Anna M.

RAKOCZY Duane OTTER

Principal Investigator Scientist
Transportation Technology Transportation Technology
Center, Inc. (TTCI) Center, Inc. (TTCI)
Pueblo, CO, USA Pueblo, CO, USA
anna_rakoczy@aar.com duane_otter@aar.com

Dr Rakoczy has more than 10 years Duane Otter, PhD, PE, has more
of research and academic than 35 years of structural and
experience. She is a member of the railroad research experience. He
TRB Standing Committee on Steel serves on several AREMA and
Bridges, a member of AREMA TRB bridge committees.
Committee 15 and ASCE.

Contact: anna_rakoczy@aar.com

1 Abstract
More than 50 percent of steel deck plate girder railway bridges in North America exceed 100 years in service.
This includes more than 14,000 spans with a total length of 145 miles that remain in service. The oldest bridges
are close to 150 years old. For these aging structures, there is a special need to develop reliable procedures to
evaluate their fitness for continued service. Simplified calculations and conservative assumptions often lead
to spurious outcomes that indicate older structures ceased to be functional decades ago. Even if a steel bridge
reaches its estimated fatigue life, the structure might be fit for future service and perhaps for a significant
period of time. Fitness for service assessments that utilize probabilistic methods, and that are informed by and
consistent with detailed physical inspections of the structures, provide a more accurate assessment of the
fitness and expected life of bridges. In this paper, a probabilistic method is demonstrated on three, riveted
deck plate girder spans that exceed 100 years of service. The spans are currently located at the Facility for
Accelerated Service Testing.
Keywords: riveted deck plate girders, fatigue life, probabilistic method, steel bridge, railway bridges,
remaining life.

The fatigue life of some steel bridges installed at

2 Introduction
Transportation Technology Center, Inc. (TTCI) is
testing five riveted steel girder railway bridge spans
for fatigue and safe service life performance. The
testing is being conducted at the Facility for
Accelerated Service Testing (FAST), located at the
Transportation Technology Center (TTC) in Pueblo,
CO, USA. These bridges carry approximately 136
MGTonnes per year of heavy axle load (HAL) traffic.
TTCI is using these bridges to investigate improved
safe service life estimates for common steel railway
bridge spans.
FAST already has been exceeded, according to the
American Railway Engineering and Maintenance of
Way Association (AREMA) Manual for Railway
Engineering, Chapter 15 calculations [1]. However,
the spans are in operation with no maintenance
required, no defects noted, and have accumulated
additional tonnage. These examples confirm that
the fatigue life evaluated using current methods is
conservative.
A description of one of the spans and a fatigue life
estimate using traditional deterministic methods
are documented in a previous publication [2].

1
1378
2019 IABSE Congress The Evolving Metropolis
September 4-6, 2019, New York City

the curve, from 69 MPa to 41 MPa, has a

Using a probabilistic method, the number of cycles
distribution with a slope of 5, similar to Eurocode
of accumulated traffic is estimated in terms of
[5]. The fatigue limit is taken to be 41 MPa. Details
probability of fatigue crack initiation. A reliability-
about the data and the process of developing the
based fitness-for-service analysis can be used for
statistical parameters is described in a previous
estimating years of service life of a bridge for
publication [4].
different levels of safety.

