Structural Health Monitoring of a Steel Railway Bridge

for Increased Axle Loads

Pradipta Banerji, Prof. and Sanjay Anil Chikermane, Civil Eng., Dept. of Civil Engineering, Indian Institute of Technology
Bombay, Powai, Mumbai, India. Contact: chikermane@iitb.ac.in
DOI: 10.2749/101686611X12994961034570

Abstract rule11 and railway code provided

fatigue curve (also referred to as SN
A large proportion of the existing bridge infrastructure network are relatively or Wöhler curve)”.
old, past their design life and have much larger loads running on them than had
been considered in their design. Predicting the present behaviour of these sys- In the present study, the assessment
tems, given these increased loads, makes structural health monitoring imperative of past traffic data has been made
as a tool to assess the present condition and the safety of the remaining life. In using the traffic details documented
this paper it is shown that using a short measurement period, that is representa- by the Indian Railway authorities.
tive of the repetitive load on the bridge, and correlating the estimated fatigue life Future traffic modelling has been
with the annual tonnage could give good fatigue assessment without having to done by extrapolating the past trends
resort to a significantly more expensive continuous monitoring of the structure. rather than by using a traffic model9.
A site-validated numerical model is built, which provides an estimate of the pres- It has been shown that short-duration
ent condition. This model is then used to estimate the loads coming on the bridge recording can be used and correlating
during the measurement period. The loads are then correlated with existing traf- this with the tonnage of annual traffic
fic data to give an estimate of the annual traffic tonnage. Good estimates of the can be used for fatigue assessment
annual traffic tonnage using this method show that the measurement period is with good results. This strengthens the
representative and hence increase confidence on the validity of the remaining hypothesis of using economical short
life estimated. monitoring programs against the tre-
mendously more expensive continu-
Keywords: Structural Health Monitoring (SHM); Fatigue; Steel Bridge; Load ous monitoring paradigms. As the axle
Estimation; Remaining Life. loads of the traffic during the monitor-
ing period was not known, they were
estimated using an inverse method.
Introduction This technique also helps in identify- This method uses the experimentally
ing damage and its progression in a validated numerical model and esti-
Monitoring the health of structures structure, so that cost-effective reha- mates the loads using an iterative pro-
using sensors placed on them is a bilitation can be done before the dam- cess. It has been shown that the loads
relatively recent phenomenon in civil age causes catastrophic consequences. estimated using this process give an
engineering. Subsequently, there have However, this requires continuous excellent correlation with the annual
been significant issues related to the monitoring, which is not very popular traffic data.
use of structures, mostly again bridges, as yet because of issues related to data
beyond their design life, which has storage and availability of robust sys- Objectives of Study
made structural health monitoring tem-identification techniques.
for current condition and remaining The main objective of the study was
life assessment almost imperative as a One of the important aspects of design to assess the effects of increased axle
bridge management tool.1 Design life life extension of bridges is the esti- loads and changing traffic patterns on
prediction of newly constructed bridges mation of remaining fatigue life. A the remaining life of a typical steel
is obviously equally important.2,3 review of literature shows that there girder bridge. Furthermore, as a corol-
are several approaches available for lary to the objective, a procedure for
Advances in sensor technology cou- estimating the remaining fatigue life
pled with better understanding and health monitoring of a steel girder
of bridges. Garg et al.7 followed an bridge for increased axle loads/traf-
ability to model real structural behav- approach of finding out an effective
iour have also led to a greater inter- fic patterns is also outlined. To illus-
stress range which signifies the actual trate the overall procedure, a typical
est in this field.4,5 One of the best uses stress ranges as per the AREMA
of structural health monitoring is the steel through-girder bridge, erected in
guidelines.8 Significantly, Szeliski and 1966, with a 30,5 m simply-supported
continuous update of the performance Elkholy9 also correlated the damage
and state of a structure, using continu- span and a rocker-and-roller bearing
due to traffic, using historical traffic arrangement was considered.
ous experimental data in conjunction data to quantify past damage and used
with system identification techniques.6 a traffic model with a 5% growth to
signify future traffic trends. Description of Bridge
Peer-reviewed by international ex- Chaminda et al.10 mentions that “most The studied bridge is located on a
perts and accepted for publication
by SEI Editorial Board of the present day used fatigue assess- main traffic line in the eastern part of
ment approaches of railway bridges India between the cities Chennai and
Paper received: July 16, 2010 are generally based on combination Howrah at chain location 830/35-831/9
Paper accepted: October 29, 2010 of measured stress histories, Miner’s (Fig. 1). The super-structure comprises

