Sei sulla pagina 1di 8

J Civil Struct Health Monit (2013) 3:317324

DOI 10.1007/s13349-013-0056-1

ORIGINAL PAPER

Probabilistic method for estimating remaining fatigue life in steel


bridges using measured strain data
Jeremiah Fasl Todd Helwig Sharon L. Wood

Received: 12 June 2013 / Revised: 12 August 2013 / Accepted: 16 August 2013 / Published online: 5 September 2013
Springer-Verlag Berlin Heidelberg 2013

Abstract Transportation officials face the difficult task of Abbreviations


maintaining the nations inventory of bridges under the A Fatigue constant for detail category, defined by
pressure of reduced budgets and an aging infrastructure. To [6]
prioritize the inspection, rehabilitation, or replacement COV Coefficient of variation for each fatigue category
schedules for each bridge, quantitative data are often nee- at 2 9 106 cycles
ded to distinguish between bridges in an inventory. Mea- dL t Induced fatigue damage at year t
sured strain data can be used to estimate the remaining dm k Induced fatigue damage determined from
fatigue life of fatigue-sensitive connections and provide measurements in year k
quantitative data to bridge owners. Deterministic approa- dmj Induced fatigue damage determined from
ches have previously been developed and presented in measurements in bin j
design codes. However, due to the inherent scatter of gN Uncertainty model for the limit state function of
fatigue data, a probabilistic approach may be more NC and NL
appropriate means of calculation. The probabilistic method nj Number of cycles in bin j
allows a bridge owner to weigh the risk of maintaining a NC Random variable describing the number of
bridge past its design service life. A probabilistic approach cycles corresponding to the capacity of the
to calculating the remaining fatigue life is discussed in this structural component
paper and applied to a fracture-critical bridge. Nf Number of cycles corresponding to failure at
stress range Sr
Keywords Probabilistic fatigue life  Fatigue  NL Random variable describing the number of
Fracture critical  Monitoring  Steel bridges  Strain loading cycles applied to the structural
response component
NLj Number of loading cycles imposed at a stress
range Srj within a stress spectrum
pf Probability of failure
r Annual rate of growth (constant)
J. Fasl (&) Sr Constant-amplitude stress range
Wiss, Janney, Elstner Associates, Inc., 9511N. Lake Creek
Parkway, Austin, TX 78717, USA Srj Average stress range of bin j
e-mail: jfasl@wje.com SrD Design stress range at 2 9 106 cycles for each
fatigue category
T. Helwig  S. L. Wood
Srl Mean stress range at 2 9 106 cycles for each
Department of Civil, Architectural and Environment
Engineering, University of Texas at Austin, 10100 Burnet Road, fatigue category
Building 177, Austin, TX 78758, USA xD Median value of D
e-mail: thelwig@mail.utexas.edu b Reliability index
S. L. Wood dD Coefficient of variation for D
e-mail: swood@mail.utexas.edu

