JEMBATAN PLAT (SLAB BRIDGE) International Journal of Concrete

Structures and Materials

RESEARCH Open Access

Lateral Load Distribution for Hollow Slab

Bridge: Field Test Investigation
Yu Zhao, Xiaozhe Cao, Yongjun Zhou* , Gangqiang Wang and Ruixin Tian

Abstract
Distribution factors (DFs) for one typical cross-section as specified in the AASHTO LRFD specification can be varied
when the bridge parameters such as span length, loading lanes and skew are changed. The diversity between design
and actual DFs may be varied as the bridge parameters changed. To study this diversity, this paper presents an evalua-
tion of lateral load DFs for prefabricated hollow slab bridges. The response of the bridge was recorded during the field
test. This field test was divided into two stages: a concentrated force loading test on the prefabricated girder that set-
tled on the bridge supports before the girders were connected transversely and a vehicle loading test after the girders
were connected transversely. The instruments used to record the response of the bridge were strain gauges and dial
indicators. The measured data in the multi-stages of the field test could be used to calibrate the support condition
of the bridge and transverse connection between adjacent girders in the finite element model (FEM) using beam
and plate elements. From the FEM, DFs for this hollow slab bridge were determined and compared with the DFs in
the AASHTO LRFD specification. A parametric study using the calibrated FEM was then used to investigate the effect
of various parameters including span length, skew and bridge deck thickness on the DFs. It was found that AASHTO
LRFD specification is conservative compared with the analysis in the FEM, while this conservatism decreased as the
span length and skew of the hollow slab bridge increased.
Keywords: hollow slab bridge, field test in multi-stage, distribution factor, support condition, transverse connection

1 Introduction bridges are usually significantly stiffer in flexure because

Many hollow slab bridges are designed to be skewed
bridges to satisfy the traffic requirements. The live-load
response of this kind of bridge could be affected by span
length, bridge deck thickness and skewed angle, etc. To
describe and simplify the effect of these parameters, the
distribution factor (DF) is proposed by assigning frac-
tions of transmitted load amongst the primary structural
members (Harris 2010; Song et al. 2003).
Most hollow slab bridges in the design plan were
designed as simply supported. However, it was reported
that even slight changes in boundary conditions have
a considerable effect on the results (Schulz et al. 1995).
Bridge test in Ontario had shown that slab-on-girder
*Correspondence: zyj@chd.edu.cn
School of Highway, Chang’an University, Xi’an 710064, People’s Republic
of China
Journal information: ISSN 1976-0485 / eISSN 2234-1315
of horizontal restraint provided by the girder bearings
(Bakht and Jaeger 1988). Emon and Nowak (2001) pro-
posed that the hinge-roller supports can be partially
fixed (frozen) due to many factors and they found that
the code-specified DF values without considering the
effect of possible support fixity can be too conservative.
The actual boundary condition for the bridge was quite
important for analyzing the bridge structural behavior.
Many field tests were conducted on the existing bridges
and the field test responses were suggested to calibrate
the FEM for further research (Seo et al. 2017; Hodson
et al. 2011, Bechtel et al. 2010). Bridge response under
loading test was the interaction of various parameters.
However, field test data on the existing bridge couldn’t
calibrate all of these parameters. As a result, parameters
had few effect on the bridge behavior were considered to
be the design value in the FEM. For hollow slab bridges,
adjacent girders were connected transversely by the hinge

© The Author(s) 2020. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and
the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material
in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material
is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the
permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativeco
mmons.org/licenses/by/4.0/.
Zhao et al. Int J Concr Struct Mater (2020) 14:22
joint to attract load together. For accurate analysis, trans-
verse connection between adjacent girders and support
condition are two important parameters that need to be
calibrated in the FEM for this type of bridge. Multi-stage
loading tests on the constructing bridge were an effective
way to calibrate these two parameters, respectively.
Many studies investigated the effect of secondary ele-
ments such as parapets, sidewalk and diaphragm on the
DFs and concluded that AASHTO LRFD specification
was conservative without considering these secondary
elements (Mabsout et al. 1997; Conner and Huo 2006;
Namy et al. 2015). Geometric dimensions of the bridge
remained unchanged in these studies. However, geo-
metric dimensions of the bridge were the variables of
the DF calculation formula in AASHTO LRFD specifica-
tion. For a certain kind of bridge, geometric dimensions
could be varied in an appropriate range. As a result, the
designed DFs for different geometric dimensions were
also changed in a certain range. So the effect of geomet-
ric dimensions, which were varied in a rational range, on
the DFs should be figured out first before discussing the
effect of secondary elements.
The aim of this study is to identify the supporting
condition and transverse connection of the hollow slab
bridge in the field tests using multi-stage construction
of full-scale test in situ. First, a concentrated loading
test was conducted on the prefabricated girder that set-
tled on the supports before the girders were connected
transversely. The response of the prefabricated girder
could be used to calibrate the support condition of the
hollow slab in the FEM. Second, a vehicle loading test
was performed on the bridge after the prefabricated gird-
ers were transversely connected. Transverse connection
between adjacent girders could be calibrated with the
294
Hinged bearing
(a) Vertical section
500
inside edge of the parapet
G1 G2 G3 G4 G5
10
250 995
(b) Cross section
Fig. 1 Field tested bridge (units: mm).
Page 2 of 13
bridge response under the vehicle loading test. In addi-
tion, a parametric study was conducted to investigate the
influence of various parameters on the DFs of this hollow
slab bridge including span length, skew and bridge deck
thickness. Finally, the effect of these parameters on the
DFs was quantified compared with the AASHTO LRFD
specification.
2 Bridge Description (STRUKTUR)
A constructing hollow slab bridge, which is located
on WeiXu expressway in Henan province, China, was
selected for this study. The superstructure was sim-
ply supported prefabricated concrete hollow slab gird-
ers, which was shown in Fig. 1. The span of bridge was
13 m in length with a 12.412 m clear span. The width
of the bridge was 13.5 m and skewed at 25 . The hollow
slab girders had a height of 0.7 m and a width of 0.99 m,
and covered 0.1 m thick layer of cast-in-place reinforced
deck. The bridge girders were prefabricated and the deck
was cast. The connection of structures in the bridge
superstructure is shown in Fig. 2. The concrete of hinge
joints and bridge deck was cast at the same time. The
concrete used in the slab girders, hinge joints and bridge
deck had a designed compressive strength of 50 MPa and
modulus of elasticity of 3.45 × 104MPa. The superstruc-
ture was composed of 13 girders, in which the leftmost
girder was denominated G1 while the rightmost girder
denominated G13.
2.1 Concentrated Force Loading Test (PEMBEBANAN)
2.1.1 General
This loading phase was performed after the prefabricated
girders were installed and the adjacent girders had not
been connected transversely. Before this loading test,
13000
100mm cast deck
700
294
Roller bearing
12500 500
100mm cast deck inside edge of the parapet
700
G6 G7 G8 G9 G10 G11 G12 G13
990×11+10×12 995 250
Zhao et al. Int J Concr Struct Mater (2020) 14:22
Fig. 2 Connection configuration of the bridge superstructure.
Fig. 3 Concentrated force loading test.
a preloading test was carried out on the tested girders
to ensure the accuracy of the instruments. G1 (exterior
girder), G4 (interior girder) and G7 (medium girder)
girders were selected to be the test girders. The loading
girder, which weighed 160.44 kN, was also a prefabricated
hollow slab girder. One end of the loading girder was
simply supported on the bearings in the mid-span of the
two tested girders and the other end of the loading girder
was simply supported outside the abutment (Fig. 3). As a
result, the analysis of the tested girder could be simplified
to be a prefabricated girder with a concentrated force on
the mid-span. So the concentrated force applied on the
mid-span of each tested girder was 80.22 kN. The con-
centrated force loading tests were performed on the G1
and G4, G4 and G7, respectively. The dimension of the
girder and material properties were measured in this
loading phrase.
2.1.2 Instrument Arrangement
During the loading test, four strain gauges were placed
on the middle section of each tested girders to measure
strains, while one dial indicator was placed to measure
the mid-span deflection. Where, two dial indicators were
Page 3 of 13
installed on the tested girders near the supports to meas-
ure the support settlement. The deflection caused by the
concentrated force on the mid-span of the tested girder
could be eventually determined. Figures 4 and 5 show
the gauge configurations. Totally, for the three slabs, the
gauges for the strain (D1–D12) were located on the top
and bottom flanges of the girder, while the dial gauges for
the mid-span deflection (DN1–DN3) were located on the
bottom flange of the girder. The strain gauges were sealed
by insulating tape as moisture-proof measurement. The
lead wires from these gauges went through the hinge
joint from the top to the bottom, so that the gauges could
be used in the next loading phase after the bridge deck
was cast. A strain gauge was placed on a stress free con-
crete as the common temperature measurement. The test
data acquired in this loading phase was expected to study
the actual support conditions of the tested bridge.
3 Vehicle Loading Test (PEMBEBANAN)
3.1 General
This loading phase was performed after the slabs were
connected transversely, the bridge deck was cast and the
concrete strength met the design requirements. In order
to excite the normal response of the bridge, Heavy truck
loaded 360 kN was used as the loading vehicle. The truck
was weighed and measured before its arrival on site. Fig-
ure 6 shows the parameters of the loading vehicle.
During the loading test, the loading truck was posi-
tioned at each of the five transverse loading cases (LC)
to produce different response (Fig. 7). When the load-
ing truck was driven to the predetermined position and
the engine was stalled, there was an interval of 5 min to
collect the strain and deflection data until the effect of
the moving truck on the bridge disappeared. Once the
collection of the data was complete in one location, the
loading truck was driven off the bridge and waited for
5 min so that the deformation of the tested girder could
be restored, then, the vehicle loading test could be per-
formed again on the next transverse location. Load-
ing and unloading tests were performed repeatedly to
acquire test data.
3.2 Instrument Arrangement
Strain gauges and dial indicators were all installed on the
bottom of the mid-span section of the girder (Fig. 8), In
which, six gauges that were mounted in the top flange of
the three girders in the former loading test were also used
in this loading phase to measure the strain of the top
flange. Again, two indicators were placed at the supports
and one indicator was placed in the mid-section for each
girder. Consequently, there were 19 strain gauges and 39
dial indicators in this loading phase.
Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 4 of 13

Fig. 4 Strain and deflection gauges.

Fig. 5 Measurement instrument position (units: mm). Note: Di: strain gauges; DNi: deflection gauges.
Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 5 of 13

the main girder and virtual transverse beam were mod-

1094

Hinged Bearing Mid Section Roller Bearing eled by the beam element while the bridge deck was mod-
69.77±N1 69.77±N1 40.±N1
G1 eled by the plate elements. In ANSYS program, beam188
G2 elements were used to model the main girder and virtual
LCA G3
G4
transverse beam, shell63 element were used to model
69.77±N1 69.77±N1 40.±N1
G5 bridge deck. For the beam element in MIDAS CIVIL and
1000×11

G6 ANSYS, each node for these elements had three transla-

G7 tional degrees of freedom and three rotational degrees of
G8
G9
freedom. As for the plate element and shell63 element,
G10 there were six degrees of freedom at each node were used
²±e
G11 to model bridge deck. Material and structural proper-
G12
G13 ties in the FEM were based on the bridge design plan as
Center line of the girder
294 well as the collected and measured information about
1094

13000
Fig. 6 Loading vehicle measurement (units: mm).
4 Finite‑Element Model
In order to assure the correction of the finite element
analysis (FEA), two different FEA programs—MIDAS
CIVIL and ANSYS were conducted. In MIDAS CIVIL,
the bridge supplemented with engineering judgment(Seo
et al. 2015). The concrete grade using for the girder, hinge
joint and bridge deck is C50, The modulus of elastic-
ity of the concrete is 3.45 × 104MPa. As specified in the
AASHTO LRFD specification, live load deflection is a
service issue, not a strength issue. So in the calculation
of the FEM, the nonlinearity of the concrete materials
was not considered. Figure 9 shows the FE models of the
bridge.

1000
7000 LC E
5000 LC D
3000 LC C
1000 LC B
LC A
G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13

(a) Transverse vehicle position

(b) Loading vehicle

Fig. 7 Vehicle loading test configurations (units: mm).
Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 6 of 13

Fig. 8 Instrument arrangement in the vehicle loading test. Note: Pi: strain point, PNi: deflection point.

Fig. 9 Finite element models.

Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 7 of 13

to simulate the effect of bridge deck. Beam elements

Ktop
Kbottom
Fig. 10 Boundary condition used in finite elements.
4.1 Boundary Conditions
Elastic spring elements were used to simulate actual
behavior of supports as shown in Fig. 10 (Eom and
Nowak 2001). To model this kind of connection, general
connection options in elastic connection were selected in
MIDAS CIVIL, while the combin14 elements were mod-
eled in ANSYS model. In this calibrating process, main
girders were not connected transversely. Elastic spring
elements were added to restrict the horizontal movement
of the bridge on the base of the simply support condition
according to the design plan. The stiffness of the spring
elements was represented by K values. The suitable K
values were found by comparing the deflections at the
mid-span from FEM analysis with those from the con-
centrated force loading test by trial and error analysis.
The K values used in the analysis are shown in Table 1,
in which, girder 1 and 13 had different K values with
the other girders because of their different section from
the others. The K values that acquired from different FE
models were the same.
4.2 Transverse Connection
After the boundary conditions of the FEM were deter-
mined, transverse connections between adjacent gird-
ers were calibrated with the field test data acquired in
vehicle loading tests. 100 mm thick plate elements and
shell63 elements were included in this modeling phrase
Ktop and beam188 elements used for transverse connection
between adjacent girders were defined as virtual trans -
verse beam elements, which had no quality but only
transverse stiffness between adjacent girders. The cross
sections of virtual transverse beams were rectangles with
a width of 1000 mm. The height of the virtual transverse
beam was iterated to fit the test results. The truck loads
were applied as concentrated loads and were positioned
on the FEM as presented in the vehicle loading test. By
trial and error, the suitable virtual transverse beam height
was determined by comparing correlation and error
between FEM and field test responses. Percent error,δp,
and correlation coefficient, ρ (Seo et al. 2017) were used
in the comparison:
Σ 2
ωf − ω a
δp = 100 · Σ 2 (1)
ωf
Σ� �
ρ= ω − ω (ωa − ωa)
Σ ��
f f
�2
ω − ω (ω − ω )2 (2)
f f a a
where ωf , ωa, ωf and ωa = field deflection response, FEM
deflection responses, sample mean values of ωf , and sam-
ple mean values of ωa, respectively.
Virtual transverse beam height with highest correlation
coefficient and the lowest percent error was determined
to be the ultimate height. Table 2 presents a summary
of the percent errors and correlation coefficients among
different virtual transverse beam heights. The correla-
tion coefficients of different transverse beam heights are
1, which means using virtual transverse beam method
could simulate transverse connection of adjacent girders
pretty well. Virtual transverse beam with 300 mm height

Table 1 Spring coefficients used in FEM.

FEA program Girder number 1# 2# 3# 4# 5# 6# 7# 8# 9# 10# 11# 12# 13#

MIDAS CIVIL/ANSYS Ktop (kN/mm) 630 770 770 770 770 770 770 770 770 770 770 770 630
Kbottom (kN/mm) 630 770 770 770 770 770 770 770 770 770 770 770 630

Table 2 Model accuracy for different virtual transverse beam heights.

Virtual transverse beam 100 200 300 400 500 600 700
height (mm)

MIDAS CIVIL δp(%) 12.97 5.88 5.32 6.46 8.06 9.75 11.37
ρ 1 1 1 1 1 1 1
ANSYS δp(%) 9.32 8.84 6.91 9.17 10.03 10.55 11.36
ρ 1 1 1 1 1 1 1
Zhao et al. Int J Concr Struct Mater (2020) 14:22
Fig. 11 The deflection of the test girder.
has the lowest percent error in both FEM analyses. So
the virtual transverse beam with a width of 1000 mm and
a height of 300 mm was found to forecast most closely
actual transverse connections between adjacent girders.
5 Live Load Distribution Factors
The actual DF of each girder can be calculated using the
experimental deflection data. The effective deflection
under vehicle loading was calculated by the following
Eq. 3 as shown in Fig. 11.
ωe ωt ωr ωc
(3)
where ωe is the effective measured deflection of girders
under loads, ωt is the total measured deflection of gird-
ers, ωr is the measured residual deflection of girders, ωc is
the measured deflection at both ends of girders.
The formula used here to calculate DF for the ith girder
is Eq. 4 (Harris et al. 2008), which has been widely used
in many live-load tests (Civitillo et al. 2014; Harris et al.
2016; Waldron et al. 2005). These DFs are representative
of a single-truck loading resisted by each girder. DFs in
five different loading conditions were shown in Fig. 12.
DFi = ωi Σ
13
× Nvehicles
ωi (4)
i=1
where, DFi is load distribution factor of ith girder; ωi is
maximum deflection of ith girder; Nvehicles is the number
of vehicles on the bridge.
It can be concluded that the girder which suffered
the traffic load directly experiences the greatest deflec-
tion, whereas girders far away from the loading position
experience less. DFs for exterior girder were 52% greater
than the DFs for the most heavily loaded interior girder,
the reason is that in this test there are no railings which
usually offered a significant stiffening contribution to the
exterior girders and the exterior girder is unable to dis -
tribute load to an interior member.
Figure 13 shows the deflections at mid-span on each
girder under different loading conditions. It can be
seen from Fig. 13 that the deflection curves were nearly
overlapped for the five loading conditions. It can be
Page 8 of 13
Fig. 12 DFs in different loading conditions.
concluded that the FEM is reliable to conduct further
researches.
Distribution factors for single and multilane loaded are
different in AASHTO specification. Only one lane loaded
was conducted in the field test. The number of lane was
changed in the calibrated FEM to observe the deflection
of girders. The loading vehicle used in the FEM was the
same as it used in field test and the vehicle wheel loads
were applied as concentrated loads in the FEM. The
bridge was divided into three 3.75 m-wide-lanes. FEM
with concentrated load was shown in Fig. 14. Loading
vehicles were positioned on the middle of each lane later-
ally. The longitudinal position of trucks was calculated as
the position producing the maximum bending moment
at midspan. The number and position of the load vehicles
were systematically varied to consider all possible load
combinations. In all cases the vehicles were placed fol-
lowing the 25 skew.
Interior and exterior distribution factors are calculated
separately in the AASHTO specification. As such, the
maximum distribution factor for the one, two and three
loaded lanes for both the interior and exterior girders
were calculated using the maximum deflection of differ-
ent girders. Calculated moment DFs for the interior and
exterior girders for each load cases are listed in Table 3.
DFs calculated by the AASHTO LRFD were also listed in
Table 3.
Table 3 shows that using AASHTO LRFD specifications
resulted in a conservative DFs estimate. The AASHTO
DFs were much higher than those calculated using the
calibrated FEM for exterior girders. For one lane loaded,
the AASHTO LRFD DFs were 1.5 times larger than that
of the FEM DFs for exterior girders and 1.2 times larger
for interior girders. For multilane loaded, the AASHTO
Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 9 of 13

Fig. 13 Deflection distribution in different loading conditions.

Zhao et al. Int J Concr Struct Mater (2020) 14:22
Fig. 14 FEM with three loaded lanes.
LRFD DFs were 1.4 times larger than that of the FEM
DFs for exterior girders and were very close to the FEM
DFs for interior girders. DFs for exterior girder were con-
trolling DFs in AASHTO LRFD specification, while the
controlling DFs were interior girder DFs in the FEM. This
is probably caused by the lack of parapets in the FEM,
which could provide stiffness for the exterior girders.
DFs for interior girder in AASHTO LRFD specification
was 1% to 18% higher than the DFs in FEM. For exterior
girder, DFs of AASHTO LRFD specification was 37% to
48% higher than the DFs in FEM.
As the loading lane increased from one to two in the
FEM, DFs increased 17% for exterior girders and 34% for
interior girder. While as the loading lane increased from
two to three, the DFs increased by 2% and 9% for exte-
rior and interior girders, respectively. DFs could be dif-
ferent for one lane loaded and two lanes loaded, but less
difference for two lanes loaded and three lanes loaded. In
multilane loaded cases, DFs of three loaded lanes were
more closed to the AASHTO LRFD specification com-
pared with the DFs of two loaded lanes. When the bridge
Table 3 Distribution factors for different number of lanes.
Number of lanes Calculation method
One lane AASHTO LRFD
MIDAS
ANSYS
Two lanes AASHTO LRFD
MIDAS
ANSYS
Three lanes AASHTO LRFD
MIDAS
ANSYS
DFEXT distribution factor of exterior girder; DFINT distribution factor of interior
girder.
Page 10 of 13
was filled with vehicles on the lanes that it can be divided
into, the DFs were more closed to the AASHTO LRFD
specification for this multilane loaded case.
6 Parametric Study
A parametric study was performed for hollow slab bridge
to investigate the parameters affecting the DFs of vehicle
loads. The effect of parameters on the DFs were quanti-
fied and compared with AASHTO LRFD specifications.
Span length, bridge skew and bridge deck thickness were
three major parameters in this parametric study. A base-
line FEM was determined using the dimensions and sup-
porting conditions of the field tested bridge. The baseline
model was adjusted to study the effect of one parameter
while the other parameters were kept in the baseline
value. While changes in certain parameters (i.e., span
length) do affect other parameters (i.e., girder stiffness),
the effect was minimal and they could be evaluated inde-
pendently (Zokaie 2000). Therefore, span length was var-
ied from 8 to 20 m, which was the specified length of the
hollow slab bridge, in the FEM. The effect of skew was
investigated by adjusting the skew angle from 0 to 60 .
The bridge deck thickness was varied from 0.1 m to 0.5 m
by adjusting the thickness of plate element. DFs calcu-
DFEXT DFINT
lated from the FEM were compared with those calculated
0.230 0.202 ones according to the AASHTO LRFD specifications, this
0.155 0.171 ratio of which could reflect the difference between FEM
0.155 0.175 and AASHTO LRFD specifications.
0.265 0.255 DFs for one loaded lane and three loaded lanes were
0.183 0.231 calculated in the Table 4 to investigate the effect of span
0.180 0.233 length. It can be seen from Table 4 that DFs for exte-
0.265 0.255 rior girder increased 10% for one loaded lane and 42%
0.193 0.252 for three loaded lanes, while for interior girder, DFs
0.190 0.252 decreased 30% for one loaded lane and 9% for three
loaded lanes. As the DFs calculated by MIDAS CIVIL
and ANSYS were really close, so the DFs in Fig. 15 were
Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 11 of 13

Table 4 DFs for span parameter in the FEM.

Span (m) 8 10 13 16 20

Calculation method MIDAS ANSYS MIDAS ANSYS MIDAS ANSYS MIDAS ANSYS MIDAS ANSYS

DFEXT-1 0.136 0.132 0.147 0.147 0.155 0.155 0.155 0.155 0.150 0.147
DFINT-1 0.205 0.209 0.177 0.174 0.171 0.175 0.160 0.163 0.148 0.145
DFEXT-3 0.148 0.145 0.175 0.171 0.193 0.190 0.200 0.197 0.210 0.206
DFINT-3 0.269 0.272 0.261 0.263 0.252 0.255 0.251 0.251 0.246 0.249
DFEXT-i distribution factor of exterior girder for i loaded lane; DFINT-i distribution factor of interior girder for i loaded lane.

lanes. The lack of parapets in the FEM, which could pro-

Fig. 15 FEM/AASHTO ratio for length parameter. Note: R
EXT-i
of exterior girder in i loaded lane; RINT-i: DF ratio of interior girder in i
loaded lane.
the average values of the DFs in two different FEM anal-
yses. In Fig. 15, DF ratios increased and were closed to
1 as the span length increased for both one loaded lane
and three loaded lanes. This trend showed that AASHTO
LRFD specification became less conservative as the span
length increased. The ratio of the FEM to AASHTO dis-
tribution factor for exterior girder ranged from 0.45 to
0.81 for one loaded lane and from 0.45 to 0.76 for three
loaded lanes. While for interior girder, the ratio of the
FEM to AASHTO LRFD specification distribution factor
was 0.76 to 0.91 for one lane and 0.84 to 0.95 for three
vide stiffness for the exterior girder, made the DF of exte-
rior girder lower than that of interior girder.
DFs in different bridge skew were shown in Table 5.
Compared with the DFs of the straight bridge (of which
the skew is 0 ), the DFs decreased 5% for exterior girder
and 4% for interior girder when the bridge skew was 5 .
For exterior girder, DFs were not changed for the skew of
10 , 15 and 25 , while the DFs decreased 3% to 11% as
the skew increased from 35 to 60 . DFs of interior girder
were not changed as the skew increased from 10 to 45
and the DFs increased 3% when the skew was 60 . DFs in
Fig. 16 were also the average DF values of two different
FEM analyses. It can be seen in the Fig. 16 that AASHTO
became less conservative as the skew increased from 10 .
: DF ratio The ratio of FEM and AASHTO was greater than 1 for
interior girder when the skew was 60 , while the DFs of
exterior girder were less than design value for each of
skew angle.
The calculation results from MIDA CIVIL and ANSYS
were pretty close. DFs in Fig. 17 were the average val-
ues of the two FEM analyses. DFs for exterior girder
increased 4% and 5% for one and three loaded lanes,
respectively, while for interior girder, DFs decreased 3%
and 2% for one and three loaded lanes, respectively. Nei-
ther the DF of exterior nor the interior girder was influ-
enced by the deck thickness significantly in Fig. 17.
7 Conclusions
A hollow slab bridge was subjected to a live-load test
and detailed FE analyses. DFs for AASHTO LRFD speci-
fication and actual bridge were compared. In addition,

Table 5 DFs for skew parameter in the FEM.

