OF RAILWAY BRIDGES
By Daniel H. Tobias, l Associate Member, ASCE, and Douglas A. Foutch,2 Member, ASCE
ABSTRACT: A newly developed method for the fatigue evaluation of riveted railway bridges is presented in
the paper. The data that were collected during a large-scale bridge instrumentation program along with the
fatigue resistance test database compiled at the University of Illinois at Urbana-Champaign, were used exten-
sively during the formulation of the method, or fatigue model. The fatigue model is based on reliability theory
due to the high degree of scatter in all fatigue strength tests and the uncertainty associated with estimating
fatigue loadings. Fatigue strengths and loadings are both described by probability distributions. The probability
Downloaded from ascelibrary.org by VISVESVARAYA NATL INST OF TECH on 03/30/17. Copyright ASCE. For personal use only; all rights reserved.
of failure is calculated using these two sets of distributions and a performance function that is a modified form
of Miner's damage law. Several parametric studies were conducted using the new fatigue evaluation method.
The studies give a broad view of the potential remaining lives of shorter-span railway bridges subjected to unit
train loadings.
IINTRODUCTION along with the response spectra for five typical riveted railway
bridges. A loading spectrum for a type of freight describes the
A large portion of the existing railway bridge infrastructure most probable range of load distribution for a rail car. A re-
im North America is of riveted steel construction. Many of sponse spectrum represents the most likely stress range distri-
tillese bridges are still in service and most were constructed bution for a bridge member or detail. The term spectrum (plu-
plrior to World War II. Some were built before 1900. Generally, ral spectra) is analogous to a probability distribution(s). This
plroblems with fatigue in riveted railway bridges has been in- information was used extensively in Phase II when this new
tlrequent (Fisher et al. 1987). However, concern in the industry fatigue evaluation methodology, or fatigue model, was for-
ii, growing because many riveted railway bridges are ap- mulated. The data from the bridge instrumentation program
p,roaching or exceeding their assumed design lives of about 80 were primarily used for development of a technique to deter-
years (Foutch 1987). Compounding this problem for older rail- mine fatigue loadings. The structural steel fatigue test database
way bridges is a current and probable future loading environ- compiled at UIUC was utilized during the development of a
ment that is quite different from that assumed when these method for quantifying the fatigwe strength of riveted bridge
hridges were designed. The historical Cooper E type loading members (Munse 1992). The UIUC database includes nearly
employed for design had comparatively heavy steam loco- all of the fatigue tests ever conducted on steel members. The
motives with a smaller uniform trailing weight (Clark 1984). instrumentation program focused mainly on riveted plate
Conversely, current trailing weights, for example, unit com- girder bridges and the fatigue stress cycles produced at center
modity consists, can regularly cause stress levels that are span. Consequently, the fatigue model focuses on evaluation
higher than those produced by today's locomotives (Tobias of these types of structures at this location. However, the de-
1994). Increases in allowable loads are also expected in the veloped model can be expanded to include all types of railway
near future. With these types of trailing weights, the number bridges and details.
of stress cycles that can potentially cause fatigue damage is
much greater than for historical loadings. Furthermore, the BACKGROUND AND PHILOSOPHY
current American Railway Engineering Association (AREA) Due to the high degree of uncertainty associated with eval-
code procedure for evaluating bridges in fatigue is difficult to
uating bridges in fatigue, the developed methodology is based
implement and realistic tests on the fatigue strength of riveted
on reliability theory. There has always been a great deal of
members have only begun to be conducted over the last 10-
scatter in the data characterizing the fatigue strengths of var-
15 years.
ious details. The future prediction and past estimation of the
A research program with a primary goal of developing a
number and magnitude of stress range cycles that any partic-
modern method for the fatigue evaluation of railway bridges
ular detail in a bridge will undergo is also very difficult. Bas-
was established in 1988. Participating researchers were from
ing a fatigue evaluation method on reliability theory allows
the University of Illinois at Urbana-Champaign (UIUC), the
for the inclusion of the uncertainties associated with strengths,
Association of American Railroads (AAR), and Clemson Uni-
bridge responses, and loadings. These inclusions allow fatigue
v,ersity (CU). The first step, or Phase I of the research effort,
endurance predictions to be more qualified and give a broader
was a large-scale bridge instrumentation program, which mea-
idea of the potential lives of bridges.