123
318 J Civil Struct Health Monit (2013) 3:317324

D Palmgren-Miners critical damage accumulation Metal fatigue is not a simple process; rather it involves
index localized damage as the metal is subjected to cyclic load-
fA Parameter for the lognormal distribution for A ing, which causes variability in the fatigue resistance. In
fD Parameter for the lognormal distribution for D many cases, a negative fatigue life may be calculated
fdL t Parameter for the lognormal distribution for using a deterministic approach, which indicates that the
dL t bridge component has exceeded the design fatigue life,
kA Parameter for the lognormal distribution for A even though the component has not failed in fatigue. For
kD Parameter for the lognormal distribution for D instance, according to the AASHTO Manual for Bridge
kdL t Parameter for the lognormal distribution for Evaluation [1], the mean number of cycles to failure for a
dL t particular connection detail can be more than 2.5 times the
lA Mean of A design number of cycles to failure. Thus, there is consid-
ldL t Mean fatigue damage derived from year t erable scatter in the estimated fatigue life, and this inherent
ldL 1 Mean fatigue damage derived from year 1 variability must be considered when evaluating an existing
bridge. A probabilistic approach provides a rational method
llog10 A Mean value of log10 A
for including this uncertainty in the analysis [2]. In addi-
rA Normalized standard deviation of A
tion, with a probabilistic approach, much of the confusion
rlog10 A Standard deviation of log10 A
of calculating a negative remaining fatigue life (bridge age
U Standard normal cumulative distribution is older than calculated fatigue life), as might occur with a
function deterministic approach, is mitigated.
As part of the research outlined in this paper, measured
strain data from an in-service fracture-critical bridge were
used to estimate the remaining fatigue life based on the
response to traffic loads. Measuring the distribution of
1 Introduction stress ranges reduces the inherent uncertainty as compared
to calculating an effective stress range based on design
Highway bridges are critical elements of the transportation specifications. By minimizing the uncertainty, a more
network, providing the means to transport people and goods representative remaining fatigue life can be estimated. A
across the nation. As more bridges in the US reach or exceed probabilistic method for estimating the fatigue life of this
their intended design lives, transportation officials must bridge is discussed in this paper, and the results are com-
make difficult decisions regarding bridges in the inventory pared with a deterministic approach that was presented in
that can be safely kept in service versus those that need to be Fasl et al. [3].
replaced or retrofitted. In the current economic environment,
transportation officials do not have the resources to replace a
bridge simply because it has reached a certain age. Instead, 2 Background
transportation officials primarily rely on qualitative data
from routine visual inspections to assess the overall condi- If the strength and complete loading history were known
tion and maximize the service life of each bridge in their for each component in a bridge, establishing the remaining
inventory. However, qualitative data are generally not suf- fatigue life for critical members would be straightforward.
ficient to distinguish among bridges in an inventory, which is However, due to uncertainty in material strength, section
why transportation officials prefer quantitative data. dimensions, loading, and many other sources, estimates of
One of the critical deterioration mechanisms in steel the number of cycles to failure contain a great deal of
bridges is fatigue-induced crack growth that can potentially uncertainty. Structural reliability is a convenient method to
lead to brittle fracture. Evaluating the remaining fatigue characterize the probability of failure considering the
life in elements of the bridge is one means of providing uncertainty from various sources. Reliability can be defined
quantitative data to bridge owners to assist in making as the probability that failure will not occur. For specimens
decisions. The fatigue life of bridge components can be designed for flexure and shear, modern codes use structural
estimated using either deterministic or probabilistic reliability methods to determine the appropriate load and
approaches. Deterministic equations that are based on resistance factors to provide a consistent level of reliability.
design approaches are frequently used because the results Reliability concepts have also been used to determine
are relatively easy to calculate and understand. Despite the the optimal inspection schedule for critical connections
ease of use, deterministic approaches may not be the most subjected to fatigue loading [2, 4]. The process involves
appropriate method because uncertainty is not included in setting a target level of reliability (btarget) and optimizing
the analyses. the inspection interval in terms of the cost of the