210 Scientific Paper Structural Engineering International 2/2011

 Element Average Std. deviation Coefficient of variation
Carbon 0,199 0,008 0,042
Manganese 0,679 0,011 0,016
Silicon 0,069 0,002 0,03
Sulphur 0,032 0,003 0,109
Phosphorus 0,031 0,002 0,051
Chromium 0,025 0,001 0,049
Nickel Traces
Aluminium 0,011 0,001 0,053
Titanium 0,005 0,002 0,395
Fig. 1: Photograph of the bridge Copper Traces
Iron 98,861 0,033 0

five spans of overall length 32,51 m Table 1: Material properties of steel

with centre to centre bearing distance
of 31,92 m. A rocker bearing at the pier
end and a roller bearing at the abut- 1
Elements
added as point masses using the mass
SEP 23 2010
ment end are used. The first span of MAT NUM
U1' 02:22:25 element MASS21.
the bridge has been taken for instru- U2'
For an ideal model the roller is con-
mentation and study. Short welded rail Y
U3'
sidered as being perfectly smooth.
of 60 kg/m weight runs on the bridge X U4'
However, in reality, there is always
on steel channel sleepers. U5'
some degree of fixity. This is modelled
L0

A cross-bracing was taken from the L1 using translational spring elements.

bridge and tested for its metallurgi- L2 These too are modelled in ANSYS
cal composition and physical proper- L3 using COMBIN14 elements (Fig. 2).
ties. Ten element metallurgical testing L4
A design train was formed compris-
was carried out at 15 locations on the L5
ing two locomotives leading, 55 loaded
member. The statistical results of the L6
wagons trailing and two locomotives
material analysis of the samples taken Fig. 2: Finite element representation of trailing at the end. The axle loads of
from various parts of the member are bridge the wagons and the locomotives were
given in Table 1. The physical proper- measured at a weigh bridge. Thus all
ties measured were Young’s modulus the loads were known.
of 213 GPa, yield stress of 272 MPa, the truss bridge using pinned ends,
ultimate tensile stress of 457 MPa and joint fixities were introduced in the To acquire the data from the sen-
an elongation of 33%. The bridge has form of rotational springs. COMBIN14 sors placed on the super-structure,
undergone regular inspection by the elements were used to model the rota- an amplifier system was used (Fig. 3).
railway authorities over its entire life tional springs. The additional mass was This data-acquisition system had
and the condition has been reported
as good with no significant levels of
corrosion. Longitudinal strain gauges Piezoelectric accelerometers in
vertical and lateral directions
at midspan and quarter span
U1 U2 U3 U4 U5
Site-Validated Numerical
Model
Although detailed structural drawings
Displacement gauges
existed for this bridge, a comprehensive in longitudinal direction
site measurement was initiated. This L0 L1 L2 L3 L4 L5 L6 at roller
was done to record any deviations
from the drawings. The numerical Displacement gauges in vertical and lateral direction

model was built using the site mea- Upstream girder (with strain gauging on top and bottom of section)

sured cross sections. The additional Axial strain gauges Piezoelectric accelerometers in
plates at joints were also measured to vertical and lateral directions
at midspan and quarter span
arrive at the total dead weight of the U1' U2' U3' U4' U5'
bridge. The total mass of the bridge
was estimated to be 58 t.
The numerical model of the bridge was
created using ANSYS software version Displacement gauges
5.0.12 The individual members were in longitudinal direction
at roller
L0' L1' L2' L3' L4' L5' L6'
modelled as 3D beam elements with
geometric offsets as exactly measured Displacement gauges in vertical and lateral direction
from the site conditions. The element
Downstream girder (with strain gauging on neutral axis of section)
BEAM44 was used for modelling the
beam elements. In place of modelling Fig. 3: Instrumentation scheme showing critical sensor locations