3 Uncertainty in Fatigue Resistance 4 Bridge Description

All three investigated spans are DPG spans with
3.1 Background open decks located at FAST on a 5-degree curve and
are superelevated by 100 millimeters [2]. The two
The current fatigue evaluation procedures in North
short steel DPG bridge spans on the West Steel
America do not account for uncertainties in load
Bridge (Figure 1) were installed in December 2014.
and resistance. NCHRP 721 [3] indicates that
A 10.1-meter span built in 1904 was shortened by
greater uncertainty is involved in fatigue
~180 millimeters at each end to fit the existing
evaluations as compared to bridge strength
opening; making it a 9.75-meter span.
evaluations or load ratings. Sources of uncertainty
in the fatigue evaluation process include the
scattered nature of the S-N curves, variable loads,
and approximations in structural analysis or load
effect estimation. Inherent uncertainties, however,
can be reduced using more refined analyses or field
measurements to better define the stress range at
the details in question. In addition, the use of
probabilistic methods is recommended to
determine the level of safety for various evaluation Figure 1. 7.3-meter and a 9.75-meter riveted DPG
cases. While a bridge might show an increase in spans in the West Steel Bridge at FAST
estimated fatigue life, the use of higher
A 9.1-meter riveted DPG span (Figure 2) was
probabilities of fatigue carries greater risk and
installed in December 2017, originally designed as
should be accompanied by more frequent
9.75-meters long in 1912. It was shortened 0.6
inspections.
meter on one end to fit the opening at FAST.
3.2 S-N curve and fatigue resistance
parameters
S-N curves present the number of cycles to failure
as a function of constant stress ranges for different
categories of details. The current fatigue design S-N
curves are based on a two standard deviation shift
from the mean values at 2 million cycles this
provides a probability of failure of 2.275 percent.
Therefore, an optimally designed detail (according
to S-N curves) that is exposed to the design stress
ranges has a 2.275 percent probability of cracking
during the specified lifetime.
Previous research on fatigue performance of full- Figure 22. A 9.1-meter riveted DPG spans in the
scale riveted bridge girders revealed that the data SOA Bridge at FAST
follows two trends [4]. The first part of the curve,
for stress ranges above 69 MPa, has a distribution
with a slope of 3, as is common; the second part of

2
1379
2019 IABSE Congress The Evolving Metropolis
September 4-6, 2019, New York City
4.1 Load on investigated bridges
Train operations at FAST include a train of primarily
143 tonnes gross rail load (GRL) cars
approximately 10 percent heavier than the
maximum interchange standard of 130 tonnes GRL.
Train length is up to 110 cars. Normal train
operations at FAST are at 64 km/h (40 mph). All
three spans are being overloaded by the HAL train
at FAST:
The 7.3-meter span is overload by 7 percent.
The 9.1-meter span is overloaded by 18 percent.
The 9.75-meter span is overloaded by 33
percent.
The overload includes the effects of unbalanced
superelevation.
4.2 Data collection at FAST
Strain gages are located on the bottom cover plates
at mid-span on the main girders of each span. For
fatigue evaluation, the following steps are taken:
1. The cycles are counted using a rain flow
cycle counting method [6].
2. Following AREMA guidelines, an equivalent
stress range is calculated using the root-
mean-cube method for slope 3 or modified
method for other slopes. The average
equivalent stress ranges for the south
girders under the high rail are:
a. 7.3-meter span 44.8 MPa
b. 9.1-meter span 48 MPa
c. 9.75-meter span 68 MPa.
3. Variations in the equivalent stresses for
these girders are from 4 percent to 7
percent. The bridges are tested in
controlled conditions; therefore, the
variations are low.
All three spans unload completely (live load stress
goes to zero) under the center of each car. This is
expected as the span lengths are less than the inside
axle spacing on the cars of ~10.5 meters. Since
fatigue is governed by stress range or magnitude of
the stress cycles, these short spans can be more
susceptible to fatigue as compared to longer spans
that do not experience full unloading as a train
traverses.
3
4.3 Summary of past load histories
The 9.1-meter span accumulated more than 736
MGTonnes of 119-tonne car traffic and 835
MGTonnes of 130-tonne car traffic in revenue
service prior to installation at FAST. The stress levels
in revenue service were 38 MPa and 42 MPa
respectively. The FAST train loading on the high rail
girder produces stresses of 48 MPa. The FAST
tonnage accumulated is about 180 MGTonnes
which corresponds to 1.26 million cycles.
The 10-meter span accumulated 816 MGTonnes of
119-tonne car traffic and 726 MGTonnes of 130-
tonne car traffic in revenue service. The stress levels
were 49 MPa and 53 MPa respectively. The FAST
train loading on the high rail girder produces
stresses of 68 MPa. The FAST tonnage accumulated
is about 536 MGTonnes which corresponds to 3.75
million cycles.
The 7.3-meter span accumulated about 883
MGTonnes during the revenue service operation,
but that corresponds to less than 1 million
accumulative cycles (888,437). At FAST, the total
tonnage accumulated is about 536 MGTonnes
which corresponds to 3.75 million cycles at a stress
of 44.9 MPa.
5 Fatigue Evaluation Using
Probabilistic Method
Advanced statistical analyses, such as the
probabilistic method, are the only methods for
evaluation of bridge fatigue life with consideration
of the uncertainties involved in the load and
resistance. Probability of fatigue crack initiation
depends on the selection of values of statistical
parameters representing a relevant fatigue
category. The limit state function for fatigue of steel
railway bridges can be expressed in terms of the
accumulated fatigue ratio [7]. The accumulated
fatigue ratio is a function of two parameters:
effective stress, S, and number of cycles, N.
Variability in stresses is related to axle load. Time
variation is related to N, number of cycles. This is
assuming that trains will continue to operate with
the same nominal axle load.
1380
2019 IABSE Congress The Evolving Metropolis
September 4-6, 2019, New York City