123
J Civil Struct Health Monit (2013) 3:317324 319

inspections, repairs, and likelihood of failure. Reliability constant (A), and the demand on the member or connection
concepts have also been used to estimate the remaining is represented by the number of loading (NL) cycles times
fatigue life [4]. the cube of the constant-amplitude stress range (Sr).
To evaluate the fatigue life from a probabilistic view- Although the fatigue curves are based on specimens
point, a limit state function involving random variables subjected to constant-amplitude stress ranges, the stress
must be developed. In terms of number of cycles, the ranges induced by traffic on bridges will vary with the
fatigue limit state function can be represented by Eq. 1. weight and length of the vehicles crossing the bridge. To
Failure corresponds to rapid growth of a fatigue crack that relate the variable-amplitude stress ranges in the bridge to
leads to fracture, or when the number of loading cycles the constant-amplitude fatigue SN curves, a cumulative
exceeds the capacity of the structural component: damage theory is needed. The most popular theory is
gN NC  NL \0 1 Palmgren-Miners rule [7], which is based on a linear
damage hypothesis. Palmgren-Miners rule is simple to use
where g(N) = uncertainty model for the limit state func- and has been shown to agree well with test results [8].
tion of NC and NL; NC = random variable describing the However, there is also uncertainty in the capacity from
number of cycles corresponding to the capacity of the using Palmgren-Miners rule (D). By following Palmgren-
structural component; NL = random variable describing Miners rule, the fatigue limit state function can be
the number of loading cycles applied to the structural rewritten (Eq. 4)
component. X
The fatigue capacity is typically characterized by SN gN D  A  NLj  S3rj \0 4
curves (stress range versus number of cycles) that were where D = Palmgren-Miners critical damage accumula-
based on experimental tests of typical steel connections tion index; NLj = number of loading cycles imposed at a
subjected to a given stress range. The majority of the tests stress range; Srj = within a stress spectrum.
conducted in the US were performed in the 1970s and According to Palmgren-Miners rule, the loading cycles
1980s, and the results indicated that the primary variables and variable stress ranges can be represented by a single
affecting the fatigue life were the stress range and type of variable (dL t), which is called the induced fatigue dam-
connection detail [5]. The AASHTO Load and Resistance age. If dL t is substituted for the summation in Eq. 4, the
Factor Design (LRFD) Bridge Design Specifications [6] fatigue limit state function can be rewritten (Eq. 5)
use a set of eight fatigue curves, Categories AE0 , to
characterize the majority of bridge details. gN D  A  dL t\0 5
AASHTO provides a deterministic equation for calcu- where dL t = induced fatigue damage at year t.
lating the number of cycles to failure (Nf) for a given stress
range (Eq. 2):
A 3 Definition of probability models for random variables
Nf 2
S3r
To calculate the probability of failure for a specific fatigue
where Nf = number of cycles corresponding to failure at life, the random variables in Eq. 5 must be characterized
stress range Sr; Nf = number of cycles corresponding to based on the fatigue limit state function. Because the
failure at stress range Sr; A = fatigue constant for detail fatigue limit state function involves products, the random
category, defined by [6]; Sr = constant-amplitude stress variables can be modeled with lognormal distribution
range. functions. As such, the reliability factor (b) can be deter-
Alternatively, the value can be determined graphically mined using Eq. 6, and the probability of failure (pf ) can be
from the SN curves. However, because the curves are calculated using a standard normal cumulative distribution
used for design, the AASHTO SN curves represent the function (Eq. 7)
5th-percentile value (95 % confidence interval that 95 % of kD kA  kdL t
the details will have a fatigue life greater than the design b q 6
number of cycles, Nf). f2D f2A f2dL t
Rearranging terms and substituting the AASHTO defi-
pf Ub 7
nition of cycles corresponding to failure (Eq. 2), the fatigue
limit state function (Eq. 1) can be rewritten as Eq. 3. where b = reliability index; pf = probability of failure;
U = standard normal cumulative distribution function;
gN A  NL  S3r \0 3
kA = parameter for the lognormal distribution for A;
As such, the capacity of the structural component is kD = parameter for the lognormal distribution for D;
represented by a random variable based on the fatigue kdL t = parameter for the lognormal distribution for dL t;