Structural Engineering International 2/2011 Scientific Paper 211

several conditioning modules for the abutment bearing and the train The moving loads were then simulated
acquisition of data viz. 64 channels trailing off the span. in the numerical model using a full
of strain, eight channels of deflection, 3. A through run of the entire train transient analysis. The time step con-
eight channels of vibration and eight load at different speeds: sidered was 0,005 s, which was identical
channels of temperature. The system (a) Crawling speed (<5 km/h) to the site data sampling period. The
and the data logger were interfaced (b) Slow speed (20 km/h) comparisons of the responses are
to a laptop computer through an USB (c) Medium speed (40 km/h) shown in Fig. 4.
port. All the data recorded through (d) Track design speed (65 km/h) The dynamic parameters were esti-
the data acquisition system was 4. Various load cases involving start- mated from the site ambient vibration
monitored and stored using a dedi- ing the train from rest with the first data using an eigenvalue realization
cated software used for data acquisi- axle of the leading locomotive in line algorithm13 and gave a very good cor-
tion, visualisation and analysis of test with the support at the pier end and relation for the first six eigenvalues
data as well as documentation of test 4 axles of the first wagon also on the compared. The results are tabulated in
results. The measurement system was span. Table 4.
configured to trigger, control and dis- 5. Various load cases involving braking
play the data. when the train is on the span. As the numerical model gives a rea-
sonably good correlation with the site
The following load cases were In addition to these, a 48 h continuous data, it has been used for further simu-
recorded on site using the design train monitoring cycle was done. The events lation purposes. This method gives a
with known loads. recorded in this period were the nor- site-validated numerical model. The
1. Locomotive static case—with the mal prevailing traffic. The numerical boundary conditions were updated
first axle of the loco directly above model was then validated with the site- considering least-square error estima-
the abutment bearing and the train measured data for the design train. tion with the site data. The good cor-
trailing on the span. Some of the results and comparisons respondence of this model with the site
2. Wagon static case—with the first with site data for static cases are given data validates further simulation exer-
axle of the first wagon directly above in Tables 2 and 3. cises done with the numerical model.

Load case Locomotive static Wagon static Slow speed through run Design speed through run
Response quantity Numerical Site Numerical Site Numerical Site Numerical Site
(mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm)
Displacement at
−9,6 −9,7 −13 −12,4 −14 −13,6 −14,1 −13,9
centre of span
Table 2: Comparison of responses—numerical model versus site data—displacement at centre span

Load case Locomotive static Wagon static Slow speed through run Design speed through run
Response Numerical Site Numerical Site Numerical Site Numerical Site
quantity (microstrain) (microstrain) (microstrain) (microstrain) (microstrain) (microstrain) (microstrain) (microstrain)
Strain in
L2–L3
117 122 147 165 161 173 164 180
(bottom
chord)
Strain in
U1–U2 −162 −176 −223 −198 −232 −214 −233 −210
(top chord)
Strain in
L0–U1 −130 −131 −165 −154 −188 −169 −191 −176
(end-raker)
Strain in
U1–L2
221 216 264 234 281 271 284 286
(first
diagonal)
Strain in
bottom
flange 160 165 176 162 262 234 266 237
of cross
girder
Strain in
top flange
−143 −138 −12 −18 −173 −180 −179 −167
of rail
bearer
Table 3: Comparison of responses—numerical model versus site data—axial strains

212 Scientific Paper Structural Engineering International 2/2011

 Vertical displacement at centre of
span–slow speed run
Axial strain in diagonal for
slow speed run
Estimation of Loads
0
300 Numerical Site A sensor was placed on the rail, which
Numerical Site

Axial strain (microstrain)