5.1 Probability method demonstrated on
the 9.1-meter span
The analysis for the 9.1-meter span, as presented in
Figure 3, shows that the span has a 17 percent
probability of fatigue crack initiation with the
current 1,734 MGTonnes when considered as
Category D, but only ~5 percent probability if
considered with slope 5 and new statistical
current 2,065 MGTonnes. When comparing the life
of the bridge to the slope 3 curve, the bridge also
has about a 65 percent probability of fatigue crack
initiation, but the probability of crack initiation is
rising faster. Further, when the bridge approaches
2,722 MGTonnes, the probability of crack initiation
exceeds the 95th percentile; while the probability
remains below the 90th percentile for a slope of 5.
-4

parameters [4]. Riveted Details, slope 5

Riveted Details, slope 3
0.9995 -3
-4
NCHRP Cat. D slope 3
Riveted Details, slope 3 0.995 NCHRP Cat. C slope 3
-2

0.9995 -3 Riveted Details, slope 5 0.95

0.9 -1
0.995 NCHRP Cat. C slope 3
0.75
-2
NCHRP Cat. D slope 3
0.95 0.5 0
0.9 -1
0.75 0.25
1
0.1
0.5 0 0.05
2
0.25
1
0.01
0.1 3
0.05 0.001
2
4
0.01
0.001
3
MGTonnes 0 500 1000 1500 2000 2500
million cycles 0 2 4 6 8 10 12 14 16 18
4