FEA program Skew 0 5 10 15 25 35 45 60

MIDAS CIVIL DFEXT 0.194 0.184 0.195 0.195 0.193 0.186 0.188 0.172
DFINT 0.252 0.242 0.253 0.253 0.252 0.254 0.253 0.261
ANSYS DFEXT 0.193 0.183 0.192 0.192 0.190 0.184 0.186 0.170
DFINT 0.254 0.245 0.255 0.255 0.255 0.255 0.256 0.263
Zhao et al. Int J Concr Struct Mater (2020) 14:22
Fig. 16 FEM/AASHTO ratio for skew parameter.
Fig. 17 DFs for bridge deck thickness parameter.
the effects of span length, skew and bridge deck thick-
ness on the DFs were also investigated in the parametric
study. From the analyses, the following conclusions were
obtained:
1. Compared witfl tfle DFs in tfle FEM of tflis flollow
slab bridge, AASHTO LRFD DFs for interior girder
(1–18%) were less conservative tflan exterior girder
(37–48%).
2. The AASHTO LRFD specification became less con-
servative (8%) wflen tfle loading veflicles were filled
in tfle lane tflat tfle bridge can be divided to for mul-
tilane loaded cases.
3. AASHTO LRFD specification became less conserva-
tive as tfle span lengtfl increased for tflis flollow slab
Page 12 of 13
bridge. DFs of exterior girder increased 10–42%
wflile of interior girder decreased 9–30% as tfle span
lengtfl increased.
4. AASHTO LRFD specification became less conserva-
tive as tfle skew was greater tflan 10 . DFs of FEM for
interior girder were 1.4 times larger tflan tfle DFs in
AASHTO LRFD specification wflen tfle skew was
60 , wflile tfle DFs of exterior girder were less tflan
design value witfl an increase in tfle skew.
5. Deck tflickness flad little effects on tfle DFs of flollow
slab bridge.
8 Discussion
The parapets have been proved to provide more stiffness
to the exterior girder compared with the interior girders
(Barker 2001; Zhou et al. 2015). As the vehicle loading
test was conducted on the bridge before the parapet was
constructed, the parapets were not modeled in the FEM.
So the effect of the parapets on DFs was not discussed in
this paper.
Acknowledgements
The authors thank Zhongguo John Ma for his assistance during writing this
paper and Wanfeng Liu and Dong Zheng for their assistance during the field
tests.
Authors’ contributions
YZ proposed the aim of the research and designed the field test. GW and RT
conducted the field test and collected the test data. XC and YZ analyzed and
interpreted the test data regarding the live load distribution. YZ was a major
contribution in writing the manuscript. All authors read and approved the
final manuscript.
Funding
Finical support by the key research platform open fund project of China
No. 310821171121. National Science Foundation of China under Grant Nos.
51678061 and 51978063.
Availability of data and materials
The datasets used and analyzed during the current study are available from
the corresponding author on reasonable request.
Competing interests
The authors declare that they have no competing interests.
Received: 6 September 2019 Accepted: 31 January 2020
References
Bakht, B., & Jaeger, L. G. (1988). Bearing restraint in slab-on-girder bridges.
Journal of Structural Engineering, 114, 2724–2740.
Barker, M. G. (2001). Quantifying field-test behavior for rating steel girder
bridges. Journal of Bridge Engineering, 6, 254–261.
Bechtel, A., McConnell, J., & Chajes, M. (2010). Ultimate capacity destructive
testing and finite-element analysis of steel I-girder bridges. Journal of
Bridge Engineering, 16, 197–206.
Civitillo, J. M., Harris, D. K., Gheitasi, A., Saliba, M. & Kassner, B. L. 2014. In-service
performance and behavior characterization of the hybrid composite
bridge system—A case study. arXiv preprint arXiv:1409.2447.
 Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 13 of 13

Conner, S., & Huo, X. S. (2006). Influence of parapets and aspect ratio on live-
load distribution. Journal of Bridge Engineering, 11, 188–196.
Eom, J., & Nowak, A. S. (2001). Live load distribution for steel girder bridges.
Journal of Bridge Engineering, 6, 489–497.
Harris, D. K. (2010). Assessment of flexural lateral load distribution methodolo-
gies for stringer bridges. Engineering Structures, 32, 3443–3451.
Harris, D. K., Civitillo, J. M., & Gheitasi, A. (2016). Performance and behavior of
hybrid composite beam bridge in Virginia: Live load testing. Journal of
Bridge Engineering., 21, 04016022.
Harris, D. K., Cousins, T., Murray, T. M., & Sotelino, E. D. (2008). Field investiga-
tion of a sandwich plate system bridge deck. Journal of Performance of
Constructed Facilities, 22, 305–315.
Hodson, D. J., Barr, P. J., & Halling, M. W. (2011). Live-load analysis of postten-
sioned box-girder bridges. Journal of Bridge Engineering, 17, 644–651.
Mabsout, M. E., Tarhini, K. M., Frederick, G. R., & Kobrosly, M. (1997). Influence of
sidewalks and railings on wheel load distribution in steel girder bridges.
Journal of Bridge Engineering, 2, 88–96.
Namy, M., Charron, J.-P., & Massicotte, B. (2015). Structural behavior of bridge
decks with cast-in-place and precast concrete barriers: Numerical mod-
eling. Journal of Bridge Engineering, 20, 04015014.
Schulz, J. L., Commander, B., Goble, G. G., & Frangopol, D. M. (1995). Efficient
field testing and load rating of short-and medium-span bridges. Struc-
tural Engineering Review, 3, 181–194.
Seo, J., Kilaru, C., Phares, B. M., & Lu, P. (2015). Live load distribution factors for
a short span timber bridge under heavy agricultural vehicles. Structures
Congress, 2015, 2164–2173.
Seo, J., Teja Kilaru, C., Phares, B., & Lu, P. (2017). Agricultural vehicle load distribu-
tion for timber bridges. Journal of Bridge Engineering, 22, 04017085.
Song, S.-T., Chai, Y., & Hida, S. E. (2003). Live-load distribution factors for con-
crete box-girder bridges. Journal of bridge engineering, 8, 273–280.
Waldron, C. J., Cousins, T. E., Nassar, A. J., & Gomez, J. P. (2005). Demonstration of
use of high-performance lightweight concrete in bridge superstructure
in Virginia. Journal of performance of Constructed Facilities, 19, 146–154.
Zhou, Y. J., Ma, Z. J., Zhao, Y., Shi, X. W., & He, S. H. (2015). Improved definition of
dynamic load allowance factor for highway bridges. Structural Engineering
and Mechanics, 54, 561–577.
Zokaie, T. (2000). AASHTO-LRFD live load distribution specifications. Journal of
Bridge Engineering, 5, 131–138.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.

KELEBIHAN & KEKURANGAN

- Berdasarkan informasi yang diberikan, gap penelitian yang diidentifikasi dalam jurnal tersebut adalah
kurangnya pertimbangan terhadap kondisi batas aktual dan parameter yang mempengaruhi respons
beban hidup jembatan pelat berongga. Uji lapangan yang dilakukan pada jembatan eksisting tidak
mampu mengkalibrasi seluruh parameter sehingga menimbulkan asumsi nilai desain pada model elemen
hingga (FEM). Kesenjangan ini menunjukkan perlunya penelitian lebih lanjut untuk menganalisis secara
akurat perilaku struktur jembatan pelat berongga dengan mempertimbangkan kondisi batas aktual dan
parameter yang mempengaruhi responsnya terhadap pembebanan.

PEMBEBANAN
- Dalam jurnal ini, terdapat dua jenis pembebanan yang dianalisis, yaitu pembebanan gaya terkonsentrasi
dan pembebanan kendaraan.
- Pembebanan Gaya Terkonsentrasi: Pada tahap uji beban gaya terkonsentrasi, beban terkonsentrasi
diterapkan pada gording eksterior jembatan menggunakan metode uji lapangan. Beban ini diterapkan
pada tengah setiap jalur jembatan secara lateral. Jumlah dan posisi kendaraan beban divariasikan untuk
mempertimbangkan semua kombinasi beban yang mungkin.
- Pembebanan Kendaraan: Pada tahap uji beban kendaraan, beban kendaraan diterapkan sebagai beban
terkonsentrasi pada FEM jembatan. Beban kendaraan ini merupakan beban yang dihasilkan oleh
kendaraan yang melintasi jembatan. Jumlah dan posisi kendaraan beban divariasikan untuk
mempertimbangkan semua kombinasi beban yang mungkin.

PERENCANAAN STRUKTUR & STRUKTUR

- Desain struktural jembatan yang dijelaskan dalam konteksnya adalah jembatan pelat berongga dengan
balok penopang prefabrikasi dan dek bertulang cor di tempat. Jembatan ini mempunyai panjang bentang
13 meter, dengan bentang bersih 12.412 meter. Lebarnya 13,5 meter dan miring 25 derajat.
- Balok pelat berongga yang digunakan pada jembatan ini memiliki tinggi 0,7 meter dan lebar 0,99 meter.
Mereka dibuat dari pabrikasi, sedangkan deknya dicetak di lokasi. Sambungan struktur pada bangunan
atas jembatan ditunjukkan pada Gambar 2 konteks.
- Beton yang digunakan pada gelagar pelat, sambungan engsel, dan dek jembatan mempunyai kuat tekan
rencana sebesar 50 MPa dan modulus elastisitas sebesar 3,45 × 10^4 MPa. Bangunan atas terdiri dari 13
gelagar, dengan gelagar paling kiri diberi nama G1 dan gelagar paling kanan diberi nama G13.
 Zhao et al. Int J Concr Struct Mater (2020) 14:22 Page 14 of 13

MANFAAT

- Memahami distribusi beban lateral pada jembatan hollow slab: Penelitian ini memberikan pemahaman
yang lebih baik tentang distribusi beban lateral pada jembatan hollow slab. Hal ini penting untuk
memastikan keandalan dan keamanan struktur jembatan.
- Validasi metode analisis numerik: Penelitian ini menggunakan uji lapangan untuk memvalidasi metode
analisis numerik yang digunakan dalam perencanaan jembatan hollow slab. Validasi ini membantu
memastikan bahwa metode analisis yang digunakan memberikan hasil yang akurat dan dapat
diandalkan.
- Peningkatan desain jembatan hollow slab: Dengan pemahaman yang lebih baik tentang distribusi beban
lateral, penelitian ini dapat membantu dalam meningkatkan desain jembatan hollow slab. Hal ini dapat
mengarah pada pengembangan pedoman desain yang lebih efisien dan aman.
- Kontribusi pada penelitian jembatan beton: Penelitian ini memberikan kontribusi pada penelitian yang
lebih luas tentang jembatan beton. Hasil penelitian ini dapat digunakan sebagai referensi untuk
penelitian selanjutnya dalam bidang ini.

CONTOH JEMBATAN :

Jembatan Bandar Ngalim Kota Kediri

JEMBATAN GIRDER

Contents lists available at ScienceDirect

Engineering Structures

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

Load-carrying capacity evaluation of girder bridge using moving vehicle

Zhen Sun a, Dionysius M. Siringoringo b,*, Yozo Fujino b
a
State Key Laboratory of Safety and Health for In-Service Long Span Bridges, Jiangsu Transportation Institute, 2200 Chengxin Street, Jiangning kexueyuan, Nanjing,
China
b
Institute of Advanced Sciences, Yokohama National University, 79-1 Tokiwadai, Hodogaya-ku, Yokohama, Kanagawa 240-8501, Japan

AR T I C LE I N F O A B S T R A C T

Keywords: Efficient and reliable load-carrying capacity evaluation of deficient bridges is essential for bridge maintenance
Bridge load-carrying capacity and management. Based on load-carrying capacity evaluation of an existing bridge, operator can formulate
Static load test strategies and implement countermeasures such as repair, strengthening or traffic regulation. Currently, static
Truck pass-by test load test is used widely for evaluation of bridge load-carrying capacity. The test, however, is often considered
Radar device
expensive and requires massive traffic closure that hinders frequent implementation on highway or bridges in the
Girder bridges
rural area. (KELEMAHAN) In this paper, an efficient method for load-carrying capacity evaluation of girder
Influence line
bridges based on displacement under moving vehicle is proposed. (KELEBIHAN) The method consists of three
main procedures, namely, radar- based dynamic displacement measurement under truck pass-by test, influence
line extraction, and estimation of the displacement under the static load test scenario. Based on these procedures,
evaluation of load-carrying capacity is conducted following the AASHTO manual. The method was verified in
numerical simulations using three-dimensional finite element model of typical girder bridge under vehicle loading.
Effects of vehicle weight, vehicle speed, and road roughness on the accuracy of displacement estimation were
investigated in the numerical simulations. Full-scale implementation of the method was conducted on a simply
supported pre- stressed concrete girder bridge to investigate feasibility of the proposed method and to provide
guidance for practical application. The results reveal that load rating factor of the girders can be reliably evaluated.
The evaluation process was also more effective and economical than the conventional method.

1. Introduction condition assessment of real bridges. Due to the variation in material

Deterioration of bridge structures is a widespread problem in many
developed and developing countries. This is mainly due to ever-
increasing traffic volume, poor quality of material, poorly designed
structure, harsh environment condition, and inadequate maintenance.
Bridge deteriorations limit the service life and pose challenges for
maintenance of infrastructure network. Generally, to prevent deterio-
ration on various levels, periodic inspection is specified in the codes of
most countries. Periodic inspection usually includes visual inspection
and/or non-destructive techniques (NDTs). From inspections, conditions
of bridges are rated at different levels according to the deficiency status.
An important result of inspection that concerns many bridge opera-
tors or authorities is the load-carrying capacity. Bridge load-carrying
capacity evaluation can be carried out by calculation-based methods,
where inspection results on local damages are incorporated to achieve
realistic model of the bridge. Brownjohn et al. [1] described a sensitivity-
analysis-based finite element model updating method and applied it to
properties, boundary conditions and composite action, load test is often
performed to update the finite element (FE) model used in the
calculation-based method. For example, Ding et al. [2] combined model
updating with field test data in a nonlinear finite element analysis to
predict the behavior of a slab-girder bridge. Nonlinear analysis is per-
formed on the updated model to evaluate the bridge load-carrying ca-
pacity, which verified its effectiveness.
A number of researchers have proposed various methods to evaluate
the bridge load-carrying capacity. Casas and Gomez [3] proposed a
simple method based on the reliability theory to consider the random
nature of the traffic load to calculate the target proof load in tests for
bridge assessment. Olaszek et al. [4] showed that diagnostic load test
can contribute to the bridge capacity assessment with examples of three
types of bridges, which are reinforced concrete, prestressed concrete,
and steel bridge. An in-depth overview of the state-of-the-art and future
research trends with regard to load test of concrete bridges is provided in
[5].

* Corresponding author at: Institute of Advanced Sciences, Yokohama National University, Japan.
E-mail address: dion@ynu.ac.jp (D.M. Siringoringo).

https://doi.org/10.1016/j.engstruct.2020.111645
Received 23 June 2020; Received in revised form 29 September 2020; Accepted 22 November 2020
Available online 16 December 2020
0141-0296/© 2020 Elsevier Ltd. All rights reserved.
Z. Sun et al.
Both time and cost-effective field test are necessary in practice for
load-carrying capacity evaluation of a bridge. For example, traffics on
vital link such as highway bridges are too busy that the full bridge
closure as required by conventional static load test are often impossible
to implement. On the other hand, budget for maintenance is often
limited to perform such testing on the bridges in rural areas and small
cities. This condition leaves bridges on rural roads lacking a reliable
load-carrying capacity rating even when their conditions are visibly
deficient.
In this paper, a reliable and efficient load-carrying capacity evalua-
tion method is proposed for girder bridge using moving vehicle. Instead
of conducting the static load test with several trucks, the proposed
method utilizes a pass-by truck to obtain displacements of the bridge
girder, and a radar-based device to measure the dynamic displacement.
The paper is organized such that the background on the load-carrying
capacity is firstly discussed. Next, the proposed methodologies for
load-carrying capacity using moving vehicle are described in detail.
Numerical simulations and parametric studies using three-dimensional
finite element model are latter presented, followed by full-scale imple-
mentation on a typical girder bridge. Finally, results of experiments are
discussed in detail, and some suggestions for practical implementation
of the method are provided. (MANFAAT)
2. Background for load-carrying capacity evaluation
In most countries, routine bridge inspection is conducted at a regu-
lated maximum interval. The interval is specified as 2 years, 3 years, and
5 years in United States, China, and Japan, respectively. Bridge operator
may also schedule special inspection within the periodic interval to
investigate a known or suspected deficiency, such as after a vehicle/ship
collision. Unfortunately, the widely used condition rating systems usu-
ally do not provide direct information on the remaining load-carrying
capacity of the severely deficient bridges. Such information is essential
for bridge operators to formulate strategies and implement counter-
measures such as repair, strengthening or imposing traffic regulation on
the bridge. Therefore, an in-depth investigation on the load-carrying
capacity is often necessary. The investigation is usually conducted
with the calculation-based method using finite element (FE) model
made using the design documents.
In the United States, the AASHTO bridge evaluation manual rec-
ommends comparison between the field tests and analytical results to
update the bridge load rating factor [6]. Several European countries
have adopted or in the process of adopting new codes for the assessment
of existing bridges such as the UK, Denmark, and Switzerland [3]. The
codes incorporate numerous advanced techniques for load rating of
existing bridges such as the use of data from weight-in-motion (WIM)
system. Planning, load application, measurement equipment and test
procedures for bridge load testing are also described in the guidelines for
existing bridges in UK [7]. Similar load test is also required in China [8]
for deficient bridges. The guidelines usually recommend static testing to
obtain responses and later compare them with the reference values
obtained analytically or by finite element model. Afterwards, the dead
load and live load effects are calculated and compared with the resis-
tance factor to obtain estimation of the remaining capacity. In Japan,
although not explicitly required in the guideline, static load test for load
rating is conducted in a special case to provide a reference for load-
carrying capacity evaluation. For example, the static load testing was
performed on Myokyo Bridge, a prestressed concrete box girder bridge,
in 2011 to provide estimate of remaining load-carrying capacity of the
then severely corroded bridge [9].
The load rating factor (RF) is adopted based on the load and resis-
tance rating factor method, which can be expressed as [6].
C — γ DC — γ DW
RF = DC DW
(1)
γLL (LL + IM)
In which, C stands for the current load-carrying capacity of the
2
Engineering Structures 229 (2021) 111645
bridge members; DC and DW stand for the dead load effects caused by
structural components and the wearing surface, respectively. LL and IM
stand for the live load effect and its corresponding dynamic effects
respectively; γDC, γDW and γLL denote the load factors under the different
load types. Furthermore, the current load-carrying capacity C is defined
as:
C = ϕ cϕ sϕRn (2)
where ϕc ,ϕs ,ϕ and are the condition factor, system factor, and the LRFD
resistance factor, respectively, while Rn denotes the nominal resistance
of bridge member.
For an existing bridge with available design documents, C can be
computed based on material properties, section dimensions, reinforce-
ment, and prestress configurations and other design information.
However, despite the rigorous theoretical background, difficulty in
determining the load-carrying capacity of a bridge may arise due to
uncertainties in concrete material property, cross-section area, rein-
forcement, and unknown deteriorated bridge condition. In such a case,
static load test is often conducted to obtain the bridge responses and
then compared with the computational results. The comparison is used
as a basis for updating the load rating factor to provide a more realistic
load-carrying capacity evaluation.
It should be mentioned, however, that a complete static load test is
labor-extensive and time consuming. For example, it takes almost one
day to instrument the bridge with various sensors, and another day to
conduct the test. Generally, four to six trucks are needed for a static load
test of a bridge. Traffic closure of all lanes is required, which is very
difficult to implement in reality. In addition, the static load test for a
common medium span bridge can be very costly. For example, it is re-
ported that the cost for such a test could be up to 6% of the bridge
replacement cost [10]. These are some reasons that make static load test
for realistic load capacity evaluation of existing bridges difficult to
implement widely.
3. Methodologies for load-carrying capacity evaluation
In this paper, an efficient method for load-carrying capacity evalu-
ation of girder bridges is proposed. The method based is on measure-
ment of bridge displacement under moving vehicle (PEMBEBANAN).
Fig. 1 describes a flowchart of the proposed method, which can be
explained as follows. First, a radar-based dynamic displacement
measurement device is selected, and careful preparation is conducted
for measurement loca- tion, test truck weight and speed determination.
Second, the truck pass- by test is performed to obtain the dynamic
displacement. Third, the dynamic displacement data is processed to
calculate the fundamental mode frequency of the bridge. Fourth, a low-
pass filter-based technique is employed to obtain the influence line for
displacement response of the bridge. Fifth, a simulated static load test is
conducted on the bridge, andthe displacement results are calculated.
There are three main procedures involved in the proposed method,
namely, the radar-based dynamic displacement measurement under a
truck pass-by test, the low-pass filter-based influence line extraction,
and the calculation with simulated static load test scenario. They are
explained with more details in the following subsections.
3.1. The Radar-based dynamic displacement measurement
In recent years, dynamic displacement is seen as a more attractive
measurand because it provides information on the deflection as well as
modal properties of the measured bridge. In this study, a commercial
interferometric radar device is used for non-contact displacement
measurement in the truck pass-by test. The radar system can measure
displacement of multiple targets at higher sampling frequency without
interrupting the traffic. This makes the device more flexible and
convenient to use than the traditional LVDT sensor or dial gauges. The
Z. Sun et al.
Fig. 1. Flowchart for the proposed load-carrying capacity evaluation method. Note that the procedures in shaded box are proposed in this study.
device employs two principles, namely, the Stepped Frequency
Continuous Wave (SF-CW) and the interferometry. The former is to
distinguish multiple targets, and the latter is to allow measurement of
high frequency. A more detailed introduction on the techniques and
application procedures was presented by Su et al. [11]. The device has
been applied for the field truck pass-by bridge tests and proved to be
efficient and reliable [12].
Before the truck pass-by test, a detail testing program should be
developed, which includes testing schedule and scenarios, such as truck
speeds and running lanes. For a girder bridge, the test truck runs over all
lanes corresponding to the regulated positions in static load test, and
displacement is measured at the mid-span of the relevant girders.
Preparation for the test includes bridge field survey, preparation of the
test truck and weight measuring. Next, radar device is deployed under
the bridge and reflectors are installed at the mid-span. The relative lo-
cations of antenna and reflector targets are measured to calculate di-
rection of the radar wave propagation. Afterwards, the test is started. To
minimize traffic interruption, the test is generally conducted during at
the nighttime when traffic volume is low. During the test, public vehicles
and pedestrians are not allowed to enter the testing area. The mid-span
displacement is measured with radar device at high sampling frequency.
3
Engineering Structures 229 (2021) 111645
3.2. Low-pass filter-based influence line extraction
Since Moses [13] introduced the idea of WIM to obtain the axles and
gross weights of vehicles in motion with sensors instrumented on a
bridge, the influence line identification has been investigated widely.
Most of the existing methods utilize bridge response under a moving
vehicle instead of the time-consuming step-by-step static test method.
Many researchers have come proposed numerous techniques to elimi-
nate the dynamic components. McNulty and O’Brien [14] adopted the
point-by-point process on the strain response graph to obtain the bridge
experimental influence line. Manual adjustment is required in this
process, however, and this can be subjective and largely depends on the
experience of the user. Wang et al. [15] proposed a fitting method for
extracting the influence line from bridge dynamic response induced by a
passing vehicle, which describes the influence line and the fluctuation
part with a piecewise polynomial and a series of harmonic sinusoids,
respectively. Sun et al. [16] derived two damage indexes from identified
strain influence line from structural health monitoring data, which can
indicate the damage existence and damage location, respectively. The
methodology is verified with measurements of a cable-stayed bridge. An
in-depth review and comparison of the existing influence line identifi-
cation methods was summarized by Zheng et al. [17].
Z. Sun et al. Engineering Structures 229 (2021) 111645

Fig. 2. A simply supported beam under a moving load at a constant speed.

Vehicle-induced bridge vertical displacement response y(x,t) is
schematically illustrated in Fig. 2, The vehicle-induced load is repre-
sented by a constant force P that moves at a constant speed v from left to
right. At the instant of force arrival, the beam is at rest without initial
defection or velocity. The beam is assumed to follow the Euler-Bernoulli
beam assumption with small deflections, and damping is assigned pro-
portional to the vibration velocity [18,19]. The governing equation of
motion of a simply supported beam with uniform cross-section can be
expressed as
4 2
EI ∂ y(x, t) + m ∂ y(x, t) + 2mω ∂y(x, t) = Pδ(x — vt)
∂x4 ∂t2 b
∂t
In Eq. (3), x denotes the coordinate in the longitudinal direction on
the beam; t is the time of the load moving;y(x, t)is the vertical
displacement of the beam at time t; L is the total length of the beam; P is
the amplitude of the load; m is the mass density for a unit length; EI is the
flexural rigidity of beam; v is the velocity of the moving load; and ωb is
the circular frequency of the beam. δ(x —vt) denotes the Dirac’s delta
function, which shows that the load P is applied only at the contact
location vt on the beam. The equation of motion is subjected to boundary
In this paper, a low-pass filter is employed to obtain the quasi-static
part as the influence line. Initially, the fast Fourier transform (FFT) was
applied to the dynamic displacement data measured by radar device,
and the fundamental frequency of the bridge is identified. It should be
noted that although the vehicle weight may affect fundamental fre-
quency of a bridge identified from moving vehicle test as reported in
[27], such variation was not observed in this case. This is probably
because in this case, the vehicle mass is much smaller than total weight
of the bridge. This information was used to determine the cut-off fre-
quency for the low-pass filter. The original signal was filtered with the
(3)
cut-off frequency in the frequency domain; and an inverse Fourier
transform is performed to obtain the filtered displacement.
3.3. Displacement calculation using simulated static load test scenario
After obtaining the displacement influence lines, the simulated static
load test is performed according to the designated truck weights and
locations in static load test scenarios. The displacement yt is estimated as
follows.
∑n

conditions with regard to displacement and rotation: yt = (Fi × di) (7)

i=1
y(0, t) = 0; y(l, t) = 0; (4.a)
In which, n denotes the total number of loads in the simulated static
⃒ ⃒ load test. Fi denotes the amplitude of the i-th vehicle load. d i denotes the
∂ 2 y(x, t)⃒ ∂ 2 y(x, t)⃒ =0 (4.b)
⃒⃒ = 0; ⃒
∂x 2 ∂x 2 ⃒ value on the influence line corresponding to the position of the i-th
x=0 x=l
vehicle load. The bridge displacement response yc is calculated under
The closed-form solution for Eq. (3) can be obtained for initial designated truck weights and locations in the finite element model based
conditions y
on the design plan. Afterwards, the factor yct is utilized to update the
∂y(x, t)⃒⃒ calculated load rating factor RF for the bridge capacity evaluation.
y(x, 0) = 0; and ⃒ =0 (5)
∂t ⃒t=0 The dead and live load effects under the design load and the load-
carrying capacity at the concerned cross-section is calculated, and the
as follows: [18,19]. load rating factor can be obtained following Eq. (1). In this equation, the
( ) RF value larger than one indicates that the bridge design load-carrying
iπx ( )
2PL3 ∑
∞
L
sin iπvt capacity is satisfied. Whereas the value smaller than one indicates that
y(x, t) = — 2{
mπ2 i=1
i (iπa) — (vL) 2 } sin L
2 the bridge design load-carrying capacity is less than required, and that
( ) (6) strategies for repair, strengthening or truck weight limitation counter-
iπx
+ 2PL3 v
4 ∑
∞
{ sin L
measures are advisable.
mπ a 2 } sin(ωit)
3 2
i=1 i (iπ a) — (vL) 4. Numerical simulations by finite element analysis
√̅̅̅ 2 2 √̅̅̅ EI
In which, a = EI and ωi = i π is the circular frequency at the i-th In order to verify reliability of the proposed load-carrying capacity
m L2 m

mode of the beam vibration.
It can be observed in Eq. (6) that the bridge response under moving
vehicle includes two parts, the quasi-static component, and the dynamic
component [19]. They are related to the vehicle speed and the bridge
natural frequency, respectively. The vehicle induced dynamic
displacement has been investigated by numerous researchers, focusing
on extraction of bridge fundamental frequency from estimated vehicle
response [20–24], detection of local stiffness loss [12,25], prediction of
bridge dynamic response [26], among others.
evaluation method, numerical investigation is conducted on a three-
dimensional girder bridge model. Static load scenario is firstly con-
ducted, which serves as a reference of comparison. Truck pass-by load
scenarios with different truck weight and speeds are then investigated,
and the results provide a guidance for selecting the truck weight and
speed in the real bridge field tests.

4
Z. Sun et al. Engineering Structures 229 (2021) 111645

Fig.3. (a) Finite element model of the bridge built in ABAQUS software, (b) Static load test scenario, (c) Location of artificial damage on girder G4 at between 7 m
and 9 m positions by reduction of Young’s modulus corresponding to 20% of stiffness loss.

4.1. Description of finite element model of the bridge

Table 2
Displacement results at mid-span of five T-girders.
A finite element bridge model was developed for simulation using
finite element software ABAQUS [28] as shown by Fig. 3(a). Details of Load scenario T1 T2 T3 T4 T5

the bridge model is described below. The deck is made of reinforced Static (mm) 2.31 2.98 2.67 3.12 2.32
concrete with following dimensions: length 25 m, width 8.5 m and
thickness 0.2 m, accommodating two traffic lanes. The concrete cover is
4.3.1. Truck model
0.035 m thick. Main girder consists of five steel I-girder beams in the
A quarter car model of 20-ton vehicle [22] is used in the analysis, as
longitudinal direction with dimensions 1.2 m high and 0.6 m wide.
shown in Fig. 4 (a). The stiffness coefficient and damping coefficient are
Thicknesses of the web and flange members are both 0.03 m. Artificial
damage of is created on girder G4 between 7 m and 9 m positions with 3.8 × 106N/m and 3.4 × 104 N⋅s/m, respectively. The equation of mo-
reduction of Young’s modulus corresponding to 20% of stiffness loss as tion for the quarter car vehicle model traversing a bridge can be
shown in Fig. 3(c). expressed as:
The transverse beam is an x-shaped beam. The top and bottom Mv Ü v + Cv U̇ v + Kv Uv = fv (8)
chords are diagonals with dimensions 100 × 100 × 9.5 and 75 × 75 ×
9.5, respectively. The deck and bearings, I-girder, and transverse beams, where Mv, Cv, Kv are the mass, damping coefficient, and stiffness of the
rebars are modeled by solid element, shell element and truss element, vehicle, respectively. Üv , U̇v , Uv are the acceleration, velocity, and
respectively. There are 11,067 elements and 15,897 nodes in total. The displacement of the vehicle, respectively. The symbol fv denotes the
boundary conditions are specified as simply supported at two ends of the vertical force on the vehicle due to interaction with the bridge at the
five I-shape girders on bearings. Details on the material properties used wheel position.
in the finite element model are provided in Table 1. It is assumed that the vehicle moves at a constant speed on a straight
path. The interaction force with the bridge is expressed as:
4.2. Static load scenario f = fv + Mv g — Cv U̇ v — Kv Uv (9)

In the static load scenario, four 20-ton trucks are placed on the bridge The road surface roughness that affects dynamic response of the
in a stationary position as shown in Fig. 3(b). In the longitudinal di- vehicle-bridge interaction system is represented by random field using a
rection, the load positions are arranged to ensure the largest mid-span stationary Gaussian random process. The road surface roughness is
deflection. In the transverse direction, the positions correspond to the described by the following power spectral density functions [26]:
two traffic lanes. To avoid local stress concentration, the four vehicle ∑N √̅̅̅
weights are applied as the surface forces. Static analysis is performed, r(x) = k=1
2Gd (nk)Δncos(2πnkx — θk) (10)
and the deflections at the mid-span of the five T girders are obtained as In which, Gd(nk) = a
(2πnk )
2 is the power spectral density function. a is
shown in Table 2. The values are 2.31 mm, 2.98 mm, 2.67 mm, 3.12 mm,
selected as 2 × 10 indicating a good road surface condition. Δn is the
—6
and 2.32 mm, respectively. It is noted that the displacement of the
damaged girder T4 is larger than that at the symmetrical location of frequency interval of the road surface wave expressed as Δn = nmax —nmin ,
in which n N

girder T2. max and n min are 1 Hz and 0.05 Hz. Nis the wave number
selected as 20 here andnk is the frequency of wave number k; while θk is
the random phase angle distributed between 0 and 2π. The selected road
4.3. Truck Pass-by load scenarios roughness profile used in the numerical simulation is illustrated in Fig. 4
(b).
In the simulated truck pass-by load scenario, parametric study In the vehicle-bridge interaction analysis, displacement compati-
related to the truck weight and speed is conducted. The results are bility at the wheel contact points should be satisfied. This means that
described in the following sections. vehicle and the bridge move together in the vertical direction, without
zero contact force. A global iteration procedure is used in the compu-
Table 1 tation [26]. The interaction force time history is calculated in MATLAB
Material properties of bridge model. with Newmark’s method and then applied as loads on the ABAQUS
Material Property Value bridge model [28].
Concrete Density (kg/m3) 2400
Young’s modulus (Pa) 2.8 × 1010 4.3.2. Load scenario description
Poisson’s ratio 0.2 The parametric study is performed with different truck weights and
Steel Density (kg/m3) 7850 speeds with the following scenarios. 1) While keeping truck speed
Young’s modulus (Pa) 2.1 × 1011
constant at 5 m/s, the truck weight is varied to 10ton, 20ton, 30ton,
Poisson’s ratio 0.3

5
Z. Sun et al. Engineering Structures 229 (2021) 111645

Fig. 4. (a) Quarter-car vehicle model and (b) Road roughness thickness used in finite element numerical simulation.