123
320 J Civil Struct Health Monit (2013) 3:317324

fA = parameter for the lognormal distribution for A; kA ln10  llog10 A 12


fD = parameter for the lognormal distribution for D;
fdL t = parameter for the lognormal distribution for dL t. fA ln10  rlog10 A 13
The lognormal parameters, D and A, can be determined While random variables are used to describe D and A, the
from the results of laboratory fatigue tests, whereas dL t induced fatigue damage can be estimated from
can be estimated using measured data. measurements. Though the fatigue damage during a
The Palmgren-Miners critical damage accumulation particular period of time can be obtained directly from
index (D) accounts for the uncertainty associated with using measured strain data, it is the accumulated damage over the
Palmgren-Miners rule and has been characterized by the service life of the bridge that influences the remaining fatigue
offshore industry. Similarly to bridges, offshore structures life. In general, traffic volume is irregular and varies with
are subjected to variable-amplitude loads. Based on a survey time of day, variations in weather, accidents, and many other
of fatigue tests on structural sections, the offshore industry factors, which makes accurate modeling of actual traffic
uses a lognormal distribution with a median value (xD ) of 1.0 patterns extremely complicated. The goal then is to use a
and coefficient of variation (dD ) of 30 % [9]. For most tests, traffic model that on average closely represents the actual
the median is 1.0 and the coefficient of variation ranges traffic patterns. The most common approach is to assume the
between 30 and 60 % [10]. Following the protocol used in annual volume of traffic increases geometrically at a rate
offshore structures, values of xD = 1.0 and dD = 0.3 were between 2 and 6 %.
assumed in this paper. The lognormal parameters associated When bridges are monitored, most measurement periods
with D are calculated using Eqs. 8 and 9: typically range between 2 and 8 weeks. These data are
kD lnxD 0:0 8 representative of year k after the bridge was first placed in
q
  service. The evaluation of the cumulative fatigue damage
fD ln 1 d2D 0:163 9 requires an estimate of the damage induced before and after
the monitoring period. If traffic is assumed to increase
where xD = median value of D; dD = coefficient of vari- geometrically at a constant rate, historic traffic counts for
ation for D. the specific site can be used to estimate a rate of growth. In
The variation in the fatigue constant (A) can be deter- addition, measured data from multiple years can be used to
mined from test results reported in the literature. After estimate the rate of growth and minimize the uncertainty
studying the data from 374 tests, Fisher et al. [11] con- from an assumed growth rate.
cluded that log10 Nf can be assumed to follow a normal Assuming the average traffic volume grows geometri-
distribution. As such, log10 A can also be assumed to cally at a constant rate each year, the cumulative damage
follow a normal distribution, while A follows a lognormal can be estimated by Eq. 14
distribution.
The regression values for log10 A corresponding to the 1 rt  1
current AASHTO SN curves for design [6] were deter- ldL t ldL 1 14
r
mined by [12, 13], and were later presented in Moses et al.
where r = annual rate of growth (constant); ldL t = mean
[14]. The regression values from [14] are expressed in
fatigue damage derived from year t; ldL 1 = mean fatigue
terms of the mean (Srl ) and design (SrD ) stress ranges at
two million cycles for a slope of -3.0 (Table 1). The damage derived from year 1.
regression values for log10 A were determined for each As seen from Eq. 14, the cumulative damage in any year
fatigue category (Table 1) using Eqs. 10 and 11 (t) depends on the mean damage in the first year of service.
  The mean fatigue damage induced during the current year
llog10 A log10 2  106 3  log10 Srl 10 (k), which is extrapolated from the measured strain data,
q can be used to approximate the damage from the first year
rlog10 A ln1 COV2 11 using Eq. 15
where Srl = mean stress range at 2 9 106 cycles for each ldm k
fatigue category; COV = coefficient of variation for each ldL 1 : 15
1 rk1
fatigue category at 2 9 106 cycles; llog10 A = mean value
of log10 A; rlog10 A = standard deviation of log10 A. The lognormal parameters for dL t can be calculated
To model A using a lognormal distribution, the distri- using Eqs. 16 and 17
bution for log10 A was transformed using Eqs. 12 and 13
  f2
[2]. The derived parameters of the lognormal distribution d t
kdL t ln ldL t  L 16
for each fatigue category are summarized in Table 2. 2