−4 250 is extremely sensitive to individual
Displacement (mm)

axle movements. From the data from

200
−8 this sensor, the axles that had crossed
150 the bridges were counted for each
−12 100
train passage. These were segregated
into different types of trains crossing
50 the bridge. The composition of traffic
−16
0 in the 48 h period is given in Table 5.
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 Information on traffic details were also
(a) Time (s) (b) Time (s) obtained from the railway authorities
and analysed. The significant informa-
Axial strain in bottom chord – slow speed run
Axial strain in top tion is given in Table 6.
Numerical Site chord – slow speed run
200 0 Numerical Site
As can be seen from Table 5, the num-
ber of passenger trains per day (22,5)
Axial strain (microstrain)

Axial strain (microstrain) −50

150 and the other trains per day (27,5)
−100 during the 48 h monitoring cycle cor-
100
responds accurately with the annual
−150 average per day as given in Table 6.
−200
Thus the composition of the traffic in
50
the 48 h continuous monitoring cycle is
−250 an accurate representation of the aver-
0 age annual traffic. From the tonnage
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
hauled by the passenger trains, deduct-
(c) Time (s) (d) Time (s)
ing the loads of the locomotives, the
Longitudinal strain in cross girder bottom Longitudinal strain in rail bearer top average axle loads for the 1304 hauled
flange – slow speed run flange – slow speed run passenger coaches is computed as
300 100 13,44 t/axle. This is computed knowing
Longitudinal strain (microstrain)

Longitudinal strain (microstrain)

Numerical Site Numerical Site

250 50
the exact loads of locomotives (6 axles
of 20,5 t) and the number of axles in an
200 0 average day.
150 −50 The traffic load details of the mea-
100 −100 sured strain time histories in the 48 h
continuous monitoring period were
50 −150
unknown. The loads coming on the
0 −200 bridge can generally be segregated
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 into:
(e) Time (s) (f) Time (s)
1. Locomotive axles
Fig. 4: Correlations of numerical model with site response
2. Wagon axles corresponding to empty
freight wagons
Mode number Frequency from numerical Frequency from site 3. Wagon axles corresponding to empty
model (Hz) estimates (Hz) passenger wagons
4. Wagon axles corresponding to brake
1 3,43 3,34
vans
2 4,05 3,87 5. Wagon axles corresponding to loaded
3 5,92 6,18 freight wagons
4 7,52 7,28
Of all these, only the wagons cor-
5 8,78 9,34 responding to loaded freight are
6 9,92 9,95 unknown. The pattern of the response
data clearly indicates whether the
Table 4: Comparison of frequencies from numerical model with site estimates
train is a passenger or a freight train,
and also if it is loaded or empty. The
Train type Number of Number of Number of Number of load estimation procedure is thus lim-
trains locomotive axles trailing axles of trailing axles of ited to estimating the loaded freight
wagons brake van wagons.
Locomotives 5 48 — — As the numerical model was giv-
Passenger 45 288 2836 2 ing excellent correlation with the
Empty goods 8 54 1422 14 site data, it was used to estimate the
loads. This estimation was done by
Full goods 42 462 8646 78
first assuming the loads to be in a cer-
Table 5: Composition of traffic in 48 h continuous monitoring period tain range, and then changing them in

Structural Engineering International 2/2011 Scientific Paper 213

 Type of train Number of trains Axles of Tonnage Total traffic in gross Estimation of Total Fatigue Life
(average per day) coaches hauled per day million tonnes The estimated fatigue lives are those
Passengers 22,5 1304 18 478 44,25 corresponding to a 2,3% probability
Other trains 27,8 — >100 000 of failure for class D structures. The
stress cycles and the total fatigue life
Table 6: Traffic details as per railway records summary are given in Table 8.
The total fatigue life is taken as the min-
imum of all these values, and this works
small increments and ensuring that the 300
Load spectrum of loaded goods trains over 48 h period
out to be 86 years. The critical member is