40ton, and 50 ton. 2) While keeping truck weight at 20 ton, the truck
speed is varied to 2.5 m/s, 5 m/s, 12.5 m/s, and 25 m/s. In each
simulated scenario, the truck runs over the two traffic lanes that
correspond to the truck transverse positions in the static load scenario.
An example of loading case for vehicle runs over traffic lane above girder
T4 is illustrated in Fig. 5.

4.4. Results of numerical simulation

In the dynamic analysis, displacement time histories at the mid-span

Fig. 5. Truck pass-by load scenario.
of the five girders are obtained after convergence is achieved in
computation. A frequency analysis is performed with Fast Fourier
Transform method, and the fundamental natural frequency is obtained
as 8.63 Hz. After sensitivity analysis on cut-off frequency, a low-pass
filter of 1 Hz is employed to extract the quasi-static component in the
displacement time series.
Displacement results measured at mid-span of girder T2 under the
scenario of 20ton truck running at 5 m/s are presented here as an
example. Both the original and filtered results are plotted in Fig. 6. The

Fig.6. Displacement time-histories at mid-span of girder T2 for the case of: (a) original response to truck pass-by over girder T2, (b) filtered response to truck pass-by
over girder T2, (c) original response to truck pass-by over girder T4, (d) filtered response to truck pass-by over girder T4.

6
Z. Sun et al.
Fig. 7. Comparison of displacements obtained from static and dynamic load
scenarios for various truck weights.
vehicle locations in the static load test scenario in Fig. 3 (b) are 8.5 m
and 16.5 m respectively, which correspond to the instantaneous time
1.7 s and 3.3 s in Fig. 6 (a), (b), (c) and (d). The comparison of dis-
placements obtained from the static and moving load with varied truck
weights at constant speed 5 m/s is illustrated in Fig. 7. In the figure, the
displacement result at mid-span of all five T-girders are obtained based
on Eq. (1). The figure demonstrates that the difference between 10-ton
truck scenario and the static load result ranges from 8.1% to 10.5%.
In contrast, the differences for all other four truck weights are between
0.32% and 4.5%. Therefore, selecting the same truck weight as in static
load case would provide better accuracy in the proposed method.
Fig. 8 exhibits comparison of displacements subjected to a 20-ton
truck loading running at various speeds of 2.5 m/s, 5 m/s, 12.5 m/s
and 25 m/s, and the corresponding displacement obtained by static
loading. It is evident from the figure that the differences with the static
result are much smaller compared to that in the case of varied truck
weights. The difference is more than 2% for the highest speed scenario of
25 m/s, which is still acceptable. It can be concluded that the proposed
method is very robust with regard to the truck running speed.
Fig. 8. Comparison of displacements obtained from static and dynamic load
scenarios for various truck speeds.
7
Engineering Structures 229 (2021) 111645
5. Full-scale implementation of the proposed method
In order to verify the efficiency and efficacy of the proposed method
in practice, a full-scale implementation was conducted. Both static load
test and truck pass-by test were carried out on a simply supported pre-
stressed concrete (PC) T-girder bridge.
5.1. Description of tested bridge
Full-scale implementation of the proposed method was conducted on
Wenxi Bridge. The bridge is a PC T-girder type in the suburb area of
Fuzhou City, China (Fig. 9(a)). It was completed in 2007 and consists of
nine 30 m simply supported spans. The bridge has a total width of 12.5
m with two traffic lanes in the middle and two 1.75 m pedestrian
walkways. The 15.7 cm thick concrete deck is supported on five PC T-
girders. Details on the prestressed concrete (PC) T-girder bridge is shown
in Fig. 9 (c), and dimension of the deck cross section is described in Fig. 9
(d).
5.2. Description of static load testing
During the static load test, the traffic was closed in both directions.
Four loading truck locations are shown in Fig. 10 (a) and photo of the
loading test is shown in Fig. 10 (b). It is noted that the static test took
about 3 h with total traffic closure, and 12 persons participated in the
test. The traffic closure was implemented under low traffic volume. The
load test was administered by the local traffic administration and
announced to local people in advance. Each truck weighs around 30 ton
as presented in Table 3. The distance between first (front) axle with the
second (rear) axle is 3.5 m. Whereas the longitudinal and lateral dis-
tances between the two rear wheels is 1.4 m and 1.8 m, respectively
(Fig. 10(a)). The ratio between the test load effect and the design load
effect is 1.01. Digital level Leica DNA-03 is used to measure the
displacement, which has a precision of 0.1 mm. The displacement
measurement results at mid-span of the five T girders are summarized in
Table 4.
5.3. Description of the truck pass-by testing
In the dynamic test, the truck ran over the two lanes respectively, and
the running routes are selected according to the truck position in Fig. 10
(a). The truck speeds were selected as 10 km/h, 20 km/h and 30 km/h
respectively. The reflectors were mounted at mid-span of girder T1 and
T2 respectively, and a schematic view of the measurement with radar
device is illustrated in Fig. 11(a). The distance between the radar device
and reflector on T1 is 14.20 m with an elevation angle of 24◦. Fig. 11(b)
shows the reflector used for the radar device. The truck pass-by test was
conducted for six times with 12 min of traffic closure involving four
persons and one test truck. The testing time was reduced significantly
from a typical duration of 3 h in the conventional static test to only 12
min. Moreover, the required number of persons involved is reduced from
12 to 4 persons. This is certainly more efficient and effective than the
standard static load test.
The bridge displacements were measured, and the fundamental
natural frequency was obtained at 4.41 Hz. A low-pass filter was used to
eliminate the dynamic components in the signal, and the cut-off fre-
quency was selected as 0.5 Hz after sensitivity analysis. Fig. 12(a) and
(b) show the original and filtered displacement time histories recorded
at girder T2 for the case of truck ran over lane 1, respectively. Whereas
the original and filtered displacement time histories recorded at girder
T2 for the case of truck ran over lane 2 are illustrated in Fig. 12(c) and
(d), respectively. Similarly, the filtered displacement signals at girder T1
and T2 under all three different speeds were obtained. As shown by
Fig. 12, the maximum dynamic displacements at the center of the girder
T1 and T2 are 1.83 mm and 3.36 mm, respectively. Whereas the
maximum filtered displacements are 1.55 mm and 2.45 mm,
Z. Sun et al.
Fig. 9. Description of object bridge. (a) Photo of Wenxi Bridge, (b) Finite element model of the bridge built in Midas Civil software, (c) T-girder beam cross section,
(f) Bridge deck cross section (unit:mm).
respectively.
The measurements results were compared with finite element model
of the bridge built in Midas/ Civil Software [29] (Fig. 9 (b)). Comparison
of the testing results with the finite element model under the same
loading condition is also shown in Fig. 12. The maximum corresponding
displacements under the test truck load at the center of the girder T1 and
T2 are 1.90 mm and 3.27 mm, respectively. The displacements obtained
from the load test are smaller than the results from finite element
analysis, which is believed due to the fact that the bridge stiffness is
larger than the designed stiffness, as is the usual case of typical bridges.
Nevertheless, the results demonstrate that bridge displacements under
loading condition obtained experimentally agree reasonably well with
the estimated displacements in the finite element model.
5.4. Analysis of measurement results
Based on the static loading truck configuration shown in Fig. 10 (a),
the truck positions were obtained, and the corresponding values on the
displacement time history were extracted. According to Eq. (1), the mid-
span displacement on girder T1 and T2 were obtained for three different
truck speeds as presented in Table 5. The table demonstrates that the
results of analysis using truck pass-by test are all smaller than that in the
static load test. The largest difference is 10.3% and 5.3% for girder T1
and T2, respectively. A possible explanation of this is that the method is
based on linear extrapolation of displacement due one pass-by truck
while the actual measurement results may include some nonlinearity
effect. It should also be mentioned that the increase in displacement may
not linearly correlate with the load amplitude increase. In a destructive
test of a similar PC girder bridge, the same nonlinearity phenomenon
was observed when the load was applied from zero to the maximum
loading [30]. This loading procedure induced nonlinearity due to
opening of cracks in the damaged section of the girder and eventually
C — (γDC )(DC) — (γDW )(DW) 8456 — 1.25 × 2323 — 1.5 × 1022
RF 1 = = = 0.98
(γL )(LL + IM) 1.75 × (1959 + 1959 × 0.2)
8
Engineering Structures 229 (2021) 111645
reduced the stiffness. The similar phenomenon of stiffness loss caused by
cracks opening was also pointed out in another investigation [31].
In principle, the proposed method utilizes a specific load level to
predict behavior of the bridge under heavier loads, which is verified to
be more efficient in time and less labor-intensive. In the application, it is
important to consider nonlinearity effects with regard to the extrapo-
lation. It is recommended that nonlinearity amplification factor of 1.1 is
adopted for PC T-girder bridges. Hence the averaged mid-span
displacement under three speeds become 6.66 mm and 7.55 mm for
girder T1 and T2, respectively. These results agree well with the previ-
ously obtained deflections in Table 4.
5.5. Estimation of load rating factor from measurement
As mentioned previously, a finite element model of Wenxi Bridge was
built in Midas/Civil software [28] as a reference to compare the mea-
surement results with theoretical values. The model was built using as-
built drawing of the bridge. Using the finite element model, the mid-
span displacements under the static loading scenario were calculated
as 8.80 mm and 9.00 mm for girder T1 and T2, respectively. The live load
transverse distribution factors for girder T1 and T2 were 0.656 and 0.523,
respectively. The load effects and load-carrying capacity were calculated
following Eq.(1) as illustrated in Table 6, where the design live load
scenario was applied according to loading pattern shown in Fig. 13.
According to the carrying capacity C defined by Eq.(2), the condition
factor (ϕc), system factor (ϕs ), and the LRFD resistance factor (ϕ) for this
bridge are 0.95, 1.0 and 0.9, respectively. For the prestressed concrete
bridge, γDC = 1.25, γDW = 1.5, γL = 1.75. Moreover, according to the
bridge deck surface condition, IM is taken as 0.2. The load rating factor
RF for girder T1 and T2 can be calculated as follows.
Z. Sun et al. Engineering Structures 229 (2021) 111645

Fig. 10. Static load test (a) Loading truck configuration (Unit: m) (b) Photo of the test.

Table 3 Table 4
Truck axle weights. Displacement results in the static load test.
Truck No. Axle weight (ton) T girder no. Displacement (mm)

Total weight Front axle Two rear axles T1 6.66

T2 7.18
1 30.69 6.14 24.55
T3 7.17
2 30.17 6.03 24.14
T4 6.80
3 29.62 5.92 23.70
T5 6.64
4 29.55 5.91 23.64

C — (γ DC )(DC) — (γ DW )(DW) 8009 — 1.25 × 2212 — 1.5 × 824

RF 2 = = = 1.22
(γL )(LL + IM) 1.75 × (1562 + 1562 × 0.2)

9
Z. Sun et al. Engineering Structures 229 (2021) 111645

Fig 11. Implementation of radar measurement system, and truck loading. (a) Schematic view of truck pass by test (b) Reflector mounted on the T girder.

Fig. 12. Displacement results at T2 mid-span under truck speed of 10 km/h and their comparison with FE model results for: (a) original displacement when truck
moves over lane 1, (b) filtered displacement when truck moves over lane 1, (c) original displacement when truck moves over lane 2, (d) filtered displacement when
truck moves over lane 2.

Table 5
Obtained displacements at mid-span of two girders. Table 6
Speed Girder T1 Girder T2 The calculated parameters for load carrying capacity evaluation (Unit: kN•m).
(km/h)
Displacement Difference Displacement Difference Parameter Girder T1 Girder T2
(mm) with static (mm) with static DC 2323 2212
result result DW 1022 824
10 5.97 10.3% 6.80 5.3% LL 1959 1562
20 6.09 8.6% 6.97 2.8% C 8456 8009
30 6.11 8.3% 6.81 5.0%

10
Z. Sun et al.
Fig. 13. Design load for Wenxi Bridge.
Finally, the load rating factor is modified with an adjustment factor
based on the load test result as:
RFT = κRF (11)
where κ is the adjustment factor obtained from comparison between the
test and analytical model and its value is taken as 1.15 according to the
AASHTO manual [6]. Therefore, the modified load rating factors for the
two girders T1 and T2 are 1.13 and 1.40, respectively. Generally, the
load rating is represented by the rating factor RF as illustrated in Eq. (1).
It provides an estimation of the bridge carrying capacity with regard to
the design load. In this example, results of analyses suggest that the two
girders T1 and T2 can respectively sustain 1.13 and 1.40 times the
design load shown in Fig. 13.
The results demonstrate that the proposed method can efficiently
provide an estimate of load carrying capacity using the truck pass-by
test. Note that while the mid-span deflection of the bridge in this case
study is only around 3 mm, the proposed method can be implemented
effectively since the radar device used here has the measurement ac-
curacy in an order of 0.01 mm [11]. This is one of the advantages of
using radar device compared to traditional displacement measurement
device such as LVDT sensor or digital level have a precision of 0.1 mm.
Therefore, under the above level of accuracy of radar device, the pro-
posed method can serve the needs in engineering practices.
Based on the load carrying capacity obtained with the proposed
method, bridge owner and roadway operator can make better decisions
for maintenance and management. For example, traffic regulation can
be imposed the original legal gross vehicle weight of 40 ton could be
regulated to 36 ton if the actual load rating factor is 0.9. The results
show that accuracy of the truck pass-by test is reasonably good
compared to the conventional static load test with less testing time and
fewer labor forces. This could be a promising application for typical
medium-scale bridges on the rural area.
6. Conclusions
Evaluation of load-carrying capacity of existing bridges is essential
for rational and economical maintenance of severely damaged bridges.
Conventional static load testing is often considered too costly and labor-
intensive thus preventing its wide application. The conventional static
load test of existing highway bridges is very difficult for highway
operator because it requires relatively long traffic closure time. In rural
bridges, they are also rarely conducted due to lack of budget.
In this paper, an efficient method for evaluation of load-carrying
capacity of existing bridge using moving vehicle is proposed. The
method comprises three main procedures, namely, radar-based dynamic
displacement measurement under truck pass-by test, influence line
extraction, and estimation of the static load. Based on these procedures,
evaluation of load-carrying capacity is conduced following the AASHTO
manual. The method was verified in numerical simulations using three-
dimensional finite element model of typical girder bridge under vehicle
loading. Effects of vehicle weight, vehicle speed, and road roughness on
the accuracy of displacement estimation was investigated in the nu-
merical simulations.
11
Engineering Structures 229 (2021) 111645
The proposed method was implemented and verified in a full-scale test
of a PC girder bridge using both static load test and truck pass-by test. The
results show that displacement estimated by the truck pass-by test is
reasonably accurate compared to the conventional static load test with
the largest difference within 10%. For better estimation, an amplification
factor is suggested to consider the nonlinearity effects under heavier loads
on damaged girder bridges. The study has demonstrated that the proposed
method can serve as an efficient substitute to traditional static load test,
especially for expressway or rural bridges. Based on the load-carrying
capacity rating factor, bridge owner and roadway operator can make
better decisions for maintenance and management. Traffic regulation can
be imposed by limiting the heavy trucks when the estimated load-carrying
capacity has reduced significantly. Efficacy of the proposed method has
been shown by reducing the required number of persons involved in the
test and the testing time. In the future more tests are necessary to quantify
the nonlinear amplification factor for other bridge types such as slab, box,
and truss girder bridges.
It should be mentioned that although the current study is conducted
for evaluation of load-carrying capacity following the AASHTO manual,
the similar approach can be adopted to other codes and guideline pro-
vided that they are based on capacity evaluation and that the load tests
of existing bridge are required. Such requirements demand an efficient
and effective method for estimation of load-carrying capacity of existing
bridge by a simple and less expensive load test procedure. Therefore, it is
anticipated that the similar study on the load rating factor will be
implemented on other codes for assessment of the load carrying capacity
of existing bridges.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
References
[1] Brownjohn JMW, Xia PQ, Hao H, Xia Y. Civil structure condition assessment by FE
model updating: methodology and case studies. Finite Elem Anal Des 2001;37(10):
761–75.
[2] Ding L, Hao H, Xia Y, Deeks AJ. Evaluation of Bridge Load-carrying capacity Using
Updated Finite Element Model and Nonlinear Analysis. Adv Struct Eng 2012;15
(10):1739–50.
[3] Casas JR, Gómez JD. Load rating of highway bridges by proof-loading. KSCE J Civil
Eng 2013;17:556–67.
[4] Olaszeka P, Lagoda M, Casas JR. Diagnostic load testing and assessment of existing
bridges: examples of application. Struct Infrastruct Eng 2014;10(6):834–42.
[5] Lantsoght EOL, van der Veen C, de Boer A, Hordijk DA. State-of-the-art on load
testing of concrete bridges. Eng Struct 2017;150:231–41.
[6] AASHTO. Manual for bridge evaluation. American Association of State Highway
and Transportation Officials. 2nd ed. Farmington Hills, MI; 2011.
[7] The Institution of Civil Engineers-National Steering Committee for the Load Testing
of Bridges. Guidelines for the Supplementary Load Testing of Bridges. London, UK;
1998.
[8] Ministry of Transport, China. Specifications for Inspection and Evaluation of Load-
bearing Capacity of Highway Bridges, JTG /T-J21–2011; 2011.
[9] Nishio M, Abe M. Operation of a damaged prestressed-concrete girder bridge with
repair in Japan. P I Civil Eng Foren Eng 2019;172(1):19–26.
[10] Faber MH, Val DV, Stewart MG. Proof load testing for bridge assessment and
upgrading. Eng Struct 2000;22:1677–89.
[11] Su D, Nagayama T, Sun Z, Fujino Y. An interferometric radar for displacement
measurement and its application in civil engineering structures. Proceedings of
Sensors and Smart Structures Technologies for Civil, Mechanical and Aerospace
Systems 2012; San Diego, CA, United states.
[12] Sun Z, Nagayama T, Nishio M, Fujino Y. Investigation on a curvature -based
damage detection method using displacement under moving vehicle. Struct
Control Hlth 2018;25(1):e2044.
[13] Moses F. Weigh-in-motion system using instrumented bridge. J Transp Eng-ASCE
1979;105(3):33–249.
[14] McNulty P, O’Brien EJ. Testing of bridge weigh-in-motion system in sub-arctic
climate. J Test Eval 2003;31(6):1–10.
[15] Wang NB, He LX, Ren WX, Huang TL. Extraction of influence line through a fitting
method from bridge dynamic response induced by a passing vehicle. Eng Struct.
2017;151:648–64.
 Z. Sun et al. Engineering Structures 229 (2021) 111645

[16] Sun SW, Sun LM, Chen L. Damage detection based on structural response induced by traffic load: methodology and application. Int J Struct Stab Dy 2015;16(4): 1640026.
[17] Zheng X, Yang DH, Yi TH, Li HN. Development of bridge influence line identification methods based on direct measurement data: a comprehensive review and comparison. Eng
Struct 2019;198:109539.
[18] Frýba L. Vibration of solids and structures under moving loads, Vol. 1. Springer Science & Business Media; 2013.
[19] Weaver JR, Timoshenko SP, Young DH. Vibration problem in engineering. New York: John Wiley and Sons; 1990.
[20] Yang YB, Lin CW, Yau JD. Extracting bridge frequencies from the dynamic response of a passing vehicle. J. Sound & Vibr. 2004;272(3–5):471–93.
[21] Lin CW, Yang YB. Use of a passing vehicle to scan the bridge frequencies - an experimental verification. Eng. Struct. 2005;27(13):1865–78.
[22] Sun Z, Nagayama T, Su D, Fujino Y. A damage detection algorithm utilizing dynamic displacement of bridge under moving vehicle. Shock Vib 2016;8454567.
https://doi.org/10.1155/2016/8454567.
[23] Wang H, Nagayama T, Nakasuka J, Zhao B, Su D. Extraction of bridge fundamental frequency from estimated vehicle excitation through a particle filter approach.
J Sound Vib 2018;428:44–58.
[24] Siringoringo DM, Fujino Y. Estimating Bridge Fundamental Frequency from Vibration Response of Instrumented Passing Vehicle: Analytical and Experimental Study. Adv Struct
Eng 2012;15(3):417–33.
[25] O’ Brien EJ, Fitzgerald PC, Malekjafarian A, Sevillano E. Bridge damage detection using vehicle axle-force information. Eng Struct 2017, 153(15): 71–80.
[26] Sun Z, Zou ZL. Towards an efficient method of predicting vehicle-induced response of bridge. Eng Computation 2016;33(7):2067–89.
[27] Yang YB, Cheng MC, Chang KC. Frequency variation in vehicle-bridge interaction systems. Int. J. of Struct. Stability & Dyn. 2013;13(2):1350019.
[28] Abaqus 6.10 [Computer software]. Dassault Systemes Simulia Corp.,Providence, RI.
[29] MIDAS.IT Corporation 2006.
[30] Budette EG, Goodpasture DW. Test to failure of a prestressed concrete bridge. PCI J 1974;19(3):92–103.
[31] Saiidi M, Douglas B, Feng S. Prestress force effect on vibration frequency of concrete bridges. J Struct Eng-ASCE 1994;120(7):2233–41.

KEKURANGAN :
Dalam abstrak jurnal tersebut, terdapat permasalahan yang dihadapi dalam evaluasi kapasitas pemikul beban
jembatan. Permasalahan tersebut antara lain adalah metode pengujian beban statis konvensional yang dianggap
mahal dan membutuhkan banyak tenaga kerja, sehingga sulit diterapkan secara luas. Terutama pada jembatan di
daerah pedesaan, keterbatasan anggaran seringkali menjadi hambatan untuk melakukan pengujian semacam itu.
Akibatnya, jembatan-jembatan di jalan-jalan pedesaan seringkali tidak memiliki penilaian kapasitas pemikul beban
yang dapat diandalkan, meskipun kondisinya terlihat buruk.

KELEBIHAN :
Dalam jurnal ini, diusulkan suatu metode yang efisien untuk evaluasi kapasitas dukung beban jembatan gelagar
berdasarkan perpindahan kendaraan yang bergerak.

PEMBEBANAN :
a. The Radar-based dynamic displacement measurement
Pembebanan dilakukan dengan menggunakan pengukuran perpindahan dinamis menggunakan perangkat
radar berbasis interferometri. Perangkat radar ini digunakan untuk mengukur perpindahan non-kontak pada
saat truk melintas di atas jembatan. Pengukuran perpindahan ini dilakukan dengan menggunakan prinsip
Stepped Frequency Continuous Wave (SF-CW) dan interferometri. Perangkat radar ini memiliki
keunggulan dalam mengukur perpindahan dengan akurasi yang tinggi, dengan ketelitian hingga 0,01 mm
b. Low-pass filter-based influence line extraction
pembebanan dilakukan dengan menggunakan metode low-pass filter untuk mengekstraksi garis pengaruh
(influence line) yang berkaitan dengan respons perpindahan quasi-statis jembatan akibat kendaraan yang
melintas. Metode ini digunakan untuk memisahkan komponen quasi-statis dari respons dinamis jembatan.
c. Displacement calculation using simulated static load test scenario
Pembebanan dilakukan melalui simulasi skenario pengujian beban statis. Dalam pengujian ini, posisi truk
pada pengujian beban statis ditentukan, dan nilai-nilai perpindahan pada waktu tertentu diambil. Hasil
analisis menggunakan metode uji lewat truk ini menunjukkan bahwa perkiraan perpindahan lebih kecil
dibandingkan dengan hasil pengujian beban statis. Perbedaan terbesar adalah 10,3% untuk girder T1 dan
5,3% untuk girder T2. Perbedaan ini mungkin disebabkan oleh metode ini hanya berdasarkan pada
ekstrapolasi linier perpindahan dari satu truk yang melintas, sementara hasil pengukuran aktual mungkin
memasukkan efek nonlinier. Selain itu, peningkatan perpindahan mungkin tidak berkorelasi secara linear
dengan peningkatan amplitudo beban. Meskipun begitu, metode ini tetap memberikan hasil yang dapat
diterima dan dapat digunakan untuk mengevaluasi kapasitas pemikul beban jembatan

12
 Z. Sun et al. Engineering Structures 229 (2021) 111645

STRUKTUR :
struktur jembatan yang dievaluasi adalah jembatan girder. Jembatan ini memiliki dek yang terbuat dari beton
bertulang dengan dimensi panjang 25 m, lebar 8,5 m, dan ketebalan 0,2 m. Jembatan ini memiliki lima balok I-
girder baja sebagai girder utama dengan dimensi 1,2 m tinggi dan 0,6 m lebar. Ketebalan anggota web dan flange
adalah 0,03 m. Pada girder G4, dibuat kerusakan buatan dengan mengurangi modulus Young sebesar 20% dari
kekakuan aslinya.

PERENCANAAN STRUKTUR :
Untuk estimasi yang lebih baik, faktor amplifikasi diusulkan untuk mempertimbangkan efek ketidaklinieran pada
beban berat pada jembatan girder yang rusak. Studi ini telah menunjukkan bahwa metode yang diusulkan dapat
menjadi pengganti yang efisien untuk uji beban statis tradisional, terutama untuk jembatan jalan tol atau jembatan
di daerah pedesaan. Di masa depan, lebih banyak pengujian diperlukan untuk mengukur faktor amplifikasi yang
tidak linear untuk jenis jembatan lain seperti jembatan plat, kotak, dan girder truss.

MANFAAT :
• Metode ini efisien dan ekonomis karena menggunakan pengukuran perpindahan dinamis dengan perangkat
radar yang cepat, menggantikan pengujian beban statis yang mahal dan memakan waktu.
• Metode ini memungkinkan pengambilan keputusan yang lebih baik dalam pemeliharaan dan manajemen
jembatan, dengan memberikan evaluasi kapasitas pemikul beban yang akurat untuk merencanakan perbaikan
atau regulasi lalu lintas yang tepat.
• Pengujian kapasitas pemikul beban jembatan menjadi lebih mudah dan cepat dilakukan, tanpa perlu menutup
jalan secara penuh, yang sulit dalam pengujian beban statis konvensional.
• Metode ini cocok untuk jembatan skala menengah di daerah pedesaan, terutama pada jembatan girder beton
prategang, dan memberikan hasil yang dapat diandalkan dalam mengevaluasi pemikul beban.

CONTOH JEMBATAN :

Jembatan girder Brawijaya Kota Kediri

13
JEMBATAN PELENGKUNG (ARCG BRIDGE)

Contents lists available at ScienceDirect

Ain Shams Engineering Journal

journal homepage: www.s cienc edirect.c om

Civil Engineering

Structural optimization of concrete arch bridges using Genetic

Algorithms
Mostafa Z. Abd Elrehim, Mohamed A. Eid, Mostafa G. Sayed ⇑
Civil Engineering Department, Minia University, Egypt

A R T I C L E I N F O ABSTRACT

Article history: Concrete bridges are used for both highways and railways roads. They are characterized by their durabil-
Received 13 April 2018 ity, rigidity, economy and beauty. Concrete bridges have many types such as simply supported girder
Revised 29 December 2018 bridges, arch bridges and rigid frame bridges. However, for very large spans, arch bridges are more eco-
Accepted 24 January 2019 nomic in addition to their beauty appearance. In this research, a geometrical structural optimization
Available online 16 February 2019
study for a deck concrete arch bridges using Genetic Algorithms technique is presented. This research
aims mainly to demonstrate a methodology to find the least cost design, in term of material volume,
Keywords: by finding the optimal profile. A Finite Element numerical model is used to represent the arch structure.
Optimization
The MATLAB programing platform is used to develop codes for Genetic Algorithms optimization tech-
Genetic Algorithms
nique and Finite Element analysis method. The resulted design from the optimization process is com-
Finite Element
Arch bridges pared to traditional design and an obvious cost reduction is obtained.
© 2019 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Ain Shams University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-
nd/4.0/).