123
J Civil Struct Health Monit (2013) 3:317324 321

Table 1 Variation in fatigue constant (A) from regression analysis non-redundant, and the failure of a flange from a longitudinal
[14] girder would be expected to cause a structural collapse. The
Category Fatigue Srl at 2 9 106 SrD at 2 9 106 COV (%) total length is nearly 82 m with 22.4-m end spans and a 38.1-
constant, cycles (MPa) cycles (MPa) m center span. The longitudinal girders in the end spans are
A (GPa3) continuous over the interior supports and extend 9.3 m into
A 8,200 228 160 21.7 the center span (Fig. 1). Because the center section is sus-
B 3,900 157 125 14.1 pended by hangers between the overhangs in the middle
B0 2,000 124 100 13.2 span, the bridge is statically determinate.
C 1,400 115 90 15.3 Strain gages were installed at multiple locations along
D 720 90 71 14.2
the length of the bridge on the top and bottom flanges.
E 360 66 56 9.7
However, the focus of this paper is on data gathered from
0 gages installed on the west side of the top flange of the east
E 130 50 40 13.2
and west longitudinal girders at Location 1 (Fig. 1). This
location corresponds to an AASHTO Category E detail. To
Table 2 Variation in fatigue constant (A) from and derived param- measure the nominal response of the girder (matches
eters for lognormal distribution assumption of AASHTO SN curve), the gages were
AASHTO Fatigue constant Standard Parameters for installed 0.6 m away from the stiffener angles.
fatigue (GPa3) deviation, lognormal To obtain an estimate of the induced fatigue damage
category rlog10 A distribution from measured strain data, the local maxima and minima in
lA llog10 A kA (GPa3) fA the dynamic strain history, and the corresponding cycle
amplitudes, must be determined. For the bridge described
A 24,000 4.37 0.21 10.07 0.49
in this paper, a simplified rainflow counting algorithm [15]
B 7,800 3.89 0.14 8.96 0.32
was used to calculate the amplitude of each cycle in a strain
B0 3,800 3.58 0.13 8.25 0.30
history, and the cycles were accumulated to form a histo-
C 3,000 3.48 0.15 8.02 0.35 gram. Hookes law and a modulus of elasticity of 200 GPa
D 1,400 3.16 0.14 7.27 0.33 were used to convert the measured strains to stresses.
E 560 2.75 0.10 6.33 0.22 An example histogram of number of cycles and stress
E0 250 2.39 0.13 5.50 0.30 ranges for the east girder is shown in Fig. 2a. This histo-
gram corresponds to service-level traffic crossing the
bridge during a single 24-hr period. There are a wide
v

u
u  2 ! variety of stress ranges captured throughout the day, with a
l
tln 1
d L t
fdL t : 17 maximum stress range of 115 MPa. The induced fatigue
rdL t damage during a measurement period in year k can be
calculated directly from the histogram by applying Eq. 18
By combining fatigue test results reported in the
to each bin of the histogram and summed (Eq. 19)
literature with field measurements, the probability of
failure for a particular connection can be calculated for 3
dmj nj  Srj 18
any given year and used by bridge owners to set priorities
among an inventory of bridges. The probability equations X
N

presented in this paper were applied to a fracture-critical dm k dmj 19


j1
bridge and are discussed in the next section.
where dmj = induced fatigue damage determined from
measurements in bin j; nj = number of cycles in bin j;
4 Application to a bridge Srj = average stress range of bin j; dm k = induced fati-
gue damage determined from measurements in year k;
A three-span, plate-girder bridge was instrumented with N = number of bins.
strain gages to evaluate its fatigue performance. The bridge is The contribution of each bin to the induced fatigue
part of a major transportation corridor in the US and expe- damage for a particular measurement period can be eval-
riences significant truck traffic. The welded connections that uated. As seen in Fig. 2b, though there were a few cycles
are most susceptible to fatigue damage were added to the with stress ranges above 100 MPa (very heavy trucks), the
bridge more than 35 years ago during the widening of a majority of damage is caused by vehicles between 35 and
bridge deck to add a shoulder. The bridge is considered to 55 MPa (typical truck traffic). In addition, the lower cycles
be fracture critical because the twin-girder system is (less than 10 MPa) did not cause a significant amount of

123
322 J Civil Struct Health Monit (2013) 3:317324

Fig. 1 Elevation of the bridge


and location of the strain gages.
Strain gages were installed at
Location 1 on the west side of
the top flanges of the
longitudinal girders