sured the current loading environment and its effects on typical
In this light, any method for predicting the useful life of
main and secondary structural members. The first phase lasted
railway bridges should be viewed in certain contexts. Exact
albout five and a half years. A database of the current loading
life predictions are not possible or practical at this juncture.
spectra from several significant types of freight was compiled,
This fact may not change in an amount of time where older
'Pres., Structuredyne Consultants, Inc., 511 North Elm St., St. Joseph, bridges will be affected. However, when a reasonable evalu-
IL 61873. ation method is applied to a large number of bridges, a ranking
2Prof., Civ. Engrg. Dept., Univ. of Illinois at Urbana-Champaign, of the structures most susceptible to fatigue damage at pre-
311298 Newmark Lab., 205 N. Mathews Ave., Urbana, IL 61801. scribed time intervals can be obtained. Rankings like this can
Note. Discussion open until October 1, 1997. To extend the closing be an important bridge management tool for planning future
date one month, a written request must be filed with the ASCE Manager
bridge maintenance and inspection intervals as well as replace-
of Journals. The manuscript for this paper was submitted for review and
possible publication on February 29, 1996. This paper is part of the Jour- ment schedules. When the fatigue evaluation method is reli-
nul of Bridge Engineering, Vol. 2, No.2, May, 1997. ASCE, ISSN ability-based, these rankings can take more useful forms. This
1084-0702197/0002-0053-0060/$4.00 + $.50 per page. Paper No. is primarily true because a bridge engineer can determine what
1~!755. level of probability of safety is acceptable for a given level of
JOURNAL OF BRIDGE ENGINEERING / MAY 1997/53
There are three primary tasks involved in the fatigue model: leo
(1) the calc~lation of fatigue loadings for each type of freight
to c~oss a bndge; (2) the determination of fatigue strengths for
a bndge; and (3) an evaluation of these two variables to es-
timate re?'laining life. Fatigue strengths and fatigue loadings J
are descnbed by probability distributions. The distributions are
FIG. 2. Generalized Flowchart of Fatigue Model for Evaluation
~valuated using a performance function. A performance func-
ofa Bridge
tIOn allows for the calculation of failure probabilities for en-
gineeri~g problems with multiple random variables (loading
cyc.les. The probability of failure is represented by the overlap
and resistance). The developed method for the determination
regIOn and IS calculated using the performance function. For
of fatigue.loadings is the most computationally complex aspect
of the fatigue model. The calculation of fatigue strength and multiple types of freight-loading a bridge, the solution process
the method used to evaluate the performance function are gen- is similar.
erally more straightforward. A generalized flowchart that presents a more detailed outline
of the three primary tasks in the fatigue model is shown in
. Fig. 1 illustrates how a bridge is evaluated in fatigue for a
Fig. 2. Each of the three main tasks are outlined in dashed
smgle type of freight loading. The solution is shown in a stan-
lin~s. As sh?wn in the figure, the subflowchart for calculating
dard SoN diag~am format. The fatigue loading distribution,
fatigue loadmgs for each type of freight to load a bridge is the
fNL(S,.)[n~(Sre)] IS shown to the left of the fatigue strength line.
most complicated. Fatigue loadings are difficult to calculate
The fatigue ~trength or resistance distribution, fNR(Sre)[nR(Sre)],
because they are functions of other probability distributions.
plots to the nght of the loading distribution. The terminologies
As shown in Box G, Monte Carlo train simulation techniques
used to express the fatigue resistance and strength distributions
are employed to calculate response spectra and fatigue loading
ar~ particul~ to reli~bility notation. Thefterms indicate prob-
distributions.