1. Introduction
In recent years, Artificial Intelligence (AI) techniques show wide
applications to facilitate human life. These techniques (such as
Evolutionary Algorithms, Simulated Annealing, Particle Swarm
Optimization, Ant Colony Optimization, Tabu Search, Fuzzy Opti-
mization and Artificial Neural Network-based methods) have been
applied individually or coupled to solve the life problems [1,2].
Genetic Algorithms technique, which is one of AI branches, is used
for solving complex engineering optimization problems. The
Genetic Algorithms (GAs) have been started in 1975 by John Hol-
land [3]. GAs have been widely used in most engineering fields
such as: chemistry and physics, mechanics, aerodynamics,
robotics, networks, architecture and civil engineering. In civil engi-
neering, GAs have been used in almost all branches; for example:
reinforced concrete flat slab buildings [4], optimization of tunnel
profile in different ground conditions [5], determining the optimal
* Corresponding author at: El-souq St., Next to Civil Registry, Maghagha, El-Minia,
Egypt.
E-mail addresses: M.Eid@mu.edu.eg (M.A. Eid), eng.mgs.2013@gmail.com
(M.G. Sayed).
Peer review under responsibility of Ain Shams University.
Production and hosting by Elsevier
cross-section of steel beams [6] and the design of water distribu-
tion networks [7]. Few structural optimization researches investi-
gated the optimal design of arch bridges. In 2008, there was an
attempt for bending moment reduction in arch using a specific
arrangement sequence for hangers [8]. Two years later, an optimal
design of steel truss arch bridges using a Hybrid Genetic Algorithm
was presented [9]. A comparison between circular and parabolic
arch was presented in the same year and it was observed that
the parabolic arch has lower internal forces, more economic than
the circular arch and it is more efficient for open bridges [10].
Nowadays, Genetic Algorithms are being applied to widespread
application in business, scientific and engineering specializations.
Finite Element Analysis (FEA) is a numerical technique that can
be used to get the accurate solution for complex engineering prob-
lems. It started since several centuries back as an idea of replacing
the complicated problem by a simpler one. The main concept of
FEA is considering the origin solution as built up of many small
interconnected sub regions [11]. It has been used for solving many
structural analysis problems [12]. Today, the Finite Element
Method is considered one of the most important analysis tools
for engineers and scientists.
The arch bridges have a main advantage over other crossing
structures; the economy. In order to increase their economy supe-
rior, some attempts were introduced to find the optimal design for
these bridges. In 1999, the optimality criteria method was used to
develop a recursive relationship for the design variables
(mainly, the depth at the crown and at the support) considering

https://doi.org/10.1016/j.asej.2019.01.005
2090-4479/© 2019 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Ain Shams University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
508 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516

displacement, stresses and minimum depth constraints [13]. In
2014, global optimization algorithms were used to minimize the
cost of super structure particularly arches and hangers of network
arch bridge by optimizing the geometric shape, rise to span ratio,
cross section of arch, and the number, arrangement and cross sec-
tional dimensions of hangers [14]. In another attempt, presented in
2015, a research proposed a methodology for optimization the lon-
gitudinal shape of open arch bridge using Simultaneous Perturba-
tion Stochastic Approximation (SPSA) algorithm [15]. In 2016, a
new research presented a new strategy for modular steel construc-
tion: Adjustable module for variable depth steel arch bridges [16].
In the same year, some researchers developed a new methodology,
which evaluates the optimum configuration of network arch bridge
schemes (hangers) [17].
The current research deals with the concrete arch bridge opti-
mization problem by developing a new analysis tool that com-
prises both Finite Analysis and Genetic Algorithms using MATLAB
programming platform. In this methodology, the arch is modelled
numerically as a continuum structure divided into 2D Constant
Strain Triangular (CST) elements joined together at corners. The
joints coordinates are the design variables for the optimization
process which aims to minimize the arch weight. The developed
Finite Element Analysis program is used to check the structural
safety issue.
2. The numerical model
2.1. Geometry
The parabolic geometry for the applied arch is presented by the
equation:
f ωx
z¼4 ðl — xÞ
l2
where
x, z are the coordinates of any point on the arch
l is the span of arch
f is the rise of the arch at the crown (at l/2)
The rise to span ratio for arches varies widely. Most arches
would have a rise to span ratio in the range from 0.16 to 0.20.
The span to crown thickness (t crown) ratio (from existing concrete
arches) can be taken between 70 and 80. The springing thickness
(t springing) to crown thickness ratio is between 1.55 and 1.72.
Fig. 1 shows the preliminary dimensions for a typical concrete
arch [18].
2.2. Structural system and applied loads
The considered arch bridge consists of two-way directions sep-
arated by an island with 2.00 m width and sidewalk in each side
with 3.00 m width. Each way has 2 lanes with lane width = 3.00 m.
So, the total width of the bridge is 20.00 m as shown in Fig. 2.
The structural system consists of a bridge slab, secondary
beams, main beams, and the arch girder as shown in Fig. 3. Second-
ary beams are distributed every 2.00 m along the bridge (2 edge
and 9 intermediate secondary beams). The main beams, with
20.00 m length, distributed at least every 4.00 m in which the
bridge slab is divided into a number of one way bays. So, the num-
ber of main beams is adaptable by the program according to the
span which is defined by the user. The two arch girders support
the bridge from both sides.
The dead loads is automatically calculated by the program con-
sidering slab thickness, main beam dimensions, secondary beam
dimensions and material properties which are given by the user.
According to the Egyptian Code of Practice for design loads
(ECP), there is a typical live loading case for design, including the
dynamic impact factor, shown in Fig. 2 [19]. Using influence line,
the worst position for vehicle to get the maximum internal forces
on arch girder is the center.
2.3. The structural analysis
Herein, FEA model is presented. This model is classified as a two
dimensional case (plane stress). The assumption of plane stress is
applicable for bodies whose dimension is very small in one of the
coordinate directions. Thus, the analysis of thin plates loaded in
the plane of the plate can be made using the assumption of plane
stress. In plane stress distribution, it is assumed that.
ð1Þ rzz ¼ rzx ¼ ryz ¼ 0 ð2Þ
where (z) represents the direction perpendicular to the plane of the
plate as shown in Fig. 4.
The main concept in FEA is to divide the structural problem to
small interconnected elements. The Constant Strain Triangular
(CST) elements are to be used to form the arch body. The CST ele-
ment is composed of 3 corner nodes; each node has 2 Degrees of
Freedom (DOF), u and v, as shown in Fig. 5. The stress-strain rela-
tion is given by:
frg ¼ ½D]feg ð3Þ
0 1
1 t 0
E B C
½D] ¼ @ 2 t 1 A 0 ð4Þ
1 t —1—t
0 0 2
where

8 9
< ex =
e = strain vector for element = : ey ;
cxy
ex = strain component in x direction.
ey = strain component in y direction.
cxy = the strain on the plane xy.8 9
< rx =
r = stress vector for element = ry
:s ;
xy
rx = stress component in x direction.
ry = stress component in y direction.
sxy = shearing stress on the plane where x value is constant and
y is the direction.
D = the elasticity matrix which expresses the properties of
Fig. 1. Preliminary dimensions for a typical arch. material
 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516 509

Fig. 2. Plan of bridge and applied loads.

Fig. 3. Structural system.

E = modulus of elasticity for reinforced concrete where

t = Poisson’s ratio for reinforced concrete
Ke = the stiffness matrix of the element.
The arch body is divided into (32) CST elements, as shown in Ve = the volume of the element = Ae * width.
Fig. 6, connected together in (34) nodes. It should be noticed that Be = the matrix which expresses the geometry of the
the arch is symmetric in geometry and loads. Nodes (1, 2, 33, element.
and 34) are fixed support. So, the stiffness matrix size; taking the
end conditions in consideration; is (30 nodes * 2 DOF) = 60 * 60. 0 1
b1 b2 b3 0 0 0
1
For a linear displacement field with in-plane loads, the resulting
B 0 0 0 c1 c2 c3 C
A ð6Þ
6 × 6 in-plane stiffness matrix for each CST element can be ½B]e ¼ @
expressed as: 2Ae c1 c2 c3 b1 b2 b3

½K]e ¼ ½B]eT ½D]½B]e V e ð5Þ where

510 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516

c1 = the horizontal distance between node 3 and node 2 =

x3 – x 2
c2 = the horizontal distance between node 1 and node 3 =
x1 – x 3
c3 = the horizontal distance between node 2 and node 1 =
x2 – x 1

From the 32 stiffness matrices of the CST elements, a global

stiffness matrix [K] can be formed through their contact points
and therefore strain vector for elements can be expressed as:
feg ¼ ½B]fdg ð7Þ

fdg ¼ ½K]—1 fFg ð8Þ

where

F = load vector applied on nodes

d = deformation vector for nodes

2.4. A verification example

Fig. 4. Example of a plane stress problem.

In order to check the developed numerical model and FEA code,

a verification example is analyzed with SAP program with the same
conditions and the results are compared as shown in Table 1 [20].
From these assumptions, the coordinates that forms the arch
body are mapped in Fig. 7.
Figs. 8–10 show the extreme values for nodal deformation, nor-
mal and shear stresses. The comparison between developed model
and plane stress SAP model is illustrated in Table 2.
The table shows that difference in results between the devel-
oped model and SAP model does not exceed 3% of values.

3. The optimization technique

In this paper, Genetic Algorithms technique is used to get the

optimal nodes coordinates to form the optimal shape of arch girder
which has the minimum weight and therefore has the minimum

Table 1
Assumed conditions.

Concrete strength for the 29419.95 kN/m 2 Poisson’s ratio 0.25

Fig. 5. The Constant Strain Triangular Element.
standard cube (fcu )
Width of arch girder 0.80 m Thickness of bridge 0.20 m
2 3 slab
x 1 z1 1 Width of secondary beams 0.30 m Depth of secondary 0.50 m
Ae = area of the element = 0.5 * det. 4 x2 z2 15 beams
x 3 z3 1 Width of main beams 0.40 m Depth of main 0.70 m
b1 = the vertical distance between node 2 and node 3 = z2 – z3 beams
Span of arch 30.00 m Rise of arch (f) 6.00 m
b2 = the vertical distance between node 3 and node 1 = z3 — z1 Crown thickness 1.20 m Springing thickness 3.00 m
b3 = the vertical distance between node 1 and node 2 = z1 – z2

Fig. 6. Finite Element model.

M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516 511

Fig. 7. Verification example dimensions and coordinates.

Fig. 8. Deformation.

Fig. 9. Normal Stresses.

Fig. 10. Shear Stresses.

512 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516

qffi ffiffiffiffiffiffiffiffi ffiffiffi
Table 2
Comparison of results for the developed numerical model.
Allowable shear stress ¼ 21:79 ω f cu =cc ðkN=m 2Þ ð13Þ

Model type Developed model SAP model

Allowable deflection ¼ l=250 ðmÞ ð14Þ
Max. Deformation 0.0074 m 0.0072 m
Max. Compression Stress 6940.916 kN/m 2 6940.9157 kN/m 2 where
Max. Shear Stress 2976.783 kN/m2 2976.7833 kN/m2
fcu = concrete strength for the standard cube after 28 days
cc = the reduction factor for the concrete strength
cost. A steady state Genetic Algorithms technique is applied for this
model [21]. The optimization process aims to consume the element 3.1. The design variables
for maximum stresses and deformations allowed by the Egyptian
specifications. So, the objective function can be formulated as fol- The design variables here are the nodes vertical coordinates.
lowing [22]: Thirty four nodes are selected to form the arch body. Sixteen nodes
X in the right side have equal coordinates to their counterpart in the
Minimize f ðxÞ ¼ V ¼ Ai ω T i i ¼ 1; 2; 3 · ·· · · · ; n ð9Þ
left side in addition to two nodes on the axis of symmetry. So, only
where 18 nodes are applied as design variables for the optimization pro-
cess with 16 alternative positions for each node. The step between
V = Total volume of structure alternative positions for each node depends on the given span as
Ai = Element Area (defined by coordinates) shown in Fig. 11. It means that the program searches in (16 18 ) dif-
Ti = Element Thickness (constant) ferent geometric design field to get the optimal safe design. Num-
n = number of elements ber of nodes and search space can be increased to give more
accurate analysis with higher computational cost.
Two constraints are applied; induced stresses and deformations
should not exceed the allowable stresses and allowable deforma- 3.2. Genetic Algorithms operators
tions defined by the Egyptian Code of Practice for design and con-
struction of concrete structures (ECP) [23]. The constraints in this Genetic Algorithms technique starts with generating a set of
model can be formed as follow: possible solutions (parents) randomly as an initial population to
the problem. Eight random solutions, which represent different
g i ¼ rall — ri ≥ 0; i ¼ 1; 2; 3; · · · · ·· ; n ð10Þ arch geometries, with length of 18 gene (nodes coordinates) are
used for this model. These design vectors are encoded to binary
gi ¼ D all — Di ≥ 0; i ¼ 1; 2; 3; ·· ·· · · ; m ð11Þ form to facilitate the application of mating operators. So, the total
where length for each solution, chromosome, is 18 * 4 = 72 bit. Genetic
Algorithm’s mating operators are crossover and mutation. Each
ri = the stresses in each CST element two solutions (parents) are combined together to create two chil-
Di = the deformation in each node dren solutions. Uniform crossover technique is used for this model.
m = number of nodes The mutation technique which is used for this model is flip bit
rall = the allowable stresses mutation with mutation probability selected by the user. All the
Dall = the allowable deformation sixteen solutions (parents and children) are collected in one pool.
All solutions are sent to the structural analysis program to check
The allowable stresses and deformation can be calculated as the safety constraints, stresses and deformations. The program
follow: reads the induced stresses and deformation values for each node
and compare them to the allowable limits, according to the ECP.
Allowable normal stress ¼ f cu=cc ðkN=m 2Þ ð12Þ Unsafe solutions gets penalty function by increasing their target

Fig. 11. The design variables.

 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516 513

weight with a suitable factor. The best 8 solutions which have less mutation probability (p m) – number of iterations (not less than
weight for the arch girder are selected to be new parents for the 1000 iterations).
next generation [24]. These operators are repeated and new gener- By using a laptop CORE I3 processor and 2 GB RAM, the devel-
ations are created until the global optimum solution is reached. oped model has been applied for 3 examples with different spans
as shown in Table 3.
4. The results After running the model for these examples, with average
4 days running time for each example, the induced optimized
The developed program gives the user the authority to change shape for each case has been shown in Figs. 12–14.
any of the following numerical model parameters: It’s noticed that thickness of arch girder is increased in the
Arch span (not less than 20.00 m) – arch width – secondary lower part of the first and last third while increasing in the upper
beam width – secondary beam depth – main beam width – main part of the middle third of the arch for all cases. A skeletal model is
beam depth – bridge slab thickness – fcu – poisson’s ratio – developed for the investigated cases using SAP program. The
induced bending moment diagrams are shown in Figs. 15–17.
It is clear that the resulted rough optimal design follow the
Table 3 moment diagram with reversed direction. This rough geometry
Input Data of Examples. can be smoothed using curve fitting (for practical considerations)
Example no. 1 2 3 as shown in Figs. 18–20.
Span (m) 20 30 53
Fig. 21 shows the progression of the minimum weight value
Arch width (m) 0.70 0.80 0.80 through 1,000,000 iterations of the optimization process. There
Width of secondary beams (m) 0.25 0.30 0.30 are many clear drops in the weight value in the first 100 iterations
Depth of secondary beams (m) 0.50 0.50 0.60 then the change in weight value occurs at long intervals until
Width of main beams (m) 0.30 0.40 0.40
Depth of main beams (m) 0.60 0.70 0.80
reaches the optimal weight.
Thickness of bridge slab (m) 0.20 The resulted optimal geometry is compared to concrete arch tra-
Concrete strength (fcu ) (kN/m 2 ) 29419.95 ditional design. This comparison is illustrated in Table 4. The rough
Poisson’s ratio 0.25 optimal design has arch weight reduction ranges from 35% to 40%
Mutation probability (P m ) 0.1 compared to traditional design arch weight. After solution enhance-
Number of iterations 1,000,000
ment, the reduction percentage became in range of 30–35%.

Fig. 12. Optimal Arch Shape for 20.00 m span.

Fig. 13. Optimal Arch Shape for 30.00 m span.

Fig. 14. Optimal Arch Shape for 53.00 m span.

514 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516

Fig. 15. Bending Moment Diagram for skeletal model of span 20.00 m.

Fig. 16. Bending Moment Diagram for skeletal model of span 30.00 m.

Fig. 17. Bending Moment Diagram for skeletal model of span 53.00 m.

Fig. 18. Enhanced arch shape vs traditional shape for 20.00 m span.

Fig. 19. Enhanced arch shape vs traditional shape for 30.00 m span.
M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516 515

Fig. 20. Enhanced arch shape vs traditional shape for 53.00 m span.

Fig. 21. Progression of minimum weight for 53.00 m span.

516 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516

Table 4
Comparison of optimal and traditional design.

Example no. 1 2 3

Arch weight (kN) Optimization model 339.80 799.83 2427.57

Traditional model 532.01 1235.64 4002.09

Weight reduction percentage 36.13% 35.27% 39.34%

Maximum deformation (m) Optimization model 0.0074 0.0098 0.0194

Traditional model 0.0048 0.0074 0.013

Max. compression stress (kN/m 2 ) Optimization model 10269.14 7699.79 9083.85

Traditional model 6863.97 6940.916 6984.59
Max. shear stress (kN/m 2 ) Optimization model 3050.99 3049.67 3048.51
Traditional model 2859.13 2976.783 2993.58

Bruno D, Lonetti P, Pascuzzo A. An optimization model for the design of

5. Conclusions network arch bridges. J Constr Steel Res 2016;170.
[18] Nettleton DA, Torkelson JS. Arch bridges. Washington, D.C. 20590: Bridge
Division, Office of Engineering Federal Highway Administration, U.S.
The developed program gives a wide variety for the user to
Department of Transportation; 1977.
assign the input values even for odd ones. It perform self- [19] Egyptian code of practice for design loads. Housing and Building National
adaptation to calculate loads and structural system according to Research center, Egypt, Code No. ECP 201; 2008.
[20] https://www.csiamerica.com/products/sap2000.
entered data. The developed Finite Element Analysis program gives
[21] Back T, Fogel DB, Michalewicz Z. Michalewicz, evolutionary computation, vol.
trusted results when compared to well-known commercial pro- 1: basic algorithms and operators. Bristol and Philadelphia: Institute of Physics
grams. The proposed optimization methodology proved to be suc- Publishing; 2000.
cessful technique for the investigated structural optimization [22] Rao SS. Engineering optimization- theory and practice. 4th ed. John Wiley &
Sons Inc; 2009.
process. It can reach sufficiently the optimal solution with reason- [23] Egyptian code for design and construction of concrete structures. Housing and
able computation cost. Although, the results shows a reasonable Building National Research center, Egypt, Code No. ECP 203; 2017.
reduction in weight for the proposed optimal solution. The reduc- [24] Eid MA. Genetic algorithms in structural engineering and its application to
tunneling. Faculty of Engineering of Minia University in partial fulfillment for
tion percentage is not affected significantly with the entered bridge the requirement of Ph.D. Degree in Civil Engineering; 2011.
span. The resulted optimal designs, after geometric enhancement,
have a reduction percentage ranges from 30% to 35% compared
to traditional designs. Considering the induced deformations and Mostafa Zaki ABDELREHIM (Prof, PhD) is recognised as
stresses in the optimal arch geometry, the shear stress is the con- an expert in tunnelling and problematic structures. His
experience spans 30+ years, including more than 15
trol parameter as the allowable shear strength is reached com-
years of experience with European universities, research
pared to normal stress and deformation. The resulted optimal centres, and international consultant companies for
profile reflects the bending moment profile which matches well structural analysis and tunnelling disciplines. His
with the common design principles. research activities, interests and publications include
Improvement of computational techniques and its
application to large complex finite element models,
References Optimization techniques and genetic algorithms for
structural analysis and design and, Developing analysis
[1] Olivas F et al. Ant colony optimization with dynamic parameter adaptation tools and retrofitting techniques of damaged structures
based on interval type-2 fuzzy logic systems. Appl Soft Comput, 53. Elsevier B. especially tunnels.
V.; 2017.
[2] Gonzalez B, Valdez F, Melin P, Prado-Arechiga G. Fuzzy logic in the
gravitational search algorithm for the optimization of modular neural Mohamed Abdelfattah EID (PhD) is an assistant profes-
networks in pattern recognition. Elsevier Ltd; 2015. sor in Civil Engineering Department, Minia University –
[3] Holland JH. Adaptation in natural and artificial systems. Ann Arbor,
Egypt. His main teaching subjects are structural analysis
MI: University of Michigan Press; 1975.
and tunnelling engineering. His reseach fields are
[4] Sahab MG, Ashour AF, Toropov VV. A hybrid genetic algorithm for reinforced
structural optimization, tunnelling engineering and
concrete flat slab buildings. Comput Struct 2005;83.
structures integreity. He also has 20 years of experience
[5] Eid MA, Zaki Abd Elrehim M. Optimization of tunnel profile in different ground
conditions using genetic algorithms. In: 3rd international conference on in concrete structures analysis and design and he
computational methods in tunnelling and subsurface engineering. Ruhr works as a consultant for several Egyptian official
University Bochum, Germany, Euro: TUN; 2013. organizations.
[6] Griffiths DR, Miles CJ. Determining the optimal cross-section of beams. Adv
Eng Inform 2003;17.
[7] Keedwell E, Khu S. A hybrid genetic algorithm for the design of water
distribution networks. Eng Appl Artif Intell 2005;18.
[8] Tveit P. The network arch: bits of manuscript in September 2008 after lectures
in 50 countries; 2008.
[9] Cheng J. Optimal design of steel truss arch bridges using a hybrid genetic
Mostafa Gomaa SAYED is a civil engineer with 6 years
algorithm. J Constr Steel Res 2010;66.
experience in road and building construction. He works
[10] Sonavane T. Analysis of arches. Faculty of the Graduate School of the
as a supervision engineer in New Minia City for New
University of Colorado in partial fulfillment of the requirements for the
degree of Master of Science; 2014. Urban Community Authority, Ministry of Housing –
[11] Rao SS. The finite element method in engineering. 5th ed. Elsevier Inc; 2011. Egypt. As a researcher, he is interested in the structural
[12] Argyris JH, Kelsey S. Energy theorems and structural analysis. Aircraft Eng vols. optimization research field.
26 and 27, October 1954 to May 1955. Part I by Argyris JH, Part II by Argyris JH,
Kelsey S.
[13] Uzman U, Daloglu A, Saka MP. Optimum design of parabolic and circular
arches with varying cross-section. Struct Eng Mech 1999;8(5):465–76.
[14] Islam N, Rana S, Ahsan R, Ghani S. An optimized design of network arch bridge
using global optimization algorithm. Adv Struct Eng 2014;17(2):pp.
[15] Pouraminian M, Ghaemian M. Shape optimization of concrete open spandrel
arch bridges. Gradevinar 2015;67(12):1177–85.
[16] Wang Y, Thrall AP, Zoli TP. Adjustable module for variable depth steel arch
bridges. J Constr Steel Res 2016;126.

[17]
 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516 517
KELEBIHAN :
Penelitian untuk mendemonstrasikan metodologi dalam menemukan desain dengan biaya paling rendah, khususnya
dalam hal volume material, dengan menemukan profil yang optimal.

KEKURANGAN :
Kurangnya penelitian tentang optimasi desain jembatan lengkung beton dengan Algoritma Genetika memerlukan
penelitian lebih lanjut untuk mengembangkan metode yang lebih efektif. Meskipun ada beberapa studi sebelumnya di
bidang ini, masih ada ruang untuk pengembangan yang lebih efisien.

PEMBEBANAN :
Beban mati dihitung secara otomatis oleh program dengan mempertimbangkan ketebalan pelat jembatan, dimensi
balok utama, dimensi balok sekunder, dan properti material yang diberikan oleh pengguna. Beban hidup mengacu
pada kasus pembebanan desain yang mencakup faktor dampak dinamis. Menurut Kode Praktik Mesir untuk
perancangan struktur, ada satu kasus beban hidup yang sering digunakan untuk menguji kekuatan struktur.

STRUKTUR :
Struktur jembatan tersebut terdiri dari pelat jembatan, balok sekunder, balok utama, dan balok lengkung. Balok
sekunder terdistribusi setiap 2,00 m sepanjang jembatan, dengan 2 balok tepi dan 9 balok sekunder di antaranya.
Balok utama, dengan panjang 20,00 m, terdistribusi setidaknya setiap 4,00 m, yang membagi pelat jembatan menjadi
beberapa bay satu arah. Ada dua balok lengkung yang mendukung jembatan dari kedua sisi.

PERENCANAAN STRUKTUR :
Sruktur jembatan lengkung beton direncanakan dengan menggunakan metode analisis elemen hingga (finite element
analysis) dan algoritma genetika (genetic algorithms). Proses perencanaan melibatkan pemodelan komputer jembatan
sebagai struktur berkelanjutan yang dibagi menjadi elemen segitiga dengan regangan konstan (Constant Strain
Triangular/CST) dalam program analisis elemen hingga. Variabel desain yang digunakan dalam optimasi adalah
koordinat simpul-simpul elemen segitiga ini, dengan tujuan mengurangi berat jembatan lengkung. Selain itu, program
analisis elemen hingga yang telah dikembangkan juga digunakan untuk memeriksa masalah keamanan struktural
jembatan.

MANFAAT :
- Pengurangan biaya
Penelitian ini menggunakan algoritma genetika untuk mencari desain jembatan lengkung beton yang lebih efisien
secara biaya. Dengan menemukan desain yang lebih optimal, penelitian ini dapat membantu mengurangi biaya
pembangunan jembatan lengkung beton
- Efisiensi struktural
Dengan menggunakan pendekatan optimasi yang canggih, penelitian ini dapat menghasilkan desain struktural
jembatan lengkung beton yang lebih efisien. Desain yang lebih efisien dapat meningkatkan kinerja struktural
jembatan dan mengurangi risiko kegagalan struktural
 518 M.Z. Abd Elrehim et al. / Ain Shams Engineering Journal 10 (2019) 507–516

CONTOH JEMBATAN :

Jembatan Pelengkung Kademangan Blitar

 JEMBATAN GANTUNG (SUSPENSION)

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

journal homepage: www.elsevier.com/locate/jweia

Investigation of dynamic wind loads on a long-span suspension bridge

identified from measured acceleration data
Ø.W. Petersen a,*, O. Øiseth a, E. Lourens b
a NTNU, Norwegian University of Science and Technology, 7491, Trondheim, Norway
b Delft University of Technology, 2628, CN, Delft, the Netherlands

A R T I C L E I N F O A B S T R A C T

Keywords: The traditional wind load assessment for long-span bridges relies on assumed models for the wind field and
Structural monitoring aerodynamic coefficients from wind tunnel tests, which usually introduce some uncertainties. Recent studies have
Wind loads shown that large deviations can exist between the predicted and observed wind-induced dynamic response of
Force identification suspension bridges. In studies of the dynamical behavior of bridges, inverse force identification methods can
Suspension bridge
therefore be an interesting tool in the assessment of possible uncertainties involved in the modeling of wind loads.
This paper presents a novel case study of the identification of the dynamic wind loads on the 1310 m long
Hardanger bridge, a suspension bridge equipped with a monitoring system for wind and vibrations. The modal
wind loads are identified from acceleration data using an algorithm for model-based joint input and state esti-
mation. Several data sets with different wind conditions are presented. The wind loads are studied in the time and
frequency domains and are compared to the mean velocity and turbulence characteristics of the wind.