Number of axles average wagon load (tons)

numerical model maps with the mea- Number of axles Average wagon load
the rail bearer, which although does not
sured strain in the same proportion 250
exhibit substantially high stress ranges,
as the known load trains. This process 200 has multiple cycles for lower ranges
was used to correlate the traffic in the
making it critical for fatigue failure.
measured cycle with the annual aver- 150
age in terms of composition and loads. An interesting feature here is that
100
In the knowledge of the authors, this the locations can be grouped into
work has not been attempted before. 50
two broad domains: locations show-
ing “global behaviour” and locations
The tonnage carried over the 48 h 0 showing “local behaviour”. Locations
period was estimated by categoris- 5 10 15 20 25 30 35 40
Train number showing global behaviour exhibit
ing the train traffic into the different
peak stress cycle ranges that are not
types of wagons hauled. As these have Fig. 5: Load spectrum of loaded goods
sensitive to the number of axle sets,
different axle spacing, the numeri- trains over recording cycle
but are sensitive to the overall load
cal model was used to simulate these
passing over the bridge. The bottom
runs. First a unit load was run on the
Estimation of Remaining Life chord and the first diagonal are loca-
numerical model at different veloci-
tions of this kind. The vertical posts
ties. The exact velocities of the actual
The remaining fatigue life estimation and rail bearers are locations showing
train were estimated using the sensors
requirement was addressed in the fol- local behaviour, where the peak stress
placed on the rail. These sensors gave
lowing manner: first, for each fatigue- cycle ranges are extremely sensitive
an influence line response for every
critical member of the open-web girder to the number of axle sets passing
passing axle. Knowing the spatial con-
span, the measured strain histories over the bridge. This is visualized in
figuration of the axles, the velocity of
were transformed into corresponding the extremely large number of peak
the train could be worked out. After
stress cycles as per the rainfall counting stress cycles seen at these locations.
this, the loads were considered to be a
algorithm presented by Downing and The cross girder is an intermediate
standard reference value and the unit
Socie;14 then, applying the procedure member where there are large num-
load influence line response was con-
of BS 5400: Part 10,15 the total fatigue ber of stress cycles seen at very low
volved with the loads using different
life for each member was estimated, stress range corresponding to its local
velocity patterns. This method gener-
considering the steel to be of the form behaviour, while the higher stress
ates the total response of the loads at
given by the metallurgical composi- cycles show a global effect.
different locations. The convergence
tion shown in Table 1, from the stress
was tested using a least square error
cycle patterns; finally, the remaining
norm. The train loads were then iter- Incorporation of Traffic Data
life of each member was arrived at by
atively changed to ensure that this in Remaining Life Estimation
subtracting the past fatigue life from
norm was a minimum. The loads cor-
the total fatigue life, where the past The remaining life estimation of a
responding to this value were then
fatigue life was estimated in an inno- bridge is complex, and it is based on
averaged with respect to the number
vative manner based on past historic several parameters. The change in
of axles and the results are given in
traffic tonnage data, and the remain- axle loads can have significant effects
Table 7. The good correspondence
ing life was corrected by incorporating in changing fatigue life, and although
of the numerical model with the site
the future traffic trends based on past a correlation is shown between traf-
data validates this analysis system. The
trend lines similar to the procedure fic density and fatigue life, if the traf-
average load for all the loaded goods
done by Szeliski and Elkholy.9 fic composition changes, this relation
trains were worked out and are shown
in Fig. 5.
From Fig. 5, the total haulage was Type of vehicle Number of axles Average load Total haulage (t)
estimated at 175,954 metric tonnes. per axle (t)
And the total load hauled over the Loaded goods wagons 8646 20,35 175 954
48 h period was thereby estimated as
Locomotives 852 20,5 17 466
240 371 t as shown in Table 7.
Passenger coaches 2836 13,44 38 116
The estimated tonnage hauled was
Empty goods wagons 1422 5,75 8177
43,87 million tonnes as against the
reported figure of 44,25 million tonnes. Brake vans 94 7 658
This corresponds to a difference of less Total tonnage in 48 h 240 371
than 1%. So the fatigue estimates from
Estimated tonnage annually 43 867 708
the 48 h period can be considered rep-
resentative of the entire year. Table 7: Estimated load hauled over the continuous monitoring period