MGTonnes 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
million cycles 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36
Figure 3. Fatigue life estimates for FAST 9.1-meter
span using probabilistic method
Despite small differences, all four sets of statistical
parameters of fatigue give similar results of
comparable magnitude in a range of 1,388-2,359
MGTonnes for Minimum Life. However, the
difference increases when higher tonnage is
considered. The Average Fatigue Life with 50
percent probability of failure will be reached at
5436 MGTonnes when considered with new
parameters including a slope of 5 and only 2354
MGTonnes when considered with parameters of
Category D.
5.2 Probability method demonstrated on
the 9.75-meter span
The analysis for the 9.75-meter span as presented
in Figure 4, shows that the bridge span has greater
than a 99 percent probability of fatigue crack
initiation with the current 2,020 MGTonnes when
considered as Category D. In comparison, when
using a new set of resistance parameters based on
the bi-linear S-N curve, the results are notably
different. The bridge experiences maximum stress
ranges below 69 MPa; therefore, it can be
compared to a slope of 5 on the S-N curve. With this
assumption, the bridge span has about a 60 percent
probability of fatigue crack initiation with the
Figure 4. Fatigue life estimates for FAST 9.75-meter
span using probabilistic method
The Average Fatigue Life with 50 percent probability
of failure was reached at 1733 MGTonnes when
considered with parameters of slope 5 and only 646
MGTonnes when considered with parameters of
Category D, which is three times more than what
was estimated using Category D. Using Category D
is too conservative for estimating the fatigue life of
this span. Fatigue life estimation using new
statistics developed for a bi-linear S-N curve lead to
results that better match the service life of the span
to date.
5.3 Probability method demonstrated on
the 7.3-meter span
The analysis for the 7.3-meter span, shows that the
bridge span has less than 1 percent probability of
fatigue crack initiation with the current traffic of at
least 1,406 MGTonnes. The Minimum Fatigue Life
with 5 percent probability of failure will be reached
sometime between 1,814-2,722 MGTonnes. A large
increase in predicted fatigue life is noticed when the
span is considered using the proposed slope of 5.
This can be an indication that the span has close to
infinite fatigue life. The Average Fatigue Life with 50
percent probability of failure will be reached at
7866 MGTonnes when considered with parameters
of slope 5 and only 3018 MGTonnes when
considered with parameters of Category D.

4
1381
2019 IABSE Congress The Evolving Metropolis
September 4-6, 2019, New York City

6 Discussion, Conclusions and 721, Transportation Research Board,

Washington D.C., USA.
Acknowledgements [4] Rakoczy A.M., D. Otter, S. Dick. 2018.
Probability of failure can provide the most versatile
estimate of the remaining safe service life of a span. Using the Probabilistic Method and New
The service life estimate depends on the fatigue
category and applied load for various levels of risk TTCI/AAR Report R-1027.
(probabilities of fatigue crack initiation). During the [5] Comité Européen de Normalisation, EN
service life of a bridge, the accumulated fatigue is 1993-1-9, 2005. Design of Steel Structures,
increasing in time at different rates, depending on Part 1-9: Fatigue, European Committee for
tonnage per year and train type. All these factors Standardization.
must be specified in order to obtain accurate results [6] Fisher, J.W., Kulak, G.L., and Smith, I.F.C.
from the reliability analysis. 1998. A Fatigue Primer for Structural
Engineers, National Steel Bridge Alliance,
Using a probabilistic method, the number of cycles
USA.
of accumulated traffic is estimated in terms of
[7] Rakoczy, A.M., Nowak, A. S., and Dick, S.
probability of fatigue crack initiation (detection).
2016. Fatigue reliability model for steel
The reliability analysis can be used for estimating
railway bridges. Structure and
remaining service life of a bridge with different
Infrastructure Engineering (2016): 1-12.
levels of safety.
It is recommended that a probabilistic method be
used for riveted steel girder spans when a higher
probability of crack initiation is accepted,
understanding that inspection efforts must be
increased accordingly. In particular, it is advised
that the method be used for steel spans that have
exceeded their fatigue life according to
conventional Category D calculations but show no
signs of deterioration and fatigue cracks are not
present.

7 Acknowledgments
This research is sponsored by the Association of
American Railroads.

8 References
[1] American Railway Engineering and
Maintenance of Way Association, Manual
for Railway Engineering AREMA. Chapter 15
Steel Structures, Lanham, Maryland,
2018.
[2] Otter, D., Rakoczy, A.M., and Dick S. 2016.
Fatigue Life and Fitness-for-Service
Analysis: 32-foot Steel Deck Plate Girder
Bridge Span at FAST. Technology Digest TD-
16-025, AAR/TTCI, Pueblo CO, USA.
[3] Bowman, M.D., Fu, G., Zhou, E.Y., Connor,
R.J., and Godbole A.A. 2012. Fatigue
Evaluation of Steel Bridges. NCHRP Report

5
1382