1. Introduction assumptions: the wind field is stationary, homogeneous, and normal to

As the development of bridges has led to longer and more slender
spans, the assessment of wind loads has become increasingly critical for
reliable structural design (Larsen and Larose, 2015). Modern wind load
and response analysis is based mostly on theories of buffeting due to
turbulence (Scanlan, 1978a) and self-excitation due to bridge motion
(Scanlan, 1978b) that have been refined in a variety of formulations:
comparisons can be found in (Chen and Kareem, 2002; Kavrakov and
Morgenthal, 2017). Although these theories are well established, the
actual parameters that go into this analysis can be a significant source of
uncertainty (Jakobsen and Tanaka, 2003; Caracoglia, 2008). The classic
wind load assessment also relies on aerodynamic coefficients and
admittance functions for the specific bridge geometry that are usually
obtained from wind tunnel tests using scale models. Simplifications and
uncertainties from these tests transfer directly to the predicted wind
loads. Although the computational methodology and technology for
wind tunnel testing have become sophisticated (Siedziako et al., 2017;
Diana et al., 2004; Cigada et al., 2001), the local conditions and
complexity experienced by an actual bridge cannot be recreated. For
instance, the lack of case-specific data usually leads to several simplified
the bridge. These assumptions neglect phenomena that can occur in re-
ality: non-stationary events, an inhomogeneous or skew wind field, and
the influence of local topology. Some full-scale studies also report dis-
crepancies between the predicted and measured responses of long-span
bridges (Fenerci and Øiseth, 2017; Wang et al., 2012; Macdonald,
2003; Bietry et al., 1995; Cheynet et al., 2016), indicating that there are
still some uncertainty gaps for predicted wind loading.
The real wind loads on large structures cannot be measured directly in
a practical manner. An alternative approach to this problem is the inverse
estimation of loads from response data, also known as force identifica-
tion. These techniques require (limited) measured vibration data and a
finite element (FE) model of the structure. A well-known challenge in
inverse force identification is issues related to the problem of ill-
posedness (Jacquelin et al., 2003; Lourens, 2012), meaning that the so-
lution is generally sensitive to numerical errors, measurement noise, and
model errors, which are inevitably present in non-synthetic data. In
recent years, many different techniques for force identification have been
proposed. Furthermore, various Kalman-type filters (Ma et al., 2003;
Azam et al., 2015; Lourens et al., 2012a, 2012b; Maes et al., 2018; Song,
2018) and frequency-domain (Liu and Shepard, 2005; Rezayat et al.,

* Corresponding author.
E-mail addresses: oyvind.w.petersen@ntnu.no (Ø.W. Petersen), ole.oiseth@ntnu.no (O. Øiseth), e.lourens@tudelft.nl (E. Lourens).
URL: https://www.ntnu.edu/kt/research/dynamics (Ø.W. Petersen).

https://doi.org/10.1016/j.jweia.2019.104045
Received 8 March 2019; Received in revised form 8 June 2019; Accepted 16 November 2019
Available online xXXX
0167-6105/© 2019 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/ ).
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

2016) and time-domain (Li et al., 2015; Qiao et al., 2016; Aucejo et al.,
2018a; Sun et al., 2015; Lai et al., 2017; Chang et al., 2019; Bernal and
Ussia, 2015) deconvolution approaches with regularization schemes
have been investigated.
Inverse estimation of wind loads on tall buildings has been studied in
wind tunnels (Zhi et al., 2018; Hwang et al., 2011) and in full-scale
conditions (Zhi et al., 2016; Kang et al., 2012; Niu et al., 2015). Recon-
struction of wind loads on small-scale guyed masts have also been re-
ported (Amiri and Bucher, 2017). While an increasing number of
cable-supported bridges are equipped with data monitoring systems
(Ko and Ni, 2005; Wong, 2004), to the best of the authors’ knowledge, no
studies have focused on the inverse estimation of wind loads. Therefore,
it is of interest to test the state-of-the-art force identification methodol-
ogies with full-scale data and to assess their utility as a tool in detailed
studies of wind loads. As will be explained, force identification does not
utilize any specific load model assumptions, which makes the method- Fig. 2. Topography of the area surrounding the bridge.
ology attractive for studies of non-stationary data, wind gusts or
abnormal events, and could help to provide a better understanding of the
mechanisms behind wind loads. Before such detailed studies can be
performed, the performance of the methods must be evaluated to identify
the weak and strong points. Although the inverse methods have been
explored in laboratory experiments, the transition to full-scale bridges
means that different conditions and uncertainties will play an important
role.
In this article, the dynamic wind loads on the Hardanger bridge are
estimated based on acceleration data. The capability of the full-field
estimation of these algorithms for long-span bridges was studied previ-
ous works (Petersen and Øiseth, 2019), showing that the methodology is
feasible, with the main limitation being the number of modal responses
that can be reconstructed.
The rest of this paper is organized as follows. Section 2 presents the Fig. 3. Positions of the sensors on the Hardanger bridge.
bridge case study and the methodology. In Section 3, results from several
data sets with different wind conditions are examined. The loads are installed sensors. Eight digital ultrasonic anemometers A1-A8 (Gill Wind
assessed in the time and frequency domains, and the influence of the Master Pro 3D) are mounted on the hangers along the bridge span.
wind characteristics on the loads are investigated. A short discussion of Twenty triaxial accelerometers (CUSP-3D strong-motion, range 4 g,
the usefulness of the methodology is provided, and conclusions are given SNR 130 dB) are located in the girder and the towers. More detailed
in Section 4. specifications of this system are given in (Fenerci et al., 2017).

2. Case study: the Hardanger bridge

2.1. The bridge and monitoring system
The Hardanger bridge (Fig. 1) is a suspension bridge with a main span
of 1310 m. Crossing the Hardanger fjord, the bridge is located in a fjord-
mountainous terrain (Fig. 2). The long-span and slender features of the
bridge make it sensitive to excitations from wind approaching from the
coastal area to the west or the inland mountainous regions to the east.
The bridge is a valuable case study for future long and slender bridges
planned for similarly complex terrains, and a structural monitoring sys-
tem has been in operation since 2013. Fig. 3 shows the positions of the
2.2. Equations for bridge dynamics in force identification framework
In the following, a short derivation of the state-space equations for the
dynamic system modeling is provided. These equations for reduced-order
multimodal systems are fairly well known but are included here for a
clear problem definition. For studies of wind-induced response in regular
in-operation conditions, a linear dynamic model is adequate, and the
equations of motion for the bridge in a FE format can be given by:
M0 u€ðtÞ þ C0 u_ ðtÞ þ K0 uðtÞ ¼ fðtÞ (1)
where the subscript ð·Þ0 denotes the still-air properties, i.e., contribution
from the structure only. The vector uðtÞ2 RnDOF is the response of the
physical DOF, and fðtÞ2 RnDOF a force vector. To reduce the model order,
the reduction basis is constructed from a set of nm mass-normalized still-
air vibration modes φj 2 RnDOF :

K0 — ωj2M0 φj ¼ 0 (2)

Further, the full response is approXimated by uðtÞ≈ ΦzðtÞ, where

Φ ¼ ½φ1 φ2 …φnm ] 2 RnDOF ×nm and zðtÞ 2 RnDOF is the generalized coordi-
nate vector. Assuming the damping matriX C0 is proportional, the
generalized equation of motion becomes:

z€ðtÞ þ 2ΞΩz_ ðtÞ þ Ω2 zðtÞ ¼ ΦT fðtÞ ¼ pðtÞ (3)

Fig. 1. View of the Hardanger bridge from the north end (Photo: Øyvind Wiig where Ω ¼ diagðω1 ; …; ωnm Þ and Ξ ¼ diagðξ1 ; …; ξnm Þ are diagonal
Petersen/NTNU). matrices assigned the still-air natural frequencies (ωj ¼ 2πfj) and

2
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

damping ratios. The measurement vector ðy Þt2Rny , which consists of Kalman-type technique, is used. This method provides minimum-
accelerations and displacements in selected structural DOFs, is given as bð Þforces (p tk ).In addition,
variance estimates of the states (x tk ) and input bð Þ
follows: a dual Kalman filter (DKF) (Azam et al., 2015) and a
time-domain sequential deconvolution method (Bernal and Ussia, 2015)
yðtÞ ¼ Sa u€ðtÞ þ S d uðtÞ (4)
are tested. For more detailed information on the estimation methods, we
refer to the cited works.
where Sa and Sd 2Rnd×nDOF are binary matrices that select the measured
Contrary to classic (forward) modeling of wind loads, no specific
DOFs. By introducing the modal state variable
time-space assumptions are imposed on the wind field or the wind forces.
xðt Þz¼t ½T ðz_Þ t ð ÞR]2nm2, the system equations is converted into state-
T T
Since the force identification is driven directly by the vibration data (and
space form:
the system model), there is no model for the wind field and no functions
x_ ðtÞ¼ AcxðtÞþ BcpðtÞ (5) are given for the aerodynamic coefficients and admittance of the bridge
deck. This is critical since the study is not constrained to cases with
yðtÞ¼ G xðtÞþ J pðtÞ (6) idealized load conditions so that one can study loading from wind fields
c c
that are skew, inhomogeneous, non-stationary or essentially unknown.

0 I
Ac ¼ 2 (7) 2.3. System model and limitations in the methodology
—Ω —2ΞΩ
The still-air modes are solved from a FE model of the bridge (Fig. 4).
0
Bc ¼ (8) The model is tuned via a model updating procedure to match modal
I
properties from operational modal analysis; see (Petersen and Øiseth,
2019) for details. Table 1 lists the modes, where the labels H, V, T and P,
Gc ¼½ SdΦ — SaΦΩ2 —SaΦ2ΩΞ ] (9) respectively, denote horizontal, vertical, torsion and pylon motions. The
identified (almost) still-air damping ratios ξj in Table 1 are adopted for
Jc ¼½ SaΦ ] (10) the state-space model.
In the discretization to an intersample time Δt ¼ 0:1 s, a first-order The system model is truncated to contain only 18 modes, a choice
hold constraint is imposed on the input. This constraint is necessary to explained in the following. In most force identification methods, the
preserve the model accuracy for the very low-frequency range, which is number of unknown forces is theoretically limited by the number of
important for analysis of the wind-induced response. After the addition of linearly independent outputs. The identification of modal forces requires
stochastic noise, this process yields the following model equations: that rankðJÞ ¼ nm in an instantaneous inversion of system (Maes et al.,
2014). Fig. 5 shows the evaluation of the singular values of J for different
xkþ1 ¼ Axk þ Bpk þ vk (11) model sizes. The clearly visible singular value drop-off for the 19-mode
model indicates close-to-linear relationships in the output data, which
yk ¼ Gxk þ Jpk þ wk (12) increases the condition number of this matriX. Therefore, no more than
18 modal forces could be identified without running into severe ill-
where the secondary state xk ¼ xk — Fpk is introduced (Aucejo et al., conditioning in the system inversion. This number of forces is notably
2018b) and the system matrices are given by: less than the upper theoretical limit governed by the aforementioned
rank-criterion, highlighting the importance of designing optimal sensor
A ¼ expðAcΔtÞ (13) locations, in particular for long-span bridges that have many modes that
contribute to the wind-induced total response. The truncated system
B ¼ ðA — IÞAc—1 Bc — F þ AF (14) model accounts for roughly 80% of the measured acceleration data.
, Some limitations exist when only acceleration data are available, and
F ¼ Ac—1 ðA — IÞ Δt — I Ac—1 Bc (15) the workarounds and effects hereof are discussed in the following. To
obtain a steady state of the filter gain matrices and reduce the compu-
G ¼ Gc (16) tational time, displacement data are generated by numerical integration
of the accelerations. The influences of these displacements are weighted
J ¼ Jc þ GcF (17) downward to rely mostly on the acceleration data. In filtering type ap-
plications, the covariance matrices are often considered tuning parame-
The vectors wk 2 R2nm and vk 2 Rny are zero-mean white noise terms ters for the system; R is set to diagonal with values of 10—6 for the
that model the uncertainties of the states and measurements. Their cor- accelerations and 10—2 for the displacements. This process is similar to
responding covariance matrices are given by: the principle of using dummy measurements for stabilization, as pro-
E wkwT ¼ Q δkl posed in (Naets et al., 2015); the displacement outputs are strongly
l
penalized to lessen their influence while maintaining a stable solution. A
l
E vkv ¼ R δkl
T
(18) —3

E wkvTl ¼ S δkl value of Q ¼to10

insensitive I is adopted
a scaling of thesefor the noise
matrices on the states.
of filters,
several orders The solution
up or down. is
Since the inverse algorithms are linear the frequency content
A clear model definition of the unknown forces is essential in the
of the estimated forces directly relates to the output data. In the very-low-
context of input estimation. For cable-supported bridges, the wind forces
frequency range, the accelerations are not very sensitive to the modal
are not localized to a small number of nodes but are distributed along the
forces. Although the installed accelerometers perform well for very-low-
entire bridge. The full force vector fð tÞ in Eq. (1) takes into account any
frequency vibrations, the data below approXimately 0.01 Hz could
type of loading that may be present (buffeting forces, self-excited forces,
contain errors or be affected by the removal of very small linear trends.
traffic forces, etc.). The system formulation in Eqs. (11) and (12) con-
The force estimates are therefore high-pass filtered at 0.01 Hz in the post-
siders these forces projected to a modal space, i.e., ΦTf ðt Þ ¼ pð Þt 2 Rnm .
processing of the results. The actual buffeting loads due to turbulence do,
In this work, the well-established joint input-state estimation algo- however, have low-frequency components below this limit. Ideally,
rithm (JIS) (Lourens et al., 2012b), which can be classified as a
actual static-sensitive measurement data could be used to obtain better

3
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

Fig. 4. Still-air modes in the system model.

3. Identification of wind loads

Table 1
Properties of the still-air modes in the system model. The symbols f j , ξj, and φj 3.1. Assessment of load characteristics
denote identified modal properties (Petersen and Øiseth, 2019).
Mode no. j Mode name fj [Hz] f j [Hz] ξj [%] MACðφj ; Reðφj ÞÞ The four selected recordings that are used in this study are listed in
1 H1 0.051 0.052 0.65 0.999
Table 2. As will be shown, these recordings reflect the variability of wind
2 H2 0.105 0.105 0.77 0.993 conditions that occur at the site. This variability should be considered not
3 V1 0.112 0.119 1.77 0.989 only because it influences the wind load characteristics, but also because
4 V2 0.142 0.142 0.65 0.995 it is imperative that the proposed methodology performs consistently
5 H3 0.185 0.183 0.77 0.993
well regardless of the wind conditions the bridge is exposed to. For the
6 V3 0.203 0.206 0.27 0.992
7 V4 0.212 0.212 0.35 0.997 considered data sets, the response is largely dominated by wind effects.
8 V5 0.276 0.276 0.26 0.998 Although some contributions from traffic could occur, the traffic density
9 H4 0.318 0.318 0.63 0.990 on the bridge is generally low, with a daily average of 1900 crossing
10 V6 0.332 0.333 0.25 0.995 vehicles. Vortex shedding effects should be minimal; the bo X girder is
11 T1 0.371 0.374 0.41 0.964
12 V7 0.401 0.401 0.24 0.997
streamlined with guide vanes mounted underneath to mitigate vortex-
13 H5 0.463 0.464 1.56 0.704 induced vibrations. The buffeting forces due to turbulence and self-
14 V8 0.468 0.471 0.26 0.991 excited forces are expected to be the most significant actions.
15 P1 0.511 0.516 0.16 0.889 For the first recording, Figs. 6 and 7 show the identified forces for
16 P2 0.518 0.529 0.22 0.859 modes 1–12 in the time and frequency-domain. The characteristics of the
17 V9 0.545 0.547 0.31 0.998
remaining modes (13–18) are similar and are not shown here. For
18 T2 0.550 0.560 0.65 0.978
brevity, the identified forces from recording 2–4 are only shown in the
frequency-domain (Figs. 8–10). In the frequency domain, the modal
forces are similar across the different recordings, although it is expected
the actual time-domain evolutions are quite different in terms of local/
global extrema, steady level/sudden increase in forces, correlation
among modes, etc. All modal forces are largely dominated by the low-
frequency components, which is consistent with buffeting loading,
although some small peaks are observed at certain frequencies, indi-
cating harmonic components, which manifests mainly for the horizontal-
type modes. Self-excitation could lead to such peaks due to natural
coupling with the motion of the structure. It should also be noted that in
inverse identification, errors in the system model typically lead to
Fig. 5. Singular values of J for different model sizes. spurious peaks in the estimated forces at the natural frequencies since

low-frequency estimates, but such data were unfortunately not available. Table 2
Furthermore, the purely static forces (f ¼ 0 Hz) are theoretically Data sets used in the identification of forces.
impossible to reconstruct from acceleration data alone (Maes et al., 2014) Recording Date and start Duration Mean wind Incoming
and are therefore not considered further. no. time velocity (A6) direction
In addition, the output data are low-pass filtered at 0.6 Hz during 1 2015-03-10 e6h e 10–28 m/s East
preprocessing since the contributions above this limit belong to modes 05:59
not accounted for by the system model (Table 1). In summary, only the 2 2015-05-05 e4h e 10–22 m/s West
dynamic force content in the range f 2 0:01; 0:6 ]Hz is considered. 14:53
½
3 2016-01-29 e8h e 15–28 m/s West
Different results could be obtained for other bridges or sensor network 12:30
configurations. 4 2016-12-26 e4h e 10–20 m/s West
18:09

4
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

Fig. 6. Identified forces for modes 1–12 in recording 1.

Fig. 7. Fourier transform of the identified forces for modes 1–12 in recording 1. The light blue curve is a smoothed FFT, the star signifies the single natural frequency,
the magenta lines signify all natural frequencies, and the green lines signify non-modeled modes. (For interpretation of the references to colour in this figure legend,
the reader is referred to the Web version of this article.)

5
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

Fig. 8. Fourier transform of the identified forces for modes 1–12 in recording 2. The light blue curve is a smoothed FFT, the star signifies the single natural frequency,
the magenta lines signify all natural frequencies, and the green lines signify non-modeled modes. (For interpretation of the references to colour in this figure legend,
the reader is referred to the Web version of this article.)

Fig. 9. Fourier transform of identified forces for modes 1–12 in recording 3. The light blue curve is a smoothed FFT, the star signifies the single natural frequency, the
magenta lines signify all natural frequencies, and the green lines signify non-modeled modes. (For interpretation of the references to colour in this figure legend, the
reader is referred to the Web version of this article.)

6
Ø.W. Petersen et al.
Fig. 10. Fourier transform of the identified forces for modes 1–12 in recording 4. The light blue curve is a smoothed FFT, the star signifies the single natural frequency,
the magenta lines signify all natural frequencies, and the green lines signify non-modeled modes. (For interpretation of the references to colour in this figure legend,
the reader is referred to the Web version of this article.)
these forces attempt to correct the unexplained behavior of the output
data. In this case, four non-modeled modes (i.e., modes not included in
the state-space model) from the FE model near 0.24 and 0.4 Hz are
indicated in the figures. The deflection of these modes is mainly in the
cables with some small movements in the boX girder. Neglect of these
modes likely causes estimation errors around these frequencies, leading
to a blunt peak for the horizontal and torsion modes. Once more, this
characteristic points to a limitation in this methodology for cable-
supported bridges with a high number of modal contributors and/or
the quality of the sensor network.
Consistently, for all recordings, a small dip is observed in the fre-
quency domain at the respective natural frequencies for modes V1–V7
and T1. This dip can be explained by the self-excited effects, which
introduce added damping in the vibration modes. Although these dips
may appear to be insignificantly small compared to the total force
magnitude, they are important since they decrease the forcing around the
natural frequency, which significantly reduces the modal response. For
most suspension bridges with boX girder decks, the vertical modes
experience the strongest influence of aerodynamic damping, which is
consistent with the observations made here. Furthermore, a sharp-
gradient drop is observed for mode T1 at f¼ 0:36 Hz, which is due to
the aerodynamic stiffness reduction of the torsional motion. However,
this phenomenon becomes miXed with the two non-modeled modes at
0.4 Hz, and the close spacing of the modes hinders a clear interpretation.
Self-excited forces are commonly modeled as proportional to the
structural displacements and velocities. These self-excited forces (still in
the modal projection space) can actually be obtained from the state es-
btimates by p se ðtÞ ¼ ½ΦT Kae Φ b ΦT Cae Φ]xðtÞ. Depending on the modeling
scheme adopted (Chen and Kareem, 2002;
2017), the aerodynamic coefficient matrices Kae and Cae2 CnDOF (ob-
Kavrakovand Morgenthal,
tained from wind tunnel tests) are generally dependent on the mean wind
velocity (meaning that they will not be time-invariant) in addition to the
b
frequency of the moving structure, so the calculation of pseðtÞ becomes a
convolution integral due to memory effects. An alternative method
7
Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045
would be to jointly estimate these aerodynamic coefficients with the
inputs and states; however, such complex approaches are considered
future work and are not pursued here.
3.2. Influence of wind characteristics
Correlating the identified loads with the characteristics of the wind is
of great interest. In this context, anemometer data are useful, although
the measurements could be seen as a sparse sampling of a large non-
sparse wind field. The limited amount of data poses a challenge. The
wind forces on the structure depend on the effective wind field along the
whole bridge, but only eight sampling locations are available at
approXimately the elevation of the bridge deck. Although some as-
sumptions or spatial interpolations can be made, the wind velocity field
does not have a known low-rank dynamic structure, so the eight mea-
surements alone cannot be used reconstruct the actual properties of the
full field. The use of LIDAR technology is an alternative approach that
potentially could fill this gap (Cheynet et al., 2017). In the present
application, the comparison of the wind measurements and identified
forces is therefore of a heuristic type more than an analytical connection.
To make use of the anemometer data, the turbulence and mean wind
velocity should be separated while making as few assumptions as
necessary. Let U ð Þt be the along-wind velocity with incoming direction
ð θÞ t , as defined in Fig. 2, and let W
ð Þt be the vertical velocity. We assume
that the wind velocity consists of a deterministic time-varying mean
value Uð Þt ; Wð Þt plus a fluctuating component uð Þt ; wð Þt that is
stochastic:
UðtÞ¼ UðtÞþ uðtÞ; WðtÞ¼ WðtÞþ wðtÞ (19)
We use empirical mode decomposition (EMD) (Xu and Chen, 2004) to
characterize the mean wind velocity from the measurement data. First
proposed in (Huang et al., 1998), EMD is a data-driven tool that can be
used to decompose non-stationary data into a basis set of intrinsic mode
functions (IMFs), denoted by cjðtÞ, through an iterative sifting process;
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

see, for example, (Xu and Chen, 2004): mentioned previously, the wind is measured at only eight locations along
X X the span (x ¼ f460; 280; 240; 200; 180; —10; —180; 420g m), so the
UðtÞ¼ Ncðj UÞðtÞþ rð UÞðtÞ; WðtÞ¼ Ncðj W Þ ðtÞþ rðWÞðtÞ (20) quantities in Eq. (22) are not fully known. Despite this limitation,
ðÞ ðÞ ðÞ ðÞ
U t forces.
ing u t and U t the
Here, w tshort-term
represent “averaged
the approXintensity”
imate intensity of the buffet-
is quantified by the
N N

help of a moving window root mean square (MW-RMS) metric, defined

j¼1 j¼1
The iterative extraction of the N IMFs continues until a stop criterion
is met. The final residuals rðUNÞðtÞ and rðNWÞðtÞ are treated as the time-
varying mean wind velocities UðtÞ and W ðtÞ, while the sum of the IMFs
ðÞ ðÞ
corresponds to u t and w t . The EMD stop criterion is formulated such
that an IMF should not have two consecutive extrema further than 30 min
apart; such behavior is viewed as a trend that is so low frequency that it is
due to a time-varying mean rather than turbulence. This corresponds to a
so-called intermittency check (Huang et al., 1999), where all IMFs consist
of frequencies greater than 1/1800 Hz. Since the wind velocity is a fairly
broadband process, the number of IMFs is typically N 6≈ 10.—
Figs. 11, 13, 15 and 17 show the wind velocity for anemometer A6 in
the middle of the bridge in recordings 1–4; similar results are obtained
for the other anemometers. The time-varying trends are captured quite
well. Adopting linearized quasi-steady buffeting theory (Scanlan, 1978a;
Jain et al., 1996), the buffeting nodal forces (f b ðx i ;tÞ2 R3) on the bridge
boX girder are approXimately proportional to the products UðtÞuðtÞ and
UðtÞwðtÞ at the coordinate xi 2 ½ —L =2; L =2] along the span L ¼ 1310 m:
2 observed.
2 3
fb;yðxi; tÞ ρBUðxi; tÞLi
fbðxi; tÞ¼ 4 fb;zðxi; tÞ 5 ¼ 2
4 2ðD=BÞCD ðD=BÞC’D — CL2CLC’L
fb;θðxi; tÞ
3
þ ðD=BÞC 2BC
M BC’ M 5
uðxi ; tÞ
D
wðxi ; tÞ
(21)
where ρ is the air density, Li is an influence length, and B and D are the
deck width and height, respectively. The overbar and prime denote the
center and gradient in the linearization of the aerodynamic coefficients
for drag, lift and moment around the mean angle of attack. The effects of
the admittance functions are not considered here. Equivalently, the total
modal buffeting forces become:
pb ðtÞ ¼ ΦT S b ½f bðx 1; tÞ ⋯ f bðx i; tÞ ⋯ f bðx M; tÞ ] T (22)
T T T
with Sb 2 RnDOF×3M selecting the set of M loaded nodal DOFs. As
for an arbitrary function aðtÞ as the integral:
sffiffiffiffiffiffiffiZffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
MW — RMS½aðtÞ] ¼ aðτÞ2 dτ (23)
tþTw =2
1
Tw t—Tw =2
where the centralized window is set to Tw ¼ 60 s. For a fair comparison
with the forces, the time series for uð tÞ and wðtÞ are also high-pass filtered
at 0.01 Hz in this calculation.
Figs. 12, 14, 16 and 18 show the MW-RMS for the wind turbulence
products UðtÞuðtÞ and UðtÞwðtÞ at anemometers A1-A8, together with the
b
identified modal forces (pðtÞ). For the latter, the MW-RMS is normalized
to unity for the individual modal component because magnitude differ-
ences prohibit a clear comparison. Note that the turbulence illustrations
in these figures do not tell the full story since their net effect on the bridge
will depend on the spatial correlations in relation to the shape of the
individual mode. Still, some interesting concurrent trends can be
In recording 1 (Fig. 12), the wind comes from the eastern sector θ 2
½45∘; 90∘]. The onset of the skew-most wind around t ≈ 1 : 30 h also
coincides with the large turbulence, in particular for A1-A5, which can
likely be explained by the path of the wind across the tall mountains on
the south side of the fjord (cf. Fig. 2). This skew wind and its load effects
are interesting since wind tunnels tests of section models with pro-
portions similar to the Hardanger bridge have indicated a significant
dependency on the skew angle and the angle of attack (Zhu et al., 2002).
However, from these load estimates, it is difficult to distinguish the ef-
fects related to a single environmental variable since the presence or
evolution of multiple variables cannot be controlled separately. In this
case, small load effects purely from the skewness are likely to be masked
by the simultaneously occurring strong turbulence, which that has a
larger influence on the buffeting forces. The clear inhomogeneities in the
wind field could also play a role. Earlier studies have indicated that in-
homogeneities occur due to the local topology, and in this regard, the
eastern and western winds tend to have different patterns (Lystad et al.,
2018).

Fig. 11. Measurements from anemometer A6 in recording 1: a) along-wind velocity; b) vertical wind velocity; c) direction.

8
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

Fig. 12. Comparison of the time evolution in recording 1: a) along-wind turbulence; b) vertical turbulence; c) identified modal forces.

Fig. 13. Measurements from anemometer A6 in recording 2: a) along-wind velocity; b) vertical wind velocity; c) direction.

9
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

Fig. 14. Comparison of the time evolution in recording 2: a) along-wind turbulence; b) vertical turbulence; c) identified modal forces.

Fig. 15. Measurements from anemometer A6 in recording 3: a) along-wind velocity; b) vertical wind velocity; c) direction.

10
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

Fig. 16. Comparison of the time evolution in recording 3: a) along-wind turbulence; b) vertical turbulence; c) identified modal forces.

Fig. 17. Measurements from anemometer A6 in recording 4: a) along-wind velocity; b) vertical wind velocity; c) direction.