214 Scientific Paper Structural Engineering International 2/2011

could change significantly. However, in 50 GMT trend from inception of bridge annual goods to passenger traffic has

Total annual load (million tons)

this study it is only the correlation of 45 been more or less unchanged over the
the remaining life with traffic haulage 40 past 40 odd years) and will remain
and density that is explored. 35 unchanged during its entire life.
30
The composition of a single train can 25
also affect the fatigue life. If empty 20
Conclusions
wagons are interspersed between 15
loaded wagons, for locations show- In this paper, a site-validated model
10
ing global behaviour, this can signifi- was obtained by running design trains
5
cantly increase the number of cycles 1960 1970 1980 1990 2000 2010 with known axle loads and by mea-
per event and would have significant Year suring various critical responses of
effects on fatigue life. However, for the bridge. This model was then used
Fig. 6: Traffic trend from inception of to solve the inverse problem of esti-
members showing local behaviour, this bridge
would not make significant difference. mating axle loads using these bridge
For this bridge, as the critical member responses during regular traffic run
is a rail bearer showing marked local over a 48 h period, and then subse-
tonnes over a period of 40 years. At quently estimating the load spectrum
behaviour, composition of this nature
the current level of traffic, this figure during that period. From the 48 h load
will not have any significant effect
would be achieved in about 18 years, as spectrum it was shown that this short-
on fatigue life. Therefore, where the
against the 42 years taken for reaching duration monitoring provides an excel-
fatigue-critical member shows local
this figure. lent estimate of the total annual traffic
behaviour, the fatigue life can be effec-
tively estimated by considering the The remaining fatigue traffic is thus haul that almost matches the actual
traffic haulage and density. 3029,6 million tonnes (3805,6 – 776). total annual traffic haul data provided
There has been a clear rise in the traf- by the Indian Railways. It has thus
The fatigue life of the bridge is given as been concluded that a site-validated
fic over the last 10 years or so. If the
86 years in Table 8. The bridge is cur- numerical model can be used to gener-
fatigue life is viewed in terms of traffic
rently 42 years old. A naïve estimate of ate the average daily load spectrum on
quantum, the remaining life at current
the remaining service life would there- any open-web girder bridge by utiliz-
traffic level is about 68 years (86 – 18).
fore be 44 remaining years for a 42 ing structural response measurements
The large difference in service life esti-
year old bridge. However, the fatigue over a representative short duration of
mation is due to the past traffic being
life of 86 years corresponds to an regular traffic flow.
significantly less than the present traffic.
annual traffic quantum of 44,25 million
tonnes. As can be seen from the graph At present day, considering a total It is also shown that, if the fatigue-crit-
in Fig. 6, although the current traffic is possible haulage of 3805,6 million ical member of the bridge is an “axle-
high, the traffic has not been consis- tonnes for service life, the accumu- sensitive” member, i.e. the peak stress
tently high through the past life of the lated damage as existing in the bridge cycle range is dependent on the num-
bridge. In fact, there has been a recent today can be considered as about ber of loaded axles over the bridge,
surge of traffic in the last few years. For 20,39%. If the trend of traffic is seen the remaining fatigue life of the bridge
a present annual traffic of 44,25 mil- from today backwards, the average can be estimated from an analysis of
lion tonnes, the estimated total fatigue rate of increase is about 6,38%. This the remaining traffic capacity of the
life is 86 years. Thus, considering a sim- corresponds reasonably well with the fatigue-critical member, provided the
ilar traffic mix for the entire life, the figure of 5% considered by Szeliski composition of traffic remains rela-
estimated total fatigue traffic is 3805,6 and Elkholy.9 tively unchanged over the life of the
million tonnes. The actual conservative structure. In this analysis, the increasing
Given the constant increase in traffic traffic and axle loads are factored in
remaining life depends upon the past
of about 6,38%, the remaining esti- the estimation of the remaining life.
traffic and possible future trends. The
mated service life is just about 25 years,
past traffic data from the inception of
accounting for the increasing traffic
the bridge is given in Fig. 6.
volume. The basic assumption here is Acknowledgements
The current traffic is clearly not a that the relative composition of traf- The authors gratefully acknowledge the per-
good representation of the past traffic fic, as given in Table 5, has been (this mission given by Indian Railways, and the
details. The total past traffic till date is corroborated by past data, which East Coast Zonal Railway in particular, to
on the bridge has been 776 million shows that the relative percentage of monitor the bridge for this study. The authors