~blhty-denslty functions and the subscript terms designate fa-
tigue loading and fatigue resistance at equivalent constant am-
Bridge Evaluation and Performance Function
p~itu~e ~atig~e loading Sr~. The mean value of the strength
~lstnbutlOn he~ on the fatigue strength line. Fatigue strength The performance function used for fatigue evaluation is
lmes are prescnbed by the AREA code as categories A through given by a modified form of the Palmgren-Miner damage law
E' for particular details. The development of these categories (Miner 1945)
was based on research conducted on welded details. The as-
sumption that the mean strength for riveted details coincides ~ NL(Sre i )
x-LJ >0 (I)
with established fatigue strength lines is thought to be reason- i_I NR(Srei )
able. Also, the AREA code has recently introduced a new bi- where NL(Srej) and NR(Sre) = random variables described ear-
lin~ar fatigue strength line approach for some riveted bridges,
lier for fatigue strength and fatigue resistance. A bridge is eval-
which closely approximates mean strength levels. This ap- uated in. fatigue for freight types 1 through n. Freight types
proach will be discussed further in a later section. The level are conSidered separately for ease of computing fatigue load-
of each distribution along a fatigue strength line is governed ings. The x term is a number between 0 and 1. It is used to
by the equivalent constant amplitude stress range, Sre, calcu- specify an assumed amount of accumulated damage. The value
lated from the response spectrum that a type of freight pro- of x can also be calculated for a bridge using this fatigue
duces from many bridge crossings. The response spectrum is model by employing historical records of revenue traffic that
superimposed on the SoN diagram. The Y-axis represents stress have crossed the bridge. When the left-hand side of (1) be-
range and the X-axis represents the number of stress range comes less than or equal to 0, a bridge is regarded as failed
in fatigue. The equivalent constant amplitude stress range, Sre,
~or the response spectrum produced by a freight-type loading
IS well known and given by the following (Manual 1996):
Sre = [(-
1 )
I-J-N"
2:.-1 (N ,S7)
n
M
] 11m
(2)
x- -
/001
~~
N R, -
~4sd3 = 0
H,~,f3
(3)
bridges ranged from 9.1 m (30 ft) to 42.7 m (140 ft). These
test runs provided a large database of some of the most real-
istic impact measurements to date and were grouped into six ommended. The accuracy of a calculated spectrum is depen-
primary categories in this study: (1) 9.1-18.3 m span ballasted dent on the number of simulations and the final calculated
deck girders; (2) 9.1-18.3 m span open deck girders; (3) probability of failure according to one method (Shooman
18.3-27.4 m span ballasted deck girders; (4) 18.3-27.4 m 1968). The method is illustrated as follows:
span open deck girders; (5) 27.4-42.7 m span ballasted deck
girders; and (6) 27.4-42.7 m span open deck girders. The data
in each category were also further subgrouped by the speed
of test train in five equal increments of 32.3 km/h (20 mph).
%err= 200
R -pt
--
nPt
where %err = percent error in a fatigue failure probability
(9)
The overall best-fit probability distribution for the data was calculation. As the number of simulations (n) and the proba-
found to be lognormal. The fitted distributions for ballasted bility of failure (Pt) become greater, the percent error (%err)
deck spans ranging from 9.1 to 18.3 m are shown in Fig. 5. declines. The probability that the actual error is less than that
Use of these distributions to sample for impact for individual calculated by (9) is 95%. A visual inspection comparing the
car or locomotive impacts in simulated trains will give a more difference between a measured response spectrum and one cal-
accurate overall fatigue life evaluation. The statistics for the culated using the train simulation algorithm is also a good way
other five categories are presented elsewhere (Tobias 1994). to judge the accuracy of the method. Fig. 6 shows one of these
The final task in the simulation method is to run the train comparisons for one of the instrumented bridges from phase
over a bridge and generate a history of moment or stress versus 1. The measured spectrum is for 62 trains with l00-t coal hop-
distance traveled (Box 7, Fig. 4). The response spectrum for pers and the calculated spectrum is from 500 simulations. As
this train crossing is generated by subjecting the stress or mo- the figure illustrates, the two spectra are quite similar.
ment history to the rainflow algorithm (Downing and Socie Generally, if a response spectrum is calculated, the equiv-
1982). After numerous simulations, a normalized response alent constant amplitude stress range will be overestimated.