11
Ø.W. Petersen et al.
Fig. 18. Comparison of the time evolution in recording 4: a) along-wind turbulence; b) vertical turbulence; c) identified modal forces.
In recording 2 (Fig. 14), a more homogeneous wind field is observed,
although the north side (A6-A8) appears to have slightly more turbu-
lence. Moreover, the influence of the turbulence for t2 0 ½ : 30; 2 : 30] h
clearly manifests in the loads. Later, some large loads occur seemingly
without any large turbulence at the time, for instance, at t ¼ approaches to check whether they are better suited. In this section, two
f 3 : 12; 4 : 42g h. The exact reason for these loads is not known, but a
closer examination of the anemometer time series for A1 and A2 in-
dicates some strong local gusts with duration of less than 5 s, which are
not easily detected with an averaging window of 60 s.
In recording 3 (Fig. 16), the load is almost zero for the first 15 min,
when the wind starts to increase. The influence of strong and along-span
correlated turbulence can be observed at several time instants. At t ¼ 2 :
20 h, the effect of an along-wind gust is clearly visible in the forces, and
again, more pronounced at t ¼4 : 20 h. Aided by strong vertical turbu-
lence, almost all the modal forces reach the maximum magnitude at the
same time. For reference, the maximum horizontal deflection in the mid-
span reaches almost 8 m, and the estimated stress in the boX girder at this
time reaches a maximum well over 100 MPa, which testifies to the harsh
impact felt by the structure. In the later period (t > 4 : 30 h), some strong
but not fully correlated gusts still occur, for instance, around t ≈ 5 : 30 h,
which still leads to some large forces but not for all modes.
In the last recording (Fig. 18), quite inhomogeneous wind conditions
are observed. Although the data from A1 are missing, the wind at A2-A6
(south side) is considerably more turbulent. Correlations in the wind and
forces are evident at the two indicated times t ¼ f 1 : 12; 2 : 35} h.
However, the strong forces occurring at t ¼ 2 : 00 h cannot be explained
directly by the turbulence in this figure.
12
Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045
3.3. Comparison with other algorithms and design wind load spectrum
The results in the previous sections show some potential issues in the
identification of the forces; therefore, it is interesting to compare other
additional inverse methods are considered. The first is a dual Kalman
filter (Azam et al., 2015), which is distinct due to its stable estimation of
the acceleration output only. In this algorithm, the force has a statistical
evolution model of the form pkþ1 ¼ pk þ ηk , where the fictitious process
ηk is a zero-mean white noise vector with covariance E½ηk ηTl ] ¼ Qp δkl .
This covariance is deemed a regularization parameter that controls the
force solution norm. In this case, a tuned value of Qp ¼ 10—1I yields the
best fit to the output data (minimum innovation norm).
Second, we apply a sequential deconvolution reconstruction (SDR) of
the inputs (Bernal and Ussia, 2015). While the JIS and DKF are based on
one-step recursive estimation, this method utilizes a sliding window for
time-domain deconvolution and is therefore computationally efficient.
The size of the window and the rate of advance are, respectively, set to
40Δt and 20Δt according to their definitions in (Bernal and Ussia, 2015).
Note that the SDR uses a deterministic system model, thus neglecting the
stochastic noise (Q; R; S 0).
¼
The results in Fig. 19 from the time period t 2 ½2 : 30; 3 : 00] in
recording 2 display a representative image of the performance. Inter-
estingly, the deterministic method SDR provides estimates that are
consistent with the JIS, which formally has a stronger capability due to
the consideration of stochastic noise. This outcome is believed to be the
result of the use of modal forces rather than localized forces. Since a
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

Fig. 19. Identified forces for modes 1–12 for a 30 min period of recording 2.

Fig. 20. Identified forces for modes 1–12 for a 30 min period of recording 2. The star signifies the single natural frequency, the magenta lines signify all natural
frequencies, and the green lines signify non-modeled modes. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web
version of this article.)

13
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

modal force theoretically is a summation of the forces in all positions, it ~ ae ¼ ΦT Cae Φ and K
~ ae ¼ ΦT Kae Φ. Utilizing z ðω Þ ¼ H
~ð ωÞ pbð ωÞ ,
where C
can more easily smooth model errors, whereas a localized force has in-
Eq. (25) can be written as:
fluence only in its vicinity. The DKF generally gives lower values, an
effect that increases for higher modes for reasons that are not well un- ~ ae þ K
~ ae ÞH
~ ðωÞÞpb ðωÞ
pðωÞ ¼ ðI þ ðiωC (26)
derstood. From Fig. 20, it can be concluded that the difference is largely
due to the low-frequency discrepancy. Another observation is that the where the generalized transfer function is
DKF does not have a dip at the natural frequency (added damping from ~ ðωÞ ¼ ½— ω2 I þ iωð 2ΞΩ — C ~ ae Þ þ ð Ω2 — K
~ aeÞ] —1 . Only wind loads on the
H
self-excitation) but instead has an increased magnitude, which is believe
girder are taken into account due to the lack of data on the wind char-
to be unrealistic. Scaling the covariance Qp changes the overall force
acteristics at the elevation of the main cables. Assuming the spectrum
magnitude but not the frequency-domain characteristics. A robust Sf bf b ð ω Þ2 CnDOF×nDOF of the buffeting forces (Eq. (21)) is available, the
method for choosing this covariance in scenarios with many forces n ×n
generalized force spectrum SppðωÞ2 C m m
can finally be calculated as:
(possibly having different evolutions) is not available at the moment.
Next, a brief comparison is made with the loads predicted from the ~ ae þ K
Spp ðωÞ ¼ ðI þ ðiωC ~ ae ÞH
~ ðωÞÞH ΦT S f bf bðωÞΦðI þ ðiωC
~ ae þ K
~ ae ÞH
~ ðωÞÞ
design wind field model based on a Kaimal spectrum (see (Fenerci and (27)
Øiseth, 2017) for details). Herein, the static coefficients are set to the
following values obtained from wind tunnel tests (Siedziako et al., 2017): Fig. 21 compares the force spectra from the forward prediction and
the inverse estimate using data from the time period2t ½2 : 30; 3 : 00] in
CD ¼ 1:05, C’D ¼ 0, CL ¼ — 0:363, C’L ¼ 2:22, CM ¼ — 0:017, C’M ¼
0:786. The bridge deck has the proportions B ¼ 18:3 m and D ¼ 3:25 m. recording 2, in which the wind is normal to the bridge. Averaging over
In this context, this classic wind load modeling approach is called the this 30 min period, the mean wind velocity is U ¼ 21:2 m/s, and the
“forward analysis”, since its basis is the wind velocity field yielding the turbulence intensities are Iu ¼ σu=U ¼ 0:096 and Iw ¼ σw=U ¼ 0:035.
loads on the bridge. In contrast, the inverse estimation of the wind loads Although the results inevitably vary for different events, the time period
is driven by the measured acceleration data. These two approaches are shown has reasonably ideal (homogeneous and stationary) wind condi-
fundamentally different, but both independently result in the wind load tions and should yield a fair comparison. However, the design wind field
quantification. In the forward analysis the wind loads can be described model has many uncertain features that do not always match the
only stochastically, so spectral densities in the frequency domain are observed wind characteristics at the site (Fenerci et al., 2017); thus, the
employed. The buffeting plus self-excited forces are defined as: presented frequency-domain inquiry is a simple test to get an idea of the
similarity and cannot be considered a strict tool for validation.
fðω Þ ¼ f b ðωÞ þ Cae u_ ðωÞ þ Kae uðωÞ (24) Overall, the degree of (dis)agreement in Fig. 21 is as expected for the
For simplicity, Cae and Kae are taken from quasi-steady theory in the inverse force identification of full-scale bridges in complex and uncertain
following (see e.g. (Kavrakov and Morgenthal, 2017)). The (total) conditions, which has not been extensively explored. For the JIS modal
force estimate, the magnitude appears to match best at the individual
generalized forces, pðωÞ ¼ ΦTfðωÞ, can then be written as:
natural frequency. In the frequency range far below the natural fre-
~ ae z_ ðωÞ þ K
pðω Þ ¼ pb ðωÞ þ C ~ ae zðωÞ (25) quency, the DKF estimate agrees better, and the JIS yields a higher
magnitude. This result could either be due to an estimation error or

Fig. 21. Comparison between the spectral density of the inverse load estimates (averaged by Welch’s method) and the forward model prediction of the wind loads.
The star and the magenta line signify the natural frequency.

14
Ø.W. Petersen et al.
actual behavior. As discussed in Section 2.3, acceleration outputs are not
always well suited for estimating very-low-frequency forces due to their
ω2-proportionality, which becomes small for close-to-static motion. The
insensitivity could lead to a (frequency-banded) low signal-to-noise ratio,
which is known to be particularly disruptive for inverse problems.
Therefore, displacements or strains are more useful for this purpose;
however, latent sensor noise is believed to play only a small role here.
The behavior can potentially be explained by some slow fluctuations in
the mean wind velocity. Such fluctuations are typically not captured in
the forward analysis that considers stationary conditions (constant mean
wind velocity).
3.4. Evaluation of methodology and usefulness of obtained load
information
No real direct validation of the estimated forces is available at this
scale, so the uncertainties are difficult to quantify. Promising progress
has been made in understanding the application to suspension bridges,
but the case study also shows some robustness issues with the estimation
algorithms, as well as the need for extensive sensor networks to achieve a
high level of confidence in the results. Regardless, application robustness
is, for this bridge, far better than that for a similar study on floating
bridges (Petersen et al., 2019), where the model uncertainties are far
greater.
Although the data of the load and response behavior can be useful, an
obvious dilemma is that measurements cannot be made before the
structure is built, at which point it has already been designed based on
wind tunnel testing and wind field analysis, and major alterations cannot
be conveniently made. Thus, the methodology is more suited to research
and validation studies on existing bridges or simply to monitoring
response behavior throughout the lifetime. The body of knowledge
gained from the lessons from existing bridges is also a valuable stepping
stone that could facilitate the design of future bridges.
The methodology is also promising for studies of the loads in more
unusual events that do not fit the ordinary set of wind modeling as-
sumptions. The load mechanisms for bridges in highly non-stationary
downbursts in thunderstorms, for example, are not well documented
(Chen and Letchford, 2005; Solari et al., 2015). In theory, the input and
state estimation could also be extended to parameter estimation to
indicate damage (stiffness reduction), but this task remains difficult
because modal sensitivity to local damage is generally very small for long-
span bridges.
4. Conclusions
This article has presented the inverse identification of dynamic wind
loads from measured response data, focusing on the Hardanger bridge, a
long-span suspension bridge instrumented with accelerometers. The
methodology does not impose any prior assumptions on the wind
loading, as loads are estimated directly from acceleration data using an
algorithm for joint input and state estimation.
Overall, the frequency-domain characteristics of the estimated loads
appear to be realistic, except for the effect of some known errors in the
reduced-order modal model. The magnitudes of the estimated loads are
higher than predicted directly from design wind field models. Buffeting
and self-excited forces are both distinctly present, but these factors are
estimated jointly and cannot currently be separated. When compared
with the anemometer data, clear simultaneous trends between the time
series of the estimated modal forces and wind turbulence are observed.
The results suggest that the methodology, if further developed and tested,
could be useful in the characterization of wind loads on bridges with
uncertain wind conditions.
Some practical and theoretical limitations lead to issues that should
be addressed. The number of modal forces that can be reconstructed is
limited by the output data. Since long-span bridges have many modes
15
Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045
with significant wind-induced excitation, a large and well-designed
sensor network is required to reconstruct the total dynamical behavior.
The estimation based on acceleration data also constitutes a limitation in
studies of wind loads, as close-to-static forces are not always well esti-
mated. This source of uncertainty could be eliminated if strain mea-
surements were added.
Acknowledgments
This work was financially supported by the Norwegian Public Roads
Administration.
References
Amiri, A.K., Bucher, C., 2017. A procedure for in situ wind load reconstruction from
structural response only based on field testing data. J. Wind Eng. Ind. Aerodyn. 167,
75–86.
Aucejo, M., De Smet, O., Deü, J.F., 2018. On a space-time regularization for force
reconstruction problems. Mech. Syst. Signal Process. 118, 549–567.
Aucejo, M., De Smet, O., Deü, J., 2018. Practical issues on the applicability of Kalman
filtering for reconstructing mechanical sources in structural dynamics. J. Sound Vib.
442, 45–70.
Azam, S.E., Chatzi, E., Papadimitriou, C., 2015. A dual Kalman filter approach for state
estimation via output-only acceleration measurements. Mech. Syst. Signal Process.
60, 866–886.
Bernal, D., Ussia, A., 2015. Sequential deconvolution input reconstruction. Mech. Syst.
Signal Process. 50, 41–55.
Bietry, J., Delaunay, D., Conti, E., 1995. Comparison of full-scale measurement and
computation of wind effects on a cable-stayed bridge. J. Wind Eng. Ind. Aerodyn. 57,
225–235.
Caracoglia, L., 2008. Influence of uncertainty in selected aerodynamic and structural
parameters on the buffeting response of long-span bridges. J. Wind Eng. Ind.
Aerodyn. 96, 327–344.
Chang, X., Yan, Y., Wu, Y., 2019. Study on solving the ill-posed problem of force load
reconstruction. J. Sound Vib. 440, 186–201.
Chen, X., Kareem, A., 2002. Advances in modeling of aerodynamic forces on bridge decks.
J. Eng. Mech. 128, 1193–1205.
Chen, L., Letchford, C., 2005. Proper orthogonal decomposition of two vertical profiles of
full-scale nonstationary downburst wind speeds [lzcl]. J. Wind Eng. Ind. Aerodyn. 93,
187–216.
Cheynet, E., Jakobsen, J.B., Snæbjo € rnsson, J., 2016. Buffeting response of a suspension
bridge in complex terrain. Eng. Struct. 128, 474–487.
Cheynet, E., Jakobsen, J.B., Snæbjo € rnsson, J., Reuder, J., Kumer, V., Svardal, B., 2017.
Assessing the potential of a commercial pulsed lidar for wind characterisation at a
bridge site. J. Wind Eng. Ind. Aerodyn. 161, 17–26.
Cigada, A., Falco, M., Zasso, A., 2001. Development of new systems to measure the
aerodynamic forces on section models in wind tunnel testing. J. Wind Eng. Ind.
Aerodyn. 89, 725–746.
Diana, G., Resta, F., Zasso, A., Belloli, M., Rocchi, D., 2004. Forced motion and free
motion aeroelastic tests on a new concept dynamometric section model of the
Messina suspension bridge. J. Wind Eng. Ind. Aerodyn. 92, 441–462.
Fenerci, A., Øiseth, O., 2017. Measured buffeting response of a long-span suspension
bridge compared with numerical predictions based on design wind spectra. J. Struct.
Eng. 143, 04017131.
Fenerci, A., Øiseth, O., Rønnquist, A., 2017. Long-term monitoring of wind field
characteristics and dynamic response of a long-span suspension bridge in complex
terrain. Eng. Struct. 147, 269–284.
Huang, N.E., Shen, Z., Long, S.R., Wu, M.C., Shih, H.H., Zheng, Q., Yen, N.-C., Tung, C.C.,
Liu, H.H., 1998. The empirical mode decomposition and the Hilbert spectrum for
nonlinear and non-stationary time series analysis. Proc. R. Soc. Lond.: Math. Phys.
Eng. Sci. 454, 903–995.
Huang, N.E., Shen, Z., Long, S.R., 1999. A new view of nonlinear water waves: the Hilbert
spectrum. Annu. Rev. Fluid Mech. 31, 417–457.
Hwang, J., Kareem, A., Kim, H., 2011. Wind load identification using wind tunnel test
data by inverse analysis. J. Wind Eng. Ind. Aerodyn. 99, 18–26.
Jacquelin, E., Bennani, A., Hamelin, P., 2003. Force reconstruction: analysis and
regularization of a deconvolution problem. J. Sound Vib. 265, 81–107.
Jain, A., Jones, N.P., Scanlan, R.H., 1996. Coupled flutter and buffeting analysis of long-
span bridges. J. Struct. Eng. 122, 716–725.
Jakobsen, J., Tanaka, H., 2003. Modelling uncertainties in prediction of aeroelastic bridge
behaviour. J. Wind Eng. Ind. Aerodyn. 91, 1485–1498.
Kang, N., Kim, H., Choi, S., Jo, Seongwoo, Hwang, J.-S., Yu, E., 2012. Performance
evaluation of TMD under typhoon using system identification and inverse wind load
estimation. Comput. Aided Civ. Infrastruct. Eng. 27, 455–473.
Kavrakov, I., Morgenthal, G., 2017. A comparative assessment of aerodynamic models for
buffeting and flutter of long-span bridges. Engineering 3, 823–838.
Ko, J., Ni, Y., 2005. Technology developments in structural health monitoring of large-
scale bridges. Eng. Struct. 27, 1715–1725.
Lai, T., Yi, T.-H., Li, H.-N., Fu, X., 2017. An explicit fourth-order Runge–Kutta method for
dynamic force identification. Int. J. Struct. Stab. Dyn. 1750120.
Larsen, A., Larose, G.L., 2015. Dynamic wind effects on suspension and cable-stayed
bridges. J. Sound Vib. 334, 2–28.
Ø.W. Petersen et al.
Li, K., Liu, J., Han, X., Sun, X., Jiang, C., 2015. A novel approach for distributed dynamic
load reconstruction by space–time domain decoupling. J. Sound Vib. 348, 137–148.
Liu, Y., Shepard Jr., W.S., 2005. Dynamic force identification based on enhanced least
squares and total least-squares schemes in the frequency domain. J. Sound Vib. 282,
37–60.
Lourens, E., 2012. Force Identification in Structural Dynamics. Ph.D. thesis. Katholieke
Universiteit Leuven - Faculty of Engineering, Leuven (Belgium).
Lourens, E., Reynders, E., De Roeck, G., Degrande, G., Lombaert, G., 2012. An augmented
Kalman filter for force identification in structural dynamics. Mech. Syst. Signal
Process. 27, 446–460.
Lourens, E., Papadimitriou, C., Gillijns, S., Reynders, E., De Roeck, G., Lombaert, G., 2012.
Joint input-response estimation for structural systems based on reduced-order models
and vibration data from a limited number of sensors. Mech. Syst. Signal Process. 29,
310–327.
Lystad, T.M., Fenerci, A., Øiseth, O., 2018. Evaluation of mast measurements and wind
tunnel terrain models to describe spatially variable wind field characteristics for long-
span bridge design. J. Wind Eng. Ind. Aerodyn. 179, 558–573.
Ma, C.K., Chang, J.M., Lin, D.C., 2003. Input forces estimation of beam structures by an
inverse method. J. Sound Vib. 259, 387–407.
Macdonald, J.H., 2003. Evaluation of buffeting predictions of a cable-stayed bridge from
full-scale measurements. J. Wind Eng. Ind. Aerodyn. 91, 1465–1483.
Maes, K., Lourens, E., Van Nimmen, K., Reynders, E., De Roeck, G., Lombaert, G., 2014.
Design of sensor networks for instantaneous inversion of modally reduced order
models in structural dynamics. Mech. Syst. Signal Process. 52, 628–644.
Maes, K., Gillijns, S., Lombaert, G., 2018. A smoothing algorithm for joint input-state
estimation in structural dynamics. Mech. Syst. Signal Process. 98, 292–309.
Naets, F., Cuadrado, J., Desmet, W., 2015. Stable force identification in structural
dynamics using Kalman filtering and dummy-measurements. Mech. Syst. Signal
Process. 50, 235–248.
Niu, Y., Fritzen, C., Jung, H., Buethe, I., Ni, Y.-Q., Wang, Y.-W., 2015. Online
simultaneous reconstruction of wind load and structural responses: theory and
application to Canton tower. Comput. Aided Civ. Infrastruct. Eng. 30, 666–681.
Petersen, Ø.W., Øiseth, O., 2019. Finite element model updating of a long span
suspension bridge. In: Rupakhety, R., Olafsson, S., Bessason, B. (Eds.), Proceedings of
the International Conference on Earthquake Engineering and Structural Dynamics.
Springer International Publishing, pp. 335–344.
16
Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045
Petersen, Ø.W., Øiseth, O., Lourens, E., 2019. Full-scale identification of the wave forces
exerted on a floating bridge using inverse methods and directional wave spectrum
estimation. Mech. Syst. Signal Process. 120, 708–726.
Qiao, B., Zhang, X., Gao, J., Chen, X., 2016. Impact-force sparse reconstruction from
highly incomplete and inaccurate measurements. J. Sound Vib. 376, 72–94.
Rezayat, A., Nassiri, V., De Pauw, B., Ertveldt, J., Vanlanduit, S., Guillaume, P., 2016.
Identification of dynamic forces using group-sparsity in frequency domain. Mech.
Syst. Signal Process. 70, 756–768.
Scanlan, R., 1978. The action of flexible bridges under wind, II: buffeting theory. J. Sound
Vib. 60, 201–211.
Scanlan, R., 1978. The action of flexible bridges under wind, I: flutter theory. J. Sound
Vib. 60, 187–199.
Siedziako, B., Øiseth, O., Rønnquist, A., 2017. An enhanced forced vibration rig for wind
tunnel testing of bridge deck section models in arbitrary motion. J. Wind Eng. Ind.
Aerodyn. 164, 152–163.
Solari, G., De Gaetano, P., Repetto, M.P., 2015. Thunderstorm response spectrum:
fundamentals and case study. J. Wind Eng. Ind. Aerodyn. 143, 62–77.
Song, W., 2018. Generalized minimum variance unbiased joint input-state estimation and
its unscented scheme for dynamic systems with direct feedthrough. Mech. Syst.
Signal Process. 99, 886–920.
Sun, H., Feng, D., Liu, Y., Feng, M.Q., 2015. Statistical regularization for identification of
structural parameters and external loadings using state space models. Comput. Aided
Civ. Infrastruct. Eng. 30, 843–858.
Wang, H., Hu, R., Xie, J., Tong, T., Li, A., 2012. Comparative study on buffeting
performance of Sutong Bridge based on design and measured spectrum. J. Bridge
Eng. 18, 587–600.
Wong, K., 2004. Instrumentation and health monitoring of cable-supported bridges.
Struct. Control Health Monit. 11, 91–124.
Xu, Y., Chen, J., 2004. Characterizing nonstationary wind speed using empirical mode
decomposition. J. Struct. Eng. 130, 912–920.
Zhi, L., Li, Q., Fang, M., 2016. Identification of wind loads and estimation of structural
responses of super-tall buildings by an inverse method. Comput. Aided Civ.
Infrastruct. Eng. 31, 966–982.
Zhi, L., Yu, P., Li, Q.-S., Chen, B., Fang, M., 2018. Identification of wind loads on super-tall
buildings by Kalman filter. Comput. Struct. 208, 105–117.
Zhu, L., Xu, Y., Zhang, F., Xiang, H., 2002. Tsing Ma bridge deck under skew winds - Part
I: aerodynamic coefficients. J. Wind Eng. Ind. Aerodyn. 90, 781–805
Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

KEKURANGAN :
Penilaian beban angin pada jembatan bentang panjang biasanya didasarkan pada model asumsi
mengenai medan angin serta koefisien aerodinamis yang diperoleh dari hasil pengujian di
terowongan angin. Namun, pendekatan ini seringkali menghadapi ketidakpastian yang signifikan.
Hasil penelitian terkini menunjukkan adanya perbedaan besar antara respons dinamis yang diprediksi
oleh model dengan respons yang diamati pada jembatan gantung akibat pengaruh angin.

KELEBIHAN :
Dalam artikel ini, beban angin dinamis pada jembatan Hardanger diestimasi berdasarkan data
percepatan. Kemampuan estimasi lapangan penuh dari algoritma-algoritma ini untuk jembatan
rentang panjang telah dipelajari sebelumnya, menunjukkan bahwa metodologi ini dapat dilakukan,
dengan batasan utama adalah jumlah respons modal yang dapat direkonstruksi.

PEMBEBANAN :
➢ Pembebanan jembatan adalah proses menentukan gaya-gaya yang bekerja pada struktur
jembatan. Dalam jurnal ini, terdapat dua pendekatan yang digunakan untuk pembebanan
jembatan: pendekatan analisis maju (forward analysis) dan pendekatan identifikasi balik
(inverse identification).
➢ Pendekatan analisis maju (forward analysis) didasarkan pada model angin yang menghasilkan
beban pada jembatan. Model ini menggunakan spektrum kecepatan angin dan koefisien
aerodinamika untuk menghitung beban angin pada jembatan. Pendekatan ini sering digunakan
dalam desain jembatan dan menghasilkan beban angin yang dinyatakan secara stokastik
dalam domain frekuensi
➢ Pendekatan identifikasi balik (inverse identification) adalah pendekatan yang digunakan
dalam jurnal ini. Pendekatan ini tidak mengasumsikan model beban tertentu, tetapi
menggunakan data percepatan yang diukur pada jembatan untuk mengestimasi gaya angin
yang berkontribusi pada respons jembatan. Metode identifikasi gaya angin ini melibatkan
pemodelan invers menggunakan data percepatan yang diukur dan model sistem yang
diketahui. Dalam metode ini, gaya angin diperoleh melalui estimasi vektor gaya modal
berdasarkan data percepatan yang diukur

STRUKTUR :
➢ Struktur jembatan yang dibahas dalam jurnal ini adalah Jembatan Hardanger di Norwegia.
Jembatan ini merupakan jembatan gantung dengan bentang utama sepanjang 1310 meter.
Jembatan ini terletak di daerah pegunungan dengan kondisi lingkungan yang kompleks.
➢ Jembatan Hardanger memiliki fitur bentang utama yang panjang dan ramping, sehingga
sensitif terhadap angin yang datang dari daerah pesisir di sebelah barat atau daerah
pegunungan di sebelah timur. Untuk memantau respons dinamis jembatan, sistem pemantauan
struktural telah dioperasikan sejak tahun 2013. Sistem ini terdiri dari anemometer ultrasonik
digital yang dipasang di hanger jembatan dan akselerometer triaksial yang terletak di balok
dan menara jembatan

PERENCANAAN STRUKTUR :
Tidak ditemukan

MANFAAT :
➢ Memahami karakteristik medan angin
Penelitian ini memberikan pemahaman yang lebih baik tentang karakteristik medan angin di
sekitar jembatan Hardanger di daerah pegunungan yang kompleks.
17
 Ø.W. Petersen et al. Journal of Wind Engineering & Industrial Aerodynamics 196 (2020) 104045

➢ Pemantauan respons dinamis jembatan

Penelitian ini menggambarkan penggunaan sistem pemantauan struktural untuk memantau
respons dinamis jembatan Hardanger terhadap beban angin.
➢ Peningkatan desain jembatan.
Penelitian ini memberikan wawasan yang berharga untuk perencanaan dan desain jembatan
panjang dan ramping di daerah dengan kondisi lingkungan yang kompleks
➢ Pengembangan metode pembebanan jembatan
Penelitian ini dapat membantu dalam pengembangan metode pembebanan jembatan yang
lebih akurat dan dapat diandalkan, terutama dalam memperkirakan beban angin.

CONTOH JEMBATAN :

Jembatan Gantung Tulungagung.

18
 JEMBATAN CABLE STAYED

Contents lists available at ScienceDirect

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

Structural performance of a steel cable-stayed bridge under blast loading

considering different stay patterns
Cyrille Denis Tetougueni, Paolo Zampieri⁎, Carlo Pellegrino
Department of Civil, Environmental and Architectural Engineering, University of Padova, Via Marzolo 9, 35131 Padova, Italy

ART I C L E I NFO ABST RA C T

Keywords: Accidental or extreme events can induce abnormal loads on structures contributing to a local damage of primary
Blast loading components or to the progressive collapse of the structure. Although hazards, which are usually difficult to be
Cable-stayed bridge predicted, increase the complexity in the design, the need to determine the consequences on structures of these
Dynamic analysis extreme events remains significant. This study aims to evaluate the damage induced by an accidental load such
Collapse mechanism as blast loading to a cable-stayed bridge. In the first part of the paper, through non-linear dynamic analyses, it
Blast fragility
was possible to evaluate the structural response of three configurations of cable-stayed bridges, then to figure out
the possible immediate and forthcoming damages considering different loading parameters and position. Finally,
a displacement-controlled static non-linear pushover analysis of the structure has been performed to evaluate the
limit state of the structure for a random position of blast loading and to predict the possible damage state during
an accidental event such as blast loading. The results indicate significant effects of the structural configuration of
the bridge and stay patterns. The Fan cable-stayed bridge is found to be the most effective among the considered
configurations in reducing the dynamic effect induced by the blast loading on the structure. In addition, the deck
is found to be less critical in terms of blast load mitigation close to the abutment and the pylon.