Fig. 2 a Histogram of stress (a) (b)


ranges captured by data
acquisition system during a
week day in the east girder.
b Transformed histogram into
fatigue damage using Eq. 18

damage [16]. By summing the damage from each bin, the Table 3 Summary of fatigue data obtained in year 37 from strain
total amount of damage for that particular day was found to gage measurements
be 0.197 GPa3. West girder East girder
A total of 71 days of data were captured at the site.
ld37 rd37 ld37 rd37
Because traffic patterns varied throughout the week [16], (GPa3) (GPa3)
(GPa3) (GPa3)
the data were grouped by the day of the week. The fatigue
damage was determined from the histogram data, and the Monday (9 days) 0.027 0.0056 0.18 0.029
means for each weekday were calculated (Table 3). The Tuesday (8 days) 0.032 0.0039 0.22 0.019
mean weekly fatigue damage was then evaluated by sum- Wednesday (12 days) 0.030 0.0031 0.22 0.014
ming the mean of each weekday. The weekly mean was Thursday (10 days) 0.029 0.0036 0.21 0.013
used to calculate the yearly mean for year 37 (ldm 37 ), Friday (10 days) 0.030 0.0023 0.20 0.015
which was used to estimate the induced fatigue damage Saturday (13 days) 0.021 0.0020 0.15 0.011
from the first year of service using Eq. 15. Sunday (9 days) 0.012 0.0014 0.098 0.008
Following the equations presented in the previous sec- Daily mean in year 37 0.026 0.0031 0.18 0.016
tion, the probability of failure can be plotted on an annual Weekly mean in year 37 0.18 0.022 1.27 0.11
basis, as shown in Fig. 3. Figure 3b plots the probability of Yearly mean in year 37 9.44 1.11 65.5 5.74
failure for the east girder, which corresponds to the right
lane for this bridge. Because more trucks cross the bridge
in the right lane as compared to the left lane, the east girder growth were correct, the probability of failure is 95 %.
is expected to have accumulated more damage than the Using a deterministic method to estimate the remaining
west girder. As expected with the higher traffic volume, a fatigue life, the design fatigue life (5th percentile) would
comparison of Fig. 3a, b shows that the east girder has a have been exceeded in year 10 for a 2 % annual growth and
higher probability of failure than the west girder for any in year 22 for a 6 % annual growth in the east girder [3].
given year. The deterministic values are similar to the probabilistic
Because the annual rate of traffic growth was not known model because a 5 % probability of failure would have
for the bridge, a range of growth rates (26 %) were con- been exceeded in year 9 for 2 % growth and year 20 for
sidered. If the 2 % model of annual growth were correct, 6 % growth. The difference in the two methods is whether
the probability of failure for the current age of the bridge is the probability of failure for a given year (probabilistic
nearly 100 % for Location 1 along the east girder. If 6 % approach) or the bridge age when the fatigue life is

123
J Civil Struct Health Monit (2013) 3:317324 323

Fig. 3 Probability of failure at (a) (b)