Stress Bottom chord First diagonal Vertical post Rail bearer Cross girder
cycle range Number Total Number Total Number Total Number Total Number Total
(MPa) fatigue life fatigue life fatigue life fatigue life fatigue life
30 987 523 559 309 2154
40 58 391 487 2584 386
50 20 640 26 227 3 361 0 86 52 136
60 0 59 0 0 103
70 0 6 0 0 0
Table 8: Fatigue computations as per BS 5400

Structural Engineering International 2/2011 Scientific Paper 215

 also acknowledge the efforts of everyone at [5] Ko JM, Ni YQ. Development of vibration- [10] Chaminda SS, Ohga M, Dissanayake R,
InI Consulting, Mumbai, who assisted in mon- based damage detection methodology for Taniwaki K. Different approaches for remaining
itoring the bridge and collecting all the data civil engineering structures. Proceeding of fatigue life estimation of critical members in
used in this study. the 1st International Conference on Structural railway bridges. Int. J. Steel Struct. KSSC 2007;
Engineering, Kunming, China, 1999, 37–56. 7: 263–276.
[6] Goltermann P, Jensen F, Andersen M, Smart [11] Miner MA. Cumulative damage in fatigue.
References structures: possibilities, experiences and ben- J. Appl. Mech. ASME 1945; 12(3): A159–A164.
[1] Yanev B. The management of bridges in New efits
SEIfrom
Datapermanent
Block monitoring. First Inter-
[12] ANSYS: Users’ manual, Revision 5.0.
York city. Eng. Struct. 1998; 20(11): 1020–1026. national Conference on Bridge Maintenance,
Swanson Analysis Systems, Inc., 1992.
Safety and Management, IABMAS, Barcelona,
[2] Li ZX, Chan THT, Ko JM. Fatigue analysis 2002. [13] Banerji P, Chikermane S. Structural param-
and life prediction of bridges with structural eter estimation of two bridges from site data
health monitoring data—part I: methodology [7] Garg VK, Chu KH, Wiriyachai A. Fatigue using an eigen value realization algorithm.
and strategy. Int. J. Fatigue 2001; 23: 45–53. life of critical members in a railway truss bridge. 4th International Conference on Structural
Earthquake Eng. Struct. Dyn. 1982; 10(6): Health Monitoring on Intelligent Infrastructure
[3] Chan THT, Li ZX, Ko JM. Analysis and 779–795.
life prediction of bridges with structural health (SHMII-4), Zurich 2009.
monitoring data—part II: application. Int. J. [8] Shakoor Uppal A. Coping with the older rail- [14] Downing S, Socie DF. Simple rainflow
Fatigue 2001; 23: 55–64. road steel bridges. AREMA Annual Conference counting algorithms. Int. J. Fatigue 1982; 4(1):
Proceedings, 2005. 31–40.
[4] Mufti AA, Rivera E, Thomson Douglas J.
Civionics specifications for fibre optic sensors [9] Szeliski ZL, Elkholy IA. Fatigue investiga- [15] BS 5400. Steel, Concrete and Composite
for structural health monitoring. Arab. J. Sci. tion of a railway bridge truss. Canad. J. Civil Eng. Bridges: Part 10: Code of Practice for Fatigue.
Eng. 2006; 31(1C): 241–248. 1984; 11(3): 625–631. British Standards Institute: London, 1980.

216 Scientific Paper Structural Engineering International 2/2011

x084.indd 349 7/21/10 1:13:05 PM