spectrum is obtained. The normalization to the number of cy- This is true because, typically, calculated time histories have
cles per MGMT or English million gross tons (MGT) is rec- greater stress ranges than measured ones even if the charac-
teristics of the trains to cross a bridge are well known. Mea-
0.20 9.1 to 18.3 m Spans sured responses are less than calculated ones because the track
Ballasted Decks structure probably acts somewhat compositely with a bridge
as trains cross over it. Standard structural methods do not ac-
~O.IS
count for this effect. The ratio of a measured Sre and calcu-
J 0.10
lated Sre is commonly referred to as IX. The adjustment of Sre
by a factor of ( l (Box K, Fig. 2) is the last step when deter-
mining the level of fatigue loading and resistance distributions
O.OS on an S-N diagram. Table 3 presents the average ( l per train
128.7-160.9 kmIh
crossing for the five plate girder spans that were instrumented
in Phase I. Generally, each individual bridge will have a
......__-.....-~
0.00 ~---+--=::::::;::::""';;~~
unique value of (l, so testing is necessary for an accurate de-
o 10 20 30 40 termination. In lieu of testing, values for ( l prescribed by the
Impact (%)
AREA code can be used. The values of ( l specified by the
FIG. 5. Lognormal Distributions Fitted to Impact Data for code are: (1) 0.70 for spans longer than 30.5 m (100 ft); (2)
Spans Ranging between 9.1 and 18.3 m 0.80 for spans between 22.9 (75 ft) and 30.5 m; and (3) 0.85
for spans less than 22.9 m.
r=-:--:-~--=-::----:---=------;.;;-::-r=====::::;-]
2000
Freight: 100 Metric Ton
Unit Coal PARAMETRIC STUDIES
~ 1500 Bridge: 12.2 m Span
Background
~
It 1000 Several of the most important parametric studies conducted
using the proposed fatigue model are presented in this section.
J! The studies were conducted to provide a broad idea of how
~
U 500 much longer riveted girder bridges may last, to study the ef-
fects of important variables on fatigue life, and to illustrate
o ~_F"P_ _~~"" the usefulness of a reliability-based method. Fatigue damage
637 8S4 1071 1288 accrues more rapidly in shorter bridges than longer bridges
746 963 1180 because they are loaded and unloaded by car trucks rather than
Moment Range (kN-m)
cars. For this reason a simply supported bridge with a span
FIG. 6. Comparison of Measured and Calculated Response length of 12.2 m (40 ft) was chosen as a benchmark bridge
Spectra for a Bridge in Vonore, Tenn. for these studies. Generally, if a longer span was chosen the
JOURNAL OF BRIDGE ENGINEERING / MAY 1997/57
15% greater than what would be required for a Cooper E80 0+-....,...::'-+---=-t!-.....--t------t--.....--4----!
loading with full impact. In all studies 45.4 MGMT of freight o 10 20 30 40 so 60
were considered to be crossing the bridge per year with a Years
standard deviation of 5 MGMT. Some railway bridges are FIG. 7. Probable Remaining Life for a Bridge SubjeCted to
loaded by a significantly larger amount of freight per year than Loadings from 91-t Unit Coal Trains
what was selected for these studies. Half of the bridge's useful
life was assumed to be previously used up, which meant that 100 ,--------::::::;;;;;;;.....- - - - - ,
the variable x in the basic performance function [(1)] had a Bridge: 12.2 m Span
value of 0.5. In most studies, a coal hopper car with center to
center (c-c) truck spacing of 11.43 m was used. The specified
l 80 ........ ~ _ ~~j}.43mex .
~ 20
open deck plate girder bridge to loadings from 91-t unit coal
trains at the rate of 45.4 MGMT per year. Half of the bridge's
life was assumed to be used up prior to the study. Subsequent
0 studies employed variations of some important variables to
Downloaded from ascelibrary.org by VISVESVARAYA NATL INST OF TECH on 03/30/17. Copyright ASCE. For personal use only; all rights reserved.
members." Proc., 12th Annu. Bridge Conf, Engineer's Soc. of Western and
Pennsylvania, Pittsburgh, Pa. <1>( ) = standard normal cumulative-density function.