Location 1 for the a west girder
and b east girder

exceeded for a given probability of failure is calculated However, the probabilistic and deterministic approaches
(deterministic approach). can only be compared at a limited number of probabilities
Considering the response of the west girder, the esti- of failure because the fatigue constant (A) for the deter-
mates for the deterministic and probabilistic approaches are ministic approach has only been calculated for probabilities
similar. For the deterministic approach, the 5th percentile of failure of 5, 16, 33, and 50 % [16]. As such, the fatigue
would have been exceeded in year 48 for 2 % annual life at many other probabilities of failure can be calculated
growth and year 51 for 6 % annual growth [3]. In contrast, with the probabilistic approach, which provides the
the probability of failure approaches 5 % for the probabi- framework for calculating the risk of keeping a bridge in
listic approach between years 42 and 48. Thus, the prob- service. Risk involves making decisions with the possibil-
abilistic method indicates that the probability of failure ity of loss. With the probability of failure calculated for
reaches 5 % prior to the deterministic approach. Some each year, the owner can weigh the potential cost from
restraint should be exercised when estimating such a long major structural problems or failure with the cost from
length into the future by considering whether traffic growth replacing the bridge. The loss can take into account user
will continue to increase at a constant rate. For this costs from the bridge being closed, cost of bridge
example, the difference between 2 and 6 % growth rates is replacement, and cost of additional inspections. Knowing
minor for the west girder. Thus, the main benefit of these all of these costs, the owner can determine the optimal
graphs is that it shows that the east girder is much more point of replacing the bridge and/or increasing the rate of
sensitive to fatigue damage than the west girder. As such, inspections to minimize the loss to the public.
more attention during inspections could be focused on the
east girder. Acknowledgments This research was funded by the National
Institute of Standards and Technology (NIST) through the Technol-
ogy Innovation Program (TIP). The cognizant program officer was
Gerald Castellucci. The opinions expressed in this paper are those of
5 Conclusions the researchers and do not necessarily represent those of the sponsor.
The researchers would like to thank the bridge owner for providing
access to the bridge. In addition, graduate students Vasilis Samaras,
Measured strain data can be useful for estimating the Matt Reichenbach, and Ali Abu Yosef made significant contributions
remaining fatigue life of fracture-critical bridges. Dynamic to the research project.
strain data can be captured on a continuous or periodic
basis, analyzed by a rainflow counting method, and then
the probability of failure can be estimated using the prob-
References
abilistic model. As additional data are captured over the
life of the bridge, the confidence of the fatigue life estimate 1. AASHTO (2011) The manual for bridge evaluation, 2nd edn.
will be improved by determining the actual rate of traffic American Association of State Highway and Transportation
growth and the impact of increases in legal vehicle loads. Officials, Washington D.C, p 574
2. Chung H (2004) Fatigue reliability and optimal inspection strat-
A probabilistic approach to calculating the remaining
egies for steel bridges. Dissertation. The University of Texas at
fatigue life handles the uncertainty directly and provides Austin, Austin, p 212
advantages compared with the deterministic approach. For 3. Fasl J, Helwig T, Wood S, Frank K (2012) Using strain data to
the bridge considered in the example, the deterministic estimate the remaining fatigue life of a fracture-critical bridge.
Transp Res Rec 2313:6371
approach produced a similar fatigue life for the east and
4. Kwon K, Frangopol D (2010) Bridge fatigue reliability assess-
west girders. As such, the probabilistic approach was able ment using probability density functions of equivalent stress
to characterize the fatigue life in a quantitative manner. range based on field monitoring data. Int J Fatigue 32:12211232

123
324 J Civil Struct Health Monit (2013) 3:317324

5. Schilling C, Klippstein K, Barsom J, Blake G (1978) Fatigue of 11. Fisher J, Frank K, Hirt M, McNamee B (1970) Effect of weld-
welded steel bridge members under variable-amplitude loadings. ments on the fatigue strength of steel beams. Transportation
Transportation Research Board, NCHRP 188, Washington, D.C., Research Board, Washington, D.C.
p 113 12. Keating P, Fisher J (1986) Evaluation of fatigue tests and design
6. AASHTO (2010) LRFD bridge design specifications. American criteria on welded details. Transportation Research Board,
Association of State Highway and Transportation Officials, NCHRP, Washington, D.C.
Washington D.C., p 1734 13. Keating P, Halley S, Fisher J (1986) Fatigue test database for
7. Miner M (1945) Cumulative damage in fatigue. J Appl Mech welded steel bridge details. Lehigh University, Bethlehem.
12(3):159164 Report 488.2
8. ASCE Committee on Fatigue and Fracture Reliability of the 14. Moses F, Schilling C, Raju K (1987) Fatigue evaluation proce-
Committee on Structural Safety and Reliability of the Structural dures for steel bridges. Transportation Research Board, NCHRP
Division (1982) Fatigue reliability: variable amplitude loading. 299, Washington, D.C., p 94
J Struct Div 108(ST1):4769 15. Downing S, Socie D (1982) Simple rainflow counting algorithms.
9. Wirsching P (1984) Fatigue reliability for offshore structures. Int J Fatigue 4(1):3140
J Struct Eng 110(10):23402356 16. Fasl J (2013) Estimating the remaining fatigue life of steel
10. Wirsching P, Chen Y (1988) Consideration of probability-based bridges using field measurements. The University of Texas at
fatigue design for marine structures. Marine Struct (1):2345 Austin, Ph.D. Dissertation, p 338

123

Potrebbero piacerti anche