1. Introduction In the worldwide existing codes, the philosophy in the design is

Structural robustness and collapse resistance are nowadays relevant
research topics both in the design of new structures and for the safety
assessment of existing ones. For most cases, a structure is considered
safe if the design actions loading is less or equal to the design capacity
of the structure. Another design approach consists of evaluating the
structural performance of a structural system under the various condi-
tions associated with hazard events. This approach becomes adequate
in the case of accidental loads derived from events such as crash,
earthquake, collision, blast, etc… where the design actions are hard to
be predicted in the design life of the structure. Moreover, the greater
attention toward understanding the phenomena associated to blast
loading and the assessment of the damage to structures or structural
components under such threats observed in recent years is particularly
justified by an increase of manmade and accidental explosions world-
wide [1–3]. In these circumstances, different causes of the collapse of
the structure are considered. Either the structure suddenly collapses
due to the damage of a key structural component (piers, slab, deck, etc.)
or the failure of the structural system is triggered by the local damage of
a primary structural elements (cables, column, pylon, etc.) known as
zipper-type collapse [4–6].
linked to the situation that structural damage of a structural primary
component against abnormal loading condition may be allowed to
occur but without triggering the collapse of the entire structure. Thus,
not only the structural response for conventional loads (self-weight, live
loads, wind loads, etc.) should be focused on but it is becoming urgent
to study intensely the response after abnormal loads such as a hurri-
cane, wind, flood, blast loading, etc. For this reason, several scientists
have begun to study the ability of a structure to limit the redistribution
of the forces of the structural elements adjacent to the element exposed
to an accidental load or to mitigate the effect of damage [5,7,8]. These
measures contribute to creating a robust structure able to resist against
extreme events. However, the robustness and progressive collapse of a
structure are two concepts closely linked. As a matter of fact, some
researchers have studied the possible progressive collapse of structures
during hazard events. Thus, the case involving a progressive collapse of
a typical steel building has been studied in [6,9]. Recent advances in
research on the progressive failure of reinforced concrete structures
have been presented in [10,11] whereas the case of cable-stayed
bridges has been studied in [4,12]. A tremendous research activities to
investigate the damage criteria of suspended bridges subjected to
aeroelastic wind effects are presented in [13,14] whereas the influence

⁎ Corresponding author.
E-mail address: paolo.zampieri@dicea.unipd.it (P. Zampieri).

https://doi.org/10.1016/j.engstruct.2020.110739
Received 28 August 2019; Received in revised form 27 April 2020; Accepted 1 May 2020
01 41-0296 /© 202 0E lse
v ie
rL td.Allrightsreserved.
C.D. Tetougueni, et al.
of the cable damage on the dynamic response of suspended bridges has
been investigated in [15–17]. In addition, a study of the dispropor-
tionate collapse mechanism of an existing cable-stayed bridge with the
aim to improve the robustness of the structure after a severe seismic
event is found in [18]. Finally, a new system of cable-stayed bridges has
emerged in the recent three decades. The under-deck cable-stayed
bridges (UDCSB) present a better structural behavior, an economical
advantage and arguably strong aesthetic characteristics than the con-
ventional cable-stayed bridges [19]. A recent study showed that, the
dynamic response of such bridges is influenced by the load speed, the
depth of the deck, as well as the structural section [20] and revealed
how important an investigation of this innovative arrangement of the
bridge under blast threats should become a matter of concern.
In a very small number of previous studies, structural safety against
blast loading that can lead to progressive collapse has been assessed in a
probability-based approach. Even though the studies are mainly fo-
cused on earthquake engineering, it is observed nowadays a significant
advance in the case of blast loading [21–25]. Many studies to define
fragility curves have been conducted either on structural elements
[26–28] or on buildings [29–32]. The study of fragility in the case of
bridges subjected to blast loading is almost non-existent in the scientific
literature. Indeed, many authors have rather addressed their researches
to study the structural response or performance of bridges [33,34]. In
particular, Hashimi et al. [35], Tang and Hao [25] presented a sa-
tisfactory research on the structural performance of the bridge’s com-
ponents for a cable-stayed bridge subjected to blast loading throughout
a dynamic analysis. The first authors focused on the variation of the
blast loading location on the bridge deck and studied also the cable loss
mechanism after a blast event while the second document is con-
centrated on the influence of the size of the blast loading source, the
stand-off distance and presented also the effectiveness of FRP
strengthening techniques after blast loading.
The review reveals that abnormal loads can lead to the critical da-
mage of structural components that could reach adjacent structural
elements and ultimately cause a collapse of the structure. Based on this
observation, the authors have found a clear motivation to focus their
work on the study of the structural performance of cable-stayed bridges
subjected to blast loads. However, the stay patterns might play an im-
portant role in the diffusion of the damage between the bridge’s com-
ponents and should be investigated in depth. Thus, the purpose of this
research is to establish direct or indirect links between the collapse
capacity of the structural elements of various cable-stayed bridges
(Harp, Fan, and Semi-fan stay patterns) subjected to abnormal loads
such as blast loading and the possible collapse of the bridge. To reach
our objectives, the structural performance of three different cable-
stayed patterns will be assessed thoroughly and only the dynamic effect
deriving from a blast loading (neglecting the air blast released and its
consequences on cars’ glazing and people to simplify the study) will be
considered. In particular, intensive static and dynamic analyses have
been carried out on various cable-stayed bridges in order to determine a
Fig. 1. Geometry and Section of structural elements of the cable-stayed bridge. a) Closed boX section for pylon, transverse beam and base pylon b) Deck section.
2
E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9
relationship between load intensity and structural response, then be-
tween structural response and the possibility of the collapse of the
bridge. Finally, an investigation of the likelihood of exceedance of the
structural capacity of the bridge’s structural components subjected to
blast loading will be presented. In the first part of the study, only the
cases of blast loading taking place on the deck will be considered.
Indeed, since vehicles are considered as the primary transportation
mean of the explosive item, the blast loading occurring is more sus-
ceptible to impact the bridge deck than other structural elements. The
document will be subdivided into three sections besides the introduc-
tion and conclusion. A description of the bridge will be presented
firstly, followed by the numerical modeling, numerical analysis, and
discussion of the results and finally, a section on the blast fragility will
be presented. In the companion study, the authors will focus on actions
on the pylon and piers. Satisfactory results from the analyses show that
the Fan configuration presents a better resistance in the case of blast
loading event even though it displays the most rigid structure. Among
other things, the possibility of total structural failure would arise be-
cause the dynamic action of the load causes not only excessive damage
to the deck but also to the stay cables adjacent to the impact area.
2. FEM analysis
2.1. Bridge benchmark Description
The cable-stayed bridge is nowadays widely used around the world
owing to the great advantages it displays for long-span bridges. A
generic cable-stayed bridge consists on deck, pylons and inclined cables
known as stays, which serve as internal support and provide additional
vertical stiffness to the deck. In this study, three different typologies of
cable-stayed bridge namely fan, semi-fan and harp cable-stayed bridges
have been considered. They differ mainly in the pattern of the cable
stays. A generic bridge under this study indeed consists of two lateral
spans of 99.2 m long and one central span of 204.6 m long. The bridge
deck is made of 1.15 m deep and 13.0 m wide steel orthotropic boX
girder, which includes three traffic lanes and two footways. It consists
of a closed hexagonal steel boX section with 20 mm thick flanges and
webs. In addition, trapezoidal stiffeners are installed on all the section
of the deck spaced 350 mm from center to center while vertical stif-
feners each 2750 mm were provided in the deck section to limit ex-
cessive displacement and increase the buckling resistance. Rectangular
closed boX sections have also been considered for the pylon, the
transverse beam, and the base pylon. The pylon is considered as the
vertical element above the steel deck whereas the base pylon is taken as
a vertical element below the steel deck. Two transverse beams are used
to connect the vertical elements and form with the vertical element the
“H” shaped pylon. The geometry and geometrical characteristics of the
structural elements are given in Fig. 1 and Table 1.
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Table 1
Dimensions of the structural elements of the cable-stayed bridge.

Transverse beam Pylon Base Pylon Steel Deck Unit

b 1800 2800 3200 13,000 mm

h 2800 2800 3200 1150 mm
t 40 40 40
A 441,600 521,600 585,600 642331,3 mm 2
ixx 4.88 × 1011 6.41 × 1011 9.47 × 1011 1.59 × 1011 mm 4
iyy 2.37 × 1011 6.41 × 1011 9.47 × 1011 9.37 × 1011 mm 4

2.2. Blast loading model

From the structural design point of view, a blast loading carried by

vehicles is considered as the relevant case since vehicles are able to
carry a large number of explosives. Blast loading compared with other
accidental events such as earthquake, impact, wind, releases a much
higher dynamic pressure than other phenomena [36]. Although the
pressure developed after the explosion attenuates quickly with the time,
it greatly contributes to localized damage to the exposed structural
members. Deng and Jin [37] stated that the maximum effect during
blast loading is observed in the positive phase since the structural mass
participating in dynamic response is engaged later due to the short
duration of the blast. When assessing the structural blast resistance, one
of the most important things is the correct modeling of the blast
loading. Indeed, it is difficult and quite complex to study the structural
response of the bridge’s component subjected to an explosive because
the ignition of the explosion and the interaction between bridge and
shock wave are included. During an explosion, energy is rapidly re-
leased, which generates a blast wave. In the consideration of conven-
tional explosive such as TNT, the process of energy released is a re-
arrangement of its atoms [3]. Many scientists [3,25,37] studied the
modeling of the blast wave and they stated that the characteristics of
blast waves are strongly dependent on the explosive material, the dis-
tance of the explosive source and how the generated blast wave is
transferred to the structure. In the blast analysis, the scaled distance
(Eq. (1)) is considered as the parameter that most affects the peak
pressure acting on structures [38].
R high vertical stiffness were used as bearing at both bridge end and
Z=
1/3
WTNT
Where Z is the scaled distance in m/kg1/3, R is the stand-off distance
known as the distance between the center of the source and the point of
impact on the structure in [m] and W the amount of the equivalent
weight of TNT in [kg]. Many scientists used empirical approaches to
develop simplified expressions of the peak overpressure released from a
blast event [39]. Friedlander’s equation is widely used to express the
relationship between the pressure intensity versus the time (Eqs. (2)
and (3)).
⎛ t − ta ⎞ −b ((t−ta)/td)
Pr (t) = Patm + Pmax ⎜
·e ⎟
⎝ td ⎠
1 1772 114
P = ·⎡ − + 108⎤
max
Z ⎣ Z
2
Z ⎦
Where Patm is the ambient pressure, ta, td are respectively the arrival
time and the duration time of the positive phase as shown in Fig. 2.
2.3. FE modelling
For such types of bridges made of steel elements, diaphragms are
used to increase the stiffness of the elements. The structural elements
for numerical simulation can then be modelled using 1D fiber beam
element since the variation in the section due to the torsional stress is
prevented. In the analysis, the structural elements are connected to
each other by a fiXed joint with zero degrees of freedom as is done in
Fig. 2. Blast pressure distribution over time.
the practice of such bridges. The base of the pylons is considered fully
restrained in rotation and translation since they are supposed to be
embedded in a massive concrete foundation at their bases. The end
spans of the bridge are connected to the end spans substructure
(abutments) with bearings. It should be highlighted that at the first
abutment, a fiXed bearing accompanying a transversal free directional
bearing was used whereas, at the second abutment, a free multi-
directional bearing with a longitudinally free bearing served to allow
the longitudinal displacement of the deck. Finally, at the internal part
of the bridge, only vertical restraint was considered for intermediate
spans. Cable stays are directly connected to pylons while rigid links are
used to create the appropriate connection between the deck and cables.
Two situations were adopted for the stays to cancel the transmission of
bending moment between the cable elements and both deck and pylons.
Initially, cables were considered as truss elements with a linear elastic
behaviour and then, to reflect the bridge’s cable performance [18],
another analyses with a tension-only cut-off bar material were per-
formed characteristics. Finally, for simplicity, abutments were not
modeled and were replaced by fiXed constraints. Then, elastic links with
above the transversal beam. High strength steel material was con-
(1)
sidered for deck, pylons, and piers whereas the stress distribution was
considered elastic in the cables and was limited to 0.55 × fu to satisfy
the fatigue criteria. On the other hand, to consider the cable’s sag effect
due to the change in the shape under varying stresses, Ernst’s equation
[40] has been used to derive the equivalent elastic modulus of the
cable. Perfect elastoplastic stress–strain diagrams for different elements
excepted cables were used in the analysis. Fig. 3 shows the perspective
view of the bridges under study, the longitudinal profile and the
transversal view of the pylon.
2.4. Material characteristics
(2)
The material model used for deck, pylon, and transverse beams
(3) follows the perfect elastoplastic law whereas the stays follow an elastic
law. It should be emphasized that the nonlinear geometric behavior of
the cable has been taken into account through the change in the cables’
elastic modulus. A summary of constitutive parameters is given in
Table 2. However, it must be pointed out that blast loading is char-
acterized as a high impact loading, which rapidly increases the strain
rate of the structural member to which the load is applied. Many au-
thors [6,22,26] stated that concrete and steel materials stresses are
modified and their ultimate and yield stresses increase due to the high
strain rate. The most common equation to consider the effect of the
strain rate is known as the Cowper-Symonds equation defined in Eq. (4)
where f d is the dynamic yield stress under high impact rate, fyd is the
yd
static yield stress, ε ̇is the strain rate in sec-1, and C and q are constants.

3
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Fig. 3. Geometrical characteristic of the bridges. a) Harp bridge; b) Fan bridge; c) Semi-Fan bridge; d) Longitudinal profile of the bridge; e) Transversal section of the
pylon from the left to the right Harp, Fan and Semi-fan system.

4
C.D. Tetougueni, et al.
Table 2
Material characteristics of the structural members.
Steel grade Structural Constitutive Yield stress Failure
element model [MPa] strain
S420 Deck Elasto-plastic 420 5%
S355 Pylon, Elasto-plastic 355 4%
Transverse beam
SWRH72B Cable Elastic 1023 4.8‰
Moreover, C = 40.4 and q = 5 for low strength steel while C = 3500
and q = 5 for high strength steel will be considered as defined in [41].
d action as shown in Fig. 4 and may accelerate secondary damage such as
f yd
= 1.0 + ⎛ ε ̇⎞
1/q
fyd ⎝C ⎠
2.5. Load analysis
In this paper, blast loads are considered as the leading variable load
whereas traffic loads are taken as accompanying actions. The self-
weight is directly provided from the software and it is a function of the
material density and the characteristics of the section. A uniformly
distributed load G2 = 48.7 kN/m is applied along the deck which re-
presents asphalt layer, safety barrier, and waterproofing layer. The pre-
tension force on the stays is calculated throughout an optimization
process in order to compensate 95% of the permanent loads with the
maximum displacement limited to 10 mm. Traffic loads are defined
according to European norms [42]. To reach our goal, load combina-
tions for accidental design situations as defined in [43] were used
where the partial coefficients are given within the code.
3. Numerical analysis and discussion
Owing to the lack of existing experimental models of such kind of
bridges, it was not able to compare the preliminary static analysis to
experimental results. The first stage of analysis consisted of determining
the actual stress of the bridge in the traffic condition. Then, a series of
dynamic non-linear analyses have been carried out on the three dif-
ferent typologies of cable-stayed bridges under study. The nonlinear
dynamic analysis of the blast action is performed on bridge models in
the stressed and deformed shape from the permanent and traffic loads.
It should be added that the implicit approach of the FEM software
package was used in the analysis. However, with this technique, the
onset of the element deletion is always difficult to implement within the
finite element analysis. The use of an explicit solver method may
overcome the situation even though displaying disproportionate results
compared to the real situation. Initially, a direct comparison is made
between the results of displacements, strains, and stresses obtained for
the three types of configurations. Then, the most unfavorable config-
uration in terms of stress is studied individually, varying different lo-
cations of the impact with the deck.
3.1. Stress analysis
From the first analyses and considering the global reference system
of the bridge defined in Fig. 3, the following observations are made. For
the blast loading occurring at x = 6.2 m from the abutment, and the
stand-off distance R = 1 (see Fig. 4), the maximum displacements
reached by the deck in a short time history after the blast detonation is
summarized in Table 3. The results show that the displacement reaches
intolerable levels for an equivalent weight greater than 750 kg. In
particular, the maximum displacement of the deck is observed for the
Harp configuration regardless of the size of the blast loading source. It
is also found that the difference in the structural response is obvious for
the weight larger than 750 kg while the displacement is quite similar
E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9
for equivalent weight lower than 500 kg as shown in Table 3. Referring
to the strain analysis, the main difference is displayed for W = 1000 kg.
Indeed, in this particular case, the maximum strain reached the plastic
strain limit for Harp configuration while for other stays patterns, it
remains below the plastic strain limit. Finally, the structural component
subjected to the blast loading reaches quickly the allowable stresses just
after detonation. Indeed, in the case studied, all the equivalent weights
of TNT considered succeed in creating stress reaching the stress limit at
a very short time step (see Fig. 4).
However, in this case, the Harp configuration seems to dissipate the
dynamic action better than the other two because the stress fluctuates
less. These fluctuations are visible only on the negative phase of the
fatigue that would result from a large variation in the internal stresses
(4) of the element.
Finally, with a random stand-off distance, the displacement, stress
and strain variation in the section close to a blast loading increase by
increasing the dimension of the blast loading source. Indeed, the da-
maging effects in the bridge structural response induced after blast
loading increase as the equivalent weight of TNT increases.
It is noteworthy that assessing the internal strain state of the section
has a critical issue. Indeed, following the blast event, the admissible
stress is quickly reached, but its consequence is critical if the state of
internal strain is higher than the maximum plastic strain. Although the
strain state seems high for the three configurations studied with most of
them exceeding the yielding strain, the worst situation is reached for
the equivalent mass of the TNT equal to or greater than 1500 kg.
Therefore, the maximum plastic strain is exceeded with the possible
breakage of the deck at this position.
Later, two other positions of occurrence of blast loading along the
deck at x = 68.2 m and x = 201.5 m (Mid-span) were studied (Fig. 5
and Fig. 6). It is observed that the displacement at time history t = 0 ms
is 192 mm at the mid-span (Fig. 6a) which are from traffic and per-
manent loads. In addition, the displacements following blast loading
are more relevant for an event occurring in mid-span as the outcome of
comparing the results of Fig. 5a and Fig. 6a for any equivalent TNT
weight.
In terms of strain, it has been observed that for a blast loading oc-
curring in the mid-span, the consequences are much more significant,
because, before the stress generated by a blast load, the deck already
has a significant level of stress and strain due to permanent and traffic
loads. As a matter of fact, in this particular case, from equivalent loads
greater than 500 kg, the plastic strain limit would be reached as shown
in Fig. 6b. When the event takes place at x = 68.2 m, the internal strain
of the deck section is higher (Fig. 5b) than if the action takes place at
x = 6.2 m. However, it would require actions with an equivalent
weight greater than 1000 kg to cause serious damages.
Finally, the stress conditions are quite similar in that the blast event,
which strongly contributes after the deflagration, to the achievement of
yielding stress in a short time. The only difference is observed in the
considerable fluctuation of the stress, especially for the action taking
place in the mid-span for low-intensity loads as shown in Fig. 6c.
A detailed analysis of the deck revealed the possibility of a gradual
collapse that could be caused by the dynamic action of a blast load. In
the case of a cable-stayed bridge, progressive collapse can be caused by
the loss of two or more cables [4]. For an extreme event such as blast
loading, the loss of cables may be due either to the consequences of an
air blast or to the reflected pressure near the stays. Other reasons could
be the reaching of the excessive tension limit of the cable or finally to
the collapse of the deck that would cause the cables near the failure
zone to break. In this investigation, the last two cases were considered.
The first case studied considers the Fan configuration where the
blast loading occurs close to the abutment. The results show a limited
displacement of the deck at the impact area, a steady condition of the
deck when moving far from the impact area, a torsional buckling mode
similar to the case of Fig. 7a. A higher dynamic sensitivity in response
5
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Fig. 4. Structural response of the three cable configuration in the case of blast at X = 6.2 m. From the left to the right displacement, Strain and Stress of the bridge’s
deck.

Table 3 to blast action is observed for Harp and Semi-fan bridges as displayed in
Maximum vertical displacement of the deck under a blast loading at stand-off Fig. 7b&c. However, in the Fan configuration, the blast action close to
R = 1 m. the abutment could result in the loss of a single cable, especially the one
Equivalent TNT weight Cable-stayed bridge configurations affected by the torsion of the deck (Fig. 7a).
Harp Fan Semi-fan The analysis also revealed that in the case of Harp bridges (Fig. 7b&
d), the dynamic action creates a torsional buckling of the deck, which
W = 250 kg — 2678 mm — 1774 mm — 2109 mm
W = 500 kg — 2297 mm — 1489 mm — 1728 mm
may result in a reduction of its structural capacity. In these cases, two
W = 750 kg — 1989 mm — 1286 mm — 1508 mm cables are likely to be damaged for the impact at x = 6.2 m (Fig. 7b)
W = 1000 kg — 1421 mm — 1126 mm — 1298 mm whereas three cables are likely to be damaged for the impact at
W = 1500 kg — 406 mm — 331 mm — 385 mm x = 68.2 m (Fig. 7d), particularly because they will have reached their
ultimate stress. This is less relevant because the action lasts for a short

Fig. 5. Structural response of the Harp pattern with blast at X = 62.8 m and R = 1 m. a) Displacement b) Strain c) Stress.

6
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Fig. 6. Structural response of the Harp pattern with blast at X = 201.5 m and R = 1 m. a) Displacement b) Strain c) Stress.

period and the damage will only be confirmed by repeated similar ac-
tions. In the case where the action is caused by a large mass (Fig. 7e),
the consequences are more severe because a total breakage of the deck
element around the impact area is observed causing a loss of the four
cables adjacent to the detonation zone. Finally, besides the global in-
stability of the deck observed after a blast event, an evaluation of the
internal distribution of stresses presents a possible local instability of
the section of the deck around the impact zone. In particular, for a
significant blast loading, a combination of positive and negative stress
was observed throughout the section as shown in Fig. 8b. For less re-
levant blast loading, the upper part of the deck is likely to be in tension
whereas the lower part being compressed Fig. 8a.
4. Fragility analysis
From the results of the analysis performed so far, it is clear that the
consequence of blast loading can be extreme for the deck and therefore
for the whole bridge through the progressive collapse mechanism. For
this reason, it is also important to associate the structural response of
the bridge deck to the likelihood of exceedance of capacity thresholds of
the deck elements during a blast event.
4.1. Methodology
The common way to study the structural performance under a non-

Fig. 7. Blast loading analysis in cable-stayed bridge. a) W = 500 kg, b) W = 500 kg, c) W = 500 kg, d) W = 500 kg, e) W = 1500 kg.

7
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Fig. 8. Total fibre stress in MPa of the deck of

Harp configuration for two different blast in-
tensities at two different locations, a) x = 6.2 m
and b) x = 201.5 m.

deterministic approach is through the fragility assessment. The method Table 4

aims to determine the likelihood of exceedance of the structural capa- Uncertainties in geometrical and material characteristics of the bridge.
city during accidental events conditioned by the intensity measure that Symbol Description Mean Unit Cv Distribution
represents the actions [30]. The performance-based design has been
widely adopted by engineers to design structures. In particular, fragility fyk yielding strength deck 420 MPa 0.10 Lognormal
E Young modulus 210,000 MPa 0.03 Lognormal
analysis has been widely performed in the last two decades [44–46].
L1 Length of the midspan 203 m 0.09 Lognormal
Recently, the method has been used also for structures subjected to H1 Height of the pylon 53.5 m 0.04 Lognormal
various natural hazards such as flood [47], windborne debris in hurri- H2 Height of the base pylon 27.5 m 0.01 Lognormal
cane mitigated regions [48], and fire [24]. The system fragility can be B Bridge width 13 m 0.01 Lognormal
obtained throughout the Monte Carlo simulation, which is the most
challenging method and expensive computationally since it takes into
account different structural components as well as the linked un- Table 5
certainties. On the other hand, for a large size of FE element models (in Uncertainties in the section of structural component.
the case of Nuclear Power Plan, or wind turbine blades etc…), this Structural components Description Mean Unit Cv Distribution
approach proved its limit [49]. Hence, the technique for the fragility
Deck Section 642331.3 mm 2 0.01 Lognormal
computation considers the Response Surface Methodology to integrate
Base pylon 585,600 mm 2 0.04 Lognormal
the influence of the variables randomness and the approach has been Pylon 521,600 mm 2 0.04 Lognormal
adopted in [50,51]. Finally, a less expensive approach uses the direct Transverse 441,600 mm 2 0.02 Lognormal
integration method and to combine different component fragility curves
to get the system fragility [44] and will be used in this research.
In the fragility analysis, the consideration of uncertainties is fun- taken higher as considered by the Cowper and Symonds model [53].
damental. Three different types of uncertainties that are regularly The main components in the analysis follow the Lognormal distribu-
considered in many engineering fields have been defined and presented tion. Table 4 summarizes the use of defined uncertainties for material
in [26]. In this study, the uncertainties in the accidental load and and geometrical characteristics whereas in Table 5, are presented the
structure will be mainly considered whereas the interaction mechanism uncertainties in structural components section.
uncertainties will be considered as the consequence of the first two.
4.3. Blast fragility analysis
4.2. Material and geometry uncertainties of the bridges
The fragility of a structure resulting from blast loading is its prob-
Several steel cable-stayed bridges were analyzed in the scope of the ability of failure made contingent on the occurrence of a blast event of a
fragility assessment. Same as the stay patterns considered and presented specific hazard intensity. If D is the structural demand owing to the
above, Harp, Fan, and Semi-fan bridges are reproduced. The bridges are random blast event of kth hazard intensity level, Ik and Cj is the
considered not to be in a seismic zone in order not to consider a case of structural capacity to resist the load effect corresponding to the jth limit
multi-hazards since the principal hazard in our study is blast loading. In state, then the blast fragility can be obtained throughout using Eq. (5).
addition to the stay patterns, three different pylon profiles were also PFjk = P (D ⩾ Cj |Ik) (5)
considered as presented in Fig. 9. Bridges modeling follows the same
approach as discussed in the previous sections of the paper. The mean The probability of failure is calculated by simulating a large number
values of the bridge geometrical characteristics, as well as the bridge’s of samples including uncertainties in material, geometry, and hazard.
structural components, are kept the same consideration in the first Morever, the vector ξ represents the weight of blast source, W and the
−
analyses. Two statistical parameters namely the mean value μ̄and the location of the blast with respect to the structure known as the stand-off
coefficient of variation cv are adopted for the variability of the geometry distance R and the planar location (xk,yk). In this study, the structural
of the structural components. demand is computed for each occurrence of the blast event with dif-
Since constructional steel is a material produced under tight quality ferent intensity and then investigated if it satisfies the defined failure
control procedures, a cv of 0.10 is used for steel profile as proposed in criterion. The blast fragility function is obtained throughout Eq. (6).
[52]. The steel grade remains the same as defined in Table 2 whose
PFjk = P (δ D ⩾ δC | W ) (6)
parameters are considered as the mean value. The steel strength is

Fig. 9. Geometrical characteristics of the bridges. From the left to the right: Inverted Y-shaped profile; H-shaped profile, A-shaped profile.

8
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Fig. 10. Flowchart for the capacity and demand assessment of cable-stayed bridges.

Using the first-order reliability approach that takes into account the
first and second moments of individual distributions, Eq. (6) is simpli-
fied into Eq. (7).
⎡ ln(μ̄D /μ̄C ) ⎤
PFj = Φ ⎢ 2
σ ln D + σ2ln C ⎥
⎣ ⎦
Where μ̄D = median estimate of the structural demand
andμ̄C = median estimate of the structural capacity.σ 2ln D = lognormal
standard deviation of the structural demand andσ2ln C = lognormal
standard deviation of the structural capacity. Cornell et al., [54] de-
rived the relationship between the estimate of the median and the in-
tensity of the actions as shown in Eq. (8).
μ̄ = aI b or ln(μ̄ ) = ln a + b ln(I j )
D j D
The methodology adopted by the authors for the derivation of the
fragility curves is outlined in the flowchart (see Fig. 10). It uses the
component-based approach in which the fragility curve of the system
results from the unconditional combination of the fragility of each
structural component of the bridge. In each case, the structural capacity
is assessed on a stock of bridges that vary in dimensions and the cross-
section of the structural elements. The same is also true for the struc-
tural demand, which is evaluated through nonlinear dynamic analysis
for different values of scaled distance, taken in this case as intensity
(7) measure.
4.4. Structural capacity
In the capacity study, more than thirty models were studied, varying
the geometry of the bridge, the characteristics of the material, the pylon
profiles (see Fig. 9), and the cross-sectional area of the deck. Then, a
series of nonlinear pushover analyses of the bridge in the vertical di-
(8) rection were performed, which corresponds indeed to the principal
direction of the damage of the deck after a blast event. From the ana-
lyses, ideal capacity curves were derived from the various pushover
performed leading to the definition of an envelope, which describes the
set of capacity obtained from the bridge deck. Then, through a statis-
tical tool it was possible to highlight the mean capacity curve of each

9
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Fig. 11. Nonlinear static analysis of the cable-stayed bridges. a) Envelope of capacity curves of the bridges’ deck; b) Bilinear of the Mean Capacity curve.

Table 6
Statistic properties for structural parameters of cable-stayed bridges.

