DOT, Nebraska DOT, Oklahoma DOT, Kan-
sas DOT, Wisconsin DOT, and the USDA Forest Products Laboratory.
References
[1] Al-Saidy AH, Klaiber FW, Wipf TJ, Al-Jabri KS, Al-Nuaimi AS. Parametric study
on the behavior of short span composite bridge girders strengthened with
carbon fiber reinforced polymer plates. Const Build Mater 2008;22(5):729–37.
[2] Seo J, Phares BM, Wipf TJ. Lateral live-load distribution characteristics of
simply supported steel girder bridges loaded with implements of husbandry.
ASCE J Bridge Eng 2014. In press.
[3] Seo J, Phares BM, Dahlberg J, Wipf TJ, Abu-Hawash A. Statistical determination
of agricultural vehicle induced bridge distribution factor threshold. Transport
Res Board 92nd Annual Meeting; 2013.
[4] Wood DL, Wipf TJ. Heavy agricultural loads on pavements and bridges. Iowa
Department of Transportation. Iowa DOT Project HR-1073; 1999.
[5] Phares BM, Wipf TJ, Ceylan H. Impacts of overweight implements of husbandry
on Minnesota roads and bridges. Minnesota Department of Transportation.
MN/RC 2005-05; 2005.
[6] AASHTO. AASHTO standard specifications for highway bridges. Washington,
DC; 1996.
[7] AASHTO. AASHTO LRFD bridge design specifications. Washington, DC; 2010.
[8] Grau RW, Della-Moretta LB. Effects of variable tire pressure on road surfacings.
Transp Res Rec 1991;1291:313–28.
[9] Chatti K, Kim HB, Yun KK, Mahoney JP, Monismith CL. Field investigation into
effects of vehicle speed and tire pressure on asphalt concrete pavement strains.
Transp Res Rec 1996;1539:66–71.
[10] Oman M, Duesen DV, Olson R. Scoping study: impact of agricultural equipment
on Minnesota’s low volume roads. Minn Depart Transp 2001.
[11] Fanous F, Coree BJ, Wood DL. Response of Iowa pavements to a tracked
agricultural vehicle. Iowa Department of Transportation. Iowa DOT Project HR-
1075; 2000.
[12] Zokaie T. Distribution of wheel loads on highway bridges. NCHRP 12-26 Final
Report Research Results Digest 187; 1992.
[13] Kim S, Nowak AS. Load distribution and impact factors for I-girder bridges.
ASCE J Bridge Eng 1997;2(3):97–104.
[14] Nowak AS, Eom J, Sanli A, Till R. Verification of girder-distribution factors for
short-span steel girder bridges by field testing. Transp Res Rec
1999;1688:62–7.
[15] Eom J, Nowak AS. Live load distribution for steel girder bridges. ASCE J Bridge
Eng 2001;6(6):489–97.
[16] Cai CS, Shahawy M, Peterman RJ. Effect of diaphragms on load distribution in
prestressed concrete bridges. Transp Res Rec 2002;1814:47–54.
[17] Cai CS, Shahawy M. Predicted and measured performance of prestressed
concrete bridges. ASCE J Bridge Eng 2004;9(1):4–13.
[18] Yousif Z, Hindi R. AASHTO-LRFD live load distribution for beam-and-slab
bridges: limitations and applicability. ASCE J Bridge Eng 2007;12(6):765–73.
[19] Cai CS. Discussion on AASHTO LRFD load distribution factors for slab-on-girder
bridges. ASCE Pract Periodical Struct Des Constr 2005;10(3):171–6.
[20] Puckett JA, Huo XS, Mertz D. Simplified live-load distribution-factor equations.
NCHRP 592; 2006.
[21] Puckett JA, Huo XS, Patrick MD, Jablin MC, Mertz D, Peavy MD. Simplified live
load distribution factor equations for bridge design. Transp Res Rec 2005;11-
S:67–78.
[22] Puckett JA, Huo XS, Jablin MC, Mertz D. Framework for simplified live load
distribution-factor computations. ASCE J Bridge Eng 2011;16(6):777–91.
[23] EUROCODE. Eurocode 3 – Design of steel structures – Part 2: Steel Bridges;
2006.
[24] Hołowaty J. Live load distribution for assessment of highway bridges in
American and European codes. Struct Eng Int 2012;22(4):574–8.
[25] BA 16/97. Design manual for roads and bridges; 1997.
[26] Kim Y, Tanovic R, Wight R. Load configuration and lateral distribution of NATO
wheeled military trucks for steel I-girder bridges. ASCE J Bridge Eng
2010;15(6):740–8.
[27] Kim Y, Tanovic R, Wight RG. A parametric study and rating of steel I-girder
bridges subjected to military load classification trucks. Eng Struct
2013;56:709–20.
[28] Bae H, Oliva M. Moment and shear load distribution factors for multigirder
bridges subjected to overloads. ASCE J Bridge Eng 2012;17(3):519–27.
[29] Hodson D, Barr P, Halling M. Live-load analysis of posttensioned box-girder
bridges. ASCE J Bridge Eng 2012;17(4):644–51.
[30] Catbas F, Gokce H, Gul M. Practical approach for estimating distribution factor
for load rating: demonstration on reinforced concrete T-beam bridges. ASCE J
Bridge Eng 2012;17(4):652–61.
[31] Devin H. Implementation of an energy-based stiffened plate formulation for
lateral load distribution characteristics of girder-type bridges. Eng Struct
2013;54:168–79.
[32] Razaqpur GA, Shedid M, Nofal M. Experimental investigation of load
distribution in a composite girder bridge at elastic versus inelastic states.
Eng Struct 2013;49:707–18.
[33] Razaqpur GA, Shedid M, Nofal M. Inelastic load distribution in multi-girder
composite bridges. Eng Struct 2013;49:707–18.
[34] Seo J, Phares BM, Lu P, Wipf TJ, Dahlberg J. Bridge rating protocol using
ambient trucks through structural health monitoring system. Eng Struct
2013;46:569–80.
[35] WinGen. WinGen manual. Boulder (CO): Bridge Diagnostics Inc.; 2001.
[36] Montgomery DC. Introduction to statistical quality control. John Wiley & Sons,
Inc.; 1997.
[37] Stallings JM, Yoo CH. Tests and ratings of short-span steel bridges. ASCE J
Struct Eng 1993;119(16):2150–68.