Stay patterns Parameter Statistical values Ductiliy

μ̄C σC μδ

Harp δv,y mm 863.8 98.9 4.47

Fy kN 29586.2 3159.1
δv,u mm 3864.1 182.4
Fu kN 65356.8 5896.7
Fan δv,y mm 1042.3 126.6 3.35
Fy kN 30894.6 3278.9
δv,u mm 3489.4 268.8 Fig. 12. Force-displacement relationship of a damage index.
Fu kN 68595.7 7212.3
Semi-fan δv,y mm 985.5 119.4 3.58
Fy kN 32489.1 3654.7 Occupancy, C – D: Life Safety, D – E: Collapse Prevention and finally
δv,u mm 3527.3 295.2 from E – F: Collapse where B, C, D, and E represent the thresholds. A
Fu kN 70125.5 6478.4
brief summary of the limit state range is presented in Table 7.

cable stays patterns (Fig. 11a and Fig. 11b). The set of performance
displays a hardening behavior with a displacement ductility (Eq. (9))
varying from 2.6 to 4.2. Afterward, the same process was done for the
other two stays patterns and the statistical parameters of mean per-
formance were collected and presented in Table 6.
The Harp configuration presents the most flexible in all with the
highest ductility. On the other hand, the Fan configuration, which had a
good structural response in terms of vertical displacement, as presented
in the previous section, is the most rigid and reduces the structure's
ability to achieve disproportionate vertical displacements.
δv,u
μδ =
δ v,y
Where δv,u and δv,y are the mean value of ultimate and yielding vertical
displacement after a nonlinear pushover in vertical direction.
4.4.1. Limit states and damage index
In the fragility analysis, structural damage after a hazard is a
function of both intrinsic and extrinsic characteristics of the structure.
In general, it is possible to study the undergone damage by classifying
them according to the damage index [55]. This index includes the
structural response observed and the definition of limit states. The limit
states often derive from experimental campaigns or computational re-
sources. Regardless of the method used, the damage indexes are often
defined by means of curvature ductility, rotation ductility, displace-
ment ductility, etc. Indeed, several authors [5,26,39] considered two
parameters as failure criterion under accidental loads: the rotation
angle (θ) of the element’s end support and the displacement ductility
(μδ) [26].
The pushover analysis performed allowed defining the limit state for
the study as presented in Fig. 12. In particular, five regions were
adopted as highlighted in [56]: A – B: Elastic Range, B – C: Immediate
4.5. Structural demand
In the analysis of the demand, a set of various blast intensities were
considered with the possibility of occurring at truly random positions
along the longitudinal and transversal direction of the bridge deck
(Fig. 13). It is worth recording that the hazard uncertainties namely the
explosive mass and the stand-off distance and highlighted in Section 4.1
will follow a normal distribution (Fig. 13) as advised in [26]. Finally,
the longitudinal position was limited to half of the bridge due to the
symmetry and in the transversal direction, the random location of the
blast event was considered. It has been considered a wide range of in-
tensity measure by varying the equivalent TNT weight and stand-off
(9) distance.
The weight of an explosive is generally estimated by taking into
account a relevant attack scenario, which would involve a vehicle or
improvised explosive device carried by one or more individuals. The
stand-off distance in the analysis of the robustness of the bridge deck is
taken accordingly to the number of explosives considered. It is taken as
the distance between the position of the center of the weapon and its
vertical projection on the deck, which is the shortest path to reach the
deck, thus causing the most intense impact on the deck. Therefore, the
weight of the explosive is taken in the range 0–10000 kg and the stand-
off distance will be taken in the range 0.2–4 m.
The results from the analysis have been displayed in the lognormal
scale as advised in [44] and the mean value of the vertical displacement
has been expressed in terms of scaled distance Z and displayed in
Fig. 14. The results show that the lognormal distribution of the vertical
displacement follows the linear regression with a low dispersion rate.
The regression line for each bridge model confirms the outcome that the
Harp bridge model displays larger vertical displacement than the two
other bridge models with a higher dispersion of data. Finally, the ne-
gative slope is mainly due to the fact that the impact on the bridge deck
is more severe by decreasing the scale distance, which is obtained either

10
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Table 7
Limit state for cable-stayed bridges under Blast loading [56]
Limit State (LS) Threshold values of ductility Structural performance registered

Harp Fan Semi-Fan

Immediate Occupancy μδ >n 1 μδ > 1 μδ > 1 Minor distortions observed

Life Safety μδ ≥ 1.5 μδ ≥ 1.1 μδ ≥ 1.2 Minor local buckling of the deck. Localized failure
Collapse Prevention μδ ≥ 3.0 μδ ≥ 2.3 μδ ≥ 2.4 Large Displacement. Flexural buckling in the localized region
Collapse μδ ≥ 4.5 μδ ≥ 3.3 μδ ≥ 3.6 EXcessive displacement; EXcessive strain

by decreasing the stand-off distance or by increasing the weight the bridge not to reach the limit states at the same trend as the other two
TNT. configurations.

4.6. Fragility curves 4.6.1. Influence of blast loading parameters

Following the results of structural capacity and demand where the
statistical parameters defining the criteria of occurrence of different
levels of damage have been identified, it is possible to draw the prob-
ability of reaching a limit state threshold of the bridge deck.
In general, the observation is similar to that from the non-linear
dynamic analysis performed in the previous sections. It is worth noting
that the most influential parameter of the collapse of the deck is the
stand-off distance. Indeed, a low probability of deck failure is observed
as soon as one detonation point moves away from the deck, i.e. for a
relatively large stand-off distance. In addition, Fan patterns present
several advantages. Although it is not favorable compared to the Harp
configuration in terms of displacement ductility, the Fan configuration
nevertheless displays a good structural response, which allows the
The mass of the equivalent TNT considered in the analysis varies
from 0 to 10000 kg (Figs. 15–17) even though it is very unlikely to have
a terrorist attack with a mass of 10000 kg because it requires a heavy
means of transport and the terrorist attack would then become very
easily recognizable and prevented. The variation in terms of TNT
weight makes it possible to highlight the robustness of structures such
as cable-stayed bridges subjected to accidental and unpredictable
events. Indeed, the situation prevailing in the various results shows the
great capacity of the deck to avoid failure for a blast loading taking
place on a random position. Obviously, the risk of collapse increases as
soon as the equivalent weight increases.
To have a possible minor local buckling of the deck or a localized
failure, the deck of a Fan configuration must be subjected to blast
loading triggered by a mass of TNT around 300 kg. However, in order to

Fig. 13. Distribution of blast parameters for blast demand.

11
C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

Fig. 14. Prediction of blast demand in cable-stayed bridges: a) Harp bridge model; b) Fan bridge model; c) Semi-fan bridge model.

Fig. 15. Blast fragility for Fan cable-stayed bridge.

have excessive displacements leading to the collapse of the deck and
subsequently of the bridge, the TNT load must be around 1000 kg. It is
important to note that a quantity of TNT lower than those mentioned
above will produce similar or advanced damage for Harp and Semi-fan
configurations.
Contrary to the influence of the mass of TNT, which is less im-
portant from what emerges in the fragility curves, the stand-off distance
has a more significant impact. It is important to highlight that the
fragility results indicate that as the detonation occurs far from the deck,
the probability of a possible failure of the bridge deck becomes lower.
Indeed, by isolating two particular cases of stand-off distance
R = 0.4 m and R = 0.8 m in a Fan configuration, we have the prob-
ability of collapse of the deck is reduced by 30% for a mass of 1000 kg
TNT, by 45% for a mass of 500 kg (Fig. 15).
On the other hand, the variability is not as important for other
cable-stayed bridge configurations. Indeed, considering the same stand-
off distances as defined above, there is observed in a Harp configura-
tion, a reduction in the probability of collapse of 20% for a mass of
1000 kg and 30% for a mass of 500 kg. This reduction will be in the
order of 25% and 37% respectively for the 1000 kg and 500 kg masses
of the Semi-fan configuration.
To calculate the limit state probability of exceedance as presented, a
random distribution of blast loading scenario was considered. In addi-
tion, the quantity of TNT ranging from 0 − 10000 kg was considered
but in reality, it is almost impossible to have an attack with a very large
quantity of explosives because the transport would require significant

12
C.D. Tetougueni, et al.
Fig. 16. Blast fragility for Harp cable-stayed bridge.
logistical means that would reveal a suspicion on the part of the in-
telligence services. Thus, for masses varying from 50 kg − 300 kg,
which still represents quantities that would go unnoticed, it is very
likely that there is no risk of collapse because the deck will succeed in
withstand the loads caused by the explosion.
5. Conclusion
The structural efficiency of cable-stayed bridges considering three
different typologies, namely (Harp, Fan, and Semi-fan bridges).
Initially, a study of the structural response of bridges subjected to blast
loadings of different intensities was carried out. Then, a study to
highlight the consequences that a blast event could have on the possi-
bility of progressive collapse. Finally, an analysis taking into account
the probabilistic parameters was done on the structural components in
order to find the likely failure. In the study, a series of non-linear dy-
namic analyses were used to find the structural demand of bridges
considering different intensity measures while in the case of structural
capacity, a pushover analysis was required to vary the geometry of the
bridge and the cross-section of the structural components. From all
these investigations, the following findings emerged:
• The Harp configuration presents a better configuration in terms of
structural response under the action of blast loading. In addition,
excessive displacement can be reached in the case of a large quantity
of TNT used and the blast loading easily causes partial or total
yielding of the bridge section closest to the point of detonation.
• Nonlinear dynamic analysis of a wide variety of bridges has estab-
lished a relationship between structural demand following an ex-
treme Hazard such as blast loading and scaled distance. This
13
E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9
relationship in a lognormal distribution describes a negative corre-
lation between the two parameters involved. Thus, the con-
sequences following a blast loading action are more serious when
the scaled distance is reduced.
• In this study, the authors also presented the need for an internal
strain study of the sections. Indeed, in a case where stress analysis
reveals the ease with which bridge sections reach their yielding and
those even for small masses, it is important to know if the action at
the limit will succeed in creating a collapse of the section.
• The study highlighted the feasibility and importance of the potential
contribution of the derivation of the fragility curves of cable-stayed
bridges. Through a probabilistic approach, the performance-based
design for blast resistant structures are made and the likelihood of
collapse for structural elements of the bridge is presented although
this practice is commonly used on structures subjected to such ha-
zards as (earthquake, flood, hurricane, wind, etc…).
• In the common case of a fragility analysis for an earthquake, gen-
erally the structural elements at high risk of exposure are piers,
abutments and bearing. On the other hand, in the case of a blast
event, the vulnerability of the deck must also be taken into account
to the extent that this structural element remains highly exposed to
action. Finally, the major difficulty that lies in the probabilistic
analysis of this type of event consists in determining the hazard
function since blast loadings are considered as a low occurrence
probability event.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to
C.D. Tetougueni, et al.
Fig. 17. Blast fragility for Semi-fan cable-stayed bridge.
influence the work reported in this paper.
References
[1] Jenkins BM, Gersten LN. Protecting Public Surface Transportation Against
Terrorism and Serious Crime : Continuing Research on Best Security Practices.
2001;(September):124.
[2] Kumar A, Matsagar V. Blast Fragility and Sensitivity Analyses of Steel Moment
Frames with Plan Irregularities. Int J Steel Struct 2018;18(5):1684–98. https://doi.
org/10.1007/s13296-018-0077-z.
[3] Son J, Lee HJ. Performance of cable-stayed bridge pylons subjected to blast loading.
Eng Struct 2011;33(4):1133–48. https://doi.org/10.1016/j.engstruct.2010.12.031.
[4] Wolff M, Starossek U. Cable-loss analyses and collapse behavior of cable-stayed
bridges. 2013:423-423. http://doi.org/10.1201/b10430-326.
[5] Formisano A, Landolfo R, Mazzolani FM. Robustness assessment approaches for
steel framed structures under catastrophic events. Comput Struct 2015;147:216–28.
https://doi.org/10.1016/j.compstruc.2014.09.010 .
[6] Elsanadedy HM, Almusallam TH, Alharbi YR, Al-Salloum YA, Abbas H. Progressive
collapse potential of a typical steel building due to blast attacks. J Constr Steel Res
2014;101:143–57. https://doi.org/10.1016/j.jcsr.2014.05.005.
[7] Olmati P, Brando F, Gkoumas K. Robustness Assessment of a Steel Truss Bridge
2013;2013:250–61. https://doi.org/10.1061/9780784412848.023 .
[8] Abruzzo J, Matta A, Panariello G. Study of Mitigation Strategies for Progressive
Collapse of a Reinforced Concrete Commercial Building. J Perform Constr Facil
2006;20(4):384–90. https://doi.org/10.1061/(asce)0887-3828(2006)20:4(384).
[9] Song BI, Sezen H. E Xperimental and analytical progressive collapse assessment of a
steel frame building. Eng Struct 2013;56:664–72. https://doi.org/10.1016/j.
engstruct.2013.05.050.
[10] Abdelwahed B. A Review on Building Progressive Collapse, Survey and Discussion.
Case Stud Constr Mater 2019;11:e00264https://doi.org/10.1016/j.cscm.2019.
e00264.
[11] Adam JM, Parisi F, Sagaseta J, Lu X. Research and practice on progressive collapse
and robustness of building structures in the 21st century. Eng Struct
2018;173(March):122–49. https://doi.org/10.1016/j.engstruct.2018.06.082.
[12] Das R, Pandey AD, Soumya Mahesh MJ, Saini P, Anvesh S. Progressive Collapse of a
Cable Stayed Bridge. Procedia Eng 2016;144:132–9. https://doi.org/10.1016/j.
proeng.2016.05.016.
[13] Ubertini F. Effects of cables damage on vertical and torsional eigenproperties of
14
E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9
suspension bridges. J Sound Vib 2014;333(11):2404–21. https://doi.org/10.1016/
j.jsv.2014.01.027.
[14] Comanducci G, Ubertini F, Materazzi AL. Structural health monitoring of suspen-
sion bridges with features affected by changing wind speed. J Wind Eng Ind
Aerodyn 2015;141:12–26. https://doi.org/10.1016/j.jweia.2015.02.007.
[15] Materazzi AL, Ubertini F. Eigenproperties of suspension bridges with damage. J
Sound Vib 2011;330(26):6420–34. https://doi.org/10.1016/j.jsv.2011.08.007.
[16] Lonetti P, Pascuzzo A. A numerical study on the structural integrity of self-anchored
cable-stayed suspension bridges. Frat ed Integrita Strutt 2016;10(38):359–76.
https://doi.org/10.3221/IGF-ESIS.38.46.
[17] Lonetti P, Pascuzzo A. Vulnerability and failure analysis of hybrid cable-stayed
suspension bridges subjected to damage mechanisms. Eng Fail Anal
2014;45:470–95. https://doi.org/10.1016/j.engfailanal.2014.07.002.
[18] Domaneschi M, Cimellaro GP, Scutiero G. Disproportionate collapse of a cable-
stayed bridge. Bridge Eng 2019;172(1):13–26. https://doi.org/10.1680/jbren.18.
00031.
[19] Ruiz-Teran AM, Apracio AC. Response of under-deck cable-stayed bridges to the
accidental breakage of stay cables. Eng Struct 2009;31:1425–34. https://doi.org/
10.1016/j.engstruct.2009.02.027.
[20] Madrazo-Aguirre F, Ruiz-Teran AM, Wadee MA. Dynamic behaviour of steel–con-
crete composite under-deck cable-stayed bridges under the action of moving loads.
Eng Struct 2015;103:260–74. https://doi.org/10.1016/j.engstruct.2015.09.014 .
[21] Sanaz R, Armen DK. A stochastic ground motion model with separable temporal and
spectral nonstationarities. Earthq Eng Struct Dyn 2012;41(11):1549–68. https://
doi.org/10.1002/eqe.
[22] Nassr AA, Razaqpur AG, Tait MJ, Campidelli M, Foo S. Strength and stability of steel
beam columns under blast load. Int J Impact Eng 2013;55:34 –48. https://doi.org/
10.1016/j.ijimpeng.2012.11.010.
[23] Mosoarca M, Anastasiadis A, Formisano A. New factors for the seismic vulnerability
assessment of reinforced concrete buildings. Int J Sustain Mater Struct Syst
2016;2(3/4):222. https://doi.org/10.1504/ijsmss.2016.078708.
[24] Lange D, Devaney S, Usmani A. An application of the PEER performance based
earthquake engineering framework to structures in fire. Eng Struct
2014;66:100–15. https://doi.org/10.1016/j.engstruct.2014.01.052 .
[25] Tang EKC, Hao H. Numerical simulation of a cable-stayed bridge response to blast
loads, Part I: Model development and response calculations. Eng Struct
2010;32(10):3180–92. https://doi.org/10.1016/j.engstruct.2010.06.007 .
[26] Olmati P, Petrini F, Gkoumas K. Fragility analysis for the Performance-Based Design
of cladding wall panels subjected to blast load. Eng Struct 2014;78:112 –20. https://
doi.org/10.1016/j.engstruct.2014.06.004.
C.D. Tetougueni, et al.
[27] Hao H, Stewart MG, Li Z-X, Shi Y. RC Column Failure Probabilities to Blast Loads.
Int J Prot Struct 2010;1(4):571–91. https://doi.org/10.1260/2041-4196.1.4.571.
[28] Momeni M, Hadianfard MA, Bedon C, Baghlani A. Numerical damage evaluation
assessment of blast loaded steel columns with similar section properties. Structures
2018;2019(20):189–203. https://doi.org/10.1016/j.istruc.2019.04.002.
[29] Olmati P, Petrini F, Bontempi F. Numerical analyses for the structural assessment of
steel buildings under explosions. Struct Eng Mech 2013;45(6):803–19. https://doi.
org/10.12989/sem.2013.45.6.803.
[30] Khan S, Saha SK, Matsagar VA, Hoffmeister B. Fragility of Steel Frame Buildings
under Blast Load. J Perform Constr Facil 2017;31(4):04017019. https://doi.org/10.
1061/(asce)cf.1943-5509.0001016.
[31] Kelliher D, Sutton-Swaby K. Stochastic representation of blast load damage in a
reinforced concrete building. Struct Saf 2012;34(1):407 –17. https://doi.org/10.
1016/j.strusafe.2011.08.001.
[32] Ding Y, Song X, Zhu HT. Probabilistic progressive collapse analysis of steel-concrete
composite floor systems. J Constr Steel Res 2017;129:129–40. https://doi.org/10.
1016/j.jcsr.2016.11.009.
[33] Andreou M, Kotsoglou A, Pantazopoulou S. Modelling Blast Effects on a Reinforced
Concrete Bridge. Adv Civ Eng 2016 2016. https://doi.org/10.1155/2016/4167329.
[34] Hao H, Tang EKC. Numerical simulation of a cable-stayed bridge response to blast
loads, Part II: Damage prediction and FRP strengthening. Eng Struct
2010;32(10):3193–205. https://doi.org/10.1016/j.engstruct.2010.06.006.
[35] Hashemi SK, Bradford MA, Valipour HR. Dynamic response of cable-stayed bridge
under blast load. Eng Struct 2016;127:719–36. https://doi.org/10.1016/j.
engstruct.2016.08.038.
[36] Primer FEMA. for Design of Commercial Buildings to Mitigate Terrorist Attacks.
Risk Manag Ser 2003.
[37] Deng RB, Jin XL. Numerical simulation of bridge damage under blast loads. WSEAS
Trans Comput 2009;8(9):1564–74.
[38] Calculation of Blast Loads for Application to Structural Components. Administrative
Arrangement No JRC 32253-2011 with DG-HOMEActivity A5 - Blast Simulation
Technology Development.; 2013. http://doi.org/10.2788/61866.
[39] Verma SV. Blast Resistant Design of Structure. Int J Res Eng Technol
2016;04(25):64–9. https://doi.org/10.15623/ijret.2015.0425010 .
[40] Longitudinal Seismic Behavior of a Single-Tower Cable-Stayed Bridge Subjected to
Near-Field Earthquakes. Shock Vib. 2017;2017. http://doi.org/10.1155/2017/
1675982.
[41] Paik JK, Thayamballi AK. A concise introduction to the idealized structural unit
method for nonlinear analysis of large plated structures and its application. Thin-
Walled Struct 2003;41(4):329–55. https://doi.org/10.1016/S0263-8231(02)
00113-1.
15
E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9
[42] European Standard. Part 2: Traffic loads on bridges. Eurocode 1 Actions Struct.
2003;3(1991-2:2003).
[43] CEN. EN 1990: 2002 - Basis of structural design. 2002:1-87.
[44] Pang Y, Wu X, Shen G, Yuan W. Seismic Fragility Analysis of Cable-Stayed Bridges
Considering Different Sources of Uncertainties. J Bridg Eng 2013;19(4):04013015.
https://doi.org/10.1061/(asce)be.1943-5592.0000565.
[45] Stefanidou SP, Kappos AJ. Bridge-Specific Fragility Analysis: When Is It Really
Necessary? 2019;Vol 17. https://doi.org/10.1007/s10518-018-00525-9.
[46] Xie Y, Zhang J, DesRoches R, Padgett JE. Seismic fragilities of single-column
highway bridges with rocking column-footing. Earthq Eng Struct Dyn
2019;48(7):843–64. https://doi.org/10.1002/eqe.3164.
[47] van de Lindt JW, Taggart M. Fragility Analysis Methodology for Performance-Based
Analysis of Wood-Frame Buildings for Flood. Nat Hazards Rev 2009;10(3):113–23.
https://doi.org/10.1061/(asce)1527-6988(2009)10:3(113).
[48] Herbin AH, Barbato M. Fragility curves for building envelope components subject to
windborne debris impact. J Wind Eng Ind Aerodyn 2012;107–108:285–98. https://
doi.org/10.1016/j.jweia.2012.05.005.
[49] Casciati F, Cimellaro GP, Domaneschi M. Seismic reliability of a cable-stayed bridge
retrofitted with hysteretic devices. Comput Struct 2008;86:1769–81. https://doi.
org/10.1016/j.compstruc.2008.01.012.
[50] Perotti F, Domaneschi M, De Grandis S. The numerical computation of seismic
fragility of base-isolatedNuclear Power Plants buildings. Nucl Eng and Des
2013;262:189–200. https://doi.org/10.1016/j.nucengdes.2013.04.029.
[51] De Grandis S, Domaneschi M, Perotti F. A numerical procedure for computing the
fragility of NPP components under random seismic excitation. Nucl Eng and Des
2009;239:2491–9. https://doi.org/10.1016/j.nucengdes.2009.06.027 .
[52] Simões da Silva L, Rebelo C, Nethercot D, Marques L, Simões R, Vila Real PMM.
Statistical evaluation of the lateral-torsional buckling resistance of steel I-beams,
Part 2: Variability of steel properties. J Constr Steel Res 2009;65(4):832–49.
https://doi.org/10.1016/j.jcsr.2008.07.017.
[53] Olmati P, Vamvatsikos D, Stewart MG. Safety factor for structural elements sub-
jected to impulsive blast loads. Int J Impact Eng 2017;106:249–58. https://doi.org/
10.1016/j.ijimpeng.2017.04.009.
[54] Cornell CA, Jalayer F, Hamburger RO, Foutch DA. Management Agency Steel
Moment Frame Guidelines. J Struct Eng 2002;2002(128):526–33. https://doi.org/
10.1061/(ASCE)0733-9445(2002)128:4(526).
[55] Ding Y, Song X, Zhu HT. Probabilistic progressive collapse analysis of steel frame
structures against blast loads. Eng Struct 2017;147:679–91. https://doi.org/10.
1016/j.engstruct.2017.05.063.
[56] FEMA. Prestandard and Commentary for the Seismic Rehabilitation of Buildings.
Rehabil Requir. 2000;(1):1–518.
 C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2
020
)11
073
9

KEKURANGAN :
Kejadian tak terduga atau peristiwa ekstrem bisa menyebabkan beban yang tidak normal pada struktur, yang dapat
menyebabkan kerusakan pada komponen utama atau bahkan keruntuhan bertahap dari struktur itu sendiri. Meskipun
bahaya-bahaya ini biasanya sulit diprediksi, pentingnya untuk menentukan dampaknya pada struktur akibat peristiwa-
peristiwa ekstrem tersebut tetap besar.

KELEBIHAN :
- Penelitian ini bertujuan untuk mengevaluasi kerusakan yang disebabkan oleh beban tak terduga seperti ledakan
pada jembatan kabel. Penelitian ini mengisi kekosongan dalam literatur ilmiah terkait dengan analisis struktural
jembatan kabel tahan ledakan. Meskipun ada beberapa penelitian sebelumnya yang telah dilakukan dalam bidang
keamanan struktural terhadap pembebanan ledakan, penelitian ini fokus pada analisis struktural jembatan kabel
tahan ledakan dengan mempertimbangkan variasi beban ledakan yang berbeda, seperti lokasi dan intensitas beban
ledakan.
- Dengan adanya penelitian ini, para perancang dan insinyur struktur dapat memperoleh pemahaman yang lebih baik
tentang perilaku struktural jembatan kabel tahan ledakan dan kemampuannya dalam menghadapi pembebanan
ledakan. Hal ini dapat membantu dalam perancangan yang lebih efektif dan evaluasi keamanan jembatan kabel
tahan ledakan

PEMBEBANAN :
- Sistem pembebanan pada jurnal ini adalah pembebanan ledakan. Penelitian ini menggunakan analisis numerik
untuk memodelkan dan menganalisis respons struktural jembatan kabel tahan ledakan terhadap beban ledakan
yang berbeda.
- Pada analisis ini, peneliti menggunakan model elemen balok serat 1D untuk merepresentasikan elemen struktural
jembatan. Model ini mempertimbangkan variasi dalam penampang struktural dan stiabilitas torsional. Di
samping itu, diaphragm digunakan untuk meningkatkan kekakuan elemen struktural.
- Pembebanan ledakan dipertimbangkan pada dek jembatan. Beban ledakan ini diterapkan pada berbagai lokasi
dan intensitas, dan dianalisis dampaknya terhadap pergeseran, regangan, dan tegangan jembatan. Hasil analisis
menunjukkan bahwa pergeseran, regangan, dan tegangan jembatan dipengaruhi oleh lokasi dan intensitas beban
ledakan. Pada beberapa kasus, pergeseran dan regangan mencapai tingkat yang tidak dapat ditoleransi, terutama
pada beban ledakan dengan bobot setara lebih dari 750 kg. Selain itu, analisis juga menunjukkan bahwa
konfigurasi kabel-kabel pengikat pada jembatan mempengaruhi respons struktural terhadap beban ledakan

STRUKTUR :
- Struktur jembatan yang diteliti adalah jembatan kabel tahan ledakan. Jembatan ini merupakan jenis jembatan
kabel-tetap yang terdiri dari dek, pylon, dan kabel-kabel miring yang disebut sebagai stay. Jembatan ini memiliki
penampang balok kotak berbahan baja dengan ketinggian 1,15 m dan lebar 13,0 m. Di samping itu, jembatan ini
dilengkapi dengan penegak-penegak vertikal untuk meningkatkan kekakuan struktural.
- Struktur jembatan ini dimodelkan menggunakan elemen balok serat 1D dalam analisis numerik. Model ini
memperhitungkan variasi dalam penampang struktural dan stabilitas torsional. Selain itu, diaphragm digunakan
untuk meningkatkan kekakuan elemen struktural.
- Jembatan ini juga memiliki tiga konfigurasi kabel-kabel pengikat yang berbeda, yaitu konfigurasi Fan, Semi-Fan,
dan Harp. Konfigurasi ini mempengaruhi respons struktural jembatan terhadap beban ledakan.

PERENCANAAN STRUKTUR :
- Perencanaan struktur jembatan dilakukan dengan menggunakan analisis numerik dan pendekatan berbasis
kinerja. Perencanaan ini melibatkan pemodelan struktur jembatan menggunakan elemen balok serat 1D dan
melakukan analisis dinamik nonlinier untuk mengevaluasi respons struktural terhadap beban ledakan.
- Dalam perencanaan ini, variasi dalam geometri jembatan, karakteristik material, profil pylon, dan luas
16
 C.D. Tetougueni, et al. E
ngin
eerin
gStru
ctu
res2
19(2020
)110739
penampang dek dipertimbangkan. Analisis pushover nonlinier dilakukan untuk menentukan kapasitas struktural
jembatan dalam arah vertikal. Dari analisis ini, kurva kapasitas ideal diperoleh dan digunakan untuk menentukan
batas kapasitas struktural jembatan
- Selain itu, perencanaan struktur jembatan juga melibatkan analisis fragilitas. Analisis ini bertujuan untuk
menentukan kemungkinan terjadinya kegagalan struktural jembatan selama kejadian ledakan. Metode fragilitas
digunakan untuk mengintegrasikan pengaruh ketidakpastian variabel dalam menentukan kemungkinan melebihi
batas kapasitas struktural.

MANFAAT :
- Meningkatkan pemahaman tentang respons struktural jembatan kabel tahan ledakan: Penelitian ini memberikan
pemahaman yang lebih baik tentang bagaimana jembatan kabel tahan ledakan merespons pembebanan ledakan.
Hal ini dapat membantu para perancang dan insinyur struktur dalam memahami perilaku struktural jembatan
tersebut dan mengoptimalkan desainnya.
- Meningkatkan keamanan jembatan kabel tahan ledakan: Dengan memahami respons struktural jembatan terhadap
pembebanan ledakan, penelitian ini dapat membantu dalam perancangan yang lebih efektif dan evaluasi
keamanan jembatan kabel tahan ledakan. Hal ini dapat mengurangi risiko kegagalan struktural dan meningkatkan
keamanan jembatan tersebut.

CONTOH JEMBATAN

Jembatan Brawijaya Kota Kediri

17